Optical communication using fibers is a major new technology which will profoundly impact telephone systems, computer interconnections and instrumentation. Optical communication is now at the threshold of major commercial exploitation with world-wide activity in engineering and field trials. For the second edition of the book, two chapters were added primarily devoted to longwavelength light sources and detectors.
Original Description:
Original Title
Semiconductor Devices for Optical Communication - Kressel
Optical communication using fibers is a major new technology which will profoundly impact telephone systems, computer interconnections and instrumentation. Optical communication is now at the threshold of major commercial exploitation with world-wide activity in engineering and field trials. For the second edition of the book, two chapters were added primarily devoted to longwavelength light sources and detectors.
Optical communication using fibers is a major new technology which will profoundly impact telephone systems, computer interconnections and instrumentation. Optical communication is now at the threshold of major commercial exploitation with world-wide activity in engineering and field trials. For the second edition of the book, two chapters were added primarily devoted to longwavelength light sources and detectors.
Optical communication using fibers is a major new technology which will
profoundly impact telephone systems, computer interconnections and in- strumentation. Fiber links provide several major advantages over conventional electronic communication systems. These include immunity to electromagnetic interference, thinner and lighter cables, lower transmission losses (especially for very high data rates) and potential kilometer-long link capabilities extending to the gigahertz region. The advances which have led to the maturity of optical communication using fibers result from major improvements made in past decades in semicon- ductor lasers and light-emitting diodes, detectors and fibers. Optical com- munication is now at the threshold of major commercial exploitation with world-wide activity in engineering and field trials. The economical ability to process and store information has dramatically increased because of the revolutionary improvements of computers and memories. It is indeed timely that a high-data-rate transmission technology has been developed which will greatly reduce the cost of transmitting data over long distances. For the second edition of the book, two chapters were added primarily devoted to long- wavelength light sources and detectors. Exceptionally rapid progress occurred in these devices since the completion of the firsl edition. This book brings together authoritative material dealing with the semicon- ductor components used in optical communication and the associated circuitry needed to build practical systems. We concentrate on the properties and modulation characteristics of laser diodes and light-emitting diodes specifically designed for optical communication, the design and construction of light transmitters, detectors, and receivers, and techniques for optical coupling of light into fibers. The material is presented at an engineering level which includes a tutorial introduction to the subject matter and sufficient depth for understanding the rapid changes which are still occurring in this field. The editor and authors are indebted to the respective organizations which have made it possible to publish this book. These include AEG Telefunken, Bell Telephone Laboratories, Hughes Research Laboratories, and RCA Laboratories. They are also grateful to Dr. H. Lotsch and Springer-Verlag for their support in completing this book. Princeton, New Jersey January, 1982 Henry Kressel Contents 1. Int roduct i on. By H. Kressel (Wi t h 6 Fi gures) . . . . . . . . . . . 1 1.1 Backgr ound . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Out l i ne . . . . . . . . . . . . . . . . . . . . . . . . . 5 References . . . . . . . . . . . . . . . . . . . . . . . . . 7 2. Laser Diodes and LEDs for Fiber Optical Communication By H. Kressel, M. Et t enber g, J. P. Wi t t ke, and I. Ladany (Wi t h 38 Fi gures) . . . . . . . . . . . . . . . . . . . . . . . 9 2.1 Mat eri al s for Het er oj unct i on St ruct ures . . . . . . . . . . . 11 2.2 The Doubl e- Het er oj unct i on Di ode . . . . . . . . . . . . . 14 2.2.1 Gener al Descr i pt i on . . . . . . . . . . . . . . . . 14 2.2.2 Cur r ent Vol t age Charact eri st i cs . . . . . . . . . . . . 16 2.2.3 Threshol d Cur r ent Densi t y and Efficiency . . . . . . . . 16 2.3 Device Fabr i cat i on . . . . . . . . . . . . . . . . . . . . 21 2.3.1 St r i pe- Cont act Di odes . . . . . . . . . . . . . . . . 23 2.3,2 Laser Cons t r uct i on and Packagi ng . . . . . . . . . . . 25 2.4 Laser Charact eri st i cs . . . . . . . . . . . . . . . . . . . 27 2.4.1 Power Emi ssi on vs Cur r ent of Pract i cal cw Lasers . . . . . 27 2.4.2 Laser Radi at i on Pat t er ns . . . . . . . . . . . . . . . 29 2,4.3 Spectral Emi ssi on . . . . . . . . . . . . . . . . . . 36 2.4.4 The Power - Cur r ent Cur ve Li neari t y . . . . . . . . . . 36 2.5 Li ght - Emi t t i ng Di odes . . . . . . . . . . . . . . . . . . 37 2.5.1 St ruct ures . . . . . . . . . . . . . . . . . . . . . 37 2.5.2 Opt i cal Power and Fi ber Coupl i ng . . . . . . . . . . . 40 2,5.3 LED- Fi ber Coupl i ng and Tr ans mi s s i on . . . . . . . . 40 2.5.4 Carri er Li fet i me and Di ode Bandwi dt h . . . . . . . . . 45 2.6 Di ode Rel i abi l i t y . . . . . . . . . . . . . . . . . . . . 52 2.6.1 Facet Damage . . . . . . . . . . . . . . . . . . . 52 2.6.2 Gr adual Degr adat i on . . . . . . . . . . . . . . . . 54 2.6.3 Accel erat ed Agi ng Test s . . . . . . . . . . . . . . . 55 2.6.4 Changes i n Var i ous Laser Par amet er s wi t h Agi ng . . . . . 58 List of Symbol s . . . . . . . . . . . . . . . . . . . . . . . 59 References . . . . . . . . . . . . . . . . . . . . . . . . . 60 V I I l ( ; o n l t ~ n / s . Photodetectors By D. P. Schi nke, R. G. Smi t h, and A. R. Har t man (Wi t h 12 Fi gures) . 63 3.1 Si mpl e Phot odi ode St r uct ur es . . . . . . . . . . . . . . . 63 3.1.1 Response . . . . . . . . . . . . . . . . . . . . . 63 3.1.2 Rever se Cur r ent s . . . . . . . . . . . . . . . . . . 65 3.1.3 Speed of Response . . . . . . . . . . . . . . . . . 65 3.1.4 Li near i t y . . . . . . . . . . . . . . . . . . . . . . 66 3.2 Aval anche Phot odi odes ( APD) . . . . . . . . . . . . . . . 66 3.2.1 Pri nci pl es o1" Oper at i on . . . . . . . . . . . . . . . 67 3.2.2 St ruct ures . . . . . . . . . . . . . . . . . . . . . 67 3.2.3 Response . . . . . . . . . . . . . . . . . . . . . 69 3.2.4 Speed of Response . . . . . . . . . . . . . . . . . 70 3.2.5 Noi se in Aval anche Phot odi odes . . . . . . . . . . . . 72 3.2.6 Li neari t y and Sat ur at i on . . . . . . . . . . . . . . . 77 3,3 Fabr i cat i on Technol ogy . . . . . . . . . . . . . . . . . . 78 3.4 Phot odi odes for Longer Wavel engt hs . . . . . . . . . . . . 80 3.4.1 Ge r ma ni um Phot odi odes . . . . . . . . . . . . . . . 81 3.4.2 l l I - V Compound Semi conduct or s . . . . . . . . . . . 82 List of Symbol s . . . . . . . . . . . . . . . . . . . . . . . 85 Ref er ences . . . . . . . . . . . . . . . . . . . . . . . . . 85 4. Receiver Design for Optical Fiber Communi cati on Systems By R. G. Smi t h and S. D. Personi ck (Wi t h 40 Fi gures) . . . . . . . 89 4.1 I nt r oduct or y Remar ks . . . . . . . . . . . . . . . . . . 89 4.1.1 Gener al Consi der at i ons . . . . . . . . . . . . . . . 89 4.1.2 Scope and Or gani zat i on . . . . . . . . . . . . . . . 90 4.2 Ti l e Basic Recei ver . . . . . . . . . . . . . . . . . . . . 90 4.2.1 Essent i al Compone nt s . . . . . . . . . . . . . . . . 90 Phot odet ect or . . . . . . . . . . . . . . . . . . . 91 Pr eampl i f i er and Ampl i f i er . . . . . . . . . . . . . . 92 Equal i zer . . . . . . . . . . . . . . . . . . . . . 93 Fi l t er . . . . . . . . . . . . . . . . . . . . . . . 93 4.2.2 Noi se Anal ysi s . . . . . . . . . . . . . . . . . . . 93 Phot odet ect or Noi se . . . . . . . . . . . . . . . . . 95 Ampl i f i er and Ci r cui t Noi se . . . . . . . . . . . . . . 97 Nor mal i zat i on . . . . . . . . . . . . . . . . . . . 98 V a l u e s o f l ~ , 1 2 , 1 3 , Z . . . . . . . . . . . . . . . . . 103 Bit Rat e Dependence of t he Noi se . . . . . . . . . . . 110 Noi se Cor ner Fr equency . . . . . . . . . . . . . . . 111 4.3 Noi se oFFET and Bi pol ar Ampl i f i er s . . . . . . . . . . . . 112 4,3.1 FET Fr ont End . . . . . . . . . . . . . . . . . . . 112 Mi ni mum VET Noi se . . . . . . . . . . . . . . . . 114 Exampl e . . . . . . . . . . . . . . . . . . . . . . 115 Effects of Ther mal and Shot Noi se . . . . . . . . . . . 115 Cont ent s 1X 4. 3. 2 B i p o l a r F r o n t E n d . . . . . . . . . . . . . . . . . 117 Op t i mu m C o l l e c t o r C u r r e n t . . . . . . . . . . . . . 119 Mi n i mu l n B i p o l a r No i s e . . . . . . . . . . . . . . . 120 Ba s e R e s i s t a n c e No i s e . . . . . . . . . . . . . . . . 121 E x a mp l e . . . . . . . . . . . . . . . . . . . . . . . . . 4. 3. 3 F E T v s B i p o l a r F r o n t E n d s . . . . . . . . . . . . . . 123 4. 4 F r o n t E n d De s i g n s . . . . . . . . . . . . . . . . . . . 124 4. 4. 1 S t r a i g h t f o r wa r d T e r mi n a t i o n . . . . . . . . . . . . . 124 4. 4. 2 Hi g h I mp e d a n c e o r I n t e g r a t i n g F r o n t E n d . . . . . . . . 125 4. 4. 3 T r a n s i mp c d a n c e Amp l i f i e r . . . . . . . . . . . . . . 126 4 . 4 . 4 E x p e r i me n t a l R e s u l t s . . . . . . . . . . . . . . . . 128 Op t i mi z e d T r a n s i mp e d a n c e Amp l i f i e r . . . . . . . . . 130 4. 5 S e n s i t i v i t y o f Di g i t a l Re c e i v e r s . . . . . . . . . . . . . . . 132 4. 5. 1 Ga u s s i a n Ap p r o x i ma t i o n . . . . . . . . . . . . . . 133 4. 5. 2 p - i - n De t e c t o r . . . . . . . . . . . . . . . . . . . 136 Bi t R a t e De p e n d e n c e . . . . . . . . . . . . . . . . 138 E x p e r i me n t a l Re s u l t s . . . . . . . . . . . . . . . . 139 4. 5. 3 Av a l a n c h e P h o t o d e t e c t o r . . . . . . . . . . . . . . 139 E x p e r i me n t a l Re s u l t s . . . . . . . . . . . . . . . . 143 Ef f e c t o f F i n i t e E x t i n c t i o n R a t i o . . . . . . . . . . . . 143 Mu l t i p l i e d Da r k C u r r e n t . . . . . . . . . . . . . . . 145 l n t e r s y mb o l I n t e r f e r e n c e . . . . . . . . . . . . . . . 146 4. 6 Ef f e c t s o f T r a n s mi t t e r a n d C h a n n e l P a r a me t e r s o n S y s t e m P e r f o r ma n c e . . . . . . . . . . . . . . . . . . . . . . . 146 4. 6. 1 Di s p e r s i o n a n d E q u a l i z a t i o n . . . . . . . . . . . . . 147 4. 6. 2 T r a n s mi t t e r P a r a me t e r s . . . . . . . . . . . . . . . 150 4. 7 An a l o g S y s t e ms . . . . . . . . . . . . . . . . . . . . . 151 4. 7. 1 p - i - n De t e c t o r . . . . . . . . . . . . . . . . . . . 152 4. 7. 2 Av a l a n c h e De t e c t o r . . . . . . . . . . . . . . . . . 153 4. 7. 3 E x a mp l e . . . . . . . . . . . . . . . . . . . . . . 153 4. 8 R e ma i n i n g Wo r k . . . . . . . . . . . . . . . . . . . . 155 Li s t o f S y mb o l s . . . . . . . . . . . . . . . . . . . . . . . 156 R e f e r e n c e s . . . . . . . . . . . . . . . . . . . . . . . . . 159 5. Lightwave Transmitters By P. W. S h u ma t e , Jr. a n d M. Di Do me n i c o , Jr. ( Wi t h 32 F i g u r e s ) . . . . 161 5. 1 C h a r a c t e r i s t i c s . . . . . . . . . . . . . . . . . . . . . . 161 5. 1. 1 P o we r . . . . . . . . . . . . . . . . . . . . . . . 162 5. 1. 2 S p e c t r a l Wi d t h . . . . . . . . . . . . . . . . . . . 163 5. 1. 3 S p e e d . . . . . . . . . . . . . . . . . . . . . . . 164 5. 1. 4 L i n c a r i t y . . . . . . . . . . . . . . . . . . . . . . 165 5. 1. 5 T h e r ma l B e h a v i o r . . . . . . . . . . . . . . . . . . 165 5. 1. 6 Re l i a b i l i t y . . . . . . . . . . . . . . . . . . . . . 166 5. 1. 7 Ot h e r F a c t o r s . . . . . . . . . . . . . . . . . . . 167 5. 2 Mo d u l a t i o n F o r ma t s . . . . . . . . . . . . . . . . . . . 167 X Contents 5.3 Source Compar i s ons . . . . . . . . . . . . . . . . . . . 169 5.4 Di gi t al Dr i ve Ci r cui t r y . . . . . . . . . . . . . . . . . . 170 5.5 LED Dr i ver s . . . . . . . . . . . . . . . . . . . . . . 171 5.6 Laser Dr i ver s . . . . . . . . . . . . . . . . . . . . . . 182 5.7 Anal og Dr i ve Ci r cui t r y . . . . . . . . . . . . . . . . . . 189 5.8 Subsyst em Consi der at i ons . . . . . . . . . . . . . . . . 194 List of Symbol s . . . . . . . . . . . . . . . . . . . . . . . 197 Ref er ence . . . . . . . . . . . . . . . . . . . . . . . . . . 198 6. Fi ber Coupl ers. By M. K. Bar nos ki ( Wi t h 11 Fi gures) . . . . . . . . 201 6.1 Sour ce- t o- Fi ber Coupl i ng . . . . . . . . . . . . . . . . . 201 6.2 Fi ber - t o- Fi ber Coupl i ng . . . . . . . . . . . . . . . . . . 207 List of Symbol s . . . . . . . . . . . . . . . . . . . . . . . 210 Ref er ences . . . . . . . . . . . . . . . . . . . . . . . . . 211 7. Mo dul a t i o n o f Laser Di ode s ByG. Ar nol d, P. Russer, and K. Pet er mann (Wi t h 25 Fi gures) . . . . 213 7.1 Backgr ound . . . . . . . . . . . . . . . . . . . . . . . 213 7.2 The Rat e Equat i ons . . . . . . . . . . . . . . . . . . . 214 7.3 Di rect Modul at i on of I nj ect i on Lasers . . . . . . . . . . . . 217 7.4 Modul at i on Behavi or of Speci fi c Laser St ruct ures . . . . . . . 224 7.5 Effect of Modul at i on on Spect r um and Near Fi el d . . . . . . . 226 7.6 Sel f-Pul si ng Phenomena . . . . . . . . . . . . . . . . . . 231 7.7 Coupl ed Laser St r uct ur es . . . . . . . . . . . . . . . . . 234 7.8 Out l ook . . . . . . . . . . . . . . . . . . . . . . . . . 236 List of Symbol s . . . . . . . . . . . . . . . . . . . . . . . 238 References . . . . . . . . . . . . . . . . . . . . . . . . . 240 8. The Ef f ect o f Junc t i on He at i ng on Laser Li neari ty and Harmoni c Di s t ort i on. By J. K. But l er (Wi t h 10 Fi gur es) . . . . . . . . . . . 243 8.1 St at i c Laser Char act er i st i cs . . . . . . . . . . . . . . . . 243 8.2 Tr ansi ent Anal ysi s . . . . . . . . . . . . . . . . . . . . 244 8.3 Har moni c Anal ysi s . . . . . . . . . . . . . . . . . . . . 249 8.4 Concl usi on . . . . . . . . . . . . . . . . . . . . . . . 256 List of Symbol s . . . . . . . . . . . . . . . . . . . . . . . 257 9. An By 9.1 9.2 Il l ustrati ve Opt i cal Communi c at i on Sys t e m J. H. Mul l i ns (Wi t h 12 Fi gures) . . . . . . . . . . . . . . . 259 Backgr ound and Pur pose . . . . . . . . . . . . . . . . . 259 9.1.1 Gener al Fi ber Opt i c Syst em and Range of Appl i cat i ons . 259 9.1.2 Economi c Consi der at i ons . . . . . . . . . . . . . . 260 9.1.3 Ot her Consi der at i ons . . . . . . . . . . . . . . . . 262 Di gi t al I nt er of f i ce Tr unki ng . . . . . . . . . . . . . . . . 263 9.2.1 The Tel ephone Tr unk Net wor k . . . . . . . . . . . 263 9.2.2 Convent i onal Di gi t al Met al l i c Faci l i t i es . . . . . . . . . 263 Contents Xl 9. 2. 3 T h e Ph y s i c a l E n v i r o n me n t . . . . . . . . . . . . . . 264 9. 2. 4 C o mp a r a t i v e E c o n o mi c s . . . . . . . . . . . . . . . 265 9. 3 T h e C h i c a g o L i g h t wa v c Pr o j e c t , S y s t e m C o mp o n e n t s . . . . . 266 9.3.1 Ba s i c S y s t e m P a r a me t e r s . . . . . . . . . . . . . . . 266 9. 3. 2 Ca b l e a n d S p l i c i n g . . . . . . . . . . . . . . . . . 268 9. 3. 3 Ca b l e T e r mi n a t i o n s a n d Si n g l e F i b e r C o n n e c t o r s . . . . 269 9. 3. 4 Op t i c a l C o mp o n e n t s - La s e r , L E D, AP D . . . . . . . 271 9. 3. 5 El e c t r o n i c C o mp o n e n t s R e g e n e r a t o r . . . . . . . . . 272 9. 3. 6 Te s t E q u i p me n t . . . . . . . . . . . . . . . . . . . 273 9. 4 C o n f i g u r a t i o n a n d I n s t a l l a t i o n o f t h e S y s t e m . . . . . . . . 276 9.4.1 Ro u t e De s c r i p t i o n . . . . . . . . . . . . . . . . . 276 9. 4. 2 S y s t e m C o n f i g u r a t i o n . . . . . . . . . . . . . . . . 277 9. 4. 3 I n s t a l l a t i o n o f t h e Ca b l e . . . . . . . . . . . . . . . 278 9. 4. 4 I n s t a l l a l i o n o f E q u i p me n t . . . . . . . . . . . . . . 279 9. 5 P e r f o r ma n c e a n d Re s u l t s . . . . . . . . . . . . . . . . . 279 9.5.1 Ca b l e Ch a r ~l c t e r i s t i c s . . . . . . . . . . . . . . . . 279 9. 5. 2 E l e c t r o - Op t i c a l P e r t b r ma n c e . . . . . . . . . . . . . 281 9. 5. 3 Cu r r e n t P e r l b r ma n c e a n d F u t u r e P r o s p e c t s . . . . . . . 281 9. 6 C o n c l u d i n g R e ma r k s . . . . . . . . . . . . . . . . . . . 282 Re f e r e n c e s . . . . . . . . . . . . . . . . . . . . . . . . . 283 10. Li ght Sour c e s - An Upda t e By H. Kr e s s e l a n d M. E t t e n b e r g ( Wi t h 3 F i g u r e s ) . . . . . . . . 285 10.1 I n Ga As P / h a P Dc v i c c s . . . . . . . . . . . . . . . . . . 285 10.2 AI Ga As De v i c e s . . . . . . . . . . . . . . . . . . . . 289 10.3 Ap p l i c z f l i o n s . . . . . . . . . . . . . . . . . . . . . . 291 Re f e r e n c e s . . . . . . . . . . . . . . . . . . . . . . . . . 291 11. Phot ode t e c t or s and Re c e i ve r s - An Updat e By R. G. S mi t h ( Wi t h 2 F i g u r e s ) . . . . . . . . . . . . . . . 293 11.1 Ma t e r i a l f or P h o t o d e t e c t o r s . . . . . . . . . . . . . . . 293 11.1.1 Ge r ma n i u m . . . . . . . . . . . . . . . . . . . 293 11.1.2 I n Ga As P . . . . . . . . . . . . . . . . . . . . 294 i 1.1.3 Ga A1 As S b . . . . . . . . . . . . . . . . . . . . 294 11.2 p i n P h o t o d i o d e s . . . . . . . . . . . . . . . . . . . 294 11.2.1 Da r k Cu r r e n t . . . . . . . . . . . . . . . . . . 295 11.2.2 Qu a n t u m Ef f i c i e n c y . . . . . . . . . . . . . . . . 296 11.2.3 S p e e d o f Re s p o n s e . . . . . . . . . . . . . . . . 297 11.3 Avalanche P h o l o d e t e c t o r s . . . . . . . . . . . . . . . . 297 11.4 Ot h e r De t e c t o r s . . . . . . . . . . . . . . . . . . . . 299 11.5 Re c e i v e r s . . . . . . . . . . . . . . . . . . . . . . . 299 11.6 He t e r o d y n i n g . . . . . . . . . . . . . . . . . . . . . 300 Rc f e r c n c c s . . . . . . . . . . . . . . . . . . . . . . . . . 301 Subj ect I ndex . . . . . . . . . . . . . . . . . . . . . . . . . 305 Contributors Arnold, Gfinther AEG Forschungsinstitut, Sedanstr. 10, D-7900 Ulm, Fed. Rep. of Germany Barnoski, Michael K. Plesscor Optronics Inc., 20200 Sunburst Street, Chatsworth, CA 91311, USA Butler, Jerome K. Southern Methodist University, Dallas TX 75275, USA DiDomenico, Moro, Jr. Bell Communications Research, Morris Research & Eng. Ctr., 453 South Street, Morristown, NJ 07960, USA Ettenberg, Michael RCA Laboratories, Princeton, NJ08540, USA Hartman, Adrian R. AT & T Technologies, 1 Oak Way, Berkeley Heights, NJ 07922, USA Kressel, Henry E.M. Warburg, Pincus & Co. Inc., 466 Lexington Avenue, NewYork, NY 10017, USA Ladany, Ivan RCA Laboratories, Princeton, NJ 08540, USA Mullins, Joe H. AT & T Labs., Crawfords Corner Rd., Holmdel NJ 07733, USA Personick, S. David Bell Communications Research, 331 Newman Springs Road, Box 7020, Red Bank, NJ 07701-7020, USA XIV Contributors Petermann, Klaus Technische Universitfit Berlin, Fachbereich 19, Institut ffir Hochfrequenz- technik, D-1000 Berlin 10, Fed. Rep. of Germany Russer, Peter Haferweg 19, D-8000 Mfinchen 81, Fed. Rep. of Germany Schinke, David Paul Bell Telephone Laboratories, Columbus, OH 43213, USA Shumate, Paul W., Jr. Bell Communications Research Inc., 435 South Street, Morristown, NJ 07960-1961, USA Smith, Richard G. AT & T Bell Laboratories, 555 Union Boulevard, Allentown, PA 18103, USA Wittke, James P. RCA Laboratories, Princeton, NJ 08540, USA 1. Introduction H. Kressel With 6 Figures 1.1 Background The emergence of opt i cal communi cat i on using fibers was made possible by t he parallel devel opment of low loss fibers, het er oj unct i on lasers and LEDs emi t t i ng in the regions of low fiber loss, and sensitive phot odet ect or s. Fi gure 1.1 shows a schemat i c of a poi nt - t o- poi nt f i ber - communi cat i on system with the light source (and its associated electronics), the fiber and the phot odi ode with the associated receiver circuitry. Thi s book is mai nl y concer ned with the semi conduct or component s, t hei r associ at ed circuitry, and optical coupl i ng to fibers. Fibers and t hei r associated t echnol ogy are beyond the scope of this book but a bri ef review is appr opr i at e [1.1]. The t hree basic types of fibers are shown in Fig. 1.2. These include step- index, graded-i ndex and single-mode fibers. The core of the step-index fiber has a const ant refractive index which is hi gher t han t hat of the cladding. The core di amet er of the widely used step-index fibers varies from 50 to 90 lam. Hundr eds of opt i cal modes may pr opagat e in st ep-i ndex fibers and pulse br oadeni ng due to differences in the pr opagat i on del ay t hr ough the fiber (modal dispersion) limits the bandwi dt h to a few tens of megaher t z in a kilometer. The modal dispersion is great l y reduced in graded-index fibers if the core has a parabol i c refractive index profile. The bandwi dt h of such fibers can reach about 1 GHz - km - a consi derabl e i mpr ovement over step-index fibers. As a result, graded-i ndex fibers are favored for hi gh-dat a-rat e, l ong-di st ance transmission. Single-mode fibers have a very small (a few mi cromet ers) core. Here, the modal dispersion is mi ni mi zed because onl y a single mode can propagat e. Such fibers can have bandwi dt hs of several GHz- km. Mat eri al dispersion, however, may limit the bandwi dt h of all fibers. Mat eri al dispersion produces pulse br oadeni ng because the pr opagat i on velocity is a function of the wavel engt h, as shown in Fig. 2.31 [1.2]. It is desirable, therefore, to minimize the spectral wi dt h of t he light source, i.e., a laser is preferable to an LED for mi ni mal pulse broadeni ng. It is also desirable to use light sources emi t t i ng in spectral regions of low dispersion. Fi gure 2.31 shows t hat the mat eri al di spersi on is mi ni mal near 1.2 to 1.3 lam maki ng light sources emi t t i ng in t hat region desirable for high bandwi dt h systems. Cont i nuous progress has been made in reduci ng the fiber opt i cal absor pt i on in the 0.8 to 1. 3gm spectral region [1.3], as shown in Fig. 1.3 [1.4]. The loss 2 I1. K ressel I N P U T S I G N A L ( E L E C T R I C A L ) S I G N A L L - J - l i , S H A P E R / E N C O D E R I I SOURCE}_~ OPTICAL DRIVER SOURCE O P T I C A L F I B E R SHAPER/DECODER Fig, I . I . S i mp l e p o i n l - l o - p o i n t optional I' iber c e mml t mi c z l t i on s y s t e m I ' ~ COUPLER OUTPUTSIGNAL (ELECTRICAL) I ..................... 1 ( o ) ~ (b) . . I ( c ) I "I > I N D E X Fi g. 1. 2a c. T h r e e b a s i c f i ber t y p e s : (a) s i ngl e mo d e ; ( b) s t e p i n d e x : a n d I t ) g r a d e d i nde x reduct i on has been achieved by purification, part i cul arl y the reduct i on of OH ions. Fi gure t.4 shows the loss spect rum of a st at e of the art step-index fiber consisting of a borosi l i cat e cladding and phosphosi l i cat e core [1.5]. The at t enuat i on at 0. 83~m is 1. 6dBkm i and 0 . 4 7 d b k m- ~ at 1.2Urn. These experi ment al values are quite close to the i nherent at t enuat i on limit due to Rayleigh scat t eri ng in the fiber. The historical reduct i on ill the fiber modal dispersion has resulted from accurat e index profiling of graded-i ndex fibers. Fi gure 1.5 shows the dispersion reduct i on duri ng the past decade as the near-ideal parabol i c index profile was achieved. Dispersion values ill graded-i ndex fiber cables under 1 ns km 1 combi ned with a loss under 5 dB ki n- t at 0.85 ~m can now be achieved, maki ng it possible to design hi gh-dat a-rat e optical communi cat i on systems with repeat cr spacings of several kilometers. The type and qual i t y of fibers needed depend on the syster~;s requi rement s. For shor t links (a few hundred meters), a loss of tens of decibels per ki l omet er o . _ 1 IO00 I 0 0 I 0 I 0.1 e I I I I 1 9 6 5 1 9 7 0 1 9 7 5 1 9 8 0 , o o ~ 5 o [ - 3 O 2 X IC \ ~ %%% o, 0. 5 o O : 1 - . , , I l I I I I I I " 1 " , , l I I 0 , 1 o . 4 " ' o . 6 " o . e i . o i , a 1 . 4 1 . 6 i . e WAVELENGTH p.Im F i g . 1 . 4 . [ nt r oduct i on 3 Fig. 1.3. Hi sl ori cal t r end in fiber l oss in t he spech' al regi on of pr esent i nt erest [1. 4] Fig. 1.4. At t enLl al i on as a f uncl i on of wavel engt h in a s t at e of t he ar t st ep i ndex fiber cons i s t i ng of a bor osi l i cat e cl addi ng a nd phos phos i l i cal c cor e 115] Fil~. 1.5. Hi st or i cal t r end in fiber moda l di sper - si on val ues (Se!/iJc refers to a gr aded- i ndex fiber t r a de ma r k) [ 1.4] E E g I 0 0 I 0 I M U L T I N O D E S T E P I N D E X F I B E R S \ _ SELFOC m 0.1 I 1970 H E A R - P A R A B O L I C o,~PROFILE I I 1975 1980 Fig. 11.5. and a pulse br oadeni ng of several nanoseconds per ki l omet er suffice. But when t ransmi ssi on links several ki l omet ers long are desired at dat a rates in excess of 10 Mbi t s - t t he fiber loss shoul d be under 10 dB kin- i and the pulse br oaden- ing shoul d be below a few nanoseconds per ki l omet er. The l at t er r equi r ement s are satisfied with graded-i ndex mul t i mode fibers or si ngl e-mode step index fibers. Owi ng to the fiber' s small core size, however, efficient and per manent coupl i ng of a di ode light source and a single-lnode fiber is relatively difficult. Therefore, such fibers will be used onl y when the highest achi evabl e dat a rate transnaission is required, leaving t he graded-i ndex fiber the most generally useful fiber for hi gh-dat a-rat e transmission. Semi conduct or laser progress has been dr amat i c since the het er oj unct i on laser was first realized with ( AI Ga) As/ GaAs st ruct ures [1.6 10]. Fi gure 1.6 sho,;vs the historical reduct i on in the t hreshol d cur r ent densi t y at r oom t emper at ur e of GaAs homoj unct i on lasers (pri or to 1967) and het eroj unct i on lasers since then. The si ngl e-het eroj unct i on "'close confi nement " laser made possible for the first time practical pul sed-current oper at i on at r oom t empera- 4 H. Kressel % u t-, F.- z g o 8 g I-- I 0 ~ 10 4 I 0 ~ i I i i H OMOJUNCTIONS SH - C O (ALGa)As HETEROJUNCTIONS DH I 0 2 I I I I I I 1 9 1 7 1 I I I I 1965 1967 1969 1973 197' YEAR Fi g, 1,6, 1 l i s t or i ct t l t r e nd in I hc G~l As t~lsel" d i o d e t h r e s h o l d c u r r e n t d e n s i t y a t r o o m t e mp e r a t u r e . V~.llucs u n d e r 4000 A/on12 ;.ire e s s e nt i a l for cw operation ture. The addi t i on of the second het eroj unct i on, to produce the double- het eroj unct i on laser (also called doubl e-het erost ruct ure laser), made it possible to reduce tile threshold current density from 8000 Ac m -2 to values under 4000 Ac m- 2 at room t emperat ure thus maki ng possible cont i nuous wave (cw) operation. The further lowering of the threshold current density of double- het eroj unct i on AI GaAs/ GaAs lasers since 1970 resulted from better materials and thinner active regions. Practical devices have threshold current densities in the 1000 Ac m -2 range, a value fully adequat e for reliable operat i on if ot her structural parameters are appropri at el y controlled. Lowering the threshold current density was onl y one of the essential ingredients in producing useful lasers. Poor reliability p]agued early semicon- duct or lasers and, prior to 1967, the phenomena were not underst ood. Systematic st udy identified the key elements which cont ri but ed to the short and generally erratic life of early lasers. These degradat i on phenomena include facet damage resulting from the excessive optical flux density in the active region ("catastrophic" degradat i on) and an internal process of lattice defect format i on (leading to "gradual " degradation). Fort unat el y, it was discovered t hat facet damage can be avoided by using dielectric facet coatings. Gr adual degradat i on, however, cannot be reduced by such simple methods. Extensive work revealed t hat the electron hole recom- bination process could increase tile density of lattice defects in the laser active Introduction 5 region, pr esumabl y by the di spl acement of at oms following nonr adi at i ve recombi nat i on. Fur t her mor e, it was found t hat devices which cont ai ned intially a high densi t y of lattice defects, such as dislocations, degraded much mor e rapi dl y t han mor e perfect devices [1.11]. Thus, efforts have been di rect ed at the const r uct i on of laser di odes as free from lattice defects as possible. This means cont r ol of the epitaxial mat eri al growt h and ext reme care in the fabri cat i on of the diodes in or der to prevent process-i nduced defects and built-in strains in the structure. These efforts have led to laser di odes of AIGaAs with mean time to failure values, at r oom t emper at ur e, est i mat ed to be in excess of 105 h. Indeed, laser di odes have now been in cont i nuous oper at i on for times in excess of 40,000 h. The use of laser di odes in opt i cal communi cat i on offers the br oadest oppor t uni t y for pr oduci ng hi gh- per f or mance systems, but specially designed het er oj unct i on light-emitting di odes (LEDs) are useful for systems with mor e modest requi rement s in terms of bandwi dt hs and fiber link distances. Bot h edge-emi t t i ng [1.12] and surface-emi t t i ng A1GaAs LEDs [1.13] are widely used. Lacki ng the t hreshol d cur r ent of the laser diode, the LED is less t emper at ur e sensitive and easier to use but, because of a br oader radi at i on pat t ern, the power l aunched i nt o fibers is l ower t han from lasers. Fur t her mor e, the LED modul at i on capabi l i t y is typically limited to values under 200 Mbi t s - 1. The het er oj unct i on laser and LED share a common t echnol ogy and the progress made in i mpr ovi ng the laser reliability has also favorabl y i mpact ed the life of LEDs. Semi conduct or phot odi ode det ect ors are widely used as fast, reliable and conveni ent det ect ors in the visible and near-infi' ared spectral regions. Silicon phot odi odes can be pr oduced with high quant um efficiencies and with very low dar k current s. These devices have been readily adopt ed for fiber con> muni cat i on and no specific i nnovat i ons were needed. Al t hough simple p-i-n st ruct ures are qui t e adequat e for less demandi ng systems, the use of aval anche phot odi odes with i nt ernal gains (which can reach values up to ~ 100) can significantly i mpr ove the per f or mance of communi cat i on systems [1.14, 15]. In summary, we see t hat by a most f or t unat e ci rcumst ance, the low at t ent uat i on region of fibers happens to mat ch the ~0. 82 ~tm emission from the most devel oped het er oj unct i on AIGaAs laser and LED as well as the Si phot odet ect or peak response region. With t he growi ng interest in the 1.2 lain spectral region, where the fiber at t ent uat i on and dispersion are l ower t han in the 0.8 ~tm spectral region, efforts are under way to devel op light sources and det ect ors made of mat eri al s whose bandgap energy is reduced to about 1 eV compar ed to ~ 1.5 eV in the AIGaAs het erost ruct ures. 1. 2 Out l i ne The first eight chapters in this book address specific areas relevant to the choice of components for a fiber-communication system. Chapter 2 reviews laser 6 I t . Kr e s s e l diodes and LEDs. Areas discussed include the t echnol ogy of het eroj unct i on structures emitting at ~0. 82 and ~l . 2t . t m, their power emission, efficiency, radiation patterns and reliability. The modul at i on of LEDs is discussed; laser diode modul at i on is discussed in Chap. 7. Phot odet ect ors are reviewed in Chap. 3. Silicon avalanche and p-i-n phot odi odes are the most developed and widely used devices for optical communi cat i on because their spectral response is very high at ~0. 82gm (the most common spectral emission value from AIGaAs devices). However, for systems operat i ng in the vicinity of 1.2 lain, the phot odet ect or must be made using semi conduct ors with lowcr bandgap energies t han t hat of silicon. Ger mani um devices are suitable but phot odi odes produced from direct band- gap materials such as l nGaAsP are potentially superior. The circuit aspects of using sources and phot odet ect ors are discussed in Chaps. 4, 5. Chapt er 4 discusses the design of the receiver which consists of the phot odi ode, amplifier and filters. The objective of receiver design is t o minimize the optical power needed at the phot odi ode to achieve a desired bit error rate in digital systems (or signal-to-noise ratio in anal og systems). In addi t i on, cost, manufact urabi l i t y and dynami c range enter in the choice of receiver. The design of a t ransmi t t er using LEDs and laser diodes is discussed in Chap. 5, including met hods of linearizing the out put of LEDs. It is evidently i mport ant to couple as much as possible of the diode' s emitted radi at i on into fibers. Chapt er 6 discusses coupling light sources into fibers by both direct means (i.e., butting) and by the use of i nt ermedi at e optics. In addi t i on, it is convenient to fabricate the light source with a short fiber length permanent l y at t ached to the diode package. This fiber is later at t ached to the long fiber cable. Therefore, the coupling losses t hat result from mechanical misalignment of two i nt erconnect ed fibers are also analyzed. A maj or advant age of the laser diode over other lasers is the ability to modul at e its out put at rates extending into the GHz range. However, owing to the complex dynami c behavior of laser diodes, high modul at i on rates are only achieved with the diode dc biased in the vicinity of the lasing threshold current. Furt hermore, resonance phenomena impact the laser modul at i on and various sources of noise exist. These and ot her effects related to laser modul at i on are discussed in Chap. 7. Optical communi cat i on systems using laser diodes (or LEDs) are most easily designed as digital rather t han anal og modul at i on systems. The reason is that the linearity of the light source is far less critical to the qual i t y of the digital system. Nevertheless, quality anal og modul at i on is possible but affected by the type of device used and its operat i ng characteristics, Chapt er 8 analyzes theoretically laser diode anal og modul at i on as limited by factors such as the laser's t hermal and electrical resistance. The above chapters are concerned with individual sections of a system. Chapt er 9 places everything in perspective by a general discussion of the economi c implications of optical communi cat i on for telephone applications. A practical system is described which was designed by the Bell Telephone Int roduct i on 7 Labor at or i es and tested in Chi cago under realistic condi t i ons. The success of this experi ment has led to an cxpanded use of opt i cal conmmni cat i on for t el ephone systems. Indeed, the opt i cal systelns promi se to rapi dl y move from the pr ot ot ype stage into pr oduct i on dur i ng the 1980s. The t ent h and eleventh chapt ers were added for the second edi t i on of the vol ume to review devel opment s in light sources and det ect ors since the compl et i on of the first edition. The great est progress has been made in the area of sources and det ect ors in the l to 1.6Hm spectral region using l nGaAs P alloys. The device st ruct ures are similar in most ways to t hose using A1GaAs bot h types of devices use het eroj unct i ons and progress has been rapid. These devel opment s are descri bed in Chap. 10 (light sources) and Chap. 11 (detectors). References 1.1 R. D. Mauer : Proc. [EEE 61, 452 (1973) 1.2 D. N. Payne, W. A. Gambl i ng: Electron. Lett. 11, 176 (1975) 1.3 D.B.Keck, R. D. Mauer, P. C. Schultz: App[. Phys. kett. 22, 307 (1973) 1.4 D. Lockie : Electro-Opt. Sys. Des. 8, 30 (1976) 1.5 M. Horiguchi: Electron. Len. 12, 310 (1976) 1.6 H.Kressel, H. Nel son: RCA Rev. 30, 106 (1969} 1.7 I.Hayashi, M.B.Panish, P.W. Foy: [EEE J. QE-5, 211 (1969) 1.8 Zh. | . Al ferov, V.M. Andreev~ E. L. Portnoi, M. K. Tr ukan: Soy. Phys. Semicon. 3, 1328 (1969) [English transl.: Soy. Phys. Semicond. 3, 1107 (1970)] 1.9 [.Hayashi, M.B.Panish, P.W. Foy, S.Sumski: Appl. Phys. Len. 17, 109 (1970) 1.10 H.Kressel, F. Z. Hawryl o: Appl. Phys. Lett. 17, 169 (1970) 1.11 H. Kressel, J. K. Butler: Semiconductor Lasers and lIeterojunction LEDs (Academic Press, New York 1977) 1.12 M. Ettenberg, H.Kressel, J. P. Wi t t ke: IEEE J. QE-12, 360 (1976) 1.13 C.A. Burrus, R. W. Dawson: Appl. Phys. Lett. 17, 17 (1970) 1.14 L.K. Anderson, M. DiDomenico, J r., M. B. Fisher : "Hi gh Speed Phol odet ect ors for Microwave Demodul at i on of Light", in Advances in Electronics, Vol. 5, ed. by L. Young (Academic Press, New York 1970) 1.15 P. P. Webb, R.J. Mclntyre, J. Conr ad: RCA Rev. 35, 234 (1974) 2. Laser Diodes and LEDs for Fiber Optical Communication H. Kressel, M. Ettenberg, J. P. Wittke, and I. Ladany Wi t h 38 Fi g u r e s The light sources discussed in this chapt er emit radiation in the spectral regions where the at t ent uat i on of optical fibers is relatively low 0.8 to 0.9 I, tm and 1.0 to 1.3 ~tm. Diodes t hat emit st i mul at ed radi at i on are called laser diodes, and those t hat emit incoherent radi at i on are denot ed light-emitting diodes (LED). These diodes are made of single-crystal materials, and for the most part consist of heterojuncti~m structures where p- and n-type layers of different energy bandgaps are combi ned to produce the desired properties [2.1]. A laser diode is characterized by a threshold current l(h for st i mul at ed emission (Fig. 2.1). Above l,h the carrier lifetime is very short (thus allowing high modul at i on rates), the spectral emission narrows to a few angst rom units, the beam becomes quite directional, and the externally measured quant um efficiency is high. On the ot her hand, the LED generally has a broad emission spect rum (several hundred angst rom units), a broad-angl e beam, and a modul at i on capability limited by the spont aneous carrier lifetime of about 1 ns under opt i mum conditions. Al t hough bot h cw laser diodes and LEDs can emit several milliwatts of total power, their applications differ. The LED is useful for applications requiring limited bandwi dt h capability and short -t o-moderat e transmission distances, whereas the cw laser diode is useful for more demand- ing hi gh-dat a-rat e and long-distance applications. Being a device with a t emperat ure-dependent threshold, the laser diode is somewhat more difficult to use t han the LED. The recombi nat i on (i.e., active) region of laser diodes used for optical communi cat i on is bounded by heterojunctions, which provide bot h carrier o ~ . ' ~ 16o' a b o ' ~6o' 4bo CURRENT (mA) Fi g. 2. 1. P o we r e mi n e d f i o m al l L E D a n d f i o m a l a s e r d i o d e as a f unc t i on of di r e c l c ur r e nt , Ti l e k n e e in t he l a s e r c u r v e is a l (l i e t h r e s h o l d c u r r e n t wh i c h d e p e n d s on t he de vi c e a n d i t s c o n s t r u c t i o n . Ti l e n o n l i n e a r i t y in l he c u r v e s at hi gh c u r r e n l s i s d u e 1o j t mc t i o n h e a l i n g 10 H. Kressel el al. confi nement and a dielectric cavity for waveguiding. It is the addi t i on of het eroj unct i ons to GaAs laser di odes t hat has led to the very large reduct i on in the t hreshol d cur r ent density at r oom t emper at ur e, maki ng the laser di ode practical. These i mprovement s reduced t he effective laser vol ume; thus the t hreshol d current densities were reduced to typical wflues as low as 1000Ac m -2, compar ed to typical homoj unct i on laser t hreshol d cur r ent densities of 50,000 Ac m -2 to 100,000 Ac m -2. The design of an LED for opt i cal communi cat i ons is consi derabl y si mpl er t han t hat of a laser di ode because of fewer design restrictions. Whereas t he i ncor por at i on of het eroj unct i ons is essential for useful laser oper at i on at r oom t emperat ure, het eroj unct i ons are not essential for LEDs, al t hough t hey great l y i mprove their efficiency and modul at i on bandwi dt h. Therefore, het er oj unct i on LEDs are generally used for optical communi cat i on. The fabri cat i on of these diodes involves epitaxial growt h of I I I - V com- pounds by vari ous met hods, including l i qui d-phase epi t axy [2.2]. vapor - phase epi t axy [2.3], and mol ecul ar - beam epi t axy [2.4]. Li qui d-phase epi t axy is the most widely used technique. The great strides made in the past decade in liquid- phase epi t axy allow the reproduci bl e gr owt h of thin layers (a few hundr ed angst r om units thick) of mul t i -el ement alloys. Laser di odes and LEDs designed for optical communi cat i on use GaAs and AIGaAs for the 0.8 to 0.9 lain spectral region, and commonl y, l nGaAs or I nGaAsP for the 1.0 to 1.3 gm region. In the forward-bi ased diode, mi nor i t y carri ers are injected into the active region, where they may recombi ne radiatively with the maj ori t y carri ers to pr oduce phot ons havi ng the appr oxi mat e energy of the bandgap. Alternatively, the r ecombi nat i on process may occur via states in the forbi dden gap or by ot her processes t hat are nonradiative, i.e., the energy is released in t he form of heat to the lattice. The average lifetime of the injected mi nor i t y carriers ("carri er lifetime") in the active region is i mpor t ant because it limits the ul t i mat e di ode modul at i on fi-equency. The carri er lifetime, as limited by the radiative r ecombi nat i on of an el ect ron in the conduct i on band with a hole in the valence band, depends on the band st ruct ure of the material. Band- t o- band radi at i ve r ecombi nat i on is favored in "di r ect - bandgap" semi conduct ors such as GaAs where the el ect ron hole r ecombi nat i on process occurs with moment um conservat i on. In such materials, the radi at i ve carri er lifetime is relatively short and the i nt ernal quant um efficiency can appr oach 100 %. Di r cct - bandgap materials are presently essential for laser diodes, thus restricting the choice of laser materials. On the ot her hand, silicon and ger mani um are "i ndi r eet - bandgap" semicon- duct or s which do not pr oduce light efficiently. Here, the radi at i ve recom- bi nat i on of an el ect ron in the conduct i on band with a hole in the val ence band is not a moment um- conser vi ng t ransi t i on. Therefore, moment um must be conserved via the emission or absor pt i on of phonons. As a result, the radi at i ve carri er lifetime for an equal popul at i on of carriers is much l onger t han in direct- bandgap materials - milliseconds instead of nanoseconds. Consequent l y, the internal quant um efficiency is low (a fcw percent at most at r oom t emperat ure), Laser Diodes and LEDs for Fiber Optical Communication 11 because before injected mi nor i t y carri ers r ecombi ne radi at i vel y, t hey are likely to di sappear via faster nonr adi at i ve r ecombi nat i on processes i nvol vi ng defect centers. We will not discuss furt her light sources using i ndi r ect - bandgap materials in the r ecombi nat i on region, al t hough such mat eri al s may be i ncor por at ed in passive regions of het er oj unct i on devices. 2 . 1 M a t e r i a l s f o r H e t e r o j u n c t i o n S t r u c t u r e s The bandgap energy of the r ecombi nat i on region cont rol s tile emission wavel engt h: ) . ~ 1.2/E~. (if Eg is in eV, 2 is in mi cromet ers. ) The key het ero- j unct i on di ode design pr obl em is to pr oduce devices having tile appr opr i at e combi nat i on of bandgap energies within and out si de the r ecombi nat i on region while mai nt ai ni ng a close lattice const ant mat ch at the interfaces. The devices of interest here have bandgap energies of 0.9 to 1.5eV in the r ecombi nat i on region. A het er oj unct i on is formed by j oi ni ng t wo layers of semi conduct ors with differing bandgap energies. When the layers have the same conduct i vi t y t ype an isotype het er oj unct i on is formed, whereas in an anisotype het er oj unct i on the l ayer conduct i vi t y t ype differs. The devices discussed here consist of com- bi nat i ons of het er oj unct i ons t hat serve t hree basic functions : l) anisotype (n-p) het er oj unct i ons increase the carri er injection efficiency from a hi gh- bandgap region i nt o a l ow- bandgap one ; II) isotype (n-n or p-p) het eroj unct i ons pr ovi de a pot ent i al bar r i er to confine mi nor i t y carriers within a l ow- bandgap regi on: III) bot h i sot ype and ani sot ype het er oj unct i ons pr oduce a difference in t he refractive index across the j unct i on. (A f our t h funct i on is commonl y provi ded by an i sot ype het er oj unct i on - the cont act resistance to a device is reduced by growi ng a surface l ayer havi ng a l ower bandgap energy t han in the underl yi ng material.) The confi nement of mi nor i t y carriers to the active region requires a het er oj unct i on barri er several times the el ect ron t hermal energy kT. To minimize the lasing t hreshol d cur r ent density, we restrict the r ecombi nat i on region thickness by spacing the het er oj unct i ons a distance d smaller t han the carri er diffusion length. High qual i t y material is needed in the r ecombi nat i on region to pr oduce efficient devices and to ensure t hei r reliable operat i on. It is essential t hat the het er oj unct i on interface be relatively defect-free to prevent excessive non- radi at i ve r ecombi nat i on of the injected carri ers at these interfaces. The nonr adi at i ve loss of carriers at an interface is charact eri zed by a r ecombi nat i on velocity S. In the typical doubl e- het er oj unct i on st ruct ure we can define an effective carri er lifetime z due to the presence of the t wo interfaces a di st ance d apar t [2.5, 6]. 1 1 2S 1 - -~ + + (2.1) - - Z " r ~ d - - - - ~ ~tlr 6 . B , i i , i i , i , i i i i i , i 6nSb AlSb 6 . 2 6.1 ~ " ~ ' ~ [ r l A s e - - ~ - - " ~ ' * ,... g . 0 z 5 , 8 5.5 6 o P 0 . 2 0 . 4 0 . 6 0 . 8 1 . 0 1 . 2 1 . 4 1 . 6 1 . 8 2 . 0 2 . 2 2 . 4 1 2 H. Kressel et al. E g - B A N D G A P E N E R G Y ( e V ) i I i n i i i = i i { 2 o ~ . o ' ' ~ 1 , 4 I .~ i.o 2.5 2 . 0 I.I 0 2 9 0 . 8 g J 5 0 . l (1~5 0 . 6 0 , 5 5 D I O D E I A V E L E N 6 T N ( p . m } Fig. 2.2. Lat t i ce c ons t a nt vs ba ndga p ener gy and di ode el ni s- si on wavel engt l a at r oom t em- per at ur e. The s haded regi on e nc ompa s s e s t he q u a t e r n a r y al- l oy I nGa As P. A l at t i ce- mat ched het er oj unct i on is pr oduc e d by c hoos i ng a l at t i ce c ons t a nt v a l l . l e and t wo mal er i al s wi t h di fferent b a n d g a p energi es. For exampl e, h l Ga As P is l at l i ce ma l c he d to l n P ( E. - - 1. 34cV) wi l h (R) /no.sGao.2Aso..~s Po.Ds ( E ~ l . l eV), t h u s yi el di ng a di - ode emi t t i ng at a b o u t 1.13/am Fi~. 2.3a, b. Misfit di sl ocat i on a r r a ys in a compos i t i oml l l y gr aded In.,Ga~_., P vapor - gr own epi t - axial l ayer on a Ga P s ubs t r at c, (a) is a t r a ns mi s s i on mi c r ogr a ph [2. 7], (b) s hows a c a t hode - l umi nes cence scan of a si mi l ar ar ea. in whi ch di s l ocat i ons near t he sur f ace a ppe a r as l l onr adi at i ve r egi ons [2. 20] where r r is the radi at i ve carri er lifetime in the absence of an interface, and r . , is the nonr adi at i ve lifetime limited by bul k defects. In mos t pract i cal devices, r . r >> G, and we can express the i nt ernal qua nt um efficiency for small d val ues as tTi -~ (1 + 2S vr / d) - ~ . ( 2 . 2 ) For exampl e, if d=0. 3l a m and z r ~ 10- s , an i nt ernal qua nt um efficiency of 50% requi res S < 2 x 104cms - l in the absence of ot her nonr adi at i ve centers. The single mos t i mpor t ant cont r i but i on to S is f r om nonr adi at i ve r ecom- bi nat i on centers i nt r oduced at the het er oj unct i on owi ng to the lattice const ant mi smat ch bet ween the t wo mat eri al s. Fi gure 2.2 shows the lattice const ant , Laser Diodes and LEDs for Fiber Optical Communication 13 bandgap energy and di ode emission wavel engt h for several of the i mpor t ant 1I[-V alloys. A viable het er oj unct i on di ode can be const ruct ed by combi ni ng materials with equal lattice const ant and dissimilar bandgaps. ~Fhe mat eri al in the r ecombi nat i on region shoul d have a di rect bandgap for efficient radi at i ve recombi nat i on. The i mpact of lattice misfit on the f or mat i on of nonradi at i ve centers may be under st ood fi'om the det ri ment al effect of "misfit di sl ocat i ons" formed to accomodat e a lattice misfit. For exampl e, for a 1% lattice mi smat ch at an interface, a di sl ocat i on will be generat ed appr oxi mat el y every 100 at om planes, or every 500A. Because of the nonr adi at i ve centers at the di sl ocat i on core, a high di sl ocat i on densi t y depresses t he device' s internal quant um efficiency. Fur t her mor e, misfit di sl ocat i ons are not always confined to the lattice- mi smat ched interface, but can pr opagat e t hr ough mul t i l ayer structures. The effect of di sl ocat i ons on radi at i ve efficiency is dramat i cal l y illustrated in Fig. 2.3, where we compar e a t ransmi ssi on el ect ron mi cr ogr aph of a di sl ocat i on ar r ay in a l at t i ce-l ni smat ched het er oj unct i on st ruct ure [2.7] with a cat hodol u- mi nescence mi crograph. The areas of low radi at i ve efficiency (dark lines) cor r espond to di sl ocat i ons t hat lie parallel and perpendi cul ar to the pl ane of t he surface viewed. Therefore, in designing het er oj unct i on st ruct ures for LEDs and lasers it is of ext reme i mpor t ance t o choose a closely l at t i ce-mat ched system. The appr oxi mat e predi ct ed rel at i onshi p between the misfit strain and the interfacial r ecombi nat i on velocity is [2.8] S-~(2 x lOV)(Aao/ao). (2.3) Experi ment al dat a for InxGa j . , P/ GaAs het eroj unct i ons [2.9], where x was varied to pr oduce vari at i ons in S due to lattice misfit, have indeed followed (2.3). Therefore, we may t ake the maxi mum misfit strain as a guideline for selecting materials suitable for het er oj unct i on st ruct ure fabrication. In general, Aao/a o must be < 10 - 3 for good initial performance, al t hough useful devices are somet i mes pr oduced with Aao/a o values as high as 5 x 10 -3. For high device reliability, values of Aao/a o even less t han 10-3 may be required. In the AlxGa l_~As system the lattice const ant varies very little at r oom t emper at ur e and is virtually const ant at the growt h t emper at ur e [2.10] as x goes from zero to unity. Het er oj unct i ons in this alloy system consequent l y have negligible st rai n-i nduced defects, and, as we shall discuss in Sect. 2.6.2, this is helpful in pr oduci ng Iong-terln reliability. Mor e typical of most t er nar y alloys, the lattice const ant varies by a few percent bet ween the bi nary alloys fi'om which they are derived. InxGa ~ ~As, InxGa ~ _xP and GaAs~P 1 - x are exampl es of such alloys with relatively large lattice const ant vari at i ons with composi t i on. As a result, het er oj unct i on st ruct ures in these mat eri al s typically cont ai n di sl ocat i ons in t he active region. In general, such lattice defects are consi dered harmful to reliability. 14 H. Kressel et al. I 1 ] ' A T T ' C I M A T C H IN G In x G O l_ x A S y P I - y ~ ' W H E N I y ~ z . ~ 6 ( , - , ) , ,,, . . . . . . . . i , : : 1 I lap I SUBSTRATE T Fig. 2.4. Lat t i ce- mat ched doubl e- hct er oj unct i on laser using I nGa As P act i ve layer and l nP confi ni ng layers [2.12] The simplest approach to obt ai ni ng het eroj unct i on devices emitting at 1.0 to 1.3gm consists of using hap and the quat er nar y alloy InxGa , _xAs , P, _ r [2.11], where the bandgap and lattice const ant can be independently adjusted over a useful range, as shown in Fig. 2.2. At some cost in simplicity, this extra degree of freedom permits the fabrication of strain-free het eroj unct i on devices. In~Ga~ _xAS~,P~ ~, can be lattice mat ched to lnP substrates when y~2. 16(1 - x ) [2.12] to produce doubl e-het eroj unct i on lasers and LEDs emitting in the one mi cromet er region. Figure 2.4 shows such a doubl e-het eroj unct i on structure where misfit strain can be completely avoi ded (except for mi nor strain resulting from differences between the t hermal expansion coefficients of expansion of InP and InGaAsP). 2. 2 The Do ubl e - He t e r o j unc t i o n Di o de 2.2.1 General Description Het eroj unct i on diodes have been produced using from one to four heterojunc- tions but onl y the doubl e-het eroj unct i on structures, which are most widely used for optical communi cat i on, will be discussed here. Single-heterojunction, as well as three- and four-het eroj unct i on lasers, are used for high-intensity, pulsed-power operat i on [2.1 ]. Figure 2.5 shows a schematic of the doubl e-het eroj unct i on (DH) laser diode consisting of either a p-n-n or p-p-n configuration. Also shown is a phot o- mi crograph of a cross section. The recombi nat i on region is bracketed by two layers having a higher bandgap energy. Typically, the refractive index at the lasing wavelength is about 5% lower in the out er layers t han in the central layer, so t hat the recombi nat i on region is also an optical waveguide. A fraction F of the optical power is located within the recombi nat i on region, with the rest spreading past the boundi ng het eroj unct i ons as schematically shown in Fig. 2.6. The optical cavity for the laser is formed by cleaving the crystal al ong two parallel crystalline planes; the change of refractive index at the semi conduct or-ai r boundari es provides the necessary mi rror reflectivity ( ~ 30 %). A typical cavity length is 300 pm, al t hough lengths varyi ng from 70 to 500 gm are sometimes used. Laser Di odes a nd LEDs for Fi ber Opt i cal Co mmu n i c a t i o n 15 ~ 1 ,U.m O.I -O,3/z m" ~tFm Nil, am u M E T ~L L I Z , d T I O N r l l l i i i i i i i i G o A s S U B S T R A T E I n ) ALo. 3 GOo. 7 A$ : Sn Or ~'e I " / / / / , I ( p) AL o . 3 GOo . 7 AS : G e l p l G a A I : G e 1 I 1 1 l , ~ O X I D E C u H E A T S I N K ~- S O L D E R I_ S T R I P E W I D T H ~ 13/.Lm E =. - E #, I . B e V - R E C O M I B I N A T I O N I R E G I O N ~ ' - g F - I . 5 5 e V ~' - - E g " 1 . 8 3 e V I 0 0 /.L m Fig. 2.5. The cr os s sect i on o[" a doubl e - he l e r oj unc t i on l aser of AI Ga As des i gned for cw oper at i on showl l in a s c he ma t i c i l l ust r at i on ( nol to scale) and a p h o t o mi c r o g r : @l of a s ampl e t hat has been pol i shed at a s hal l ow angl e to pr oduc e hi gh magni f i cat i on in t he t r ans ver s e di r ect i on. The "t er r aci ng" effect evi dent on t he l ower por t i on of t he mi c r ogr a ph is a gr owt h ilrlifacl t hai caus es SOllle i nt erface r oughne s s I I I N D E X , ~ n I C/O~T d LOSSES "~ [ aEND ~OUT Fig. 2.6. Schemal i c cr oss sect i on of a doubl e - he l e r oj unc t i on l aser s howi ng t he refract i ve i ndex di s t r i but i on per pendi cul ar to t he j unc t i on pl ane a nd i nt ensi t y di s t r i but i on of t he fundaxnei l t al t r ans ver s e mode will] a fract i on I " of t he r adi anl ener gy wi t hi n t he r e c ombi na l i on r egi on of wi dt h d. :% and %, . ar e t he a bs or pt i on coelficielltS wi t hi n a nd beyond t he r e c ombi na t i on regi on ~STRIPE WIDTH W i M E T A L / ~ SiOz B. RE GI ON STRIPE-CO TAOT B R O A D - A R E A Fig. 2.7. Schemal i c d i a g r a ms of a br oa d- a r e a and a s t r i pe- conl act di ode wi t h SiO2 i sol at i on 16 1I. Kr e s s e l et al. The active width of the di ode depends on the met hod of j unct i on area delineation. In the simplest st ruct ure (no l onger widely used), called a br oad- ar ea l as e r (Fig. 2.7), the sides of the laser are usually cut with a wire saw. However, practical devices use a stripe cont act [2.13] to nar r ow this width, as discussed furt her in Sect . 2.3.2. The st ri pe-cont act const r uct i on not onl y permi t s conveni ent cont r ol of the di ode width, but is also preferable for high reliability because defects cannot pr opagat e into the active region from the di ode' s sidewalls, as is the case with t he br oad- ar ea const ruct i on. 2. 2. 2 Current Vol tage Characteristics Under a forward bias V~,, holes and el ect rons are injected into the recom- bi nat i on region. The current up to t hreshol d is I ~ I o {expEaj (V, - I R~) ] - 1 } , ( 2 . 4 ) where I o is the sat urat i on current , a; is a di ode par amet er and R~ is the di ode' s series resistance. In typical AIGaAs di odes at r oom t emperat ure, Rs-~l to 10 ohms, and for a di ode area Ad, I o / A a _ ~ 2 x 10 - 11A c m- 2 ; a i ~ 3 0 V Equat i on (2.4) is onl y appr opr i at e in the oper at i ng region b e l o w lasing t hreshol d because abov e t hreshol d the j unct i on vol t age of the "i deal " laser remains fixed at its t hreshol d value. Al t hough practical devices do not always exhibit this effect, one may r easonabl y appr oxi mat e the cur r ent - vol t age rel at i onshi p above t hreshol d by t he expression, V,, ~- Eg/ e + I R~, (2.5) where I is the cur r ent and Eg is the bandgap energy. Figure 2.8 shows an illustrative curve of the power emission as a funct i on of cur r ent and the concur r ent curve of the appl i ed vol t age as a funct i on of current . (The series resistance of this di ode is 10 ohms.) 2. 2. 3 Threshold Current Density and Efficiency The lasing t hreshol d current density is reached when the gain coefficient, pr oduced by the injected carriers, equals the cavi t y losses. The dependence of the gain coefficient on the injected carri er densi t y varies with the material and the t emperat ure. Since the injected carri er-pai r density is related to the cur r ent density flowing in the device, a rel at i onshi p can be deri ved for the t hreshol d cur r ent density. Below t hreshol d the densi t y of el ect r on- hol e pairs in the r ecombi nat i on region N~, increases linearly with the current density J J T N ~ - - - (2.6) e - - ~ d ' La s e r Di o d e s a n d L E Ds f or F i b e r Op t i c a l C o mmu n i c a t i o n 17 12 E - I0 " 8 :7 o _o 6 I. u 4 bJ 0 Q. 0 0 / AIGoAs OH LASER [ k = 8270 A [ W= 12~m [ 20 40 60 81 CURRENT (mA) laJ ~9 O > r-~ tM m Fi g. 2. 8. P o wc r e mi s s i o n f i ' om o n e s i de o f a s t r i p e c o n t a c t , c w AI Ga As l a s e r d i o d e a n d t h e d i o d e v o l i a g c a s a f l unct i on o f t he d i o d e c u r r e n t . T h e s t r i p e wi d t h is 12 tam where e is the el ect ron charge, d is the r ecombi nat i on region width and r is the carri er lifetime for spont aneous r ecombi nat i on. Referring to the schemat i c di agr am of Fig. 2.6, we can estimate the opt i cal losses in the device t hat must be over come for t hreshol d to be reached. We define ,%, as t he gain coefficient at t hreshol d, F g t l , = a,,ut(1 - F) + 2 ~ ln(1/R 1 R 2 ) "q- F ~ f c , (2.7) where ~,,,,t is the absor pt i on coefficient of the material out si de the recom- bi nat i on region at the lasing phot on energy, arc is the absor pt i on within t he r ecombi nat i on region (due to free carriers), L is the mi r r or ( Fabr y- Per ot ) spacing, R 1, R 2 are t he respective reflectivity values of the t wo mi r r or facets, and F is the fraction of the wave energy within the boundar i es of t he r ecombi nat i on region. F can be comput ed from a sol ut i on to Maxwell' s equat i ons knowi ng the dielectric steps at the het er oj unct i on boundar i es and the het er oj unct i on spaci ng d. For AI,.Ga~ xAs/ GaAs het eroj unct i ons, the index step An at 2=0. 9 pm is rel at ed t o t he composi t i onal change by An~0. 62x. Fi gure 2.9 shows F as a funct i on of d/2 (where 2 is the lasing wavelength) for various An values I-2.1 ]. As d / ) ~ is reduced, increasing values of An are needed to effectively confi ne the wa v e . Theoret i cal l y I-2.14], the gain dependence on cur r ent density in GaAs is of the general form g = f l , ( J - j ~ ) b where J~, fl~ and b are const ant s t hat vary with dopi ng in the mat eri al and t emperat ure. Typi cal l y, DH lasers designed for the 18 H. Kre.s.wl et al . 1. 0 0 . 9 0 . 8 [.-, ~ 0.6 J - O . 5 z ELI ILl Z 0 . 3 z 0 o 0 2 --_ _ 0 --_- -_ o , o , - - 0. [ O. 2 0 . 3 0 . 4 0. 5 0 . 6 0 . 7 0 8 0 . 9 1.0 0..._~9 d ( ~m) X Fig. 2.9. Conf i ne me nt f act or 1' ffract i on of t he r adi at i on wi t hi n t he r e c ombi na t i on region), as a f unct i on of d~2 I~r wl r i ous val ues of t he step/1H at t hc t wo het cr oj t u) ct i ons o f a doubl e- het er oj t mc- t i o n l a s e r [2. I ] p r e s e n t a p p l i c a t i o n h a v e l i g h t l y d o p e d r e c o m b i n a t i o n r e g i o n s , a n d b,-~ 1 i n t h e g a i n r a n g e o f i n t e r e s t ( 30 1 0 0 c m- 1 ) . A s i r npl e e x p r e s s i o n f o r J, h as a f u n c t i o n o f k e y d e v i c e p a r a m e t e r s is t h e n o b t a i n e d i f p e r f e c t c a r r i e r c o n f i n e me n t is a s s u me d . At 300 K d,h= 7 ~ o u , +~-~ln(1/RIR2)+cq~ +180 [ A c m - 2 ] , ( 2 . g ) a t 350 K d'h = ~l~26d[%"'(~FF) + 2 ~ l n ( l / R ~ R ' ) + q c + [ A c m - 2] (2.9) ( h e r e d is in u n i t s o f mi c r o me t e r s . ) T h e i n t e r n a l q u a n t u m e f f i c i e nc y ~1~ is g e n e r a l l y e x p e r i me n t a l l y d e d u c e d t o b e i n t h e r a n g e o f 0. 6 t o 0. 7 a t r o o m t e mp e r a t u r e , a l t h o u g h i t is v e r y l i ke l y t h a t t he t r u e v a l u e in t h e b e s t d e v i c e s is d o s e t o 1. As a n i l l u s t r a t i o n , c o n s i d e r a t y p i c a l c w A I G a A s ! a s e r wi t h o u t f a c e t c o a t i n g s wh e r e d ~ 0 . 2 lain, An ~ 0. 2, F = 0. 5, c, o m= c~rc = 1 0 c m - 1 , R1 = R 2 = 0 . 3 2 , L = 300 btm ; h e n c e ,%1 = 58 c m - 1, F r o m (2. 8) we e s t i ma t e J , , = 1240 A c m - 2 Laser Diodes and LEDs for Fiber Optical Communication 19 3 0 0 0 ~ E > - G ) Z I - - Z I.iJ o r I . D t::3 3 : I : Z -T- I - - J 2500 2000 1500 1000 0 / / / ' ~ ~-~-t h = 4 8 0 0 A/cm2P'm A 1,xG O l- x A s D H L A S E R S f o ~ E X P E R IM E N T I o A x = 0 . 6 5 - + 0 . 0 5 5 0 0 ro 0 Ax = 0 . 2 5 - + 0 . 0 5 + ~ x = 0 . 5 0 : t 0 . 0 5 / An = 0 . 4 F, o : , . . . . , . . . . , . . . . , . . . . , . . . . . . . . . O . I 0 .2 0 .5 0 4 0. 5 0 .6 HETEROdUNCTI ON SPACING d ( / z ml Fig. 2.10. Threshold current density as a ftmction ofl he hcterojunction spacing d for AI.,Ga i _.,As/GaAs double-hetero- junction lasers with various A1 concen- tration differences Ax at the hctcro- junctions. For the theoretical curves the rcfi'active index step dn~(). 62 A\ has been assumed [2.15] (assuming unity internal quant um efficiency), in reasonable agreement with the best experimental values for such a structure. Because of the decrease in F with d, an opt i mum d value exists which minimizes Jth for a given An value. Fi gure 2.10 shows experimental J t h dat a for Al xGaj _: , As/ GaAs DH lasers [2.15] where bot h d and An were changed. We see t hat for sufficiently large d values where F is not a sensitive function of d, J.h increases linearly with the active region width, Jth/d~4800 Ac m 2-I.tm. This value is close to the theoretical J,h value from (2.8) with r/i = 1. For a given An, a mi ni mum Jth value is observed with decreasing d. The lowest report ed J,h value of 475 A cm- 2 is obt ai ned with d,~0.11am and An~0. 4 [2.15], which requires a high AI cont ent in thc boundi ng regions (x ~0.65). Ot her i mpor t ant laser paramet ers are the differential quant um efficiency, q~xt, measured above threshold [2.16], 1 ~ln(1/R1R2) ~ / ~ , t = ~ / ; ( 2 . 1 0 ) ~ln(1/R,R,)+ F l ~ f ~ ) ~ f c Jr- 0~ou t 2 0 H. Kressel e t aI . and t he t ot al power conversi on efficiency ~/v, Po Y/ p - - 2 ' (2.1l) I R. + IEg/e where Po is the emi t t ed power at cur r ent I. Owi ng to oper at i on in the st rongl y st i mul at ed emission regime, the r/' i val ue in (2.10) may be hi gher t han the r/i value in (2.8). The differential quant um efficiency can be high ; values as high as 50 % have been measured at r oom t emperat ure. The power conversi on efficiency, how- ever, is low near t hreshol d, but increases rapi dl y with drive to reach a maxi mum at 2 to 3 times the t hreshol d current . In practice, the al l owabl e power level depends on the laser' s oper at i ng dut y cycle and peak power because excessive t hermal dissipation or radi ant out put from the facets can damage the device. Oper at i ng in the pul sed-current mode, power conversi on efficiency values as high as 20% (emission from bot h facets) [2.17] have been measured from AI GaAs/ GaAs het er oj unct i on lasers. Typi cal values for devices designed for optical communi cat i on are nearer 5-10 % (Sect. 2.4.1). It is experi ment al l y found t hat the t hreshol d cur r ent density of many laser di odes increases exponent i al l y with t emper at ur e J. , w_ exp(T/To) (2.12) where the value of T o and the range of validity of the expression varies. A common observat i on is t hat T o ranges f r om ~4 0 t o ~200 K. If we compar e the t hreshol d cur r ent at 70C to t hat at 22"C, then T o = 40-200 K cor r esponds to Ith (70 ~'C)/lth (22 C) = 3.5 t o ~ 1.3. Pl ot t ed in Fig. 2.1 l a is this rat i o as a funct i on of het er oj unct i on barri er height AEg for doubl e- het eroj unct i on lasers made from vari ous materials. The t emper at ur e sensitivity decreases with increasing AE~ because of the i mproved carri er confi nement . Equat i on (2.12) expresses empirically the combi ned cont r i but i on of many factors t hat increase Jth'With t emperat ure. Neglecting j unct i on heating, the least Jm change with t emper at ur e occurs in devices where the radi at i on and carri er confi nement s [-2.18, 19] are t emper at ur e i ndependent , and where the defect densi t y in the r ecombi nat i on region is low, maki ng the i nt ernal quant um efficiency t emper at ur e insensitive. Thus, high qual i t y DH lasers with high het eroj unct i on barri ers have a weak t emper at ur e dependence of the t hreshol d cur r ent density. For AI GaAs/ GaAs lasers, this occurs for a bandgap step of >0. 4e V and results in a rat i o J,h (70 C)/Jth (22 C) of 1.25-1.50, Fig. 2.1 la. The cw and pulsed t hreshol d cur r ent of a given device may diverge with increasing t emper at ur e because the increased t hermal dissipation in cw oper- at i on increases t he j unct i on t emper at ur e relative to the heat sink t emperat ure. However, Fig. 2.11b shows t hat in the highest qual i t y laser diodes, the pulsed and cw t hreshol d current s are within 10% up to ~70 C. Laser Di odes and LEDs l ot Fi ber Opt i cal Communi cat i on 21 3.C A rd OJ Z .01 (J o o .1= " ~ I 5 - 1 . 0 0 a I I I I ! I t E3 o I I l I 0 . 1 0 . 2 0 . 3 0 . 4 AEg ( e V ) X - ( A I , G o ) A s o - - ( A L Ga } A s A- - ( TnGa) P/ ( ZnGa) AI r - I - - zn P / ZnGoAIP I I I 0 . 5 0 . 6 0 . 7 0 . 8 6O E izJ n- =: 50 ..J o I - ,,x b / / / , / / x / / CW / / z ~ . / / . ) - / . , , / ~ ' ' / PULSED ~" Il h a[ expl T/166K) AI GoAs ~ LASER 12Fro x 130/~m X(20C) ~. 8370~ 2 o i i [ i HEAT SINK TEMPERATURE (C} Fig. 2.11. (a) The rat i o . l , h ( 7 0 ' C ) / J u , (22 ~C) vs a funct i on of t he bandgap st ep at t he act i ve regi on of doubl e- het er oj t mct i on lasers. ( x ) [2.193; (L~) [2.15] ; (/5, L_) [2.63]. (b) Dependence of t he laser t hr eshol d cur r ent on heat sink t emper at ur e for pul sed and for cw oper at i on. The AI GaAs laser di ode has a 12 gm oxi de- def i ned st ri pe and is 1301.tin l ong 2. 3 Devi ce Fabri cat i on The most developed devices are A1GaAs heterojunction structures produced by liquid-phase epitaxy [2.2]. Here, growth is performed in a furnace (in a hydrogen atmosphere) over a temperature interval typically between 800 and 900C. The growth technique consists of sliding a polished GaAs substrate wafer into successive bins containing solutions of Ga, GaAs, AI and various dopants to produce either n-type or p-type layers. The alloy composition and a ~2, 5f i re, d ~ 0 . 4 F m { b / / / / / / / / . , - / / / / ~ , , ,~, / / / / / / / / / / / / InGoAsP: Zn p~ 1018cm "3 I n P : Zn p ~ I OI Bcm' 3 ~ . . = ~: - . - 7: - - ~ ~ _ ~ - ~ ~7~n ~ - ~ Au + Sn / Au+Zn / i S i 0 2 ~InGoAsP -tACTIVE [ REGI ON Fi g. 2. 12. (a) Doubl e - he t e r oj unc - l i on. s t r i p e - c o n t a c l l as e r d i o d e of l n G a A s l ) / I n P d e s i g n e d for c w e mi s s i o n at a b o u t l . I to 1.3;am. (b) Power emi s s i on vs c ur r e nt for an I n Ga As P / I n P c w l as er pr o- duced by v a p o r - p h a s e e p i t a x y [2.25] i 5 0 5 2 - 4 V P E I n G o A s P / I n P L A S E R I O / z . m S T R I P E 2 7 5 p . m L O N G ) . = 1 . 2 5 p . m 22 II. Kressel el al. I I I I I I I 0 0 2 0 0 3 0 0 4 0 0 D C C U R R E N T ( m A ] conducti vi ty type and level are accurately controlled, using phase diagram and segregation coefficient data, see, e.g. [Ref. 2.1, Chaps. 10- 12] . The planarity of the j unct i ons is very important. Di sturbances are due to improper nucleation, oxygen cont ami nat i on in the furnace, and lack of precise temperature control over the wafer. A form of nonplanarity of the interface called "terracing" consists of steps in the plane of the active region. These steps, spaced several micrometers apart, vary in height between hundreds of angstrom units to under 50/~. Large steps produce significant light scattering losses and result in high threshold current densities [2. 21]. Fortunately, the step height of the terraces can be controlled by careful growth technique to less than 50/~ [2. 22], where their effect is negligible. Another critical growth parameter is the layer thickness. The available technol ogy makes it possible to reproducibly grow doubl e-heterojuncti on di odes with active regions as thin as 0.03 tam, al though typical values are 0.2 to 0.3 tam. Laser Diodes and LEDs for Fiber Optical Communication 23 Doubl e- het er oj unct i on st ruct ures of I nGaAs P/ I nP are also successfully grown by liquid-phase epi t axy [2.23]. Fi gure 2.12a shows a typical cw laser st ruct ure with an I nGaAs P p-t ype "cap" layer to facilitate ohmi c cont act . The I nGaAs P layer is gr own from a sol ut i on of In, GaAs and haP, whereas the InP is gr own from a sol ut i on cont ai ni ng In and haP. Here, the t emper at ur e range for growt h is in t he vicinity of 700 C. The pr obl em of cont r ol is mor e difficult t han with AIGaAs alloys because the quat er nar y layer must have a specified composi t i on to be accurat el y lattice mat ched to l nP (Sect. 2.1). Such st ruct ures also are grown by vapor - phase epi t axy [2.24]. Bot h liquid- and vapor - phase epi t axy have pr oduced devices emi t t i ng at 1.1 to 1.3 i_tm with J,h ~ 2000 Ac m- 2 which are thus sui t abl e for r oom- t emper at ur e cw oper at i on. Fi gur e 2.12b shows the power emi ssi on vs cur r ent curve fi'om a cw I nGa As P/ I nP laser pr oduced by vapor - phase epi t axy [2.25]. The lowest LPE- gr own laser t hreshol d current density values (pulsed operat i on, br oad- ar ea devices) are 730A cm 2 at 2 =1 . 3 ~ m with d=0,11.tm, and 2000 A c m- 2 at 2 = 1.06 txm with d =0. 2- 0. 3 lain [2.26]. 2.3.1 Stripe-Contact Diodes Laser di odes are prepared by cleaving t wo parallel facets (usually { 110} crystal planes) to form the mi rrors of the Fa br y- Pe r ot cavity. Al t hough the sides of the active region can be made by et chi ng or sawing, they are preferabl y defined by a stripe cont act for the following reasons: I) The radi at i on of the stripe laser is emi t t ed f r om a smaller region, t hus i mprovi ng coupl i ng of the r adi at i on i nt o fibers ; II) The operat i ng cur r ent can be reduced by defining a small active ar ea; 111) The t hermal resistance of t he di ode is generally reduced because the heat -generat i ng active region is embedded in a larger, inactive semi conduct or medi um : IV) Wi t h a smaller active di ode area it is easier to obt ai n a reasonabl y defect-free ar ea; V) The active region is isolated from an open surface al ong its t wo maj or dimensions, a fact or essential for reliable l ong-t erm oper at i on (Sect. 2.7); VI) A small stripe width permi t s fundament al lateral mode operat i on. Many i ngeni ous schemes have been described in the l i t erat ure to form st ri pe-cont act laser diodes. Fi gure 2.13 shows some of t he mor e widely studied ones. In the planar stripe lasers (Fig. 2.13a), the ohmi c cont act is defined by an isolating oxi de layer [2.13]. Alternatively, pr ot on bombar dment can be used to form resistive regions everywhere on the device surface except in the stripe area (Fig. 2.13e) [2.27]. Selective diffusion of t he p- t ype dopant into an n-t ype surface l ayer also can form an i sol at ed area (Fig. 2.13b) [2.28]. Mes a stripe laser types include area definition by etching (Fig. 2.13c) and the bur i ed- het er oj unct i on version [2.29] where, after mesa etching. AIGaAs is 24 H. Kressel et al. ~ 2 ( ~ .__] DIFFUSED P-REGION F METAL Si 0 2 , ~ x x .. F META L ' " ' " 'o oo'o' , i I l l , l , J I GaAstNorP)~ I ( A l GalAs (P) (APGa)As / / / / / / / / / / / / / / p ) . F - ~ t / . / / / / : ' t ' : ' : ' : / e ' : " / / I , ( A L G a ) A s ( N ) GaAs(N or ~ ( A L G a ) A s ( N ) I GoAS(N) ~ GOAs (N1 J .-L ._L (o) (b) GoAs ( N o r p ) F uETA" ~L- a S .j---METAL ~ o x , o E I I _L ..J_ _-- - . [ c ) (d) GoAs ( P) I HI GH RESI ST I VI T Y ( A I G ) A s ( P ] ~ I I . . ~R EGIONS b t / \ ---META" z ~ . . . . . . . . . GoAs ( N o r P ) (AI, Go )As ( N ) Go As ( N ) _ L (e) Fig. 2. 13a-e. Vari ous , net hods for pr oduci ng st r i pe- cont act laser st r uct ur es [2.1 ] grown i nt o the moat s between the di odes (Fig. 2.13d). The et ched mesa t ype is undesi rabl e because the excessive sur f ace- t o- vol ume rat i o of the device is det ri ment al t o reliability. The bur i ed- het er oj unct i on laser overcomes this handi cap. Here, the AIGaAs layer grown into the grooves passivates the sidewalls of the device. The het er oj unct i on at the sidewalls also const i t ut es a dielectric step which provi des st rong lateral radi at i on confi nement . Different st ri pe-cont act st ruct ures pr oduce different degrees of lateral cur r ent and radi at i on confi nement . The pl anar stripe st ruct ures lack lateral bl ocki ng regions at the r ecombi nat i on region edges. Therefore, t hey cannot be made arbi t rari l y nar r ow because the cur r ent and radi at i on spread out si de the defined cont act area. The rat i o J ( h / J t h ( ~ : . O is useful in compar i ng the different st ri pe-cont act st ruct ures in t erms of their ability to confine laterally t he cur r ent and radi at i on. Jib(m) is the t hreshol d current densi t y of the mat eri al prepared in the form of a (broad) sawed-side laser and J,h is the t hreshol d cur r ent density of the st ri pe-cont act device, cal cul at ed on t he basis of the stripe area. Ideally, J t h / . l t l , ( W J ) = 1; the great er the current and radi at i on spread beyond the stripe area, the larger the rat i o becomes. Fi gure 2.14 compar es oxide-isolated [2.30], Laser Di odes and LEDs for Fi ber Opt i cal Co mmu n i c a t i o n 25 4 " - 3 E I S O L A T E D M E S A S T R I P E - J O I I I I I I I I I I I I I I I I I I I I I I I I 2 4 6 S I 0 1 2 1 4 1 6 1 8 2 0 2 2 2 4 S T R I P E W I D T H ( / ~ . m ) Fig. 2.14. The rat i o J , h / J , l , ( ~ ) as a f uncl i on of st r i pe wi dt h of oxi de- def i ned [2. 30], pr ol on- bomba r de d [2. 31], and et ched mesa [2. 32] l aser di odes of AI Ga As pr ot on- bor nbar ded [2.31], and et ched mesa [2.32] devices. For stripe widths great er t han about 20 lain, all st ruct ures pr ovi de J t i , / J m ( o o ) rat i os close to unity, but with decreasi ng stripe widths, the oxi de-i sol at ed and pr ot on- bombar ded st ruct ures gradual l y lose t hei r effectiveness compar ed to the et ched mesa structures. Buri ed-het eroj unct i on st ruct ures appr oach the per f or mance of the et ched mesa device and, with stripe widths of t to 2 pro, t hreshol d current s as low as 10 mA have been obt ai ned. However, such nar r ow stripe widths limit the useful opt i cal power level (Sect. 2.6.1). Pract i cal st ri pe-cont act structures, designed to emit about 10roW of cw power (with t hreshol d current s of 40 to 150mA) requi re a stripe width of about 10 tam. For this purpose, the pl anar stripe lasers are the easiest to produce. Not e t hat with the oxi de-st ri pe st ruct ure it is possible to cont r ol to a significant ext ent the degree of lateral current confi nement by cont r ol of the sheet resistance of the material between the ohmi c cont act and t he r ecombi nat i on region. A high sheet resistance minimizes t he cur r ent spread [2.33]. 2.3.2 Laser Construction and Packaging Fi gure 2.15 shows a cross section of an A1GaAs, oxi de-defi ned st ri pe-cont act laser designed for low t hreshol d cur r ent cw oper at i on at r oom t emperat ure. The thin "cap" of GaAs permi t s a good electrical cont act t o be made, while 26 H. Kressel et al. p - G o A s - ] 2~,,, t d ~ O . I / x m { ~ l / x rn. W=12/xm . Z n - D I F F U S E D I,I--- ~ . . . ~ ~ - Si 0 2 :> "..-'/:/"/~./'E~.-.~ .~4"~- ~- G,z~.4J4,,cr - k . P" " . / - ' 1 ~ ~ OI Sem - s p - A | 0 356 o065As ( ~ 1018c ~ 51~ .: . . . . . " . . . . _- . . . . . . . . ~ AIO.08G%92 A s n - A10.35 Go065 A s n - Go As SUBSTRATE (100) I ~1 Fig. 2.15. Schemat i c cr os s sect i on of AI Ga As l aser di ode wi t h oxi de- def i ned st r i pe conl act ] Fig. 2.16. Di ode package wi l h a s hor t fiber l engt h "pi gt ai l " for easi er c oupl i ng to fiber cabl es [2. 34] C O N T A C T . C O N T A C T , . ~ ~ . ~ ~ \ G R O O V E D C L A M P I N G / LM I \ ~ B L O C K S U R F A C E / ~ C A P D C O P P E R / B L O C K H E A T S I N K I I I 0 " O r " 0 . 2 " minimizing cur r ent spreading. A layer of SiO 2 several hundr ed angst r om units thick is deposi t ed at low t emper at ur e on the wafer aft er t he compl et i on of t he epilaxial gr owt h process. Using phot ol i t hogr aphy, a stripe 8- 10gi n wide is defined in the oxide, after which the oxi de in the stripe region is removed. The wafer is then sealed i nt o a quar t z ampoul e t oget her with a source of Zn m or der to diffuse a shallow p+ region into t he exposed GaAs stripe to reduce the cont act resistance. Fol l owi ng metallization, the wafers are cleaved i nt o strips and then subdi vi ded by furt her cleaving or sawing bet ween the stripes. The typical cw laser di ode chip is a 500 x 300 pm rectangle, of which onl y the small por t i on under the stripe is active. The diodes are mount ed on headers. A commonl y used package is shown in Fig. 2.16 where the laser is mount ed with i ndi um solder p-side down on a copper block. The case can be hermet i cal l y sealed with a cap cont ai ni ng a glass window. Alternatively, a short length of fiber can be perl nanent l y bonded to the package and subsequent l y at t ached to a long fiber cable link [2.34]. In hybri d circuits, the laser di ode (or LED) may be mount ed with the ot her circuit elements on metallized ceramic. The device in Fig. 2.15 has a distance of ~ 2 lam between the r ecombi nat i on region and the heat sink. Such a thin l ayer is desirable to minimize the t hermal resistance of the structure. The thin cap reduces bot h the lateral cur r ent spread beyond t he stripe edges and the electrical resistance of the diode. Due to its i mpor t ance for cw oper at i on and its effect on ot her laser propert i es, we now consi der heat flow in injection lasers in mor e detail. Laser Diodes and LEDs for Fiber Optical Communication 27 Most of the heat in injection lasers (or LEDs) is gener at ed in the vicinity of the r ecombi nat i on region. Thi s heal escapes i nt o the heat sink by conduct i on t hr ough the semi conduct or mat eri al and the sol der interface. Since the distance to the heat sink is much less t han the typical emission wi dt h ( > 10 lain), an appr oxi mat i on to the t hermal resistance is obt ai ned by cal cul at i ng the one- di mensi onal heat flow, while t aki ng the effective stripe width (i.e., t he width of t he light-emitting region) as the laser width. Three-di mensi onal spreadi ng is al l owed in the heat sink. The t her mal resistivity of A1GaAs increases with AI cont ent from about 2cm- deg W - I in GaAs to 8cl n-deg W -1 for Alo. sGao. sAs [2.35]. Therefore, t he thickness of the p-t ype A1GaAs l ayer shoul d be minimal, with the rest ri ct i on t hat the optical field spreadi ng from the r ecombi nat i on region not be allowed to reach the highly absor bi ng GaAs '~cap" layer. A l l am thick Alo.35Gao.65As l ayer is generally adequat e for this purpose, but it cont r i but es about half the t hermal resistance of t he device. For a nomi nal stripe width of 12 lain and an effective wi dt h of 20 lain, the t her mal resistance cont r i but i on of t he AIGaAs l ayer is about 13 K W 1. A 1 tam thick GaAs "cap" cont r i but es about 2 K W- ~, a i tam thick i ndi um sol der layer 3 K W- l, and the copper heat sink 5 . 5 KW i for a t ot al of ~ 2 3 KW -1. This is not far from typical experi ment al values of 20 to 30 K W i for a 300tam l ong pl anar stripe laser. Therefore, with a power i nput t o the laser of 0.4 W, the j unct i on t emper at ur e is about 10 C above the heat sink t emperat ure. Of course, the t hermal resistance varies appr oxi mat el y inversely with the di ode length, and a 150 lain laser with a similar stripe const r uct i on will have a t her mal resistance of ~ 50 K W ~. So far the emi t t i ng region was assumed to be of a uni form t emperat ure. A mor e accurat e analysis [2.363 shows t hat t wo-di mensi onal heat flow has t o be consi dered, especially when t he stripe width appr oaches the distance between the j unct i on and the heat sink. Lat eral and upwar d heat flow reduces the actual t hermal resistance, and causes a t emper at ur e gradi ent al ong the j unct i on plane. Thr ough the t emper at ur e dependence of the refractive index, this cont ri but es to mode guiding under t he stripe. Recent experi ment s [2.37] have shown t emper at ur e nonuni f or mi t i es cor r el at ed with sol der voi ds arising in the mount i ng pr ocedur e and t emper at ur e peaks of appr oxi mat el y 5 K at the t wo facets. 2 . 4 L a s e r C h a r a c t e r i s t i c s 2. 4. 1 Po we r Emi s s i on vs Current o f Pr ac t i c al c w Las ers The t hreshol d cur r ent of a st ri pe-cont act laser is cont rol l ed by : I) The initial qual i t y of the epitaxial st ruct ure, i.e., the t hreshol d cur r ent densi t y as measured in br oad- ar ea structures. This depends oil t he paral net ers discussed in Sect. 2.2 ; 28 H. Kressel et al. , 2 0 I ~ 14 ~EEI 00 " ' 80 ~ nr t ) 6 0 ~ 1 Q o 4 o W r~ 1- 2 0 I - o , , , , , tO0 200 500 LASER LENGTH L (p. m) Fig. 2. 17. Th r e s h o l d cur r ent as a fu.~ctiou of ~ 4 I l aser l engt h L for l asers havi ng t he s t r uct ur e of Fig. 2.15 ~ 2 [ Fi R. 2. 18. Power emi s s i on f r om one facet as l,- a f unct i on of c ur r e nt t hr ough a wel l - behaved O' cw l aser o1" the t ypc s hown in Fig. 2.15, as a 0 f unct i on of heat si nk t e mpe r a t ur e AtGoAB OH LASER ] / / / ~ / / %:-tram1 2 0 4 0 6 0 8 0 I 0 0 DIODE CURRENT (mA} 70 60 E l - 50 t.d J ,,, 4 0 Z 0 i D 30 I.- 0 a: 20 W 0 13_ 10 0 0 , ~ , I 1 ~ I ~ ~ / / / , , ~ -CALCULATED i I / / / ~ /. 1 / i , , ; I / / +o5O/o DUTY CTCt E / / / / oo,o% DuTY ~C~E I f / ' - : c w , f / / ,Y o/E I I I I I I 50 tO0 150 200 250 500 550 CURRENT (mA) 400 Fig. 2.19. Power emi s s i on f r om one facet as f unct i on of cur r ent for a l aser di ode oper at ed cw a nd pul sed at dul y cycl es of 10 and 50%. The st r i pe wi dt h is 121am, t he di ode l engt h is 125pro. t he series r esi st ance is 4 o h ms , t he t he r ma l r esi st ance is 80 K/ W and t he thres!~old cur r ent i ncr eases with t e mpe r a t ur e as exp (T/80) Laser Di odes and LEDs for Fi ber Opt i cal Co mmu n i c a t i o n 29 I1) The effective st ri pe-cont act area, which takes i nt o account the cur r ent and r adi at i on spread beyond t he stripe edges and t he length of the laser; III) The t emper at ur e of the r ecombi nat i on region at the oper at i ng cur r ent and dut y cycle. Here, the electrical and t her mal resistance are key paramet ers. The st ruct ure shown in Fig. 2.15 provi des adequat e lateral current confine- ment and yields st ri pe-cont act lasers with t hreshol d current s under 100mA. Fi gure 2.17 shows the dependence of the t hreshol d cur r ent on laser length L; with L = 1001am, the t hreshol d current is 40 to 60mA I-2.38]. Fi gure 2.18 shows the power emission from one facet as a funct i on of the current f r om a represent at i ve laser di ode oper at i ng cw at heat sink t empera- tures bet ween 0 and 70 C. (This laser was not provi ded with a refl ect or on the back facet.) A reduct i on in the t hreshol d cur r ent as well as a subst ant i al i mpr ovement in the power efficiency is obt ai ned by addi ng a reflector. Met al reflectors (such as Au) are t r oubl esome because they are easily damaged duri ng di ode fabrication. To eliminate this probl em, a mul t i pl e-l ayer dielectric re- fl ect or of evapor at ed Si and AI203 has been devel oped; with six layers, reflectivities in excess of 96 % are achieved [2.39]. At the maxi mum power level ( 30r oW of emission), power efficiency values of 14% have been obt ai ned. However , a reasonabl e cw oper at i ng power level for l ong-t erm reliability is closer to 10 mW for a typical device and t he power efficiency at this level reduces to about 7 %. Hi gher peak power values are at t ai nabl e from a di ode in pulsed oper at i on, but the power-current curve is dut y-cycl e dependent because the increasing j unct i on t emper at ur e changes the t hreshol d cur r ent and differential quant um efficiency. The out put curve as a funct i on of dut y cycle can be predi ct ed from the known t emper at ur e dependence of lth and the di ode' s electrical and t hermal resistance. The power dissipation Pdi,,s increases with I , P a i s ~ - - I E g / e + 1 2 R ~ . Fi gure 2.19 compar es the measur ed and calculated power curve for cw oper at i on and at 10 and 50% dut y cycles. For this laser, T o in (2.12) was about 80 K, the t her mal resistance was ~ 80 K W- ~, and the series resistance R~ was 4 ohms. The power curves are essentially i nvari ant for dut y-cycl e values under 10% because of the negligible j unct i on t emper at ur e changes. The above discussion has concent r at ed on AIGaAs lasers, but similar consi derat i ons appl y to ot her lasers. A not ewor t hy difference between I nGaAs P/ l nP and A1GaAs lasers is in the t emper at ur e dependence of the two types of structures. I nGaAs P/ I nP lasers designed for emission at about 1.2 gm have a het er oj unct i on barri er height of about 0.3 eV. Therefore, their t hreshol d cur r ent is mor e t emper at ur e sensitive t han t hat of AIGaAs lasers with barri ers of 0.4 eV (Fig. 2.1 la). 2. 4. 2 Las e r Radi at i on Pat t e r ns M o d e s . The allowed el ect romagnet i c modes of tile laser cavity are separabl e into t wo i ndependent sets, havi ng ei t her t ransverse electric (TE) or t ransverse 30 14. Kressel et al . magnet i c (TM) pol ari zat i on. The mode number s m, s, and q give the number of ant i nodcs in the opt i cal field al ong the t hree axes of t he cavi t y (Fig. 2.20), transverse, lateral and l ongi t udi nal , respectively. In the present not at i on, the f undament al mode is denot ed 1. The allowed longitudinal modes are det ermi ned by the effective index of refract i on and the cavity length. The resul t ant Fa br y- Pe r ot mode spacing is several angst r om units in typical laser diodes. The lateral modes are dependent on the met hod used to define the t wo edges of the di ode and on the t emper at ur e and gain profile across the diode. General l y, in nar r ow st ri pe-cont act lasers onl y l ow- or der modes are excited. The transverse modes depend on t he dielectric vari at i on perpendi cul ar to the j unct i on plane. In the devices discussed here onl y t he fundament al t ransverse mode is excited, a condi t i on achieved by restricting the wi dt h of the wavegui di ng r ecombi nat i on region d (i.e., het ero- j unct i on spacing) to a few t ent hs of a mi cromet er. Therefore, the far-field radia- tion pat t er n consists of a single l obe in the di rect i on perpendi cul ar to the j unc- tion, Fig. 2.21a. ( Hi gher - or der t ransverse modes give rise to ' ~rabbit-ear" lobes which are undesi rabl e for fiber coupling.) The far-field pat t ern in the di rect i on perpendi cul ar to the j unct i on pl ane reflects t he radi at i on di st ri but i on in the vicinity of the active region. This, in t urn, depends on d and on the refractive index step An at the het eroj unct i ons. Fi gure 2.22 shows the dependence of the beamwi dt h 0~ (full angul ar width at the half-intensity points) on d/2 for vari ous An values. ( For Al~Ga~ _xAs/ GaAs het er oj unct i ons An ~ 0.62x.) The decrease in 0 for small d reflects the reduced radi at i on confi nement as t he wavegui de becomes narrower. For efficient coupl i ng t o fibers one may even allow a somewhat higher t hreshol d [because of reduced F (2.8)] to achieve nar r ower beam pat t erns. Fundament al lateral mode oper at i on can be obt ai ned by sufficiently restricting the width of the active region. The lateral modes depend on the dielectric profile in the plane of the j unct i on and thus, on the t echni que used for j unct i on area definition. The simplest case occurs when t here are t wo high (and equal) steps An perpendi cul ar to the j unct i on pl ane formi ng a rect angul ar box cavity. Then the device can be model ed in t erms of " box modes" [2.1]. Consi der a di ode of width W. Fr om a simple analysis of the critical angle for t ot al internal reflection at t he sidewalls, we find t hat t he highest lateral mode number Sm capabl e of pr opagat i ng in the st ruct ure is \ n / j ' (2.13) Therefore, to oper at e in the fundament al lateral mode, the index step condi t i on is n 8 " (2.14) , I I _ - d c~ d AJ.IS N3 J.NI 2 " o z o" A.LISN3.LNI . L H g I q 7 I - ( . 9 . . . . 1 h i U l n . " n ~ I - -
0 o = #_ z ] . 8 5 m A ~~ ~ - ~ o o = ~ . - = , : o = < .-._~ ~ ~ w ~ . ~ _ ~ - ~ : > - , , ' - ; . ~ ~. ~, ~ -~ T~ r, # , ~ , I [ I t 1 . 6 T m ^ 32 1t. Kressel et al. 6 0 5 O W ~ D mw4O 3 0 ~D O I I I I I 0 0.1 0.2 0.3 0.4 0.5 0.6 0 . 9 d E F F E C T I V E C A V I T Y W I D T H ( p . m ) [ - ~ L ] Fig. 2.22. Angular width at the half-power point as a function of d/2 and the index step An for double-heterojunction lasers For GaAs, n = 3.6 and 2 ~0. 9 lam ; hence Ant n < 7.8 x 10- 3 W- 2, where Wis in mi cromet ers. Thi s step-index, step-sidewall model is onl y appr opr i at e for sawed-side, etched-side, or bur i ed- het er oj unct i on devices. Analysis of t he pl a na r stripe lasers is more difficult because the shallow dielectric profile is rel at ed to the cur r ent and gain di st r i but i on; hence, it is subject to change with current and optical power level. Experi ment al l y, f undament al lateral mode oper at i on is frequent l y obt ai ned in st andar d ( < 131am) pl anar stripe lasers, at least near t hreshol d, but hi gher - or der modes generally reach t hreshol d with increasing current , pr oduci ng a mul t i -l obed, far-field pat t er n as shown in Fig. 2.23 where we see the second mode reachi ng t hreshol d. Mode gui di ng in pl anar stripe lasers results from the combi ned cont ri - but i ons t o the dielectric profile of several elements. Al t hough t he specific cont r i but i on of each is not readily calculated, the following are rel evant : I) increasing gain near the stripe cent er produces a cor r espondi ng increase in the i magi nary part of the dielectric const ant ; II) local heat i ng rel at ed t o t he cur r ent di st ri but i on and power dissipation increa,ses the real part of the dielectric const ant ; IlI) however, acting in the opposi t e di rect i on to reduce the dielectric const ant under the stripe is t he di st ri but i on of free carriers, i.e., the hi gher t he carri er density, t he l ower the index This effect is closely rel at ed physically to (l). ~= 2 0 E 02 [al o Q. I-- n o ku 13 Laser Diodes and LEDs for Fiber Optical Communication 33 I I I ! - / 3 0 0 35O 4 0 0 L A S E R DRI VE CURRE NT , r n A REL AT I VE I NTENSI TY N "I- o I,- u'~ 3 > . / / / / / / i I I I i - 4 2 0 mW i I l I i - 2 0 + 2 * 4 B E A M A NGL E , DEG. Fig. 2.23. (a) Power emission as a function of current for a laser showing a kink in its power curve. The oxide-defined slripe width is 25 p.m. The dotted line is the relative noise power (at 60 MHz) which peaks at the threshold of each of the two modes. (b) Far-field in the junction plane at two power levels (9 and 20 roW) showing the change from the fundamental to the next lateral mode A shallow maxi mum in the dielectric profile under the stripe results from the combi ned effect of the above three factors. With changes in current and optical power density, the local heat i ng and gain coefficient profile are likely to change, thus affecting the ability of the various lateral modes to propagate. For example, spatial hole burni ng can occur in the region of the most intense optical excitation with the result t hat the local gain coefficient is depressed I-2.40]. This may lead to more efficient coupl i ng of the gain profile to the s econd mode t han to the fundament al lateral mode. Fundament al lateral mode operat i on is desirable not onl y because it eases coupling into low numerical apert ure fibers (including single-mode fibers), but also because mode changes with current are frequently accompani ed by kinks in the power out put vs current curves, as shown in Fig. 2.23a. The key principle which governs the restriction of laser operat i on to a single mode is t hat the differences between the propagat i on losses of the fundament al and the higher- order modes be as large as possible. Restricting the stripe width of buried- het eroj unct i on lasers to very small values of 1 to 2pm is one met hod of achieving this objective, al t hough at the expanse of the useful power from the device. Anot her approach consists of i ncorporat i ng regions in the device t hat produce a greater internal absorpt i on coefficient for the hi gher-order modes 34 H. Kresse! et aI, OHMIC CONTACT,~ /Zn DIFFUSED 1~\\\\~\\~ p- 6,A, (cAP} . ~ . ' " ~] ~- p- At o. 33 Gao.67 As j_kx,~z,,-, s~ e,,~ x ~ x x x x L n - At o 05 Ge095 As (ACTIVE) ~0,4/Lm F 1 " " T I \ / - - ' ~ n - At 0 33 Gao 67 A s I , l \ n- GoA's (SUB;TRATE) a )HMIC CONTACT, x .Zn DIFFUSED ~, , , . \ I , \ ~ .~ .,,p -GoAs (CAP) ~ r . p - A ' L O . 3 5 6 o 0 . s 5 A s ~ 5 ~ . . . . . x, , ~N~i . . p _ At o8 % ~2 A, (ACTIVtl f %1. " T - n - Ga At (SUBSTRATE) L I b 30 25 E v 2 o o t5 o IO o C C D H - A 1 C.W 20" C 7 0 " C 5 0 I 0 0 150 2 0 0 DC DRI VE CURRENT ~ l ( mA ) Fig. 2, 24a- c. Sc he ma t i c cr os s sect i on of t wo l aser s t r uc t ur e s of AI Ga As des i gned for s us t a i ne d f unda me nt a l l at eral mo d e oper a- t i on. In (a) t he l osses of t he f unda me nt a l mode ar c l ower t h a n t hos e of hi gh- or de r mo d e s becaus e of i t s field s pr eads beyond t he c ha n- nel r egi on i nt o t he a bs or bi ng n- t ype Ga As s ubs t r a t e [2. 41]. In (b) t he t hi cknes s of t he acl i vc r egi on is co~sl r i ct ed onl y in t he desi r ed l asi ng regi on. The t hr es hol d for hi gh- or de r l at eral mode s c a nnot be r eached if t hc ext ent of t he cons t r i ct ed regi on is suffi ci ent l y smal l [2. 42]. Not e t hat bot h s t r uc t ur e s r equi r e t he gr owt h of epi t axi al l ayers over n o n p l a n a r r egi ons. (e) Power cur ve of t he s t r uct ur e m (b) s howi ng excel l ent l i neari t y. ( The bendi ng at hi gh c ur r e nt s is due to j unc t i on heat i ng. ) I - 2 . 4 2 ] t han for the f undament al one. Schemes i nvol vi ng buri ed channel s [2.41-1, con- stricted active region thickness [2.42], and angled cont act s [2.43] have been proposed in or der t o achieve this objective. Fi gure 2.24 shows the cross section of t wo st ruct ures designed to pr omot e fundament al lateral mode operat i on. In the channel - s ubs t r at e- pl anar st ruct ure in Fig. 2.24a [2.41], the fundament al lateral lasing mode i~ confi ned to the channel region because the absor pt i on losses ext ernal to t hat region are high. Laser Di odes a nd LEDs for Fi ber Opl i cal Cor nmt mi c a t i on 35 Thi s is due to the del i berat el y small spaci ng between the active layer and the subst rat e in the regions out si de t he channel, which results in a significant fract i on of the radi at i on spreadi ng from the active l ayer into the subst rat e in these regions. Thus, the losses of the hi gher - or der lateral modes (which spread beyond t he nar r ow channel region) are hi gher t han for t he fundament al mode. Not e t hat the channel wi dt h must be relatively nar r ow ( < 10pm) in or der for the desired effect to occur. Fi gure 2.24b shows an al t ernat i ve st ruct ure suitable for mode cont r ol where the thickness of the active region is const r i ct ed onl y in the desi red lasing region [2.42]. Thus, the carrier densi t y for a given current density is sufficient for t hreshol d to be reached onl y in the cent ral region under the current -i nj ect i ng stripe cont act . If the wi dt h of t he const ri ct ed region is sufficiently nar r ow, high- or der modes cannot reach t hreshol d, t hus pr oduci ng a mode-st abi l i zed laser. However, as is the case for the st r uct ur e in Fig. 2.24a, hi gh- or der - mode oper at i on is possible with this st r uct ur e when t he const ri ct ed active region wi dt h becomes excessive. Ther ef or e, bot h t ypes of st ruct ures are limited in the power level which can be reliably pr oduced in the fundament al lateral mode. Because of t he mode stability in t hei r safe oper at i ng regions, bot h st ruct ures exhibit linear power - cur r ent curves. Fi gure 2.24c shows such a curve of the st ruct ure in (b). Fiber Coupling. Coupl i ng of the r adi at i on from nar r ow st ri pe-cont act laser di odes i nt o typical mul t i mode step-index or graded fibers presents little difficulty. About haft of the power from one facet can be i nt r oduced i nt o the fiber by careful alignment. But the use of slnall-diameter, si ngl e-mode fibers i nt roduces addi t i onal coupl i ng probl ems. An accurat e sol ut i on to the elec- t r omagnet i c pr opagat i ng wave pr obl em for the si ngl e-mode case has been made by Kapany [2.44], and Snitzer [2. 45], but it is mat hemat i cal l y compl ex. However, the fundament al mode can be appr oxi mat ed by the Gaussi an field di st ri but i on of Kogelnik and Li [-2.46]. If one appr oxi mat es the beam froln the source as Gaussi an [2.47], the coupl i ng efficiency i nt o the single fiber mode can be comput ed [2.48]. Present single-mode fibers have core di amet ers in the 3 to 8 pm range, and far-field beam spreads of the or der of 3 to 7 '~. (A consi derabl e fraction of the opt i cal energy in a si ngl e-mode fiber is carri ed out si de the core, leading to effective Gaussi an mode di amet ers larger t han the physical core size.) Thus, since the appr oxi mat el y I t, tm thickness of a typical DH laser emi t t i ng region is less t han the fiber core size, its beam spread angle is much greater, requi ri ng some form of opt i cal mat chi ng el ement to maxi mi ze coupl i ng to the fiber. In the laser j unct i on plane, stripe widths are generally great er t han the fiber mode di amet er, while the beam angle is perhaps 2 to 3 times the fiber accept ance angle. In this case, no opt i cal elements can coupl e all the laser radi at i on to the fiber, since the radi ance of the system (energy per unit area per unit solid angle) cannot be increased with optics. About 1 mW of power can be coupl ed from a 10p, m-wide stripe laser i nt o a si ngl e-mode fiber by simple techniques. 36 H. Kressel et al . 2.4.3 Spectral Emission The wavel engt h of the emission depends on the bandgap energy, the dopant concent r at i on in the r ecombi nat i on region and the j unct i on t emperat ure. For commonl y used DH lasers with lightly doped r ecombi nat i on regions, the lasing peak energy is 20 to 30meV below the bandgap energy [2.1]. The spect rum shifts t owar d lower energy with increasing t emper at ur e at a rat e of ~ 0 . 5 me VK- 1 ( ~ 3 AK ~at ~8500A) . The spectral width of the laser emission depends on the number of l ongi t udi nal modes excited. The evol ut i on of the spect rum from a cw laser di ode with increasing cur r ent is shown in Fig. 2.21b. Near t hreshol d, the emission is relatively wide, with many l ongi t udi nal modes. With increasing drive, a few of the modes become domi nant , and in some lasers a single mode domi nat es. Mor e typically, however, one iliads 3 to 4 lines with a t ot al spectral width of 5 to 10, x. within the half-intensity points. Lasers with very short cavities (tinder 100gm) have great er l ongi t udi nal mode separat i ons, and oscillate in a single l ongi t udi nal mode mor e frequent l y t han l onger lasers. The cw power emi t t ed in a single mode in such short lasers somet i mes exceeds 10 mW. However, if the drive cur r ent of such a laser is widely varied, the out put wavel engt h may shift, and the overall linewidth may br oaden. Therefore, mai nt ai ni ng spectral puri t y in pulsed laser oper at i on is difficult unless the cur r ent range is restricted. 2.4.4 The Power-Current Curve Linearity The power - cur r ent curves of laser di odes frequent l y exhi bi t slope changes (denot ed kinks) havi ng vari ous origins. Defects in the j unct i on region, which can pr oduce lasing in filaments, may cause kinks. A second established cause of kinks results from lateral mode changes in lasers with self-induced wavegui di ng (i.e., those lacking st rong lateral mode guiding). As not ed earlier, nar r ow pl anar stripe lasers frequent l y oper at e in a single lateral mode near t hreshol d, but mor e modes appear as the cur r ent is increased. As shown in Fig. 2.23, the power curve shows a slope change at the cur r ent where the t hreshol d for the second mode is reached. The power out put in thc fundament al mode appears to sat urat e, while the second mode becomes domi nant . Ext ended linear power curves are expect ed in devices where : I) many modes can be excited (thus pr oduci ng a smoot h t ransi t i on from one mode to the next), such as in wide-stripe lasers; II) a single lateral mode (which can be high order) remai ns domi nant in a wide cur r ent range where dielectric sidewalls are built into the laser st ruct ure. I nt er modul at i on measur ement s are one measure of linearity (see Chap. 9). Indeed, one finds experi ment al l y t hat linear power curves are mor e commonl y found in lasers with nar r ow stripes when a single mode domi nat es (as in Fig. 2.24c), or in wide pl anar stripe lasers ( > 20 ~tm), where many modes are easily excited. Laser Diodes and LEDs for Fiber Optical Communication 37 2. 5 Li ght - Emi t t i ng Di ode s The distinguishing features of LEDs designed for opt i cal communi cat i on include high modul at i on rat e capability, high radiance, high reliability, and emission wavelengths restricted to the near-i nfrared spectral regions of low at t enuat i on in fibers. The earlier LEDs, based on GaAs or GaAs P homo- t hey have some limited appl i cat i on, For t unat el y, t he advances in the het er oj un- bot h lasers and LEDs share a c ommon t echnol ogy. they have some limited appl i cat i on. For t unat el y, the advances in t he het eroj unc- tion laser di ode t echnol ogy coul d be appl i ed to these new LEDs, and cur r ent l y The wavelength of LED emission is i mport ant , not onl y because of fiber at t enuat i on, but also because of the reduced fiber spectral dispersion at 1.3 ~m compar ed to 0.81.tin [2.49]. Plastic fibers are also suitable for short optical links. They have at t ent uat i ons of a few hundr ed decibels per ki l omet er in t hei r region of low at t enuat i on (typically near ) ~ 0 . 6 6 ~na) ~, but their large numeri - cal apert ures make LED light coupl i ng easy. However, nei t her semi conduct or lasers capabl e of reliable cw emission at r oom t emper at ur e nor specifically designed het er oj unct i on LEDs are yet avai l abl e near 0.661am. Convent i onal red-l i ght -emi t t i ng GaAs P di odes designed for display purposes can be used if t he power level is restricted to the safe oper at i ng region. Mor e recently, however, plastic fibers [2.50] have been devel oped with mi ni mum loss at 0.7 ~tm where A1GaAs LEDs can be empl oyed. 2.5.1 Structures The materials most used for communi cat i on LEDs are Al . , Gal _xAs , l nxGa ~ ~As and In Ga~_xAs.,.P~_r; AI~Ga~ . , . As devices emi t t i ng at ~0. 82 to 0.85 ~tm are the most highly devel oped at this time. The internal quant um efficiency of st at e-of-t he-art devices is >50% at r oom t emper at ur e but the ext ernal efficiency is much lower owi ng to t he high index of refract i on of I l I - V compound semi conduct or s ( n~3. 0 to 4.0). The light spont aneousl y creat ed near a p-n j unct i on or het er oj unct i on is emi t t ed i sot ropi cal l y in t he j unct i on plane. Even for light rays striking a device interface with air at angles close to nor mal incidence, the reflectivity is high, --~30%, owi ng to the high index of refract i on of the semi conduct or. Fur t her mor e, rays striking the surface at angles great er t han about 16 from the nor mal will be compl et el y reflected back i nt o t he bulk of the device. Assuming a single opt i cal wave t ransi t across the LED pr i or to loss by internal absor pt i on, onl y a few percent of the internally generat ed radi at i on strikes the crystal surface at an angle less t han 16 , and can t hereby exit t hr ough a sinyle pl anar surface into air. However, by the use of spherically shaped emi t t i ng surfaces and reflecting side boundari es, t he ext ernal emission efficiency can subst ant i al l y exceed this value. Fur t her mor e, if the internal r eabsor pt i on of 1 These fibers are produced by ~he DuPom Corporation. 38 H. Kre.~sel et al. ,L- . . . . . . . . { I / / I / I ~ n EPOXY [ ~ / F I B E R Rp /n \ [ ~ , \ t l / t , ~ . . . . . . . Fl U . . . . . 7 a 7 E IC (:3 w F., 2 C ACE EMITTER I = I , I , I ~ I iO0 200 BOO 400 500 CURRENT (mA) Fig. 2.25. (a) Et ched-wel l LED (surface emi t - ter) of t he Burrus t ype; (b) st r i pe- cont act edge- emi t t i ng LED. (c) Tot al power emi t t ed from best surface emi t t er of t he Burrus t ype [2.58] and emi ssi on from one facet of a st ri pe- cont act , cdgc emi t t er [2.59]. The bandwi dt h of t he surface emi t t er is about 17 MHz, wher eas t hat of t he edge emi t t er is about 100MHz. (d) Temper at ur e dependence of t he power emi ssi on from edge- emi t t i ng AIGaAs LED, (e) Temper at ur e dependence of t he power emi ssi on from edge- emi ui ng l n Ga As P LED 1.4 1.2 1.0 0 . 8 o 0,6 o. 0.4 0.2 0 0 AJ.Go As )~ ( 2 0 C ) = 0 . 8 5 / z m 1 1 ~ I I I0 "C 2 0 = C 30 C 4 0 C ' x ~ 5 0 ~ 6 0 C ~ I I I f I 50 {00 150 200 d I I I 250 300 350 400 CURRENT (mA) Laser Di odes and LEDs for Fi ber Opt i cal Communi cat i on 39 1.4 i I I I i I I n G o A s P / I n P O o C ~ X (20 C) = 1.2~,~rn IOC\ Oc 1.2 - . 3 0 C~ / 0.4~ 0.2 0 50 I00 150 200 250 ~,00 350 400 CURRENT {mA) Fig. 2.25. re) Temper at ur e dependence of t he power emi ssi on from edge- emi t t i ng l nOa As P LED the radi at i on is minimized by the use of GaAs :Si recombi nat i on regions (where the emission energy is slightly less t han the energy bandgap of the absorbi ng regions), overall quant um efficiency values from all sulfaces of the di ode cat] reach 30% [2.51]. Al t hough power efficiency is i mport ant , it is not the key paramet er for LEDs for the present application. Here, the optical power coupled into the desired fiber for a given electrical power input is the quant i t y to be maximized. For example, GaAs or AIGaAs diodes with dome-shaped emitting surfaces can be very efficient (10 15 %) [2.52], but t hey are of little use with typical fibers because their large emi t t i ng area ( > 400 lain diameter) makes coupling to small di amet er fibers ( <85 gm) very inefficient. Het eroj unct i on structures are most commonl y used to optimize LEDs for coupling into fibers. We will discuss AI,.Ga[ _.~As/AI3.Ga I _ rAs (x <y) double- het eroj unct i on LEDs most extensively. Ot her useful III V het eroj unct i on structures involve similar design principles. As in the DH laser (Sect. 2.2), the injected carriers in the LED are confined to a narrow recombi nat i on region between the heterojunctions. Furt hermore, because the surroundi ng material is of higher bandgap, internal optical absorpt i on is reduced compared to t hat of a homoj unct i on structure. For fiber coupling, light can be extracted from either one surface or one edge. The two maj or confi gurat i ons are the etched-well surface emitter (or "Burrus LED") [2.53], and the st ri pe-cont act edge emitter [2.54, 55], which is similar to the stripe-contact laser diode. These structures are illustrated in Fig. 2.25a, b. 40 I-1. Kr e s s e l el al. In the Burrus st ruct ure, the cont act di amet er is typically smaller t han the fiber core di amet er, and an et ched well in t he subst rat e is used to bri ng t he fiber close to the emi t t i ng area. Ther e are also simpler surface-emitting st ruct ures made by diffusion of p-t ype dopant into an n-t ype layer to form a simple homoj unct i on emi t t er [-2.56]; ot her surface emitters have been made by growi ng a thick AlxGa ~_xAs l ayer to act as a hi gher bandgap nonabsor bi ng subst rat e [2.57]. In general, st ruct ures where the active region is close to the heat sink can be oper at ed at the highest cur r ent densi t y (to ~ 10,000 Ac m- 2) because t hei r t her mal resistance is minimized. Hence, such st ruct ures are preferred where the highest possible power level is needed for the fiber t ransmi ssi on system. 2.5.2 Optical Power and Fiber Coupling The t ot al optical power emi t t ed from the best surface [2.58] and edge emitters [2.59] is compar ed in Fig. 2.25. Not e in Fig. 2.25c t hat the power out put increases subl i nearl y with cur r ent at high cur r ent because of j unct i on heating. However, a compar i son of the r a d i a n c e of t he sources [-2.60] is mor e meal > ingful t han cornpari ng their t ot al powers. The radi ance (brightness) is t he opt i cal power emi t t ed from a unit area of the device into a unit solid angle and is expressed in units of W c m 2-sr 1 In Tabl e 2.1, we list the best r epor t ed LED radi ance values, quant um efficiencies, power out put s, and modul at i on bandwi dt hs. The fiber charact eri s- tics assumed in this compar i son (a numeri cal aper t ur e of 0.15 and an 801am core di amet er) are typical of fibers which have so far given the best overall per f or mance for l ong-di st ance (1-10 km) appl i cat i ons. The t emper at ur e dependence of the power emission from LEDs is relatively small compar ed to t hat of laser diodes. Fi gure 2.25d, e shows dat a for an edge- emi t t i ng AIGaAs LED and for an I nGaAs P/ I nP doubl e- het er oj unct i on edge- emi t t i ng LED emi t t i ng at 1.23 p.m (at 20 C). Between 20 and 70"C the power emission from the latter device dr ops (at const ant current ) by a fact or of about 2, whereas the change in the emission from t he AIGaAs LED is onl y a fact or of 1.7. The difference is believed to be due to the smaller het er oj unct i on barri er height in the I nGaAs P/ I nP di ode and resul t ant part i al carri er confi nement loss at elevated t emperat ures. 2.5.3 LED-Fiber Coupling and Transmission In this section, we discuss fact ors affecting the optical power coupl ed i nt o fibers. ( Fur t her discussion will be found in Chap. 6. ) The emi t t i ng area of a surface- emi t t i ng LED is det ermi ned by the cont act area, chosen in accnr dance with the core area of the fiber. The fiber tip is mount ed close enough to the LED to prevent beam spreadi ng between the emi t t i ng region and the fiber tip. The L a s e r D i o d e s a n d L E D s f o r F i b e r O p t i c a l C o l n m u n i c a t i o n 41 ~ ~ o . ~ . 2 o .= 0
= e'~ ;-q. C x z ' . ~ e'~ . g ==,= o r - ~ ~1 r - ~ r " n t ' - n ~ t " , , l I . . . ~ - , i t ~ " - 1 42 H. Kressel et al. beam pat t er n of a surface emi t t er is r oughl y Lambert i an, with an angul ar i nt cnsi t y dependence given by 1(0)2 1 o cos0. (Here 0 is the angle bet ween t he emission di rect i on and the normal to the emi t t i ng surface.) The radi ance of a surface-emi t t i ng di ode is det ermi ned by the thickness of the r ecombi nat i on region, the cur r ent density, the i nt ernal quant um efficiency, and i nt ernal absorpt i on. The combi nat i on of these factors makes the radi ance of the edge emi t t ers higher t han t hat of t he surface emitters. The highest r epor t ed r adi ance of a Bur r us- t ype LED is 200 W cm 2 sr This DH st ruct ure has a 2.5 lam het er oj unct i on spacing and a dopi ng level of 5 x 10 ~ v c m- 3 in the r ecombi nat i on region [2.61]. The highest radi ance values for AIGaAs edge emi t t ers arc in t he 1000 W c m - 2 s t - l range [2.34]. Mor eover , as discussed in Sect. 2.5.2, t he radi ance must be consi dered in the cont ext of the modul at i on capability, and t he bri ght est surface-emi t t i ng diodes are relatively slow 2. A doubl e- het er oj unct i on edge emi t t er can pr oduce a di rect i onal beam per pendi cul ar to the j unct i on pl ane which rivals t hat of the laser di ode ; tiffs is because the same t ransverse modal pr opagat i on characteristics, det er mi ned by the DH st ruct ure, cont r ol bot h the coher ent and i ncoherent emission [2.34, 64]. For Al GaAs LEDs, the full beamwi dt h at the half-intensity poi nt perpendi cu- lar to the j unct i on pl ane is a funct i on of the al umi num concent r at i on difference Ax at the doubl e- het er oj unct i on boundar i es d apar t [2.65], Ol ~ 20(Ax)d/2 [ r a d] . (2.15) For example, with 2=0. 81am, d=0. 1 ~tm, and Ax=0. 2, 0 , 2 2 9 . Fi gure 2.26a shows the beam pat t er n measured on such a structure. Edge emitters usually have the same stripe cont act s as lasers, but lacking the optical feedback of the laser, this does not result in a nar r ow beam in the pl ane of the j unct i on. A cos0 di st ri but i on of emi t t ed radi at i on is found in this plane. The combi nat i on of beam nar r owi ng in the per pendi cul ar plane and lack of nar r owi ng in the j unct i on pl ane causes the emi t t ed beam t o have a quasi- elliptical shape. Therefore, an anamor phi c optical system is desirable for opt i mal optical coupl i ng i nt o a (cylindrically symmetric) fiber. The effective emitting area of an edge-emi t t i ng LED cannot be conveni ent l y t ai l ored to mat ch the fiber core. Per pendi cul ar to the j unct i on plane, the light comes from a region appr oxi mat el y as thick as the r ecombi nat i on region, with some spreadi ng into adj acent layers as det ermi ned by the light guiding propert i es of the DH structure. This typically results in an optical source 1 to 2 ~tm thick. The stripe cont act confines the r ecombi nat i on region to a thin stripe across the di ode facet. The typical stripe widths of 10 to 50 ~tm are smaller t han the core di amet er of a mul t i mode fiber. For mu[t i mode fibers with core di amet ers of 60 to 1201am, a nar r ow (10~tm) stripe LED is al most a poi nt 2 Owi ng to t he rapi d t echnol ogi cal evol ut i on in t hi s field, definite device limits cannot be given. Dat a pr esent ed here are for general gui dance only. L a s e r Di o d e s a n d L E Ds f or F i b e r Op t i c a l C o mmu n i c a t i o n 43 [ . > . - I . . - Z I . . g I - . - Z ~ 0 . 5 I - ' , : : I - I I . t . I 0 9 0 EDGE >- 1.0 'EMITTING ~ 0 , 5 I - - _ . 1 n r " i i r I t ~ ' ) 4 5 O 4 5 c' 9 0 9 0 " BEAM ANGLE FROM NORMAL . _ _ 9 5 ~ . 0 , , ~ 120" 4 5 * 0 4 5 9 0 B E A M ANGLE FROM NORMAL Fi g. 2 . 2 6 a , b. F a r - f i c l d f r o m (a) a n e d g e - c mi l t i n g L E D h a v i n g a v e r y n a r r o w d o u b l e - h e t e r o j u n c t i o n s p a c i n g a n d (b) a s ur h~ce e mi t t e r source. By maki ng the cont act st ri pe wider, the emi t t i ng regi on can be made to ext end compl et el y across the fiber core, gi vi ng a "l i ne" source r at her t han a "poi nt " source. To keep t he di ode radi ance const ant , the current woul d have to i ncrease with i ncreasi ng width, t hereby i ncreasi ng the power demands on the dri ver electronics. Often, this is not accept abl e. For ma xi mum coupl i ng efficiency i nt o the fiber, the nar r owest st ri pe wi dt h shoul d be used consi st ent wi t h adequat e heat si nki ng at the desired current i nput or t ot al light out put . Onl y mul t i mode fibers can be consi dered pract i cal for LEDs. A mul t i mode fiber has di mensi ons and refract i ve indices t hat al l ow mor e t han one opt i cal wavegui de mode to pr opagat e. A useful a ppr oxi ma t e t r eat ment , appl i cabl e to the "weakl y gui di ng" fibers t hat are of most interest, has been given by GIoge [2.66]. It is useful to char act er i ze t he fiber by its numer i cal aper t ur e (N.A.) and a V par amet er , defined by V= 27ta(N.A.) 2 ' (2.16) here a is the core radi us and 2 is t he free-space wavel engt h. In t erms of this pararnet er, a mul t i mode fiber has r oughl y N, , = V2/ 2 pr opagat i ng modes. Thus, the numbe r of modes i ncreases with the core di amet er and with the "st r engt h" of the gui di ng as measur ed by t he fi ber' s numer i cal aper t ur e. For a fiber with N. A. =0. 15 and a core di amet er of 90 pro, V ~ 5 2 , and about 1340 modes can pr opagat e at a wavel engt h of 0.82 pm. For such hi gh mode number s and large core size rel at i ve to 2, r ay opt i cs give a valid pi ct ure of the pr opagat i on of light. In addi t i on t o these pr opagat i ng modes, energy can somet i mes be t r ansmi t - ted down a fiber in t wo ot her t ypes of modes : "l eaky" modes and "cl addi ng" modes. In an ideal, lossless fiber, these l at t er modes are not t rul y bound, and suffer st r ong at t enuat i on as t hey pr opagat e. Wi t h sui t abl e pr ovi si ons for mode st ri ppi ng, the cl addi ng modes can be at t enuat ed t o negligible levels in a very 44 H. K r e s s e l el al . shor t distance, and hence need not be considered. Tile l eaky modes, on the ot her hand, which cor r espond t o skew rays pr opagat i ng in a helical fashion down the fiber, under favorabl e ci rcumst ances can travel significant distances wi t hout appreci abl e loss. Because of them, it becomes somewhat ar bi t r ar y how "coupl i ng efficiency" is defined. Many experi ment al met hods to measure fiber coupl i ng cont ai n a significant cont r i but i on from such modes, especially if a relatively shor t length of fiber is used in the measur ement ; this is t he case despite at t empt s t o limit the measured t ransmi ssi on t o bound modes by limiting the aper t ur e of t he beam from t he end of t he fiber. Nevertheless, in many systems, the t ransmi ssi on distance is so short t hat energy from the source reaches the det ect or via leaky modes, and practically, these modes shoul d be t reat ed as t rul y pr opagat i ng ones. Therefore, one shoul d use care in using the t erm "coupl i ng efficiency" to make clear its oper at i onal significance. The simplest coupl i ng scheme is to place the end of the fiber against, or in pr oxi mi t y to the emi t t i ng LED surface. Colvin [2.67] calculated the coupl i ng bet ween surface emi t t ers with vari ous charact eri st i cs and step-index fibers. The cal cul at i ons are based on the assumpt i on t hat ray optics are valid, and indicate the i mpor t ance of skew and l eaky rays. For a small core-t o-source di amet er ratio, small f i ber - LED separat i ons, and low numeri cal apert ures, skew rays make a significant cont r i but i on t o coupl i ng efficiency. Consi deri ng the compl exi t y of t he probl em, it becomes clear t hat vari ous appr oxi mat e, simplified model s can be useful in est i mat i ng coupl ed power levels to be expected under vari ous condi t i ons, but t hat detailed measurement s will pr obabl y provi de the onl y really reliable values. One simple model for step- index fibers assumes t hat all light rays incident on t he exposed end of the core, and maki ng a (meridional) angle with the fiber axis less t han the critical angle for t ot al internal reflection, 0~, are pr opagat ed, while light rays with an angle 0 > 0c are rejected. Here, the critical angle is defined as 0c =Si l l - ' ~12~i't2 =s i n 1(N.A.). (2.17) For a source smaller t han the core and in cont act with tile fiber end, this gives a coupl i ng efficiency r/c oc I(0) sin 0 dO 0 q c = n,,2 I (0) si n 0 dO 0 (2.18) For a Lambert i an source, r / c=s i nZ0 =(N. A. ) 2. (2.19) For a fiber with N. A. =0. 15, qc=2. 2%, ( - 16.5dB). Laser Diodes and LEDs for Fiber Optical Communication 45 As ment i oned earlier, the beam pat t er n of edge-emi t t i ng LEDs and lasers call be qui t e di rect i onal , l eadi ng to l ower coupl i ng losses. Na r r ow- be a m LEDs, for exampl e, can give coupl i ng losses of about - 1 l dB, or mor e t han 5dB bet t er t han for Lamber t i an sources [2.34]. Ther e is a small addi t i onal Fresnel reflection loss at the ai r-fi ber interface. In a sur f ace- emi t t i ng LED, this can be el i mi nat ed by filling the space bet ween the di ode surface and the fiber tip with a mat er i al of sui t abl e index. However , t here can still be a significant (5 to 20%) reflection loss at t he LED i nt erface with this filler. To el i mi nat e bot h Fresnel i nt erface losses, the refract i ve index of the mat er i al mus t be high, ~2. 3, and the f i ber - LED spaci ng must be careful l y cont rol l ed. LEDs can be ant i refl ect i on coat ed with l ayers of 2/4 thick AI 20 3 or ot her sui t abl e mat eri al to reduce their Fresnel loss at an LED- a i r i nt erface to about 1%. Wi t h no mat eri al bet ween the LED and a fiber, tlle second (air-fiber) i nt erface loss is at mos t 4% for a fiber wi t h N. A. =0. 15. It shoul d be r emember ed t hat the coupl i ng efficiency given ill (2.18) is for a st ep-i ndex fiber, where the effective numer i cal aper t ur e is const ant over the face of the core. For a par abol i cal l y gr aded- i ndex fiber, a given (on-axis) numer i cal aper t ur e cor r esponds to onl y hal f as ma ny pr opagat i ng modes as in t he st ep- index case, and the coupl ed power is onl y about hal f as much. Losses in hi gher - or der modes are often significantly gr eat er t han t hose of l ower - or der ones. Thi s combi ned with mode coupl i ng effects at fiber i mperfec- t i ons and bends, call lead to a changi ng modal energy di st ri but i on wi t h pr opagat i on pat h length, and hence, to what can be i nt erpret ed as a changi ng effective numer i cal aper t ur e. Thus, meas ur ement s of coupl ed power made on short lengths of fiber can give mi sl eadi ng results. As t he accept ance angl e of t he fiber is increased, the coupl i ng efficiency of a surface emi t t er increases relative to t hat of an edge emi t t er. In Fig. 2.27a, we c ompa r e the cal cul at ed coupl i ng efficiency as a funct i on of numer i cal aper t ur e for a t ypi cal sur f ace- emi t t i ng and edge-el ni t t i ng LED and a t ypi cal cw i nj ect i on laser. It is difficult to exper i ment al l y c ompa r e the var i ous sources avai l abl e, but in Fig. 2.27b, we c ompa r e the cal cul at ed and exper i ment al power coupl ed i nt o a fiber as a funct i on of the numeri cal aper t ur e. For these cal cul at i ons, we use the best r epor t ed power out put s for t he surface and edge emi t t ers [2.58, 59]. For Fig. 2.27a, b we assume t hat the sour ce size is smal l er t han the (step-index) fiber di amet er. (As not ed earlier, gr aded- i ndex fibers of the same N. A. will accept about hal f the power. ) Also shown in Fig. 2.27b are dat a poi nt s for coupl ed power f r om the literature. It is seen t hat the edge emi t t er is advant ageous for small numer i cal aper t ur e fibers ( N. A. <0. 4) , whereas the surface emi t t er, because of its hi gher power out put , is mor e useful for l arge N. A. fibers. 2. 5. 4 Carri er Li f e t i me and Di o d e Bandwi dt h The emi ssi on f r om LEDs is modul at ed by cur r ent vari at i ons. For low cur r ent levels, the modul at i on capabi l i t y may be l i mi t ed by the j unct i on capaci t ance 4 6 21 14. Kressel ct al. E 15 o D 0_ 9 o o m 6 b. Y ~ . / SURFACE i n "BURRUS" LED a Q - - I I I I I I 0 OI 0 2 0. 3 0, 4 0. 5 0. 6 FIBER NUMERICAL APERTURE 0.01 0 0 2 0 . 0 5 OI 0. 2 0 . 5 I . O 0 7 E 2 E o I O z 0 , 5 0 2 ~ILATED I ~ i l I CALC EXPERIMENTAL SURFACE ~ ' ~ . . . . EMI TTER- o / t EDGE / / / - - - E M I o x I / I / / / o / b O l , i ' ' i I 0 OI OP- 0. 3 0. 4 0 5 0 6 0. 7 FIBER NUMERICAL APERTURE Fig. 2.27. (a) Coupling efficicncies from a typical laser diode and surface and doublc-hetcrojunction edge-emilting LEDs inlo step-index fibers of varying numerical aperture. The range of ex- pected values is indicated by the shaded regions. (b} Calculated coupled power as a funclion of fi ber N.A. for the best reported edge and surface emitters as well as experimentally determined values from the literature [2.58, 59] I- 2. 68] . Ho w e v e r , in hi gh f o r wa r d - b i a s o p e r a t i o n t h e l i g ht m o d u l a t i o n i s l i mi t e d by t he l i f e t i me o f t he c a r r i e r s i nj e c t e d i n t o t he r e c o mb i n a t i o n r e g i o n . I f t he c u r r e n t i s mo d u l a t e d at f r e q u e n c y 09, t he l i g ht o u t p u t Po(m) wi l l v a r y wi t h f r e q u e n c y [ 2 . 1 , 6 9 ] . Po(O) P ( ( I ~ ) = [ 1 + ( ( o r ) z ] 1/ 2' ( 2 . 2 0 ) wh e r e P0(0) i s t h e l i g ht i n t e n s i t y w i t h o u t mo d u l a t i o n a n d z i s t he mi n o r i t y c a r r i e r l i f e t i me . F i g u r e 2. 28 s h o w s t he r a t i o Po(oJ )/Po(O) f or t hr e e d i o d e s h a v i n g t h e z v a l u e s i n d i c a t e d . Laser Diodes and LEDs for Fiber Optical Communication 47 I . O . I . ~ =:p 0 Q . I L l n - W N ._1 a: 0 3 0 ' 6 . 4 n s I I I I I i I I I I i i i i i i i I0 I00 I000 M O D U L A T I O N F R E Q U E N C Y , M H z F i g . 2 . 2 8 . Normalized re- sponse as a function of fre- quency for lhree LEDs having the minority carrier lifetime values r indicated. The solid curves show the [ 1 +{~or) 2] ~z dependence [2.34] The di ode' s mo d u l a t i o n l i mi t is us ual l y defi ned in t er ms o f t he e l e c t r i c a l p o w e r a t t h e d e t e c t o r . Thi s is p r o p o r t i o n a l t o t he s q u a r e of t he opt i cal i nt ensi t y. The b a n d wi d t h capabi l i t y of t he di ode, defi ned as t he f r equency wher e p2(o))__ L 2 - - 2 P o ( 0 ) , is f r om (2.20), 1 .1; - 2rcr' (2.21) Cons i de r i ng onl y b a n d - t o - b a n d (i.e., r adi at i ve) r e c ombi na t i on, t he car r i er lifetime in a di ode vari es wi t h cur r ent J [ 2. 10] a c c or di ng t o e d r = ~ j (P0 + no) {El + 4 J / e B r d ( p o + no) z] 1 / 2 - - 1 } , (2.22) wher e Po and n o ar e t he el ect r on and hol e c onc e nt r a t i ons wi t hout i nj ect i on, d is t he wi dt h of t he r e c o mb i n a t i o n r egi on, e is t he el ect r on char ge, and B r is t he r e c ombi na t i on coeffi ci ent , whi ch var i es wi t h t he s e mi c onduc t or a nd t e mpe r a - t ure. I n Ga As , Br - ~10 ~cm3 s -~ at r o o m t e mpe r a t ur e [2. 70]. Cons i de r a p- t ype r e c ombi na t i on r egi on wher e Po >> no. For a l ow i nj ect ed car r i er c onc e nt r a t i on r el at i ve t o t he b a c k g r o u n d c onc e nt r a t i on, t he car r i er lifetime is i nde pe nde nt of t he c ur r e nt densi t y, r - ~ ( B r P o ) - 1 . Hence,./~ = ( B r p o / 2 ~ ) . At hi gh i nj ect i on levels, however , t he lifetimes becomes i ndependent of t he r e c ombi na t i on r egi on dopi ng, r ~ ( e d / B r d ) x / z a nd (2.23) I ( B r J ] 1/2 .l~ = 2 ~ \ e d ] (2.24) 48 H. Kr e s s e l et al. Thus, t here are t wo ways of reduci ng the carri er lifetime in or der to increase .['~ : I) use high dopi ng in the r ecombi nat i on region, or II) oper at e at high cur r ent densities and small d in or der t o obt ai n a high i l ! j e c t e d carri er concent rat i on. If heavy dopi ng is empl oyed, nonr adi at i ve centers are i nt r oduced when the dopi ng is increased to > 101~ c m - 3 [2.71]. For example, tile radi at i ve lifetime in p-t ype material with 1019 holes cm -3 is about 10 . 9 S, but the nonr adi at i ve lifetime is also short ened to about t he same magni t ude (or less). This significantly reduces the i nt ernal quant um efficiency since r/i--(1 +rr/z,,r) -1. Also, the internal absor pt i on of the radi at i on is increased with heavy dopi ng, furt her cont r i but i ng to a reduct i on in the ext ernal di ode efficiency. Experi ment al evidence of these effects is shown, for vari ous p-t ype (Ge) dopi ng levels in a doubl e- het er oj unct i on Burrus-t ype AIGaAs LED, in Figs. 2.29, 30. Figure 2.29 shows how the carri er lifetime decreases and the bandwi dt h increases with increasing dopi ng level [2.61]. Fi gure 2.30 shows the accom- panyi ng decrease in radiance. It is clear t hat the choi ce of an LED shoul d be guided by the requi rement s of the system, because the use of high-speed di odes general l y entails a sacrifice in available opt i cal power. The LED response is not the onl y fact or t hat can limit system bandwi dt h. The dispersion in the fiber, due to the modal dispersion present in mul t i mode fibers and t he spectral dispersion present in bot h single and mul t i mode fibers, may become limiting. Modal dispersion is a pr oper t y of the fiber only, and is not affected by the source. In sufficiently l ong fibers, all sources will event ual l y, t hr ough scattering, fill all the modes of t he fiber. However, with pr oper fiber index profiling, the modal dispersion can be limited t o less t han 0.5 ns k m- I-2.72]. Spectral dispersion on the ot her hand, depends on the wavelength and spectral width of the source [2.49]. Figure 2.31 shows the pulse br oadeni ng per ki l omet er of fiber as a function of wavelength. This plot points to the usefulness of LED sources emitting at 1.2 to 1.3 lain where the spectral dispersion appr oaches zero. It also shows why lasers are useful for very high bandwi dt h systems ; the < 20 A typical spectral width of injection lasers pr oduces onl y a few pi coseconds per ki l omet er of spectral dispersion. Unf or t unat el y, the usefulness of sources at 1.2 to 1.3 gm is limited by the available aval anche and p-i-n det ect ors in this spectral region. However, at 2 < 1.06 lam Si det ect ors can be used. At 0.8 to 0.9 gin, where bot h sources and Si det ect ors are highly devel oped, the spectral dispersion is significant, as shown experi ment al l y for di odes of varyi ng spectral width in Fig. 2.32. The bandwi dt h of the I km- l ong fiber fixed by modal dispersion is shown as a dashed line. The modul at i on capabilities of the LEDs used were well in excess of this modal dispersion limit. Therefore, the bandwi dt h limitations of the LED-fi ber system illustrated are fixed by t he varyi ng spectral widths of the LEDs and the spectral dispersion effects in the fiber. The light pulse br oadeni ng At, assumi ng a Gaussi an pulse shape, is given by ( A l ) 2 = ( A l ) ~ < , r . ~ , ~ 2 + (At)m,,,i,,I disp. (2.25) i o 3 N " r i o 2 CI 2: oB Laser Diodes and LEDs for Fiber Optical Communication \ " ~ x x T DH LED o o o o o o I 0 I I I I I I I I 1 I I , , , , , J l L I 0 i 7 1018 1019 Fig. 2 . 2 9 . D O P I N G D E N S I T Y ( G e ) , c m - 3 30 IO n E F, LG l - - IJJ h . . J n" UJ n~ laJ > I - i i i I . i J 1 0 2 0 4 9 4 0 0 0 . 2 C 0.15 g : o . l o : 7 0 m 0 . 0 5 hl 0- 0 _ J -0.05 -O,I 0.6 \ 2 0 E I 0 n~ W g sl < m I - - ( -CX._ ) d 2 n d k 2 I IO D IAM ETER =50/.tm A ~, I I I I I I I I I I i I 20 50 I00 200 BANDWIDTH ( M H z ) \ Z e r o M a t e r i a l D i s p e r s i o n at 1.27Fm I I I I I I I I I O.B 1 , 0 1 . 2 1 . 4 1 . 8 W A V E L E N G T H ( p . m ) - 2 0 0 _ ~a E u i - - L o o .~ B a - ~ 7 ~ 5 0 < n ~ 2O I 40O Fig. 2.30. Fig. 2.29. LED bandwidth and el z fective carrier lifetime z as a fanction of the doping density in the recom- bination region ofdouble-heterojunc- lion, Burrus-typc structures [2.61] Fig. 2.311. Optical power and radi- ance of Burrus-type double-hete,' o- junction LEDs its a function of the bandwidth [-2.61] Fig. 2.31. Calculated and experimen- tal material dispersion of silica-rich fiber core [2.49] . ~ ~ ,,.,? ~ ~ . , ~ ~, ~ ~ I = / o / / / ! ~ J J ~ d J . , 1 - +o o .~ / / ~! i / &$ / ; / / I ~ + / o ~ / / L L ~ _ . L i r I I 3 S N O~ S 3 ~ 3 A LLV"T~ ~ <3 0 J o , U ,.-t {3 t b , 4 # - W u4 ,-4 "6 o C~ ! Laser Di odes a nd LEDs for Fi ber Opt i cal Co mmu n i c a t i o n 51 I00 > 8o E 60 40 20 i r r i i i i I i l r l l r l l I i I i i i i i i I r i i i i i i pl_ 500 K Go As~Ge / ? _ A / GaAs :Zn 0 I I I i l l t l l i i i , , l l , l i I t r l l l l I I I I I I I I 1017 1018 1019 HOLE CONCENTRATI ON (crn - 3 } Fig. 2.33. Ahv ,'it t hc hal f - i nt ens i t y poi nt of t he phot ol umi ne s c e nc e at 300 K of Ga As : Ge Da t a for Ga As : Zn ar e s hown for c o mp a r i s o n [2.7 l ] The resul t ant bandwi dt h is ,,~(2At)-1. Tabl e 2.2 shows the good agreement between the observed and cal cul at ed frequency response for the LED- f i ber system of Fig. 2.32. To ext end the i nf or mat i on bandwi dt h with AIGaAs LEDs, nar r ow spectral widths are desirable. Thi s means low dopi ng densities in the active region because the spectral width of the source depends on the dopi ng densi t y in t he active region. Fi gure 2.33 shows the spectral full width at half maxi mum as a function of hole concent r at i on (for phot ol umi nescence emission) from GaAs doped with Zn and Ge [-2.71]. The emission from edge and surface LEDs generally follows these dat a, with some al t erat i on due to i nt ernal absor pt i on effects. We are t herefore faced with a compr omi s e: light dopi ng leads t o high efficiency and small spectral dispersion but also low modul at i on capabi l i t y if t he injected carri er densi t y is low, whereas heavy dopi ng leads to low efficiency and large spectral di spersi on but high modul at i on capability. The r educt i on of the carri er lifetime by the use of nar r owl y spaced doubl e- het er oj unct i ons offers a useful compr omi se in design. In this way, we can pr oduce large bandwi dt hs wi t hout i mpai ri ng the quant um efficiency. The limitations here are t wofol d : I) it is essential t hat the interracial r ecombi nat i on at t he het eroj unct i ons due to defects be very low. This is evi dent from (2.2) where we see t hat a reduct i on in d requi res t hat the interfacial r ecombi nat i on velocity S also be decreased in or der t o mai nt ai n the internal quant um efficiency const ant ; II) the modul at i on dept h of the di ode out put will be somewhat limited because of the varyi ng carri er lifetimes with cur r ent level. 52 11. Kressel ct al. o 0 . u~ 0 . 4 0 F, w 0 . 2 > F- < .J i . O ~ i , i { i I i I i I o 4 1 - o-o-8.o , + . A T : 4 0 0 m , ~ . o ~ - o . " ~ -~ ~ ' t ~ - [ - " F : ~ ' 0 0 m A O O :~ O ~ * ~ '+ % 0 I = l O 0 mA 0%8 +~ , e8 o~ ~, 0 . } I I I I I I I I I I I 0 2 0 5 0 1 0 0 2 0 0 MODULATtON F R E Q U E N C Y , M H z % Fig. 2.34. Nor mal i zed r e s pons e of a doubl e - he t e r oj unc t i on di ode bi ased at i ncr eas i ng c ur r e nt s in or der to i ncr ease t he i nj ect ed car r i er dens i t y in t he r ecombi l mt i on regi on [2. 34] Experi ment al dat a showi ng t he expect ed effect on the frequency response of changi ng the diode' s bias current level and superi mposi ng a small modul at i ng cur r ent are present ed in Fig. 2.34. The di ode studied was a DH A1GaAs st ruct ure with a very small value of d = 0.05 pro. Indeed, the frequency response increases as p, z , in accor dance with (2.24) for devices operat i ng in the high injection bi mol ecul ar r ecombi nat i on region. 2 . 6 D i o d e R e l i a b i l i t y Laser life may be limited by ei t her facet damage or internal "gr adual " degradat i on. The first depends on the opt i cal flux densi t y (and the pulse length) and consists of damage to the mi r r or facet ; the second is mai nl y a funct i on of the cur r ent density (i.e., el ect r on- hol e r ecombi nat i on rate), the dut y cycle, and the details of the laser fabri cat i on process. The degr adat i on phenomena of LEDs are the "gr adual " type, because optical power densities are relatively low. In any discussion of laser di ode reliability, it is i mpor t ant t o r emember t hat laser degr adat i on easily occurs as a result of t ransi ent pulses of the ki nd often found on t urni ng on or off di rect -current power supplies. These pulses, which may reach ten times t hreshol d, are not usually det ect ed in rout i ne use, but may dest r oy a l ow-power laser di ode designed for cw operat i on. 2.6.1 Facet Damage Compl et e or part i al laser failure may occur as a result of mechani cal damage of the facet in a region of intense opt i cal flux. The damage t hreshol d is reduced if flaws, which pr emat ur el y initiate damage, exist at t he laser facet. aY 2.0 o 1.0 o as O.I O Laser Diodes and LEDs for Fiber Optical Communication 53 5.0 A ~ ~u Fig. 2.35. Emined power level (per unit junction widlh in the junction plane) at which facet FH damage occurs as a function of the pulse width. Data shown for , , , , I , , . . . . I , , , . . . . . single-heterojunction lasers (SH) I00 I000 I0000 and four-heterojunction lasers PULSE WIDTH I ns) (FH)[2. 1] For a given device, the critical damage level Pc (in W cm t of emi t t i ng facet) decreases with increasing pulse length T as T - 1/ 2 [2.1]. This is shown in Fig. 2.35 for vari ous het er oj unct i on lasers. The t hreshol d level for damage is increased by deposi t i ng antireflecting fihns [2.733 such as SiO or AI20 3 on the laser facet to l ower the rat i o between the opt i cal flux densi t y inside and out si de the crystal. However, this l owered reflectivity also increases J,h [see (2.8)]. It has been suggested [2.74] t hat facet damage occurs at a const ant opt i cal electric field for a given pulse length and t hat the rat i o of critical power levels with and wi t hout films, P' J Pc , for a facet reflectivity R can be expressed as P: (1 - R) p~ = n (1 + R1/ 2) 2 ' (2.26) where n is the GaAs refractive i ndex ( ~ 3.6) and Pc is t he measured val ue for the GaAs- ai r interface. The avai l abl e dat a [2.73] are consi st ent with (2.26). Lasers oper at i ng cw also can suffer f r om facet damage if the opt i cal flux densi t y is excessive and the facets are uncoat ed. For lasers with stripe widths of 13 and 50 lain the values found were in t he range of 2 to 4 mW per lam of stripe wi dt h or 2 to 4 x 105 W cm - 2 [2.75]. (The above laser diodes were DH lasers, d~0. 2 to 0.31.ma, with Alo. lGao. gAs in the r ecombi nat i on region and Alo.3Gao.vAs in the adj oi ni ng n- and p-t ype regions.) The facet damage is generally initiated in the cent ral por t i on of t he active region under t he stripe cont act where the optical flux densi t y is highest. The discussion so far concer ned facet damage which occurs in a short peri od of time. A milder form of facet damage ("erosi on") can appear over a long operat i ng time for di odes oper at ed bel ow P,,. This effect is accel erat ed by moi st ure on t he facet and may result f r om an oxi dat i on process opt i cal l y accel erat ed in the course of laser oper at i on [2.76]. The use of half-wave thick dielectric facet coatings (not abl y A1203) eliminates facet erosi on [2.77] wi t hout increasing the t hreshol d current . 54 H. Kressel el al . 2.6.2 Gradual Degradation The gradual degr adat i on observed in laser di odes wi t hout evi dence of facet damage is usually charact eri zed by a reduct i on in t he differential quant um efficiency and an increase in J,h ; in LEDs, the efficiency decreases with time. Errat i c reliability was charact eri st i c of earl y diodes, suggesting imperfect cont r ol of the device met al l urgy, but long-lived devices have become a reality as the t echnol ogy has i mproved. Gr adual degr adat i on results from defects i nt r oduced i nt o the recom- bi nat i on region which serve as nonr adi at i ve r ecombi nat i on centers and reduce the internal quant um efficiency. However, for a given defect density, the LED efficiency may be more affected t han the lasing propert i es because the carri er dynami cs are different. A fact or of t wo reduct i on in the LED spont aneous emission efficiency typically cor r esponds to onl y a 20% increase in the t hreshol d cur r ent densi t y [-2.78]. Many i nvest i gat ors have studied laser and LED degr adat i on phenomena and the following summari zes common observat i ons [2. 1]: I) With degradat i on, the t hreshol d cur r ent densi t y increases while the differential quant um efficiency frequent l y (but not always) decreases; I1) Fl et eroj unct i on di odes similarly oper at ed degrade whet her lasing or not (as long as facet damage does not occur). The degr adat i on rat e (measured in LEDs) increases with the oper at i ng cur r ent density J as J"', where m = 1.5 to 2 [_2.77, 80] ; III) Ther e are successive degr adat i on stages. In the early stages, when the densi t y of newly formed nonradi at i ve r ecombi nat i on centers is low, the spontaneous efficiency decreases with onl y mi nor effects on the lasin9 pr oper - ties. As degr adat i on proceeds, "dar k lines" may appear when t he emission is viewed t hr ough the surface of the di ode [2.81]. The "dar k lines" are regions of concent r at ed nonr adi at i ve cent ers which are believed responsible for the spot t y near-field emission pat t ern of degr aded lasers and the rapid fall-off in the out put of cw lasers at r oom t emperat ure. The concent r at ed nonr adi at i ve centers can have several origins, including i mpuri t y preci pi t at i on at existing dislocations. In some st ruct ures exami ned by t ransmi ssi on el ect ron mi croscopy, t he dar k regions were di sl ocat i on net works which grew duri ng laser oper at i on [-2.82]. Ther e is overwhel mi ng evidence t hat nonradi at i ve r ecombi nat i on centers are i nt r oduced into the active region of degr aded devices dur i ng operat i on. Vacancies and/ or interstitials appar ent l y form or diffuse i nt o the recom- bi nat i on region, as suggested by the observed gr owt h of di sl ocat i on net works by a "cl i mb" process t hat requires vacancy or interstitial at om diffusion. The origin of these poi nt defects is still uncertain. Gold and Weisberg [2.83], in t hei r GaAs tunnel di ode degr adat i on studies, suggested t hat nonr adi at i ve el ect r on- hol e r ecombi nat i on at an i mpuri t y cent er on a lattice site coul d result in its di spl acement into an interstitial posi t i on, leaving a vacancy behind. This is the basic "phonon- ki ck" model where mul t i - phonon emission gives an Las er Di odes and LEDs for Fi ber Opt i cal Co mmu n i c a t i o n 55 intense vi brat i on of the r ecombi nat i on center. Appl yi ng this mechani sm to el ect rol umi nescence, it is assumed t hat the vacancy and interstitial at oms formed have a large cross section for nonr adi at i ve recombi nat i on. Repeat ed "ki cks" woul d gradual l y move t hem to i nt ernal sinks. Experi ment al dat a suppor t the hypot hesi s t hat the energy released in nonr adi at i ve el ect r on- hol e r ecombi nat i on can enhance poi nt defect displace- ment. For example, Lan 9 and Kimmerlin 9 [2.84] showed t hat the act i vat i on energy for the di spl acement of lattice defects i nt r oduced into GaAs di odes by i rradi at i on with 1 MeV el ect rons was subst ant i al l y reduced under forward bias. Experi ment al di ode studies where defects were del i berat el y i nt r oduced show t hat certain crystal defects initially present in the di ode can accel erat e the degr adat i on rate. These defects include a high di sl ocat i on densi t y [2.85] and cont ami nant s such as copper [2.86]. Therefore, it is i mpor t ant to use het eroj unc- tion st ruct ures with t he lowest possible misfit di sl ocat i on densities and to minimize cont ami nant s. In addi t i on, many process-i nduced defects can accel- erat e the degr adat i on rate. For exampl e, an excessively high zinc concent r at i on in t he r ecombi nat i on region is det ri ment al [2.87], as is device strain pr oduced by mount i ng the device on heat sinks with "har d" solders [2.88]. Therefore, a soft, ductile sol der such as i ndi um is commonl y used for di ode assembly. Exposed edges, where el ect r on- hol e r ecombi nat i on occurs, can great l y accelerate degr adat i on, as demonst r at ed by t he per f or mance of compar abl e oxi de-st ri pe and br oad- ar ea AIGaAs cw laser di odes cut from the same wafer [2.89]. Br oad- ar ea light-emitting diodes which have sides formed by sawing (and thus, were highly damaged) degr aded in a few hours, whereas the stripe- cont act di odes were qui t e stable for t housands of hours [2.79]. Ther e are i mpor t ant differences in the degr adat i on rates of di odes made using different alloys. Di odes where the r ecombi nat i on region consists of AI~Ga~ _~As (with x~0. 05 to 0.1) are less susceptible to gradual degr adat i on t han compar abl e diodes with GaAs in the r ecombi nat i on region [2.78]. As a result, all widely used laser di odes and LEDs have 5 to 10at . %A1 in the r ecombi nat i on region. For mat eri al s with l ower bandgap energies t han GaAs, it has been found t hat the degr adat i on rat e of l nGaAs homoj unct i on di odes decreases with increasing In cont ent [2.90]. 2.6.3 Accelerated Aging Tests Accelerated aging tests are commonl y used to predict the mean time to failure of semi conduct or component s. Insofar as the degr adat i on process is t emper a- t ure dependent , it is possible to design tests where the device is oper at ed at elevated t emper at ur es and the measur ed lifetimes are proj ect ed back t o r oom t emperat ure. The t hermal accel erat i on fact or must be experi ment al l y established. For the design of reliable aging tests for laser di odes and LEDs, the possible failure modes must be carefully consi dered. An increase in the cw laser di ode 56 H. Kressel et al. t hr eshol d cur r ent may result f r om facet damage, i nt ernal f or mat i on of non- r adi at i ve centers due t o lattice defects, or an i ncrease in the electrical or t her mal resistance. As not ed earlier, facet damage or erosi on, which are rel at ed to the opt i cal flux density, are cont r ol l ed by the use of facet coat i ngs and by oper at i ng lasers at moder at e power emi ssi on levels. The i nt ernal da ma ge mechani s m (gradual degr adat i on) is caused by nonr adi at i ve el ect r on- hol e r ecombi nat i on t hat can accel erat e defect mot i on. The rat e of this di spl acement process is expect ed to i ncrease with t emper at ur e. Because gr adual degr adat i on is a process t hat depends on the el ect r on- hol e r ecombi nat i on rate, it occurs si mi l arl y in bot h t he l asi ng and the i ncoher ent emi ssi on st at e of di odes at compar abl e cur r ent densities. Therefore, devices shoul d exhi bi t an i ncreasi ng gr adual degr adat i on rat e wi t h t emper at ur e, and such tests can be conduct ed bot h in the lasing and the i ncoher ent emi ssi on mode. One has the choi ce of moni t or i ng t he change in the t hr eshol d cur r ent with t i me at the el evat ed test temperature or aft er r et ur ni ng the laser to r oom t emper at ur e. Similarly, for an LED, the out put can be measur ed at the test t emper at ur es or at r oom t emper at ur e. Because the LED out put is rel at i vel y insensitive t o t emper at ur e c ompa r e d to the laser t hreshol d current , t he meas ur ement t emper at ur e is less i mpor t ant for the LED t han the laser. Aging meas ur ement s at el evat ed t emper at ur es have been per f or med under varyi ng condi t i ons. It is found t hat the degr adat i on rat e i ncreases wi t h t emper at ur e fol l owi ng an expressi on of the f or m exp( E/ kT) where E is an "act i vat i on energy". Fi gur e 2.36 shows the t i me requi red for a 20% pul sed t hr eshol d cur r ent i ncrease ( measur ed at r oom t emper at ur e) as a funct i on of the heat sink t emper at ur e for a gr oup of oxi de-defi ned, AI 20 3 facet -coat ed, AI GaAs laser di odes ; here E~0 . 9 6 +_0.2 eV [2.9]. Fi gure 2.37 shows the t i me requi red for a 50% reduct i on in the s pont aneous out put f r om the same device, where E = 1.1 _+0.2 eV. The aver age t i me requi red for a 20% t hreshol d cur r ent change i ncreases f r om about 500 h when oper at i ng at 100 '~C, to about 5000 h at 70 "C. At 22 '~C, the ext r apol at ed t i me is a bout 106 h. The t i me needed for a 50 % s pont aneous emi ssi on reduct i on is of the same or der of magni t ude. A laser' s cw t hr eshol d current may increase far mor e t han its pulsed t hreshol d cur r ent if the electrical or t her mal resi st ances increase. Exper i ment al dat a al so have been r epor t ed [-2.92] where AI Ga As / Ga As DH laser di odes were oper at ed cw at var yi ng t emper at ur es until t hey no l onger funct i on (despite current increases). The t i me to end of life defined this way decreases exponen- tially with t emper at ur e [ocexp( E/ kT) ] with E-~0. 7 eV, It has been found t hat laser failures follow a Gaus s i an or nor mal distri- but i on in a l ogar i t hmi c fashion. Thi s behavi or is si mi l ar to the failure mode for ot her s emi conduct or devices. In Fig. 2.38, we pl ot the failure rat e at 70'~C as a funct i on of t i me for oxi de-defi ned, st r i pe- cont act AI,.Ga~ xAS DH lasers [-2.93]. The failure here is defined as the poi nt when the laser ceases to oper at e cw at 70"C. The mean t i me to failure is 4200 h, in good agr eement with the previ ousl y descri bed tests of Fig. 2.36 and in agr eement wi t h the best dat a for AI GaAs pr ot on- bomba r de d st r i pe- cont act lasers [2.94]. iO 7:92 - i o 6 . w m \ < ~A r r 0 ~ 1 0 5 0 o I 4 w I ~ 3 2 Io 3.4 Laser Diodes and I.EDs for Fiber Optical Communication 57 A M B I E N T T E M P E R A T U R E ( C ) 3 0 4 0 5 0 6 0 7 0 8 0 9 0 I 0 0 I I I I I I I I \ \ \ \ I I 3 . 3 3 . 2 TLASEROEex p (E/kT) E = 0 , 9 5 + 0.2eV \ \ ' x \ % I I I I I &l 3. 0 2, 9 2. 8 2. 7 ( I O 0 0 / T ) K - I 2.6 Fig. 2.36. Time needed for a 20",, increase in the threshold current. nleasured at 22 C, as a function of the hcat sink (ambicnl} tem- peraturc. The dashed line is an cxtrapolation [2.91] A M B I E N T T E M P E R A T U R E ( * C ) "~ 107 22 30 4 0 1-- O_ \ 0 106 " x \ Z \ i % ,1 \ 0 n-" r,, 1 0 5 o 0 2 ,04 U J I I..- '~ i O 3 ...1 102 I I 3, 4 3 . 3 3 . 2 5 0 6 0 7 0 8 0 9 0 I 0 0 I I I I I I "ELED O[ exp ( EIkT ) E:l.l+-O.2eV \ \ \ \ \ \ J . I I I I I 3,0 2.9 2.B 2.7 2.6 ( I O0 0 / T ) K - I Fig. 2.37. Time nccdcd to ob- serve a 50% reduction in the spomaneous emission from AIGaAs stripe-contact diodes as a function of the heat sink (ambient) temperature. The dashed line is an extrapolalion [2.91] 58 II. Kressel et aI . 104 r~ @ 0 I-- z O 3 .) I k, IO 2 E ND OF L I I A T 7 0 C 3 dB DROP I N L I GH T OU T P U T AT C ON S T A N T C U R R E N T X T m = 4 2 0 0 , eft" = I. 3 O T m" 2 0 0 0 , O' ' 1 . 4 5 2 I 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 F A I L U R E S ( ~/ o ) , o ~ - /) -i- o @ M O I - r-, O n , IO 5 X bJ z I - ,< n~ laJ O .) - 1 i o " 9 8 F i g . 2 . 3 8 . C u mu l a t i v e r a t e o f f a i l u r e a s a f u n c t i o n o f o p e r a t i n g t i me o f l a s e r s a t 7 0 ' C . A I G a A s o x i d e - d e f i n e d , s t r i p e - c o n t a c t d i o d e s c o n s i s t i n g o f a g r o u p o f 4 0 d e v i c e s [ 2 . 9 3 ] Room t emper at ur e oper at i ng dat a ext end t o about 40, 000h for oxide- defined stripe A1GaAs LEDs and lasers. These tests at constant input current show little or no change for the LEDs and less t han a fact or of t wo dr op in out put for the lasers. The difference in out put of these t wo devices is simply due to the great er sensitivity of the laser to changes in t he t hermal resistance and ot her par amet er s because the laser out put is so sensitive to small changes in J,h when the cur r ent is kept const ant . 2. 6. 4 Cha ng e s in Var i ous Las e r Pa r a me t e r s wi t h Agi ng The stability of the power out put from a laser is not the onl y r equi r ement for utility in communi cat i on systems. We also requi re i nf or mat i on concerni ng some ot her i mpor t ant factors : I) Is the out put linearity const ant ? II) Are t here lateral mode changes which coul d reduce the coupl i ng efficiency i nt o fibers? Ill) Does the spectral width change with time'? IV) Are t here oscillations i nduced in the optical emi ssi on? Definitive answers concerni ng the above quest i ons cannot as yet be given because the aging propert i es depend on the t echnol ogy used to pr oduce the devices. However, some i nf or mat i on is available which reveals pot ent i al probl em areas. La s e r Di o d e s a n d L E Ds f or F i b e r Op t i c a l C o ml n u n i c a t i o n 59 Consi der first the quest i ons of laser linearity, lateral mode stability and spectral stability. Available dat a on oxi de-defi ned lasers obt ai ned at RCA Labor at or i es suggest t hat subst ant i al power degr adat i on (due t o an ~20% t hreshol d increase) frequent l y also pr oduces lateral mode changes at a given cur r ent above t hreshol d. Since lateral mode shifts with cur r ent may also pr oduce kinks in the power emission vs cur r ent curves (see Sect. 2.4.4), the linearity of these degraded lasers may change t owards t he end of life. However, the coupl i ng efficiency into mul t i mode fibers need not be subst ant i al l y changed because the angul ar di vergence of the lateral beam undergoes onl y modest change. Also, the spectral width may increases with time. It is not uncommon to find in subst ant i al l y degraded lasers t hat the nul nber of excited l ongi t udi nal modes increases. With regard to laser oscillations, Paoli I-2.95] r epor t ed on pr ot on- bombar ded stripe lasers. He found t hat some devices initially free from oscillations devel oped oscillations even t hough onl y mi nor changes in thresh- old cur r ent or differential quant um efficiency occurred. These self-pulsations (at frequencies between 300 and 600 MHz) occur r ed duri ng cont i nuous operat i on. It was also not ed in t hat st udy t hat the power emi t t ed from the t wo facets changes asymmet ri cal l y with aging. This effect coul d be very t r oubl esome in systems with "opt i cal feedback" where the light emi t t ed from the back of the laser is used t o mai nt ai n the dc bias at t hreshol d. Since onl y some lasers exhibit this effect, it is not an i nherent laser pr oper t y but r at her a defect -i nduced phenomenon which will be event ual l y identified and eliminated. List of Symbols ~ X f c ~ d ~ o u I X j //, F An 11 I ] c x t q~ tip 0j_, 011 2,21, T Tr " ~ n r T 1 a ~ c r " C L E D b B~ d Fr e e c a r r i e r a b s o r p t i o n c oe f f i c i e nt a n d a b s o r p t i o n c oe f f i c i e nt i n pa s s i ve r e gi ons , r e- s pe c t i ve l y [ c m- 1] Di o d e c o n s t a n t C o n s t a n t i n ga i n r e l a t i o n wi t h c u r r e n t d e n s i t y F r a c t i o n o f wa ve i n t e n s i t y i n s i d e t h e a c t i v e r e g i o n Re f r a c t i v e i n d e x s t e p a t h e l e r o j u n c t i o n C o u p l i n g ef f i ci ency of d i o d e i n t o f i ber Di f f e r e nt i a l e x t e r n a l q u a n t u m ef f i ci ency ( l aser ) I n t e r n a l q u a n t u m ef f i ci ency P o we r ef f i ci ency Be a m wi d t h a t h a l f - i n t e n s i t y i n t h e d i r e c t i o n s p e r p e n d i c u l a r a n d p a r a l l e l t o t h e j u n c t i o n p l a n e Emi s s i o n wa v e l e n g t h Mi n o r i t y c a r r i e r l i f e t i me Ra d i a t i v e l i f e t i me N o n r a d i a t i v e l i f e t i me O p e r a t i n g t i me f or 20% l a s e r t h r e s h o l d c u r r e n t i n c r e a s e O p e r a t i n g t i me f or a 50% o u t p u t r e d u c t i o n f r o m a n L E D La t t i c e c o n s t a n t ( Aao/ a o is t h e mi s f i t s t r a i n ) E x p o n e n t i n g a i n r e l a t i o n wi t h c u r r e n t d e n s i t y R e c o mb i n a t i o n coef f i ci ent [ c l n 3 s - 1] Ac t i v e r e g i o n t h i c k n e s s 60 t l . Kr e s s e l et al. e El e c t r o n c h a r g e Eg B a n d g a p e n e r g y [ e V] .~ Di o d e b a n d w i d t h .q G a i n coef f i ci ent [ c m - l ] l , l , T h r e s h o l d c u r r e n t [ A ] J, h T h r e s h o l d c u r r e n t d e n s i t y [ A c m 2] L F a b r y - P e r o t c a v i t y l e n g t h N. A. Nu me r i c a l a p e r t u r e o f f i be r N~ I n j e c t e d e l e c t r o n hol e p a i r d e n s i t y [ c m a] P0 P o we r e mi t t c d po, no E q u i l i b r i u m c a r r i e r c o n c e n t r a t i o n s i n a c t i ve r e g i o n ( hol es . e l e c t r o n s ) R F a c e t r e f l e c t i vi t y R Di o d e s er i es r e s i s t a n c e S Sur f a c e r e c o mb i n a t i o n ve l oc i t y [ c m s ~] T O P a r a me t e r d e s c r i b i n g t e mp e r a t u r e d e p e n d e n c e of t h e t h r e s h o l d c u r r e n t (2.12) [ K ] V, Ap p l i e d v o l t a g e W S t r i p e wi d t h Re f e r e nc e s 2.1 A d e t a i l e d a n a l y s i s o f l a s e r d i o d e s a n d h e t e r o j u n c t i o n L E Ds is p r e s e n t e d i n H. Kr e s s e l , J . K . B u t l e r : S e mi c o n d u c t o r Las e r s and l t et er r ~j unct i on L E Ds ( Ac a d e mi c Pr es s , Ne w Yo r k 1977) 2.2 H. Kr e s s e l , H. Ne l s o n : " P r o p e r t i e s a n d Ap p l i c a t i o n s of I I I V C o m p o u n d F i l ms De p o s i t e d by Li q u i d P h a s e Ep i t a x y " , i n Ph y s i c s o f Thi n Fi l ms, ed. by G. I l a s s , M. H . F r a n c o m b e , R. W. Ho f f ma n { Ac a d e mi c Pr e s s , Ne w Yo r k 1973) 2.3 F o r a r e vi e w o f t hi s e x t e n s i v e l i t e r a t u r e , see J . . l . Ti e t j e n, V.S. Ba n, R. E. E n s t r o m, 1). R i c h ma n : ,1. Vac. Sci. Tc c h n o l . 8, 56 (1971) 2. 4 A . Y . C h o : J. Vac. Sci. Te c h n o [ . 8, 531 {1971} 2.5 A. Ma n y , Y. Go l d s t e i n , N . B . G r o v e r : Se mi c onduc t or Su#Ji~ces { No r t h - Ho l l a n d , A ms t e r d a m 1965) 2.6 M. E t t e n b e r g , H. Kr e s s e l : J. Appl . Plays. 47, 1538 {1976) 2.7 G . H . O l s e n : Un p u b l i s h e d 2.8 H. Kr e s s e l : J. El e c t r o n . Ma t e r . 4, 1081 (1975) 2.9 M. Et t e n b e r g , G . l l . O l s e n : J. Appl . Phys . 48, 4275 (1975) 2. 10 M. E t t e n b e r g , R. J . Pa f f : J. Appl . Phys . 41, 3 9 2 6 ( 1 9 7 0 ) 2. I I See, f or e x a mp l e , R. S a n k a r a n , G. A. Amy p a s , R. L. Mo o n , J. S. Es cher , L . W. J a me s : J. Vac. Sci. Te c h n o l . 13, 932 {1976) a n d r e f e r e nc e s t h e r e i n t o L P E g r o wt h of I n G a A s P 2.12 C . J . Nu e s e : J. El e c t r o n . Ma t e r . 6, 253 (1977) 2.13 .I.C. D y m e n l : App[ . Phys . Let t . 10, 84 (1967) 2 . 1 4 F . S t e r n : I E E E J . QE- 9, 290 (1973) 2.15 l l . Kr es s el , M. E t t e n b e r g : J. Appl . Phys . 47, 3533 (1976) 2. 16 Th i s is a mo d i f i e d v e r s i o n o f t h e e x p r e s s i o n d e r i v e d by J . R. Bi a r d, W. N . C a r r , B.S. Re e d : Tr a n s . A I M E 230, 286 (1964) 2.17 H. Kr cssel , H. F. Lo c k wo o d , F . Z . H a w r y l o : J. Appl . Phys . 43, 561 (1972) 2.18 D. L. R o d e : J . Appl . Phys . 45, 3887 (1974) 2.19 A. R. Go o d wi n , J. R. Pet er s , M. Pi o n , G. H. B . T h o mp s o n , J. E. A. Wh i l e a wa y : J. Appl . Plays. 46, 3126 (1975) 2. 20 E . l , e v i n e : U n p u b l i s h e d 2.21 G . H . B . T h o m p s o n , P. A. Ki r k b y , J . E . A . W h i t e a w a y : l E E E J . Q E - l l , 481 (1975) 2.22 H . F . L o c k w o o d : Pr oc . 5t h Bi e nni a l El ect . Eng. Conf . ( Co r n e l l Un i v e r s i t y , I t h a c a , Ne w Yo r k 1975) p. 127 Laser Diodes and LEDs for Fiber Optical Communi cat i on 61 2.23 J. J. Hsi eh: Appl. Phys. Lett. 28, 283 (1976) 2.24 K.Sugiyama, H. Kojima, II. Enda, M. Shi bat a: Jpn. J. Appl. Phys. 16, 2197 (1977) 2.25 G.H.OIsen, C.J.Nuese, M. E!.tenberg: Appl. Phys. Lett. 34, 262 (1979) 2.26 J. J. Hsi eh: Unpublished 2.27 J. C. Dyment , L. A, D' Asaro, J. C. Nort h, B.I.Milter, J.E. Ri pper: Proc. IEEE 60, 726 (1972) 2.28 H.Yonezu, I. Sakuma, K. Kobayashi , T. Kamej i ma, M. Unno, Y. Nannichi : Jpn. J. App[. Phys. 12, 1585 (1973) 2.29 T. Tsukada: J. Appl. Phys. 45, 4899 (1974) 2.30 I . Ladany: Unpubl i shed 2.31 B. W. Hakki : J . Appl. Phys. 46, 2723 (1975) 2.32 T. Tsukada, R. l l oh, H. Nakashi ma, O. Nakada: IEEE J. QE- 9, 356 (1973) 2.33 l . Ladany: J. Appl. Phys. 48, 1935 (1977) 2.34 J.P.Wittke, M. Ettenberg, H.Kressel: RCA Rev. 37, 159 (1976) 2.35 A. J. Afromowitz: J. Appl. Phys. 44, 1292 (1973) 2.36 W.B.Joyce, R.W. I)i xon: J. Appl. Phys. 46, 855 (1975) 2.37 T. Kobayashi , G. l wane: Jpn. J. Appl. Phys. 16, 1403 (1977) 2.38 M. Ettenberg, H.F. Lockwood: To be published 2.39 M. Et t enberg: Appl. Phys.- Lett. 32, 724 (1978) 2.40 See, for example, P. A. Kirkby, A. R. Goodwi n, G. H. B. Thompson, P. R.Selway : IEEE J. QE- 13, 705 (1977) 2.41 K. AiM, M. Nakamura, T. Kuroda, J. Umeda, R. Itoh, N. Chi none, M. Maeda : IEEE J. QE-14, 89 (1978) 2.42 D. Botez: App[. Phys. Lett. 33, 87 {1978) 2.43 D.R.Scifres, W.Streifer, R. D. Bur nham: IEEE ,1. QE-14, 233 (1978) 2.44 N.S. Kapany : Fiber Optics (Academi c Press, New York 1976) 2.45 E. Sni t zer: J. Opt. Soe. Am. $1, 491 (1961) 2.46 H.Kogelnik, T. Li : Pr oc. l EEE54, 1312 (1966) 2.47 L. Cohen : Bell Syst. Tech. J. 51, 573 (1972) 2.48 H. Kogelnik: Proc. Syrup. on Quasi-Optics, ed. by J. Fox (l' olyteclmic Press, Brooklyn, NY 1964) 2.49 D. N. Payne, W. A. Gambl i ng: Electron. Lett. I I . 176 (1975) 2.50 K.S. Kamm, H. M. Schl ei ni t z: Conf. on Laser and Electron-Optical Systems, San Diego, CA (1978) 2.51 I. Ladany: J. Appl. Phys. 42, 654 (1971) 2.52 E.G.Dierschke, L.E.Stone, R.W. Haisty: Appl. Phys. Left. 19, 98 (1971) 2.53 C.A.Burrus, B.l.Mitler: Opt, Commun. 4, 307 (1971) 2.54 M. Et t enbe, g, K. C. Hudson, H. F. Lockwood: IEEE J. QF,-9, 987 (1973) 2.55 H. Kressel, M. Eltenberg: Proc. IEEE 63, 1360 (1975) 2.56 A. W. Mabbi t t , R. C. Goodfellow: Electron. Lett. 11, 274 (1975) 2.57 M.Abe, I. Umeda, O. Hasegawa, S. Yamakoshi, T. Yamaoka, T. Kolani, H. Oskada, H. Takahashi : 1EEE T,'ans. ED-24, 990 (1977) 2.58 F. D. King, J.Straus, D.1.Szentesi, A. J. Spri ngl horpe: Proc. IEEE 123, 619 (1976) 2.59 M. Ellenberg, H.Kressel. J. P. Wi t t ke: IEEE J. QE-12, 360 (1976) 2.60 D. Marcuse: IEEE. I. QE-13, 819 (1977) 2.61 T.P.Lee, A. G. Dent ai : IEEE J. QE-14, 150 (1978) 2.62 A.G. Dentai, T.P. Lee, C.A.Burrus, E. Buehler : Electron. Left. 13, 484 (1977) 2.63 C.J.Nuese, G. l l . Ol sen: Unpubl i shed 2.64 J. P. Wi t t ke: RCA Rev. 36, 655 (1975) 2.65 W. P. Dumke : [ EEE J. QE-11, 400(1975) 2.66 D. Gl oge: Appl. Opt. 10, 2252 (1971) 2.67 J. Col vi n: Oplo-e[ectronics 6, 387 (1974) 2.68 T. P. Lee: Bell Syst. Tech. J. 84, 53 (1975) 2.69 Y.S.Liu, D. A. Smi t h: Proc. IEEE 63, 542 (1975) 2.70 H. Namizaki, H. Kan, M.Ishi, A. l t oh: Appl. Phys. Lett. 24, 486 (1974) 2.71 H.Kressel, M. Ettenberg: Appl. Phys. Lelt. 23, 511 (1973) 62 H. Kressel et al. 2.72 D. Gl ogc: I EEE Trans. MTT-23, 106 (1975) 2.73 M. Et t enberg, lq.S. Sommcrs, Jr., lq. Kressel, H. F'. Lockwood: Appl. Phys. Lett. 18, 571 1971) 2.74 B. W. Hakki , F. R. Na s h : J. Appl . Plays. 45, 3907 (1974) 2.75 H. Kr essel , I . La da ny: RCA Rcv. 36, 230 (1975) 2.76 T. Yuasa, M. Ogawa, K. Endo, H. Yonezu : Appl. Phys. Lett. 32, 119 (1978) 2.77 1. Ladany, M. Et t enberg, H. F. Lockwood, 1t. Kressel : Appl. Phys. Lett. 30, 87 (1977t 2.78 M. Et t enberg, H. Kressel, I-I.F. Lockwood : Appl. Phys. Lett. 25, 82 (1974) 2.79 H. Kresscl, M. Et t cnber g, H. F. Loc kwood: J. El ect ron. Mat er. 6, 467 (1977) 2.80 D. Newman, S. Ritchie, S. O' Har a: I EEE J. QE-8, 379 (1972) 2.81 B. C. DeLoach, B.W. Hakki , R.L. Har t man, L. A. D' As a r o: Proc. IF, EE 61, 1042 (1973) 2.82 P. Petrofl, R. L. Ha r t ma n: Appl. Phys. Lett. 23, 469 (1973) 2.83 R. D. Gol d, L. R. Wci sber g: Sol i d-St at e El ect ron. 7, 811 (1964) 2.84 D.V. Lang, L.C. Ki mmer l i ng : Plays. Rcv. Le~t. 33, 489 (1974) 2.85 H. Kressel , H. Bycr, H. F. Lockwood, F. Z. Hawr yl o, H. Nel son, M. S. Abr ahams, S.H. McFar l ane: Met. Trans. 1, 1635 (1970) 2.86 A.A. Bergh : Proc. 8th Annual Reliability Phys. Conf. ( I EEE Cat al og 70C59- Phy, 1970) p. 48 2.87 1. Ladany, H. Kresscl : Appl. Plays. Lett. 25, 708 (1974) 2.88 R. L. Har t man, A. R. I qar t man : Appl. Plays. Lett. 23, 147 (1973) 2.89 I. Ladany, II. Kressel : Gallium Atwenide and Related Compounds 1974 (Inst i t ut e of Phys. Conf. Series No. 24, 1974) p. 142 2.90 M. Et t enbcrg, C.J. Nuese : J. Appl. Plays. 46. 2137 (1975) 2.91 H. Kressel , M. Et t enberg, I. Ladany : Appl. Plays. Lett. 32, 305 (1978) 2.92 R.L. Har t man, R.W. Di xon: Appl. Plays. Lett. 26, 239 (1975) 2.93 M. El t enber g: Unpubl i shed 2.94 R.L. I l ar t man, N. E. Schumaker , R. W. Di xon : Appl. Plays. Lett. 31, 756 (1977) 2.95 T. L. Paol i : I EEE J. QE-13, 351 (1977) 2.96 R. C. Goodf el l ow, A.W. Mabbi t t : El ect ron. Lett. 12, 51 (1976) 3. Phot odet ect ors D. P. Schinke, R. G. Smith, and A. R. Har t man With 12 Figures Thi s chapt er describes phot odet ect or s useful in fiber opt i c systems. These det ect ors conver t the received opt i cal power into an electrical cur r ent which is amplified and processed to deliver i nf or mat i on in a useful format. The bulk of this chapt er deals with solid-state det ect ors fabricated in silicon since t hey are t he det ect ors of choi ce at opt i cal wavelengths of 0.8 to 0.9 gin, t he oper at i ng wavel engt h of cur r ent fiber systems. Both simple phot odi odes and aval anche phot odi odes with i nt ernal cur r ent gain are discussed. Each has appl i cat i ons t hat are suited to individual propert i es. The basic physical charact eri st i cs are discussed in rel at i on t o t hei r effect on system performance. Longer optical wavelengths of 1.0 to 1.6 gin are of interest because the glass fiber propert i es of at t enuat i on and dispersion are bet t er in this range of wavel engt hs [3. 1, 2]. A l at er section reviews cnrrent t rends in phot odet ect or s suitable for these wavelengths. Of course, t here is a much br oader range of phot odet ect or s t hat is not discussed here. The reader is referred elsewhere for review articles [3. 3-6]. 3.1 Simple Photodiode Structures Silicon phot odi odes which per f or m in the visible and near i nfrared have been commerci al l y available for many years. The st ruct ure and fabri cat i on of these devices have reached a mat ur e stage of devel opment following st eady advances of silicon t echnol ogy. Thr ee st ruct ures are shown in Fig. 3.1 : a) the p-n j unct i on ; b) the p-i-n di ode havi ng a high resistivity "i nt ri nsi c" layer bet ween p+ and n + cont act s; and d) the met al - s emi conduct or diode. The p-i-n can also be i l l umi nat ed f r om the side (c) r at her t han t hr ough a cont act . 3.1.1 Response Quant um efficiency is the first consi derat i on in a phot odi ode, i.e., the effective col l ect i on of phot ogener at ed carriers by a depl et ed volume. A reverse-biased j unct i on depleted to a dept h W capt ures [-1 - e x p ( - cz W)] of the light t hat enters t hat vol ume where ~ is the absor pt i on coefficient of t he material. A gr aph of the absor pt i on efficiency of silicon vs wavel engt h is shown in Fig. 3.2 using W as a paramet er. The gr aph uses t he absor pt i on dat a of Dash and Newman [3.7] as present ed by Sze [3.8]. One can see t hat a depl et i on width of 30 to 50~tm is 64 D, P. Schinke et al. / M E T A L A N T I R E F L E C T I O N o / CONTACT\ COATING ( t 025 A si02) / / P + P + .... i"X , i slo2-.'-:-.'--~-,: ~ ,~:. ...... 5 ; - LAYER n + " (a) ( b ) A N T I R E F L E C T I O N / M E T A L ' - . h ~ ~J C O A T I N G i ~ " . . . . ; . . . . . . . . . . . J ] ~ o .~r/'za ................... n+~R E ~ ) ( T H I N n + , S E M I T R A N S P A R E N T F L E C T I O M E T A L L A Y E R C O A T I N G ( c ) ( d ) Fig. 3. 1a-d. Cons t r uct i on of di fferent hi gh- speed phot odi odes . (a) p - n di ode; ( b) p - i - n di ode: (c) p i n di ode with i l l umi nat i on paral l el to the j unct i on; (d) met al s emi conduct or di ode [3.3] 10 ? I' ua 0 . t E t t l g o . 01 - I - I 0 . 5 0 . 6 0 . 7 0 . 8 0 . 9 1 . 0 t , 1 1 0 0 ~ m 50/xm WAVELENGTH (MICROMETERS) Fig. 3.2. Abs or pt i on efficiency of a wi dt h W in silicon at 2 5 ' C as a funct i on of t he opt i cal wavel engt h. r h e curves are col nput ed usi ng tile dat a of l-3.7] and do not i ncl ude surface refl ect i ons n e e d e d f or h i g h a b s o r p t i o n e f f i c i e nc y a t 0. 8 gi n. T h e e f f i c i e nc y fal l s r a p i d l y f o r 2 > l . Og m, a l t h o u g h s i l i c on d i o d e s wi t h d e e p d e p l e t i o n l a y e r s c a n be us e d t o d e t e c t l i ght a t 1.06 g i n [ 3. 9] . T h e ne t q u a n t u m e f f i c i e nc y r / mu s t a l s o a c c o u n t f o r t h e r e f l e c t i v i t y o f t h e e n t r a n c e f a c e r~ a n d t h e l os s o f c a r r i e r s a b s o r b e d i n t h e f r o n t c o n t a c t o f t h i c k n e s s d 1. T h u s r / = ( 1 - r l ) ( e - ~"') (1 - e - ~ w ) ( 3 . 1 ) Photodetectors 65 in a device using a single pass of t he light beam. A subst ant i al fract i on of the carriers pr oduced in the front cont act may be collected by diffusion. The responsi vi t y is q2 R=q h c [ A. W- ~] . (3.2) A conveni ent bench mar k is R= I . 0 for r / =l . 0 at 2=1. 24gm. In practice, silicon phot odi odes with thin cont act s (dr < 1 lam) and an ant i refl ect i on coat i ng on the ent r ance face can achieve efficiencies q >0. 9 at fixed wavel engt hs in the 0.8 to 0. 9gi n spectral region, with cor r espondi ng responsivities of 0.5 t o 0.6 A- W- L For a given reverse bias voltage, t he depl et i on width W is di rect l y pr opor t i onal to the square r oot of resistivity ~?/2, and efficient oper at i on at low bias vol t age requires high resistivity material. For p-t ype silicon with resistivity ~ 100f~-cm, 10V bias depletes the j unct i on to a dept h W,,~ 10gm. 3. 1. 2 Reverse Currents The reverse cur r ent of the device in the dar k arises from t wo sources: I) gener at i on- r ecombi nat i on current s in t he depleted vol ume and mi nor i t y col l ect i on at t he edge of the depl et i on region and II) surface leakage currents. Cur r ent s generat ed in bul k silicon can be kept low by using high puri t y silicon and using care in processing t o avoi d i nt r oduci ng crystalline defects. Bulk current s as low as 2 x 10 ~ A. mm -3 can be achieved. Surface leakage on silicon det ect ors can be reduced by passi vat i on techniques. Cur r ent s as low as 2 x 10- ~ A. mm- ~ of peri phery at 10V bias have been achieved [3.291. Since the t ot al dar k cur r ent of the phot odi ode produces shot noise <i 2> = 2qIDB , where B is t he effective bandwi dt h of the receiver, this cur r ent must be as low as possible. In practice, small area, well-designed silicon phot odi odes are not the noise limiting el ement in communi cat i on systems. A mor e compl et e discussion of receivers and receiver design can be found in Chap. 4. A system designer must be aware t hat dar k cur r ent is dependent upon t emperat ure. The exact vari at i on of dar k cur r ent with t emper at ur es is det ermi ned by the device structure, p-i-n di odes fabri cat ed by a process, to be descri bed later, have Iocexp-(A/ l <T) where A =0. 7 eV, k=8. 62 x 10 5 eV. K ~ and Ti s the absol ut e t emperat ure. In the range of r oom ambi ent to 70C, this cor r esponds to an or der of magni t ude increase in dar k cur r ent for each 30 C rise in t emperat ure. 3. 1. 3 Speed of Response The internal device characteristics which affect speed of response are the width of t he depl et i on region and the magni t ude of the electric field within t hat region. In a fully depleted p-i-n di ode Wi s fixed by the wi dt h of t he i layer. The 66 D.P. Schinke c t a]. drift velocity of carri ers is linearly pr opor t i onal to electric field for E less t han 104V-cm ~. At hi gher fields, the el ect rons and holes appr oach maxi mum velocities of 8. 4x l0 ~ and 4. 4x 10~'cm s -~, respectively. Thus, for a lightly- doped i region, t ~ = W/ v . . . . and 1 ns response in a device with W= 50gna requires ~ 50 V. Capaci t ance is anot her fact or which must be reduced for a good receiver performance. The capaci t ance of a pl anar area A depleted to a dept h W is C=~; o KA. W I wh e r e ~ : o =8 . 8 5 x 1 0 - 1 2 F- m ~ and K=1 1 . 7 for silicon. Chi p capaci t ances of 1 pF are readily achievable. The ext ernal l oad resistance R also affects the response time t hr ough the chip capaci t ance t = R c . (3.3) The thickness of the active region of a device is a t rade-off between the compet i ng effects of fast transit time requi ri ng a nar r ow depl et i on region and the combi nat i on of quant um efficiency and low capaci t ance which requi re a wide depl et i on region. In addi t i on, a desire for low dar k cur r ent requires a srnall area and mi ni mum depleted vol ume, but the effective coupl i ng of light from an optical fiber or fiber bundl e places a mi ni mum bound on the area. A small det ect or area also minimizes capaci t ance allowing a larger load resistance for a given response time. 3.1.4 Linearity Reverse-biased phot odi odes have a highly linear cur r ent out put which is pr opor t i onal to i nput optical power over 6 decades or mor e of phot ocur r ent . The first effect of high optical power is a decrease of the electric field across the depl et i on region. This does not initially reduce the quant um efficiency, but can decrease the carri er velocity and l ower the frequency response. Details of the sat ur at i on behavi or depend upon the individual device structure, resistivity, cont act resistance and illumination condi t i ons. Sat urat i on is not expect ed to be a pr obl em for systems which oper at e at received opt i cal power levels of 1 mW or less. 3. 2 Aval anche Phot 0di odes ( APD) Phot odi odes which cont ai n a region of high electric field show aval anche mul t i pl i cat i on of phot ogener at ed carriers. The cur r ent delivered to the di ode cont act s is hi gher than the pr i mar y cur r ent generat ed by the incident light. This internal cur r ent multiplication, which can at t ai n values of 100 )r mor e in well- designed silicon diodes, subst ant i al l y increases the sensitivity of an optical receiver above t hat of a receiver cont ai ni ng a nonmul t i pl yi ng phot odi ode. The chip generally is the same small size, but operat es under different bias I n I p X t X 2 M U LTIP LIC A TIO N - ~ I p+ 4 - x 3 x 4 d 1< DRIFT ~l DEPLETION Phot odet ecl or s 67 l Fig. 3.3. Schemat i c r cpr esent al i on of a reach t hr ough avakmchc di ode st r uct ur e condi t i ons. The aval anche det ect or requi res hi gher bi as vol t ages to mai nt ai n a hi gh electric field. The i nt ernal cur r ent gai n is not a l i near funct i on of the appl i ed vol t age and is sensitive to t emper at ur e. The aval anche process i nt roduces noise in excess of the shot noise due to the cur r ent flowing in the device. Thi s excess noise depends upon the mat eri al , device st ruct ure, gain and i l l umi nat i on condi t i ons and is the f act or which ul t i mat el y limits the device per f or mance at high gain. Webb et al. [3.10] have descri bed the pr oper t i es of aval anche di odes wi t h det ai l ed consi der at i on of silicon devices. A compr e- hensive revi ew of aval anche phot odi odes which includes devices for several wavel engt hs is given by Slillman and Wo![e [3.11]. 3.2.1 Principles of Operation I mpact i oni zat i on is the basi c phe nome non for mul t i pl i cat i on of el ect rons and holes in high electric fields. The pr obabi l i t i es t hat an el ect ron and a hol e will have an ionizing collision in a di st ance dx are ccdx and #dx, respectively, where c~ and p are the i oni zat i on coefficients. These coefficients are exponent i al funct i ons of field E [3.12 16], and explicit expressi ons i ncl udi ng the t emper a- t ure dependence in silicon and ger mani um are given by Crowell and Sze [3.17]. In or der to have an i oni zi ng collision, the car r i er must at t ai n a t hr eshol d energy which is about 1.5 to 1.8 times the bandgap energy. Since the pr obabi l i t y for a scat t eri ng collision is much gr eat er t han for an ionizing collision, the i oni zat i on coefficients decrease wi t h i ncreasi ng t emper at ur e and the aval anche gai n al so decreases for a given electric field. In silicon, c > fl by an or der of magni t ude or mor e for fields less t han 3 x 10 5 V c m- ~. The magni t udes and rat i os ofc~ and fl are consi der abl y different in ot her mat eri al s. Chynoweth [3.18] has revi ewed i oni zat i on phenomena in semi conduct or s. 3. 2. 2 Structures Consi der tile schemat i c aval anche phot odi ode st r uct ur e of Fig. 3.3"' which consi st s of an n + cont act bet ween x] and x 2, a p- t ype mul t i pl yi ng region 68 D P Schinke cl al. SILICON n + - p - T/ ' - p + AVALANCHE PHOTODIODE hz.,' /-~ A NT I REFLECTI ON COATI NG r / / / / / ~ i -/ ~ ~"/ / / / J g / I . - " " ; / / A g,4 I I"i g. . 3. 4. Cr oss- sect i onal view of 50/ . Z 77- E p i , / O > 3 0 0 9 . c m p + S UB S T RA T E epimxialsiliconn + p - x p~ awt- lanchc photodiode made for illu- mination t hr ough 1he n + conlacl layer. DilllOllSions ~lrC? 11(/I t('J SC~.llc ( Copyr i ght 1978 AT & T Co. Re- pr i nt ed by per mi ssi on tom t he Bell System T~,chnical JolrrlT~g) between x 2 and x3, a drift space from x~ to x 4 where tile p+ cont act begins. Aval anche occurs in tile high-field region. In oper at i on under reverse bias the depl et i on region ext ends compl et el y from the n + to the p+ cont act s. When a phot o- gener at ed electron hole pair is created, el ect rons are injected i nt o the mul t i pl yi ng region if all light is absor bed to the right of the mul t i pl yi ng region ( x>x3) . Similarly, hole injection occurs for light absor bed to the left ( x<x2) and mixed injection of bot h carriers occurs when light is absor bed within the aval anche region. In general, lowest noise oper at i on occurs when onl y the carri er with the highest i oni zat i on coefficient is injected into the mul t i pl yi ng region. The shape and position of t hat region as well as the absor pt i on of light must be opt i mi zed for each mat eri al to achi eve the highest performance. Silicon det ect ors are generally designed so t hat el ect rons are the principal carri er to under go multiplication. Silicon aval anche diodes have been fabricated in a variety of st ruct ures : p-n j unct i ons [3.19, 23], mesas [3.24], Schot t ky barri ers [3.25, 26], and met al - oxi de semi conduct or st ruct ures [3.27]. At wavelengths in the 0.8 to 0.91.un spectral region, the large optical absor pt i on dept h in silicon makes a R e a d [3.28] or r e ac h- t hr ou. qh [3.10] st ruct ure an opt i mal structure. Thi s st ruct ure has a wide drift region for collecting phot ocar r i er s alld a smaller aval anche region ar r anged for maxi mum injection of electrons r at her t han holes. A cross section of a r each- t hr ough det ect or fabricated on high resistivity epitaxial silicon material [3. 29, 30] is shown in Fig. 3.4. A diffused guard ring prevents edge br eakdown at the peri met er of the mul t i pl yi ng j unct i on. A p+ channel st op sur r ounds the det ect or to prevent surface inversion layer leakage and to limit the lateral spread of the depl et i on layer. A wide depl et i on region, relatively thin n ~ cont act layer and antireflection coat i ng form a highly efficieht det ect or. The principal carriers reachi ng the j unct i on are el ect rons al t hough some holes are injected by light absor bed near the surface and t hr oughout t he mul t i pl yi ng region. A well-controlled ion i mpl ant at i on and diffusion form the mul t i pl yi ng p Photodetectors 69 I00 i i , { i i i 1 i i 1 I i i | 0oe Z _o I - -3 tO ' 1 . e I 1 i i i i i I I I I I i I 0 I00 200 300 BIAS VOLTAGE Fig. 3.5. Photocurrcnt muhiplication as a flmction of reverse bias in the reach-through structure of Fig. 3.4. Unity multiplication is taken at FR =20 V. region. Appr opr i at e nitride passi vat i on and metallized cont act s compl et e the structure. The met al l i zat i on is ext ended beyond the n + guar d ring and p+ channel st op to form field plates t hat reduce the surface electric fields and i mpr ove the reliability of the aval anche phot odi ode. Because of the high operat i ng vol t age of up to ~400 V, aval anche det ect ors are usually packaged in a hermet i c envi r onment [3.29]. The n +-p-a-p + aval anche phot odi ode can bc realized in several st ruct ures by the use of al t ernat e processing technologies. In addi t i on to the epitaxial st ruct ure of Fig. 3.4, the APD can be fabri cat ed in bulk, float zone material which is t hi nned to ~ 75 IJm t hr ough polishing and etching. The device can also be mesa et ched to reduce surface electric fields and capaci t ance. Mor e details of the fabri cat i on t echnol ogy are present ed in Sect. 3.3. 3.2.3 Response The phot ocur r ent mul t i pl i cat i on (gain) as a funct i on of vol t age in the above st ruct ure is shown in Fig. 3.5. The responsi t i vi t y R - = ~ l M ( V ) q ) o / h c ; ~l is the device efficiency with uni t y gain. At low reverse bias, the p region is onl y part i al l y depleted, the device has 11o gain and carri ers are not collected at high speed from undepl el ed material. As the vol t age is increased, the depl et i on layer "sweeps out " the p region and the device gain rises in a sharp knee. As t he voltage is furt her increased, the depl et i on l ayer "reaches t hr ough" the lightly doped 7r region and the device operat es efficiently at high speeds. When the mot i on of the depl et i on region becomes pi nned by the p+ cont act , the applied 70 D.P. Schinke e t al . vol t age raises the internal electric field in the p region as well as in the rt region and increases the gain cont i nuousl y to values of several hundr ed near reverse br eakdown. A functional form for the mul t i pl i cat i on curve has been given by M i l l e r [3.31] M - (3.4) ' ( L; where Vbr is the reverse br eakdown vol t age and n is an empirically det er mi ned exponent (n < 1) which depends upon the device st ruct ure, the det ect or material and the i l l umi nat i on condi t i ons. Equat i on 3.4 provi des a good appr oxi mat i on to the gain above the "knee". The exponent n for mul t i pl i cat i on in an abr upt j unct i on has been related to ~, [: and the j unct i on width for a wide range of phot oexci t at i on [3.32]. Subst ant i al mul t i pl i cat i on is obt ai ned in the n +-p-n-p + di ode over a wide range of voltages because the ~ region is much wider t hau the mul t i pl i cat i on rcgion. The phot ocur r ent gain depends upon t emper at ur e, in general decreasi ng for an increase ill t emperat ure. Temper at ur e effects in reach- t hr ough st ruct ures have been studied by C o n r a d i [3.33] who descri bed tile gain curves using the explicit t emper at ur e dependence of e and fl and a detailed knowl edge of device structure. Excellent agreement was found bet ween t heor y and experi ment . Al t hough the pr ocedur e yields accurat e results, the com- put at i ons are lengthy. An appr oxi mat e descri pt i on of the t emper at ur e de- pendence results from i nt roduci ng T linearly into the br eakdown vol t age and tile exponent n V~,r = Vu,, + a ( T - T o ) n = no + b( T - To). ( 3 . 5 ) In r each- t hr ough structures, a and b are positive and can be readily obt ai ned from experi ment al l y measured curves. 3. 2. 4 Speed of Response Oper at i on of phot odi odes at t he highest speeds requires full depletion. If carriers are generat ed in undepl et ed material, they are collected relatively slowly by diffusion over times compar abl e with the mi nor i t y carri er lifetime and pr oduce a long "diffusion tail" on a short optical pulse. When fully depleted, most det ect ors oper at e at electric fields E > 104V. cm -1 and carriers move at sat ur at i on limited velocities. The response time of aval anche phot o- diodes is limited by t wo effects : a) the time for carri ers to compl et e the process of mul t i pl i cat i on ; and b) the t ransi t time of carri ers across the depl et i on width. Aval anche phot odi odes typically exhibit it slightly asymmet r i c pulse shape with Photodeteclors 71 5 . 0 i . ~ 3 . 0 m 2 . 0 k 1 . 0 - r " ~ 0 . 5 121 Z m~0. 2 ~0. 1 ,'-, . 0 5 L I , , I N J . 0 2 o z . 0 1 .0O - ~ . 5 . t 0 3 _ ~ 2 ' 1 0 3 M o = a / / ~ "" ~ \ 5 . 1 0 2 ~\ z.lo z - % N ~ t 0 2 N N % , , . \ 5'101 ELECTRON INJECTION, HOLE I NJ E CT I ON, /~/cz 2"101 [ I I L I l l l l I I I I l t i l l 1 ~ 1 t " l ~ t l 5 10 5 0 1OO 5 0 0 1 0 0 0 AVALANCHE MU L T I P L I C A T I ON M 0 Fig. 3.6. Theoretical 3 dB bandwidth B times 2nt,,Jt,,,. = average carrier transit lime) of an awdanche photodiode plotted as aft, nction of the Iow-fi'cqucncy multiplication factor Mo, for wlrious values of 7/[L Above the dashed curve, bandwidth is nearly independent of gain; below the dashed curve, a constant gain bandwidth product applies [3.34] a fast rise t i me as el ect rons are col l ect ed and a fall t i me det er mi ned by the t ransi t t i me of holes movi ng at a sl ower velocity. E m m o n s [3.34] has cal cul at ed the - 3 d B bandwi dt h B of a mul t i pl yi ng regi on nor mal i zed to t he aver age carri er t ransi t t i me with t he rat i o of hol e injection to el ect ron injection as a par amet er . These results are shown in Fig. 3.6. The dashed curve is for M o = o~/[L When M o > ~//3 (in tile regi on bel ow tile dashed line), the curves are st rai ght lines i ndi cat i ng a const ant gai n bandwi dt h pr oduct and a depeudence of mul t i pl i cat i on on frequency of the f or m m 0 M(o~) ~, [ l . . . . . 211/ ' ' (3.6) where z I is an effective t ransi t t i me t hr ough the aval anche region. Tile si mpl est funct i onal f or m t hat descri bes the effective t ransi t t i me is r I = N ( ~ / f l ) r 2 , (3.7) where r 2 is the act ual t ransi t t i me t hr ough the aval anche region and N is a numbe r sl owl y varyi ng f r om N = 1/3 to 2 as [3/oc ranges f r om 1 to 0.001. Signal di st or t i ons due to the hi gh-fi ' equency r esponse of silicon aval anche di odes have been studied by K r u m p h o l z [3.35]. Di odes with gai n bandwi dt h 72 D.P. Schinke et al . product s of 200 GHz have been const ruct ed which showed pulse halfwidths of 0. 6ns for M up to 330 using light with 2 = 0 . 9 g m [3.36]. Kaneda et al. [3.37] have r epor t ed measurement s of the aval anche bui l d-up time as a funct i on of the width of the aval anche region and the i l l umi nat i on wavelength. They find r l ~ 5 1 0 ~3s which relates to the total aval anche bui l d-up time t=r.~M. Similar measurement s have been performed on ger mani um APDs which show z I ~-5 10-~-' s [3.383. In practice, highly efficient silicon devices oper at i ng in high sensitivity fiber opt i c systems will be limited mor e by the RC-t i me const ant and the carri er t ransi t time t han by the aval anche gain bandwi dt h product . Novel st ruct ures have been fabri cat ed which have short pat hs for carri er collection but possess relatively long optical absor pt i on distances creat ed by multiple internal reflection or diffraction gratings [3.39,403. 14owever, fiber systems oper at i ng at bit rates of several hundr ed Mbi t , s - l or less can accomodat e efficient, low-noise st ruct ures t hat have pulse rise times of 1 i ' l S . 3 . 2 . 5 N o i s e i n A v a l a n c h e P h o t o d i o d e s When the aval anche di ode is oper at i ng at an average gain of M, not all phot ogener at ed pairs are multiplied by exactly M. Ther e is a di st ri but i on of gains which is pr oduced by the statistical nat ur e of the aval anche process. This di st ri but i on of gains i nt roduces noise into the mul t i pl i cat i on process. Mclntyre [3.41] has derived an expression for the noise spectral densi t y in a spatially uni form aval anche diode. This noise spectral densi t y can be calculated if the i oni zat i on rates and the electric field are known at every poi nt within the depl et ed region. If a pr i mar y cur r ent I 0 is injected into the aval anche region, the spectral densi t y of the shot noise in the multiplied cur r ent is given by d r!l (i2> = 2ql (M2)' ( 3 . 8 ) where (M 2) is the mean square value of thc carri er multiplication. This expression can be put in an equi val ent form d d l ~ (i2> = 2ql MZF(M)' (3.9) where M = ( M) , and F(M), known as the excess noise factor, is the amount by which the noise exceeds t hat of an ideal cur r ent multiplier. The value of F(M) depends upon the detailed characteristics of the j unct i on and upon the nat ure of the pr i mar y exci t at i on : hole injection, el ect ron injection or mixed injection. By appr opr i at el y weighting the i oni zat i on rates over the electric field profile, an effective rat i o of i oni zat i on coefficients, k<,rr = (/]/c~)~ff can Photodetectors 73 be defined. When the aval anche is initiated with the most highly ionizing carri er [3.40], F ( M) =M 1 - ( 1 k e r r ) )2 (3.10) which is given to a good appr oxi mat i on by F( M) = 2(1 - k~rr) + k~rrM . (3.11) A graph of F( M) vs M for several values of k c f r is shown in Fig. 3.7. Experi ment al results have been shown to be in good agreement with this anal yt i cal form. A low excess noise fact or requires a large rat i o of the i oni zat i on coefficients of t wo carri er species and also requires t hat t he aval anche be initiated by the carri er with the largest i oni zat i on coefficient. In silicon, this requires a device design which initiates the aval anche with electrons. Anot her charact eri st i c of the statistical nat ure of the aval anche multipli- cat i on process is the nongaussi an di st ri but i on of aval anche gains. The shape of the di st ri but i on funct i on is such t hat aval anche gains larger t han the average val ue occur with great er probabi l i t y t han t hat charact eri st i c of a gaussian process with a vari ance given by (3.8). Theoret i cal expressions for the distri- but i on of aval anche gains have been deri ved by Mc l n t y r e [3.42] and experi ment al l y confi rmed by Com' adi [3.43]. Under some circumstances, this nongaussi an char act er can affect receiver performance. In pract i cal digital fiber systems, the principal effect is a reduct i on in the opt i mum aval anche gain and a shift in the opt i mum decision level. As discussed in Chap. 4, the use of a gaussian appr oxi mat i on is valid under most circumstances. The excess noise fact or for an APD is not an i nvari ant characteristic. It depends upon the mat eri al propert i es t hr ough the electric field dependent i oni zat i on rates. It also depends upon the st ruct ure and shape of the j unct i on. A simple exampl e of the shape effect is illustrated in Fig. 3.8 which shows an ideal mul t i pl yi ng region in silicon of width W with a const ant electric field. The numeri c eval uat i on was made using the expressions of Webb et al. [3.10]. For each width IV,, the magni t ude of the electric field is adjusted to give an average gain of 100 for t wo cases: I) onl y el ect rons injected i nt o the region as woul d be obt ai ned by i l l umi nat i on fi'om the back; and II) mixed injection of carriers t hr oughout the st ruct ure t hat is pr oduced by light at 0.8 lain incident t hr ough the n + cont act (see Fig. 3.3 also). Even for pur e electron injection, a t hi nner j unct i on produces mor e noise t han a wi der j unct i on because :~ and fl for silicon become nearl y equal at the high electric fields requi red to pr oduce an average M = 100 in a short distance. In the case of mixed injection, the excess noise fact or is higher t han for pure el ect ron injection but it goes t hr ough a mi ni mum. If the mul t i pl yi ng region is very broad, a large fraction of the mul t i pl i cat i on results from holes and the excess noise fact or increases. For a given j unct i on 7 4 D.P. Schinke el al . 1 0 0 5 O rY I- k e f f =/1 o 2 O , , < LLI m 1 0 5 z 5 (.rJ W C) x w 2 J I ~ l l I I I I I I I I I I I I 5 '10 2 0 5 0 4 0 0 : ~ 0 0 5 0 0 A V E R A G E G A I N M F i g . 3 . 7 . E x c e s s n o i s e f a c t o r I , " a s a f u n c t i o n o f a v e r a g e p h o l o - c u r r e n t g a i n w i t h k , . a s a p a r a m - e t e r . L o w - n o i s e o p e r a t i o n a t l a r g e w d u e s o f M r e q u i r e s k~u 2 2 1 1 ' 'FIEt..D PROFILE n ' p - i - p * S ' A'PD I ' t o b e a s s m a l l a s p o s s i b l e [ 3. , 0~] / 1 1 8 a l I I , o L i l F I l l I I I III t 6 t - I I | I ~ - - F R O N T ILLUMINATION li~ x : e o o o ~ , n = 9 ~ 9 O / o 1 4 r i i / - - - - P U R E ELECTRON IN JEC TIO N 8 X k \ 6 4 2 0 I 0 \ %. M = 1 0 0 I i I L I i I I 2 4 6 8 '10 w {y,m) F i g . 3 . 8 . D e p e n d e n c e o f t h e e xc e s s noi s e ; a c t o r o n s t r u c t u r e o f t h e m u l t i p l y i n g r e g i o n . For e a c h w i d l h 14/, t h e a m p l i t u d e o f t h e e l e c t r i c f i e l d i s a d j u s t e d 1o g i v e a n a v e r a g e g a i n o f 100 Phot ode t e c t or s 75 16 I f f / ~ SILICON n + - p - ' r r - p + 4 2 ~ r ~ - AVALANCHE PHOTODIODE "~ 0 | I I I I I I I I I I I I 0 20 40 60 80 100 120 140 t 60 180 2 0 0 220 240 MULTIPLICATION (M) Fig. 3.9. Excess noi s e fact or F(M) as a f unct i on of aver age mul t i pl i cat i on [o1 opt i cal exci t at i on at di fferent wavel engt hs ( Copyr i ght 1978 AT & T Co. ) shape, the excess noise fact or depends st rongl y on wavelength due to the vari at i on of the absor pt i on coefficient and hence, the fract i on of hole injection. Measurement s of the excess noise fact or on a part i cul ar device with the reach- t hr ough st r uct ur e of Fig. 3.4 are shown as a funct i on o1" wavel engt h in Fig. 3,9. This analysis has been applied to ot her st ruct ures [3,44, 45], and effective noise factors as low as l%rr~0.01 have been r epor t ed [3.46]. Silicon aval anche det ect ors for fiber optics appl i cat i ons in the 0.8 to 0.9 tana spectral rcgion have l%f r in the range of 0.01 to 0.08. For small l%r r and large M : F ~ 2 + I % r r M . Det ect ors in fiber opt i c t ransmi ssi on systems are packaged to oper at e in a mode in which no backgr ound light is present and onl y the signal light is incident on the mul t i pl yi ng region of t he det ect or. The mean square noise pr oduced by an APD uni forml y mul t i pl yi ng at an average gain M can be wri t t en as ( i ) 2 = 2 q [ l a ~ + (l.~ig + I db)M2F(M)], where (3.12) q is the el ect roni c char ge, Id~ is t hat component of the det ect or dark cur r ent which is not multiplied ; l,ig = R P o , the pr i mar y signal phot ocur r ent ; I db is t hat component of det ect or dar k cur r ent which is to be multiplied ; and F ( M) is the excess noise fact or descri bed above. In an aval anche det ect or, the bulk dar k cur r ent compet es directly with the signal cur r ent and must be kept small for sensitive detection. In silicon, sources 76 D, P. Schinke et a l . 1o-3 10"4 t 0 - 5 t 0 - 6 t 0 - ; ' 10 - 8 1 0 - 9 i 0 - I ~ t 0 - I ] 1.8 (*C) 250 7'00 150 120100 80 60 40 23 I l 1 I i i I I I DARK CURRENT AVALANCHE PHOTODIODE ~ , ~ S i n * - p - ' r r - p + - ~ r ~ T o t f f l i n t r ~ n i ( T ) \ ' a I I I I 2 . 0 2.2 2 . 4 2.6 O -20 I ~ ' ~ C ~ ( I D i f i l e x l r ~ D ~ n t z ( T ) \ ' ~ o \ 2.8 3.0 3.2 3.4 3.6 3.8 4.0 x l O 3 / T ( K ) Fig. 3.10. A plot of dark currcni as a function of reciprocal tcm- peralure for an avalanche pho- todiode with 100/am dia. gain region and an overall diameter of 360 pm of fully depleted region (40 gm thick) [3.30-1 (Copyright 1978 AT & T Co.) of this current are crystal imperfections, impurities within the depl et ed volume, and diffusion cur r ent t hr ough its boundar y. By t he use of i mpuri t y get t eri ng and well-controlled processing l db~2 X 10 - l i A. mm 3. General l y, Id~ is sub- stantially larger t han I , . , . 1 , ~ is generat ed in the depl et i on region and at surfaces out si de the n + cont act in the peri met er of the aval anche phot odi ode. Dark current in an aval anche device depends upon t emperat ure. In Fig. 3.10, the t emper at ur e vari at i on of a well-gettered APD with 100gin di amet er active region is pl ot t ed as a funct i on of reci procal t emper at ur e [3.30]. The dar k current ar ound r oom t emper at ur e is due to carri er generat i on from i mpuri t y centers with energy levels close to the middle of the bandgap. The act i vat i on energy of these levels is est i mat ed to be 0.66 t o 0,69 eV. At hi gher t emper at ur es between 40 and 150'~C, the dark current increases with t emper at ur e in pr opor t i on t o the square of t he intrinsic carri er density. Based on numeri cal estimates and measurement s on devices with different diameters, it is suggested t hat this dark current component is due to diffusion of el ect rons out of the high resistivity zt-typc sidewalls of the device. For t emperat ures above 150 C, the intrinsic carri er concent r at i on exceeds the dopi ng in the 7z region. The leakage resulting from diffusion out of the APD' s peri met er then increases in pro- por t i on t o the intrinsic carri er concent rat i on. The mi ni mum de{.~etable optical power in a given system is dependent upon the receiver design and system par amet er s and is discussed in Chap. 4. Experi ment al fiber opt i c systems current l y empl oyi ng APDs oper at e with mul t i pl i cat i on of 100 or less. Photodetectors 77 The above t heor y and resulting equat i ons appl y to uni forml y mul t i pl yi ng j unct i ons. In practice, spatial nonuni formi t i es in the dopant s in a p - n j unct i on and crystalline defects or metallic precipitates which penet rat e the high-field region can cause localized areas of nonuni f or m mul t i pl i cat i on and mi cro- plasmas. These defects are electrically noisy and must not be present in the j unct i on area. Silicon t echnol ogy has advanced bot h mat eri al perfect i on and processing to permi t aval anche devices to be fabri cat ed with economi cal l y at t ract i ve device yields and excellent reliability. 3 . 2 . 6 L i n e a r i t y a n d S a t u r a t i o n Aval anche phot odi odes can show excellent linearity of multiplied phot ocur r ent from very low incident light levels of less t han a nanowat t up to levels of several mi crowat t s. Fi ber systems based on light-emitting di odes have i nput opt i cal power levels of 10 - to 5 x 10 4W, while lasing di odes can inject 10 - 3 to 1 0 - 2 W into fibers. The highest aval anche gain t hat can be achieved in a det ect or is reduced by several fact ors when oper at i ng at high incident light levels : I) The bias voltage can decrease due to an external load resistor or series resistance in the semi conduct or cont act s; II) The vol t age across the mul t i pl yi ng j unct i on can decrease due to the internal series resistance of the device or space-charge effects in which carriers drifting in the depl et i on region decrease the electric field at the j unct i on; III) Hi gh-opt i cal power levels can heat t he j unct i on decreasi ng t he gain by the t emper at ur e dependence of mul t i pl i cat i on discussed above. The sat ur at i on of t he multiplied phot ocur r ent in an APD due to these effects have been consi dered by Webb et al. [3.10]. They show t hat at high mul t i pl i cat i on levels, the multiplied cur r ent increases as the square r oot of t he pr i mar y phot ocur r ent [ p . Empirically, the series resistance of the devices has been i nt r oduced into (3.4) by Melchior and Lynch [3.47] who wrot e 1 M - (3.13) I _ ( V - I R I " ' \ v~,r ) where I =MIp is the t ot al multiplied current . The maximuna at t ai nabl e gain varies as (Ip)- ~/2 when the dar k cur r ent is negligibly small. An exampl e of the decrease in measurcd mul t i pl i cat i on M = I/Ip due to i nt ernal series resistance in a r each- t hr ough device is shown in Fig. 3.1 t. The mul t i pl i cat i on is shown as a funct i on of the t ot al device current . Not e t hat at gains M~ 100 the device shows some nonl i neari t y at pr i mar y phot ocur r ent s of less t han 10 - ~ A. This t ype of nonl i neari t y depends upon the device structure, the resistivity and the width of the depleted layer, and t he size of the incident beam. In a reach- 78 D.P. Schinke et al. t 0 0 z o_ 10 . J o,_ ~k=O.88/.z.m SPOT DI A = 9 2 ~ m 1 1 = = = = I = = = ~ [ i i t L I t 0 - 6 t 0 - 5 t 0 - 4 i 0 - 3 TOTAL CURRENT (AMPS} Fig. 3.1 I. Me a s ur e d mul t i pl i cat i on M as a f unct i on of t ot al devi ce cur r ent MIp ( I t : pr i mar y phot ocur r ent ) . De- vice mul t i pl i cat i on decr eas es at hi gh- cur r ent densi t i es t hr ough st ruct ure, pr i mar y phot ocur r ent is generat ed in a ci rcul ar area of di amet er d. The pr i mar y electrons pr oduce M secondar y el ect rons and holes at the j unct i on which flow from the ci rcul ar region into the drift space of resistivity ~ and width W. The device internal series resistance is in the form of a spreadi ng resistance [3.48] 0 t ( 4W) (3.14) R = ~ - tan ~ / . A smaller focused spot sat urat es the det ect or at progressively lower light levels. A nonl i neari t y due t o space-charge effects is fast and shoul d follow the incident phot ocur r ent up to the limit i mposed by carri er velocities. The effects of sat ur at i on are : 1) an ampl i t ude compressi on of the detected pulse in a digital com- muni cat i ons syst em; and If) di st ort i on in an anal og system empl oyi ng ampl i t ude modul at i on over a wide dynami c range. One measur ement of di st or t i on in an APD has been r epor t ed [3.49] which showed second har moni c di st ort i on to be less t han - 6 0 d B up to an i nput power level of 10- s W. Details of the device st ruct ure and i l l umi nat i on condi t i ons were not given. If t he optical power is consistently at a level which sat urat es the APD, a p-i-n coul d be used with little or no loss in performance. 3.3 Fabrication Technology Silicon phot odi odes have been made in a vari et y of st ruct ures with material requi rement s and processes devel oped for part i cul ar devices. The st art i ng material can be high resistivity (~)> 103 ~q-cm) intrinsic silicon. The wafers can Photodeiectors 79 be t hi nned to the di mensi ons desired for depl et i on wi dt h and processed to form j unct i ons, ohmi c cont act s and passivated opt i cal wi ndows using a wide range of procedures. A r each- t hr ough st ruct ure which is amenabl e to fabri cat i on on large di amet er epitaxial silicon wafers has been r epor t ed by Mel chi or et al. [-3.29, 30] and is shown schemat i cal l y in Fig. 3.4. The device is processed as follows. Epitaxial mat eri al of 0 > 300f ~. cm is gr own on di sl ocat i on free p~ subst rat es and forms the rt drift region. The light dopi ng insures full depl et i on to the p+ substrate. The epitaxial l ayer is sufficiently thick to absor b mor e t han 95 9~, of the incident radi at i on. The st r uct ur e is formed by first diffusing the n + guard ring and p+ channel stop. The guard ring prevents br eakdown at the edge of the n +p j unct i on, provi des a large radi us of cur vat ur e at the n +~z j unct i on, and reduces const rai nt s at the met al cont act s by provi di ng a relatively deep j unct i on under the cont act . The p+ channel st op diffusion cuts off surface inversion channel s and limits the lateral spread of the depl et i on region. The p mul t i pl i cat i on region is formed in the cent er of the APD by a precise bor on ion i mpl ant at i on followed by cont r ol l ed diffusion. A heavily doped phos phor ous layer is diffused i nt o the back of the wafer to "get t er" impurities which have deep energy levels in silicon. The get t eri ng subst ant i al l y reduces the dar k current . A shallow n + cont act is i mpl ant ed and diffused in the cent er t o form the high field j unct i on. This st ruct ure has light incident t hr ough the n ~ cont act . Thi s cont act is thin in or der to mi ni mi ze the phot ocar r i er s lost due to r ecombi nat i on and the hol e injection which would increase the excess noise fact or of the device. The diffusion time of the n + cont act is adjusted to cont r ol the cur r ent gai n- vol t age charact eri st i c and the reverse br eakdown voltage. The wafers are anneal ed in an HCI ambi ent to reduce the mobile ion cont ent of the surface oxide, and silicon nitride is deposi t ed to passivate the st ruct ure against ionic cont ami nat i on such as Na +. The get t eri ng layer is r emoved from the back of t he wafer and an ohmi c cont act is formed by ion i mpl ant at i on of bor on. The wafer is about 450 Ixm thick and retains mechani cal st rengt h for handling. The front surface met al l i zat i on is Ti - Pt - Au formed over Pt-Si, and the back met al is Ti-Au. The metal cont act s are ar r anged to overl ap the metallurgical n-~ and 7r-p + j unct i ons. These "field plates" reduce the radius of cur vat ur e of electric field at tile silicon dielectric i nt erface and prevent the accumul at i on of charges on the ext ernal surface of the nitride. A bui l dup of charge t here coul d induce sufficiently large electric fields to cause bursts of aval anche or Zener current s [3.50]. High levels of surface charge coul d ul t i mat el y lead to increased surface l eakage cur r ent and reduce br eakdown voltages, p-i-n phot odi odes can be obt ai ned from this process sequence by omi t t i ng the bor on ion i mpl ant at i on and diffusion. They exhibit similar low dar k current s and excellent reliability. In or der to achieve a br oad j unct i on profile or a mul t i pl i cat i on region shaped to achieve low noise performance, the r each- t hr ough st ruct ure has also been fabri cat ed by a t wo-st ep epitaxial process. A por t i on of tile drift space is grown by epitaxy, a cont r ol l ed p charge is i mpl ant ed in a pat t er n and the epi t axy is compl et ed to the desired thickness and dopi ng levels [3.46, 51]. 80 D. P. Schinke ctal. Aval anche phot odi odes with the st r uct ur e of t hose in Fig. 3.4 have shown an est i mat ed medi an t i mc to failure of about 105 hour s at 170<'C and 300V reverse bias when packaged hermet i cal l y. Assumi ng an act i vat i on energy of 0. 7eV, which is det er mi ned f r om the agi ng of several gr oups at different t emper at ur es, the medi an t i me to failure of the chi p is about 10 '~ hours at r oom t emper at ur e. The pri nci pal modes of failure are high l eakage cur r ent and an increase in the magni t ude and frequency of aval anche noise pulses near br eakdown [3.52]. Of course, devices may al so fail for ot her reasons t hat are related to the i ndi vi dual packagi ng t echnol ogy. 3. 4 Photodiodes for Longer Wavelengths Si l i ca-based glass fibers show a low opt i cal at t enuat i on in tile spect ral regi on bet ween 1.0 and 1.61am [3.53]. Tot al losses of less t han 0 . 5 d B. k m -1 are at t ai nabl e. In addi t i on, pul se di sper si on is mi ni mi zed near 1.27 gm [3.54]. Thi s combi nat i on of l ow fiber loss and a mi ni mum in the fiber di spersi on offer the possi bi l i t y of hi gh- bandwi dt h syst ems oper at i ng over l onger di st ances t han are achi evabl e in the 0.8 to 0.9 gm region. Several aut hor s have assessed al t er nat i ve fiber syst ems [3.55, 56], and a fiber t r ansmi ssi on exper i ment at ) . = 1.29 lain has been r epor t ed [3.57]. Thi s wavel engt h is sufficiently r emoved from the 0.8 to 0.9 gm spect ral band t hat different mat eri al syst ems and st ruct ures are requi red for sources and det ect ors. A compr ehensi ve review of the l i t erat ure (up to the year 1975) concer ni ng det ect or s for this spect ral band can be found in [_3.11]. However , there is act i ve research i nt o new sources and det ect ors, and these areas are rapi dl y evolving. Det ect i on at 1.0 to 1.6 pm requi res a mat er i al t hat can efficiently abs or b light. Silicon, which has an indirect band gap, has an absor pt i on coefficient which decreases rapi dl y t owar ds 1.1 lain. These det ect or s can pr ovi de useful det ect i on Icvels to about 2 = 1.1 lam which includes the Nd : YAG laser wave- length of 1.06 pin. The wide depl et i on l ayer of ~ 1 mm or mor e requi red for efficiency yields l onger carri er t ransi t times and reduces hi gh-speed perfor- mance. To i mpr ove the speed of response, the silicon det ect or can be side i l l umi nat ed, al t hough t hat results in i ncreased fabri cat i on compl exi t y. Al t ernat i ve det ect or mat er i al s t hat have a nar r ower band gap include ger- mani um or mixed al l oys of silicon and ger mani um and Il l V c ompound semi conduct or s. Extrinsic phot oconduct or s (i.e., semi conduct or s with i nt en- t i onal l y doped i mpur i t y levels) have been used t hr oughout the i nfrared spect r um. Absor pt i on in these det ect or s is due to el ect roni c t ransi t i ons bet ween the gr ound st at c of the i mpur i t y and its excited st at es as well as the associ at ed conduct i on or val ence band. However, to achi eve backgr ound- l i mi t ed perfor- mance, the det ect ors are t ypi cal l y cool ed to cr yogeni c t emper at ur es to reduce gener at i on- r ecombi nat i on noise pr oduced by the i mpur i t y states. Because the nl axi munl abs or pt i on coefficient is limited by the quant i t y of dopant which can be i nt r oduced into the lattice, ext ri nsi c phot oconduct or s havc smal l er abs or p- Photodetectors 81 tion coefficients t han intrinsic det ect ors and thus requi re a larger physical size for efficient det ect i on. The above factors are consi derabl e limitations in the near infrared where intrinsic semi conduct or s can be used. A review of the physics and t echnol ogy of i mpur i t y ger mani um and silicon mfi' ared det ect ors has been given by Bratt [3.58], and Kruse et al. [3.59]. 3. 4. 1 Ge r ma ni um Pho t o di o de s The absor pt i on coefficient of ger mani um permi t s efficient phot odet ect i on to 2 = 1.8gm. Pl anar passivated p-i-n phot odi odes with guard rings have been fabri cat ed in high-resistivity ger mani um [3.60]. Due to the smaller bandgap (0.67 eV), ger mani um diodes have subst ant i al l y higher dar k current s per unit vol ume t han silicon det ect ors (1.11 eV band gap). Di ode reverse current s were domi nat ed by bul k generat ed dar k cur r ent r at her t han surface cur r ent for all but the smallest ( ~1 ram) devices. A cur r ent density of 2. 5mA- cl n -2 was report ed at r oom t emper at ur e for a device typically 1501am thick. The reverse leakage cur r ent was found to increase exponent i al l y with t emper at ur e with an act i vat i on energy of 0.4eV. The device showed rise and fall times of a few nanoseconds at 15V bias. The t echnol ogy coul d also be ext ended to larger devices with multiple arrays. Ger mani um has also been fabri cat ed into aval anche phot odi odes [3.38, 47, 61]. Bot h etched mesa and pl anar j unct i ons with a diffused guard ring have been made. Ando et al. [3.62] have fabri cat ed an n+p pl anar aval anche phot odi ode with 100 gm di amet er active region and a diffused guard ring. Since the absor pt i on coefficient of ger mani um is the or der of 10"~cm ~ at wave- lengths shor t er t han 1.5 gm [3.7], all phot ocar r i er s are generat ed within I to 2gi n of the surface. The quant um efficiency is st rongl y affected by surface r ecombi nat i on of electron hole pairs. The quant um efficiency peaked near 1. 6gm and was above 80% for a device using an SiO 2 antireflection coating. For a 100gi n di amet er diode, the reverse cur r ent of 0. 2gA consisted of 0.1 gA flowing t hr ough the guard ring peri phery and 0.1 I-tA flowing t hr ough the j unct i on. The values of these current s can be det ermi ned using the dependence of reverse cur r ent on gain. The dar k cur r ent is st rongl y dependent on t emperat ure, and varies as e x p ( - Eg/2kT) charact eri st i c of bulk generat i on. The measured excess noise fact or was pr opor t i onal to M 3 at a wavelength of 1.32 ~am, and coul d be descri bed using an effective k = 1.0. The device at t ai ned a high-speed pulse width of 200 ps with a cor r espondi ng frequency bandwi dt h of 2GHz . Passi vat i on of the surface is a pr obl em in ger mani um devices, and recent efforts have been r epor t ed by Na,qai [3.63], and Yashiro et al. [3.64]. Ger mani um aval anche phot odi odcs will also exhibit a larger amount of excess noise t han a silicon aval anche det ect or. Thi s is a consequence of t he near equal i t y of el ect ron and hole i oni zat i on rates fl/:~, 1. Thus, t here is little differential in the i mpact i oni zat i on t hat coul d be opt i mi zed by st ruct ural design. Ot her aut hor s have report ed the shot noise due to phot ocur r ent as: <i2) = 2qlpBM" where the exponent x is appr oxi mat el y 3 [-3.65]. 82 D.P. Schinke el al. 6. 6 o< 6. 4 I-- 6. 2 m 6. 0 (.3 m 5. 8 5 . 6 _J 5.4 5. 2 . / ~ . . . - - - " ~ S b / / , / InP / ~ I nAs AQ-As "~'~ GoAs Ge d" q GOP SI , I , I I I I I , I i I i I i I , I I 0. 4 0. 6 0. 8 1.0 2. 0 4.0 6. 0 8. 0 WAVELENGTH CORRESPONDING TO ENERGY GAP ( / xrn) Fill. 3.12. l.atlicc constants and wavelengths corresponding to energy gaps for several Ill V c{~mpound semicondtlctor materials [3.56] 3.4.2 III-V Compound Semiconductors Semi conduct or materials formed from elements in groups III and V of the peri odi c t abl e have been investigated and fabricated into phot ocat hodes, phot odi odes and aval anche phot odi odes. An interesting feature of these materials is the rel at i onshi p bet ween bandgap and composi t i on which gives the device designer the ability to cont r ol the spectral response by changi ng the composi t i on. Also, the mat eri al s can be pr epar ed in mul t i l ayer st ruct ures by liquid or vapor phase epitaxial growt h. Thus, there exists a wide range of possible composi t i ons, absor pt i on coefficients, i oni zat i on propert i es, and device structures. The rel at i onshi p between bandgap, composi t i on, and lattice const ant are represent ed in Fig. 3.12. Since these compounds have di rect transitions, the absor pt i on coefficient rises steeply at the band edge to values of 104cm ~ or more. These phot odi odes can absor b efficiently in a layer onl y a few mi crons thick, leading to a small device st ruct ure and a fast carri er transit time. Conversel y, the high absor pt i on coefficient increases the number of carriers lost in an undepl et ed cont act region, and the thin depl et i on region has a higher capaci t ance per unit area t han a mor e deepl y depleted device. Ther e are several material systems which pr ovi de high det ect i on efficiency in this range. Dark current s in these materials due to generat i on r ecombi nat i on in the depl et ed vol ume are pr opor t i onal to the intrinsic carri er concent r at i on n~. They shoul d be smaller t han ger mani um but larger t han silicon since ni~s e x p ( - E J 2 k T ) . Actual values of the bulk dar k cur r ent vary inversely with the carri er lifetime and are sensitive to the growt h and processing of the material. Dark current s as low as 10-'~ A are r epor t ed in a device 150tam in di amet er fabricated in GaAsSb [-3.66]. Using the aut hor s' estimates of depl et ed volume, lb.~k ~ l 0 - 2 A- cm 3, The l ong-t erm stability of surface leakage cur- Photodetectors 83 rent s in I l l V mat er i al s is not well charact eri zed. At present , t here is little publ i shed l i t erat ure on passi vat i on t echni ques on mat er i al s ot her t han GaAs. hnpa c t i oni zat i on phe nome na in I I I - V mat er i al s are consi der abl y mor e di verse t han in silicon and ger mani um. For exampl e, earl y results on Ga As i ndi cat ed t hat el ect ron and hol e i oni zat i on rat es were nearl y equal [3.67]. Lat er wor k oll aval anche devices woul d be consi st ent with a hole i oni zat i on rat e great er t han t hat of el ect rons [3.68]. Recent wor k [3.69] has shown t hat i oni zat i on rat es are st r ongl y dependent upon the el ect roni c band st r uct ur e of the crystal. Furt her, the i oni zat i on rat es depend upon the or i ent at i on of the electric field with respect to the cryst al di rect i on [3.70]. In Ga As for E al ong the <111> di rect i on holes have a hi gher i oni zat i on rat e t han el ect rons over the r ange of electric field 6 x 1 0 s > E > 3 x l 0 5 V. c m l. In the <100> di rect i on, holes have a hi gher rat e t han el ect rons onl y for E <4. 5 x 105 V . c m- 1 but have a l ower rat e at ot her values. The r at i o of i oni zat i on rat es depends upon the al l oy compos i t i on as well. By mi xi ng Sb on the V subl at t i ce of GaAs , the rat i o ~ / [ ~ can be vari ed f r om 0.25 to 2.5 [3.71]. Ther e are i ndi cat i ons t hat the dependence on the V subl at t i ce is al so found in the c ompounds of nar r ower bandgap. Aval anche noise meas ur emcnt s in haas show t hat hole i oni zat i on rat es are ~ I0 times l arger t han t hose of el ect rons 1_3.72, 73]. However , noise and mul t i pl i cat i on meas ur ement s in l nSb aval anche di odes in the (100) pl ane are consi st ent with el ect ron rat es which are higher t han t hose of hol es and depend onl y weakl y on electric field [3.74]. These few exampl es illustrate the rich var i et y of i oni zat i on phenomena to be found in these mat cr i al systems. Consi der abl e t heoret i cal and exper i ment al wor k r emai ns to form a compl et e char act er i zat i on of aval anche processes in these mat eri al s. Devices have been fabri cat ed in several mat eri al syst ems and an at t empt to review all the publ i shed l i t erat ure woul d be a large task. The fol l owi ng di scussi on of some r epr esent at i ve devices in different mat er i al syst ems i ndi cat es the cur r ent st at e of t he ar t in mat eri al s, st r uct ur e and per f or mance. The devices are or der ed by decreasi ng bandgap, or by i ncreasi ng ma xi mum wavel engt h of efficient oper at i on. Uni f or m aval anche Schot t ky bar r i er phot odi odes have been f or med on Ga As usi ng a thin l ayer of Pt as the bar r i er met al [3.69]. The devices empl oyed ei t her epi t axi al r egr owt h or diffusion t o f or m a guar d ring. Aval anche mul t i pl i cat i on over 100 was obt ai ned and showed good uni formi t y. The noise power vari ed as M 2'1 and was consi st ent with hol e mul t i pl i cat i on great er t han el ect ron mul t i pl i cat i on. The oper at i onal wavel engt h ext ended to 0.9 i.tm due to the band edge of GaAs, but an enhanced response was obt ai ned at t he highest bi as vol t ages due to the F r a n ~ Ke l d y s h effect. At a mul t i pl i cat i on of 100, det ect or pulse response has 0.1 ns rise t i me and 0.3 ns fall time. Wavegui de det ect or s for )o = 1.06 I-tin have been f or med by epi t axi al l ayers of Ga As which serve as wavegui des i-3.75]. Absor pt i on at the l ong wavel engt hs was due to a combi nat i on of el ect r o- absor pt i on and def ect - t o- band absor pt i on. The device showed an i nt ernal qua nt um efficiency of 45 % and had a r esponsi vi t y of about 0 . 3 A - W i a t 1 . 0 6 1 . t m . 84 D.P. Schinke el al. By i ncor por at i ng i ndi um into the GaAs system, the wavelength response can be ext ended to about 2 = 1.15 gm. Schot t ky barri er st ruct ures made from ln~Ga~ xAs alloy layers grown epitaxially on GaAs subst rat e showed high aval anche gain and fast response similar to GaAs [3.76]. It was found t hat for x <0. 08 the strain due to lattice mi smat ch between the epitaxial layer and t he subst rat e remai ned elastic r at her t han being relieved by t he creat i on of dislocations. The consi derat i on of lattice mat chi ng is a limiting fact or in ext endi ng the wavelength response of this system. Ioni zat i on rates have been measured in this system and showed f l / c~2. 5 at E ~ 3 x l 0 s V- c m- 1 [3.77]. The wavelength response of the GaAs system can also be ext ended by alloying the V sublattice with ant i mony. A sensitive GaAs~ _xSbx i nvert ed mesa st ruct ure has been fabri cat ed by liquid phase epi t axy on GaAs [3.66]. This grown het er oj unct i on det ect or was "backsi de" illuminated with light in t he 1.0 to 1.2 gm spectral band passing t hough the t r anspar ent substrate. Il l umi nat i on t hr ough a subst rat e of wi der bandgap produces a det ect or with a response "wi ndow" as a funct i on of wavelength. Short wavelength radi at i on is bl ocked by the substrate, while light with a wavel engt h between absor pt i on edges of the subst rat e and epitaxial l ayer is detected, and l onger wavelengths are lost by t r anspar ency of the entire structure. Dar k current s as low as 1 nA were observed and low capaci t ance ~ 0 . 1 p F i nsured high-speed response. The usable aval anche gain over the entire det ect or area was < 10 due to spatial nonuni formi t i es in the aval anche multiplication. The principle of"sel f-fi l t eri ng" has been ext ended to the 3.1 to 7.0lam spectral region using the lnAs~ .~Sb~- InAs system [3.78]. A mat eri al system t hat can be fabri cat ed with lattice mat chi ng and can provi de high responsi vi t y in the 1.2 to 1. 6gin spectral region is the I nGaAs P/ l nP system. The bandgap of I nGaAsP can be varied cont i nuousl y from 1.34eV (0.921am) to 0. 78eV (1. 6gm) in composi t i ons t hat are lattice mat ched t o l nP [3.79, 80]. Aval anche phot odi odes have been fabri cat ed as mesas et ched in epitaxial layers of Gao.2,~lno.vc~Aso.ssPo.42 grown on InP substrates by liquid phase epitaxy. The mul t i pl yi ng j unct i on was formed in the epitaxial l ayer by diffusion. Low reverse current s of < 10 -8 A were found and some devices showed uni form aval anche gain of 10. The low bias quant um efficiency of the uncoat ed devices was 45% at 2=l . 15! am. Some shift of the absor pt i on edge t owards longer wavel engt hs occurred at high bias voltages pr esumabl y due to the Fr anz- Kel dys h effect. The small det ect or showed rise times of 150 ps which coul d be limited by the rise time of the light source, and longer fall times ~ 1 ns due to the diffusion of a fract i on of the carriers from undepl et ed material. At this early stage of devel opment , the characteristics of efficiency, dark cur r ent and response time are compar abl e to or bet t er t han ger mani um detectors. Ther e are i ndi cat i ons t hat the i oni zat i on rates for electrons and holes are not equal in the I nGaAs P system. I t o et al. [3.81] fabri cat ed a wi ndow st ruct urc l nP/ I nGaAs P/ I nP into a mesa of 140gm dia. The device br eakdown voltage was 52 V and had a dark cur r ent of 0.1 nA a t VBr/2. Thei r measur ement s Phot odet ect or s 85 using incident light from different spectral bands showed t hat c~ was great er t han [~ by a fact or of 3 to 4. The I I I - V compound semi conduct or phot odi odes are an active area of research and devel opment . Subst ant i al quest i ons concerni ng material proper- ties, gr owt h and processing of materials, surface passivation, opt i mal device st ruct ure, and l ong-t erm stability and reliability remai n to be answered before these phot odet ect or s can be included in fiber systems on a rout i ne basis. List of Symbols A Di ode ar ea B Effective bandwi dt h of the recei ver C Di ode capaci t ance d Di ode di amet er d~ Thi ckness of front conl act E~ Bandgap energy l eVI F(M} Excess noi se fact or h Pl anck' s cons t ant l D Phot odi ode dar k cur r ent fat , Compone nt of t he det ect or dark cur r ent whi ch is t o be mul t i pl i ed I,~ Compone nt of t he det ect or dar k cur r ent whi ch is not mul t i pl i ed 1 o Pr i mar y cur r ent in di ode (in aval anche regi on) l~i ~ The pr i mar y signal phot ocur r ent k~,~ Effective rat i o of t he i oni zat i on coef- ficients 7/fl M Aval anche mul t i pl i cat i on value (M 0 is t he dc value) q El ect ron charge R Responsi vi t y, al so l oad resi st ance t'~ Reflectivity of surface t Response t i me t , . Average carri er t ransi t t i me Vb, Reverse br eakdown vol t age W Wi dt h of t he j unct i on depl et i on regi on :~ Abs or pt i on coefficient [ cm ~]; I.c. elec- t r on i oni zat i on coefficient [/ Hol e i oni zat i on coefficient t/ Quant um efficiency ..~ Resistivity of t he dri ft regi on in di ode r Transi t t i me {with subscr i pt lbr appr o- pi l at e di ode region) References 3.1 M. Hor i guchi , H. Os ani : El ect ron. LeU. 12, 310 (1976) 3.2 R. Ol shansky, D.B. Keck: Appl. Opt . 15, 483 (1976) 3.3 H. Mel chi or, M.B. Fi sher, F. R. Ar ams : Proc. I EEE 58, 1466 (1970) 3.4 H. Mel chi or : J. Lumin. 7, 390 (1973) 3.5 H. Mel chi or : in Laser Handbook, ed. by F. T. Arecchi, E. O. Schul z- Duboi s (Elsevier, Nor t h- Hol l and, Amst er dam 1972) 3.6 R, K. Wi l l ardson, A.C. Beer (eds.): Semiconductors and Semimetals, Vols. 5 and 12 (Academi c Press, New York 1970, 1977) 3.7 W. C. Dash, R. Ne wma n: Plays. Rev. 99, 1151 (1955) 3.8 S. M. Sze: Physics of Semiconductor Devices (Wiley, Interscience, New York 1969) p. 661 3.9 RCA Inc. Mont r eal ( Manuf act ur er ' s Dat a) 3.10 P. P. Webb, R. J . Mcl nt yr e, LConr a di : RCA Rev. 35, 234 (t974) 3. ! 1 G. E. Sl i l l man, C. M. Wolfe: In Semiconductors and Semimetals, Vol. [ 2, ed. by R. K. Wi l l ar dson, A.C. Beer (Academi c Press, New York 1977) 3.12 C. A. Lee, R. A. Logan, R. L. Bat dorf, J. J. Kl ei mack, W. Wei gmann : Phys. Rev. 134, A 761 (1964) 86 D.P. Schinke et al. 3.13 R. Van Over st r act en, H. De Man: Solid St at e El ect ron. 13, 583 (1970) 3.14 W, N. Gr a n t : S o l i d St at e El ect ron. 16, 1189 (1973) 3.15 G. A. Bar af f : Phys . Rev. 128, 2507 (1962) 3.16 G. A. Baraff: Phys. Rev. 133, A26 (1964) 3.17 C R. Crowel l , S. M. Sze: Appl. Phys. Lett. 9, 242 (1966) 3.18 A. G. Chynowet h: In Semiconductors and Semimetals, Vol. 4, ed. by R. K. Wi l l ar dson, A.C. Beer (Academi c Press, New York 1968) p. 263 3.19 L. K. Ander son, P. G. McMul l i n, L. A. D' Asar o, A. Goet zber ger : Appl. Plays. Lett. 6, 62 (1965) 3.2(/ A. D. Lucas : Opt o- el ecl r oni cs 6, 153 (1974) 3.21 J. A. Rai nes: 111 Solid State Devices, Proc. 2nd Solid St at e Dev. Reas. Conf. 1972 (hast. Plays., London 1973) p. 225 3.22 K. Ni shi da, T. Takekawa, M. Naki j i ma: App[. Plays. Left. 25, 669 (1974) 3.23 Devices fabri cat ed for commer ci al offeri ng i ncl ude: Texas I nst r ument s Types TI XL 55, 56, 59; Ni ppon Electric Co. APD 200A, 200B; EMI Types $3050 $3054 3.24 Gener al Electric Type 50EHS 3.25 M. V. Schnei der : Bell Syst. Tech. J. 45, 1611 (1966) 3.26 C. T. Wang, S.S. Li: I EEE Trans. ED-20, 522 (1973) 3.27 N. A. Fos s , S. A. Wa r d : J . AppI. Phys. 44, 728 (1973) 3.28 W.T. Read: Bell Syst. Tech. J. 37, 401 (19581 3.29 H. Mel chi or, A. R. Ha r t ma n : I EEE Intern. El ect ron Dev. Meet i ng, Washi ngt on, D.C.(1976), paper 17.5 3.30 1-1, Mel chi or, A. R, Har t man, D.P. Schinke, T. E. Seidel : Bell. Syst. Tech. J. 57, 1791 (1978) 3.31 S. L. Mi l l er : Phys. P, ev. 99, 1234 (1955) 3.32 J. Urge/l, J. R. Laguer r e: Solid St at e El ect ron. 17, 239 (I 974) 3.33 J . Conr adi : Solid St at e El ect ron. 17, 99 (1974) 3.34 R.B. Emmons : J. Appl. Phys. 38, 3705 (1967) 3.35 O. Kr umphol z : AEU (Asia El ect ron. Uni on) 27, 361 (1973) 3.36 K. Bcr cht ol d, O. Kr umphol z, J. Sur i : Appl. Plays. Lett. 26, 585 (1975) 3.37 T. Kaneda, H. Takanashi , H. Ma t s umot o, T. Ya ma oka : J. Appl. Plays. 47, 4960 (1976) 3.38 l I . Takanashi , T. Kaneda, l l . Sei : Fujitsu. Sci. & Tech. J. 10, 119 (1974) 3.39 A. At aman, J. Mul l e r : J . Appl. Phys. 49, 5324 (1978) 3.4(/ J. Mul l er : I EEE Trans. ED-25, 247 (1978) 3.41 l ~, . J. Mcl qt yre: I EEE Trans. ED-13, 164 (1966) 3.42 R_l . Mcl nt yr e : I EEE Trans. ED-19, 703 (1972) 3.43 J . Conr a di : I EEE Trans. ED-19, 713 (1972) 3.44 l l . Knabe, T. Ki mur a, Y. Mi zus hi ma: I EEE Trans. ED-24, 713 (1977) 3.45 K. Ni s hi da: El ect ron. Le~t. 13, 4/ 9 (1977) 3.46 J . J . Goedbl oed, E. T. J . M. Smeet s : El ect ron. Lett. 14, 67 (1978) 3.47 l | . Mel chi or, W.T. Ly n c h : l EEE Trans. ED- 13, 829( 1966) 3.48 R. H. Cox, H. St r ack: Solid St at e El ect ron. 10, 1213 {1967) 3.49 T. Ozeki , E. H. Ha r a : El ect ron. Lett. 12, 80 (1976) 3.50 S.R. Hofst ei n, G. War f i el d: I EEE Trans. ED-12, 66 (1965) 3.51 K. Ni shi da, K. Ishi i , K. Mi nemur a, K. Taguchi : El ect ron. Left. 13, 280 (1977) 3.52 I ) . P. Schi nke: Unpubl i shed results 3.53 H. Osanai , T. Shi oda, T. Mor i yama, S. Araki , M. Hor i guchi , T. I zawa, H. Takat a : El ect ron. Lett. 12, 549 (1976) 3.54 D. N. Payne, W. A. Ga mbl i ng: El ect ron. Lctt. I1, 176 (1975) 3,55 J. Conr adi , F. P. Kapr on, J . C. Dyme nt : I EEE Trans. ED-25, 180 (1978) 3.56 T. Ki mura, K. Da i koku: Opt . and Qua nl um El ect ron. 9, 33 (1977) 3.57 .1. Yamada, M. Sar uwat ar i , K. Asat ani , I-I.Tsuchiya, A. Kawana, K. Sugi yama, T. Ki mur a : I EEE J. QE-14, 791 (1978) 3.58 P. R. Brat t : In Semiconductors and Semimetats, Vol. 12, ed. by R. K. Wi l l ar dson, A.C. Beer ( Academi c Press, New Yor k 1977) p. 39 3.59 P. W. Kruse, L. D. McGl auchl i n, R.B. McQui st an : Elements ~[" h![i'ared Technology (Wiley, New York 1962) Photodetectors 87 3.60 .I. Conrad( : Appl. Opt. 14, 1948 (1975) 3.61 T.,Shibala, Y.lgarashi, K. Yano: Rev, Eleclr. Commun. Lab. 22, 1069 (1974) 3.62 H. Ando, H. Knabe, T. Ki mura, T. Yamaoka, T. Kaneda: IEEE J. QE-14, 804 (1978) 3.63 K. Nagai : E.C.L.Tech. J. NTT, Jpn. 20, 927 (1971) 3.64 T. Yashiro, K.Nagai, K. Yano : E.C.L.Tech. J. NTT, .Ipn. 20, 949 (1971) 3.65 C.A. Van' t Hof, T. Bakker: AppI. Sci. Res. 29, 211 (1974) 3.66 R. C. Eden: Proc. IEEE 63, 32 (1975) 3.67 R. A. Logan, S. M. Sze: J. Phys. Soc. Jpn. Suppl. 21, 434(1966) 3.68 G. E. Stillman, C.M.Wolfe, J.A. Rossi, A.G. Foyt : Appl. Plays. Lett. 24, 471 ([974) 3.69 G.E. Stillman, C.M.Wolfe, A.G. Foyl. W.T. IJindley : Appl. I'hys. Letl. 24, 8 (1974) 3.70 T.P.Pearsall, R. E. Nahory, J. R. Chel i kowsky: Phys. Rev. Lell. 39, 295 (1977) 3.71 T.P.Pearsall, R. E. Nahory, M. A. Pol l ack: Appl. Phys. Left. 28, 403 (1976) 3.72 M. P. Mi khai l ova, N. N. Smi rnova, S. V. Slobodchikov : Soy. Phys. Semicond. 10, 509 (1976) 3.73 M.P. Mikhailova, S.V.Slobodchikov, N. N. Smirnova, G. M. Filaretova :Sov. Phys, Semicond. 10, 578 (1976) 3.74 R.D. Baertsch: J. Appl. Plays. 38, 4267 (1967) 3.75 K.H.Niehols, W. S. C. Chang, C. M. Wol fe: Appl. Phys. Lelt. 31,631 (1977) 3.76 G.E. Stillman, C. M. Wolfe, 1. Melngailis : Appl. Phys. Lett. 25, 36 (1974) 3.77 T.P.Pearsall, R. E. Nahory, M.A.Pollack : Appl. Plays. Lelt. 27, 330 (1975) 3.78 D.T.Cheung, A. M. Andrews, E. R. Gert ner, G. M. Williams, J.E.Clarke, J. G. Pasko, J.T. Longo: Appl. Phys. Lett. 30, 587 {1977) 3.79 C.E.Hurwitz, J.J. Hsieh: Appl. Plays. Lell.. 32, 487 (1978) 3.80 R. E. Nahory, M.A.Pollack, W. D. Johnson, Jr., R.L. Barns: Appl. Phys. Lett. 33, 659 (1978) 3.81 M. Ito, T. Kaneda, K. Nakajima, Y.Toyoma, T. Yamoka, T. Kot ani : Electroll. Lett. 14, 418 (1978) 4. Receiver Design for Optical Fiber Communication Systems R. G. Smi t h and S. D. Per soni ck Wi t h 40 Fi gur es The pur pos e of t hi s chapt er is to pr ovi de t he r eader wi t h a basi c unde r s t a ndi ng of t he opt i cal recei ver mad t he i nt er pl ay bet ween t he c o mp o n e n t s of t he recei ver as well as t he i nfl uence of t he sour ce a nd t r ans mi s s i on medi um, The a ppr oa c h t aken will be t o pr esent t he mat er i al in a s t r ai ght f or war d engi neer i ng ma nne r , a nd t o avoi d, wher e possi bl e, t he us e of l ong ma t he ma t i c a l der i vat i ons. In t hi s way it is hope d t hat an i nt ui t i ve feel for recei ver desi gn will be gai ned. Thos e r eader s i nt er cst ed in t he ma t he ma t i c a l det ai l s will find t he m in t he references cited. 4. 1 I nt r o duc t o r y Re ma r k s 4.1.1 General Considerations In the design of an opt i cal fiber communi cat i on syst em, whet her for use in l ong di st ance communi cat i on [4,1 8] or for bussi ng of dat a over shor t di st ances, [ 4. 9- 12] and whet her oper at i ng at low or high dat a rates, one of the key el ement s of the syst em is t he receiver. The basi c pur pos e of t he recei ver is to det ect the light i nci dent upon it and to conver t it to an electrical signal cont ai ni ng the i nf or mat i on i mpressed on the light at the t r ansmi t t i ng end. The receiver is t hus an opt i cal to electrical conver t er or O/ E t ransducer. In t he same way the t r ansmi t t er funct i ons as an E/ O t ransducer. The opt i cal receiver, to be descri bed in this chapt er, consi st s of a phot ode- t ect or and an associ at ed ampl i fi er al ong wi t h necessary filtering. The funct i on of the phot odet ect or is to det ect the i nci dent light signal and conver t it i nt o an electrical cur r ent ; t he ampl i f i er conver t s this cur r ent i nt o a usabl e signal while i nt r oduci ng the mi ni mum a mount of addi t i onal noise t o cor r upt t he signal. The f undament al goal in the design of an opt i cal receiver is to mi ni mi ze the a mount of opt i cal power which must reach t he recei ver in or der t o achi eve a given bit er r or rat c (BER) in digital syst ems or a given signal to noise rat i o (S/N) in an anal og system. Thi s power, c ommonl y referred to as the sensitivity, and usual l y measur ed in dBm ( 0 d Bm= 10 - 3 W) of opt i cal power, depends upon the det ect or t ype and charact eri st i cs - ei t her p-i-n or aval anche phot o- di ode - as well as the design of the amplifier. Wher eas achi evi ng opt i mum sensi t i vi t y in the design of a receiver is the basi c goal, ot her consi der at i ons influence the details of the design, and in fact may result in a pract i cal receiver wi t h less t han the opt i mum in sensitivity. Exampl es of such consi der at i ons include achi evi ng a wide dynami c range, 90 R. G. Smith and S. D. Personick Fig. 4.1. Block divgram of optical fiber receiver oper at i on from specified, and perhaps nonopt i mum power suppl y voltages, achieving a manuf act ur abl e and cost effective design, and the design of a "uni versal " receiver which mi ght be opt i mum at one bit rate and nonopt i mum at others. In or der to assess the t radeoffs involved in the design of an opt i cal receiver, it is first necessary to undcr st and t hebas i cs of the oper at i on of such a receiver Thi s requires a knowl edge of the sources of receiver noise and their minimi- zat i on, the role played by the det ect or, the effect played by the source and the t ransmi ssi on medi um, and thc i nt erpl ay between all of them. 4.1.2 Scope and Organization This chapt er is organi zed i nt o six sections as follows: Sect. 4.2 i nt roduces the basic receiver and present a discussion of tile vari ous component s of which it is comprised. Next, the general quest i on of noise sources is i nt r oduced and t hei r cont r i but i ons to system noise evaluated. In Sect. 4.3 this general noise analysis is appl i ed specifically to the cases of FET and bi pol ar i nput transistors. Section 4.4 deals with the opt i mi zat i on of receiver design and will i nt r oduce t he "hi gh i mpedance, " or "i nt egrat i ng" amplifier appr oach as well as the use of t ransi m- pedance amplifiers. Fol l owi ng these discussions of the noise charact eri st i cs of the receiver, Sect. 4.5 deals with t he sensitivity of t he receiver using bot h p-i-n and aval anche phot odi odes as det ect ors, and presents experi ment al as well as t heoret i cal results on receiver sensitivity. Section 4.6 deals with t he quest i on of the effects of the t ransmi ssi on medi um and t he source on receiver performance. Finally a bri ef account of anal og systems is prese.ated in Sect. 4.7. Thr oughout this chapt er the emphasis is placed on t he propert i es of the amplifier and the ampl i f i er - det ect or combi nat i on. The details of the det ect or can be found in Chap. 3. In addi t i on to the references cited above t he r eader is referred to [4.13, 14] for ot her reviews on the subject of receivers and to [4. 15-17] for related discussions of fiber systems. 4. 2 The Basi c Recei ver 4.2.1 Essential Components The basic optical receiver is shown in Fig. 4.1 and consists of a phot odet ect or , a low noise preampl i fi er and post amplifier, an equal i zer and a filter. A receiver of this confi gurat i on woul d be useful for bot h anal og as well as digital appli- L " V B I A S Receiver Design for Optical Fiber Communication Systems 91 PREAMP R L Fig. 4.2. Schcnmtic representation and equiwdent circuit of photodetector and bias circuit cations. In the case of a digital system an addi t i onal function, t hat of a deci der or decision circuit woul d be required. Such a circuit funct i ons at high signal levels and does not influence the basic amplifier design and thus will be omi t t ed f r om consi derat i on at this time. That por t i on of the receiver shown in the figure is referred to as the linear channel. Photodetector In most , if not all, opt i cal fiber systems, t he phot odet ect or is a p-i-n or aval anche phot odi ode, al t hough in pri nci pl e a phot ot r ansi st or or a phot o- multiplier coul d be used. At the present time these phot odi odes are made from silicon which provi des an excellent spectral mat ch to the emission fi'om sources made from t he AI~Ga~ xAs system [4.18, 19]. In the future, when fiber systems move to the region of 1.2 1.6~un, these det ect ors will have to be made from ot her materials, e.g., Ge [4.20], GaA1AsSb [4.21], GaAI Sb [4.22], I nGaAs P [4.23] and Gal nAs [4.24, 25], which absor b in this spectral region. See Chap. 3 for details on phot odet ect or s. In oper at i on these detectors, which are reverse biased, absor b the incident radi at i on, generat i ng el ect r on- hol e pairs which drift t o t he det ect or el ect rodes generat i ng a current in the ext ernal circuit. An excellent equi val ent circuit of the det ect or shown in Fig. 4.2 is a current source shunt ed by the depl et i on capaci t ance, C a , of the det ect or. Series resistance and shunt conduct ance can also be added to t he equi val ent circuit but bot h are small, havi ng negligible effect on performance, and are omi t t ed from the equi val ent circuit for convenience. 92 R. G. Smith and S. D. Personick For a p-i-n phot odi odc the current , i o, generat ed by the incident light of power, P, is given by ! l q p (4.1) 1 = hv where q is the quant um efficiency of t he di ode and hv/q is the phot on energy in el ect ron volts. For a wavelength of 825nm, hv/q= 1.5eV. At this wavelength and for a det ect or with r/ =0. 75, i 0 =0. 5 P. The responsivity, R o, of the diode, defined as the current pr oduced per unit optical power incident, is given by qq R - hv =0. 805 q2 [ gm] (4.2) and for the above exampl e Ro=0. 5( A. W- ~) . Typi cal values for the rc- sponsivity at 2 = 825 nm are 0.4 t o 0.6 A- W- The aval anche phot odi ode multiplies the pr i mar y current , i o, by a fact or ( M) , where <M) is the average val ue of the gain (which is a r andom variable). In this case the cur r ent source in Fig. 4.2 has a value i =(M)i o=(M) ~q P=<M)RoP. (4.3) P r e a m p l i f i e r a nd A m p l i f i e r The phot ocur r ent generat ed by t he det ect or must be conver t ed to a usable signal for furt her processing with the mi ni mum amount of noise added. The preampl i fi er is here defined as the first stage or stages of ampl i fi cat i on following t he phot odet ect or . It will thus be the domi nant source of noise added to the signal and hence its design will be the principal fact or in det ermi ni ng t he sensitivity of the receiver. The bulk of this chapt er deals with the analysis of this part of the receiver and with the mi ni mi zat i on of the noise. The postamplifier, shown in Fig. 4.2, provi des the r emai nder of the amplifi- cat i on of the signal. It may precede or follow the equal i zer or may be i ncor por at ed into the funct i ons of the equal i zer and filter. Because it will not cont r i but e significant noise in a well-designed system, it is shown schemat i cal l y as appear i ng between the preamplifier and equalizer. In practice the post ampl i fi er may cont ai n features such as aut omat i c level cont rol to keep the out put signal level const ant as well as cl ampi ng ci rcui t ry to reference the signal to part i cul ar vol t age levels. The design of this part of t he linear channel, al t hough not discussed here, is nevertheless ext remel y i mpor t ant to the funct i on of an opt i cal receiver [4.5, 8]. Receiver l-)esign for Optical Fiber Conlmunication Systems 93 is{t)+ i DETR I BIAS CKT POSTA M P EQL F LT i a PREAMP AMPLR V s ( t ) Fig. 4,3. Equivalent SIGNALOUT circuit of front end Equalizer As will be seen later, the t ransfer funct i on of the pr eampl i f i er - post ampl i f i er combi nat i on may be such t hat the signal waveshape is di st ort ed, or the incident signal may be di st ort ed due t o dispersive effects in t he fiber medi um. The role of the equalizer, Fig. 4.2, is t o r emove these (linear) signal di st ort i ons, provi di ng a reasonabl e pulse shape t o the final filter. In actual fact the equal i zer and filter may compri se a single net work, but for purposes of this discussion, these funct i ons are separated. Filter The final el ement in the receiver is the filter. It will oper at e on bot h the signal and the noise sources, and its mai n funct i on will be t o maxi mi ze t he signal-to- noise rat i o while preservi ng essential features of the signal. For anal og systems the requi rement s of the filter may be to hol d ampl i t ude and phase responses within certain limits, while in digital systems it may be to minimize i nt ersymbol interference. Wi t hi n these and perhaps addi t i onal const rai nt s, t he filter will be chosen to reduce the amount of i n-band and out - of - band noise. As will be seen, the requi red filter funct i on will be defined in t erms of t he i nput pulse shape, the desired out put response, expressed ei t her in the time or frequency domai n, and t he frequency response of the rest of the linear channel. The remai nder of this section deals with the eval uat i on of the signal and the noise as functions of the par amet er s of the linear channel. 4.2.2 Noise Analysis We now consi der the linear channel in mor e detail, eval uat i ng the effect of the noise sources on the overall system noise. The upper por t i on of Fig. 4.3 shows the front end in mor e detail, including t he vari ous noise sources; the l ower por t i on shows the r emai nder of the linear channel. The signal cur r ent i j t ) generat ed by the incident light is i ndi cat ed at the left. The capaci t ance C a is the depl et i on capaci t ance of the det ect or and C, is the st ray capaci t ance associated 94 R. G. Smith and S. D. Personick with the i nt er connect i on of the det ect or to the i nput of the amplifier. The resistor R e represents the resistive por t i on of the circuit used to bias the det ect or and the i nput transistor. The generat ors i a, i 1, i, and e~ charact eri ze the vari ous sources of noise in t he system as discussed below. The preamplifier por t i on of the linear channel shown on the ri ght -hand side of Fig. 4.3 is model ed as an ideal, noise-free amplifier with shunt and series noise sources i~ and e,, respectively, and an i nput admi t t ance Y.,. The t ot al i nput admi t t ance is thus given by ! Y i , , ( " ) ) = Y , ( t ' ) ) + R, +J)(Ca + C~). l = Rin - Fj oJCT, (4.4) where Ri~ j is the resistive por t i on of the admi t t ance and C T is the t ot al capaci t ance including cont r i but i ons from the det ect or, amplifier and parasitics. Let the t ransfer funct i ons of the preamplifier and post ampl i fi er be A a ( u J ) and A 2(o)), respectively, and t hose of t he equal i zer and filter be E(~)) and F(/~)). The signal voltage at the out put of the linear channel, V~(c~), is then given by E( ( ' ) ) = Zrr((' )) I & ) ) , ( 4 . 5 ) where ZT((~)) is the transfer funct i on of the system defined by, A 1 ( o ) ) A 2 ( o ) ) E ( ~ ) F ( o : O ZT(~)) = (4.6) and I ~ ( ~ o ) is t he Four i er t r ansf or m of the signal cur r ent i,(t). The noise sources included in this model include cont r i but i ons f r om the cur r ent flowing in the phot odet ect or , i d, t hermal noise associated with the biasing resistors, ij, and noise sources of the preamplifier i, and e,. Noi se sources associated with subsequent por t i ons of t he linear channel are assumed to be small and are neglected. The sources described here fall i nt o t wo general cat egori es: Those which depend on the incident optical signal and t hose which are i ndependent of the signal level. The latter includes t he cont r i but i ons of the amplifier (neglecting nonlinearities) and the biasing circuit as well as signal i ndependent current s flowing in the det ect or, i.e., dar k c u r r e n t s ; t h e f or mer involves the phot o- generat ed current flowing in the det ect or in response to the incident optical signal which will be time varying. We consi der this noise source first. The reader who is not interested in the details of the deri vat i on may refer to (4.46-52) where the signal and cont r i but i ons of the noise sources are sum- mari zed in t erms of i nput signal and noise current s and to Figs. 4. 6-17 where the integrals and sums involved are plotted. Receiver Design for Optical Fiber Communication Systems 95 Photodetector Noise The si gnal - dependent noise in the phot odet ect or results froltl the r andomnes s associ at ed wi t h the rat e of arri val of the phot ons at t he det ect or, and when an aval anche phot odet ect or in used, t he r a ndom nat ure of t he aval anche process i nt r oduces addi t i onal fl uct uat i ons in the signal cur r ent which ent ers the ampl i fi er [4.18, 26] 1. We begin by consi deri ng a digital syst em with a si gnal i ng rat e B and a t i me slot of wi dt h T - B - 1 . At each t i me slot t he t r ansmi t t ed opt i cal power is assumed to t ake on one of t wo discrete values with a pulse shape h p ( t ) . T h e opt i cal power i mpi ngi ng on the det ect or is t hen of the f or m P( t ) = ~ bkhp(t -kT), (4.7) k - c o where k is a vari abl e denot i ng tile t i me slot and b k t akes on one of t wo di scret e values, b(0) and b(1) where it is assumed b( 1) > b(0). 2 The pr i mar y phot ocur r ent gener at ed by t he phot odet ect or in r esponse to the opt i cal power consi st s of a series of pulses each due to the gener at i on of an el ect r on- hol e pai r by an abs or bed phot on. Wi t hi n a given t i me slot the numbe r of pr i mar y el ect r on- hol e pai rs is a r a ndom vari abl e with a Poi sson di st ri but i on char act er i zed by a mean val ue pr opor t i onal to the aver age incident power. When an aval anche phot ode- t ect or is empl oyed, each pr i mar y el ect r on- hol e pai r is mul t i pl i ed by the aval anche process to pr oduce M s econdar y pairs. The aval anche gai n M is al so a r a ndom vari abl e with a mean val ue <M) and a mean squar e val ue <M2). When a p-i-n det ect or is used, <M) = 1 and <M 2) = 1. Each abs or bed phot on t hus pr oduces a pulse of char ge Mq which ent ers the pr eampl i f i er and cont r i but es t o the out put vol t age. The out put vol t age is t hen given by t he sum of all t he i nput pulses, modi fi ed by t he t ransfer funct i on of t he l i near channel . The out put vol t age will t hen be char act er i zed by an average, or expect ed val ue as well as fl uct uat i ons f r om this average, resul t i ng f r om the r andomnes s of the arri val rat e of the phot ons as well as the addi t i onal r andomnes s i nt r oduced by the aval anche process. Omi t t i ng the s omewhat l engt hy de- ri vat i on found in [4.26] the aver age val ue of t he out put vol t age resul t i ng f r om an i nci dent power P(t) is given by (4.8) where ZT(t ) is the i mpul se r esponse of the syst em given by t he Four i er t r ansf or m of ZT(~O) and I 0 is t hat por t i on of the phot ocur r ent not di rect l y rel at ed to the incident signal, i.e., dar k cur r ent or phot o- gener at ed current resul t i ng f r om st r ay light, and referred to as the dar k current . I The analysis below follows closely that presented in I-4.26]. 2 In this and similar expressions kT is not to be confused with the Boltzmann energy. 9 6 R. G. S mi t h a n d S. D. P e r s o n i c k The fl uct uat i on in the out put voltage or the noise resulting from the di ode current is dcfined by < , @ ) > = < ( v ~ ( t ) - < v ~ ( t ) ) ) : > = < v ~( 1) > - ( < , ~ ( t ) > ) : (4.9) and in t erms of the par amet er s of the system (4.10) The expression for this por t i on of the noise can be written in terms of the Four i er t ransforms as follows" (2n) z _,~, [ \ h v J P ( w ) + t0(w ) [ZT(W )*ZT(w)] e3' ~' dw, ( 4 . 1 1 ) where ZT(~J)) has been defined before, P(w) is the Four i er t r ansf or m of P( t ) , I o( o) ) is the Four i er t r ansf or m of the dar k current , assumed to be const ant in time. The quant i t y Z-r(~n)*ZT(O)) is the convol ut i on of ZT(W) defined by Z l , ( ( t ) ) , Z T ( ( t , ) ) = Z T ( ( D ' ) Z T ( ( D - - ( I f ) d w ' (4.12) Consi der now the cont r i but i on to ( n ~ ( t ) ) resulting from the const ant dar k cur r ent I o ( t ) = I o. Subst i t ut i ng I0((~) ) = I o 2n6(w) i nt o (4.12), using t he definition of the convol ut i on, and not i ng t hat Z T ( - - w ) = Z * ( w ), this noise t erm becomes <n2)a~k . . . . . . t = q l o ( M 2 ) i s 2 d w = ' i 2 q l o ( M a ) [Z,r(~))l 2 d r , 0 where (4.13) w=2 x f . In the last expression use has been made of the fact t hat IZT(W)I 2 is an even funct i on of w in or der to change the limits of integration. Equat i on (4.13) is a st at ement of the familiar relation t hat the mean square out put noise is given by the pr oduct of the mean square i nput noise current densi t y and the integral of Receiver Design for Optical Fiber Communication Systems 97 the squared magni t ude of t he t ransfer funct i on of the system. The i nput noise cur r ent density, given by d ( i Z ) - 2 q l o ( M 2 ) . (4.14) 4f is recogni zed as the familiar shot noise expressi on modified by t he mean square aval anche gain [4.27]. For conveni ence it has been assumed t hat the full dar k current undergoes aval anche gain. In practice a por t i on of t he cur r ent may not flow t hr ough the aval anche region and under go gain. For this por t i on of the current , { M 2 } = 1. De n o t i n g I m and I n as t hose por t i ons of the dar k current which are multiplied and not multiplied, respectively, the expressi on for the i nput noise cur r ent densi t y becomes d( i2} = 2q[lm( M 2 } + I .]. ( 4 . 1 5 ) ~f Amplifier and Circuit Noise The cont r i but i ons to the out put noise f r om the amplifier and bias circuit are descri bed by the shunt noise current and series noise vol t age generat ors depi ct ed in Fig. 4.3. The shunt cur r ent generat ors associ at ed with t he bias circuit and the amplifier are connect ed to t he same node and may be l umped t oget her i nt o an equi val ent shunt cur r ent gener at or with a noise spectral densi t y given by d ( i 2 ) ~ q _ 4 k T d ( i ~ ) + (4.16) d f R df The out put noi se resulting from these noise sources is given by o df IZv())lEdf' (4.17) where we have explicitly i ncl uded t he frequency dependence of d ( i 2 ( m ) ) / d f t o account for the possibility t hat the noise does not have a white spect rum. When d ( i Z ( c o ) } / d f is const ant or t he vari at i on of ( i z ( c o ) } is small over tile range of ]Zv(OJ)] 2, d ( i 2 > e q 0o ( n 2 ( t ) ) ~ , . , . , - j " IZT(~O)t2df. (4.18) d f o We will discuss this rel at i on in mor e detail later. 98 R. G. Smith and S. D. Personick The out put noise due to the series noise voltage generat or is given by d < e ~ ( , , ) ) ) < 1 1 2 > s e r ie s = i! d f ] Yi,(c'DI 2 [ Z T ( ( t ) ) [ 2 d f . (4.19) Substituting from (4.4) for Y~,, (4.19) becomes i ~I<<+~> . . . . 4f i n o d<e~> .... +(27CCT)2 d f ! f a l Z v ( m ) 1 2 d f ' (4.20) where wc have made the simplifying assumpt i on t hat the noise spectral density d < ~ 7 ,( o D > / 4 ] is independent of frequency over the range of interest which is generally the case. When this assumpt i on is not true the explicit frequency dependence can be included within the integral. Examining (4.20) it is seen t hat the series noise generat or contributes two terms to the out put noise voltage, one proport i onal to the bandwi dt h of ZT(O)) and the second proport i onal to the cube of the bandwi dt h. This latter term will turn out to be the limiting factor in the opt i mi zat i on of optical receivers. T.he total mean square out put noise, assumi ng the noise sources are uncorrelated, is thus < n 2 ( t ) > ~ - < # ] 2 > s l l u n t - I f - < # ] 2 > s e r l e s (4.21) and expressions for each of these terms are found above. The time dependence of the noise, implicit in the above expressions, results from the fact t hat <n2(t)) and hence < h a ( t ) ) depends on the incident light level; < l ' / 2 > s h u n ! a n d <n2)series will generally be i ndependent of time. No r ma l i z a t i o n The preceding expressions for the signal and the noise depend explicitly on the transfer function of the system, ZT(~) ). Thus far not hi ng has been said concerning its properties. We begin by returning to (4.7) for the optical power P(t) incident on the detector. The pulse shape h p ( t ) is normalized such t hat c~, T __~s h p ( t ) d t = 1. (4.22) With this normal i zat i on the energy associated with a given received pulse is b k T where bk=b(0) or b(l), the two discrete states. The quant i t y b has the Receiver Design for Optical Fiber Communication Systems 99 d i me n s i o n s o f p o we r wi t h t hi s n o r ma l i z a t i o n . ~ I f t h e s t a t e 0 is t r a n s mi t t e d wi t h p r o b a b i l i t y p(O) a n d s t a t e 1 wi t h p r o b a b i l i t y p ( l ) t h e n t h e a v e r a g e o p t i c a l p o we r is gi ve n b y /5 = b(O) p(O) + b(1)p(1). (4. 23) Wh e n t h e t wo s t a t e s a r e e q u a l l y p r o b a b l e , as is g e n e r a l l y t h e c a s e in c o mmu n i c a t i o n l i nks, /5= 1/2[b(O)+b(1)], p ( 0 ) = p ( 1 ) = 1/2 (4. 24) a n d f u r t h e r wh e n b ( 0 ) = 0, / 5 _ b(1) ~ , b ( 0 ) = 0 . (4. 25) I n r e s p o n s e t o t h e i n p u t p o we r P( t ) , (4.7), t h e o u t p u t v o l t a g e will be o f t h e f o r m v j t ) = ~ s k h o u t ( t - k T ) + n o i s e , (4. 26) wh e r e s k t a k e s o n o n e o f t wo d i s c r e t e va l ue s , s(0) o r s(1). T h e s h a p e o f t h e o u t p u t p u l s e hou,(t ) is c h o s e n s uc h t h a t its m a x i m u m va l ue , o c c u r r i n g a t t = 0, is uni t y, i . e . , ho. , (0) = 1. (4. 27) Us i n g t hi s n o r ma l i z a t i o n t h e e x p e c t e d o r a v e r a g e v a l u e o f t h e o u t p u t s i gnal a t t = k T d u e t o a si ngl e i n p u t p u l s e is e i t h e r s(0) o r s( l ) . Le t Hp((O) a n d Hou, (. ) ) b e t h e F o u r i e r t r a n s f o r l n s o f h p ( t ) a n d ho.,(t). r e s p e c t i v e l y , a n d de f i ne ZT((O ) = R T HT(~O ) (4. 28) wh e r e R T is a c o n s t a n t wi t h d i me n s i o n s o f r e s i s t a n c e a n d H,r((O ) c o n t a i n s al l t he f r e q u e n c y d e p e n d e n c e o f ZT(~)). Us i n g t h e d e f i n i t i o n s o f t h e v a r i o u s q u a n t i t i e s , 3 This choice of normalization differs from that used in [4.26]. 100 R. G. Smith and S. D. Personick equat i on (4.5) for the response to a single i nput pulse reduces to . s 'H , , , , ( c ') ) = ( M ) (~Tqv) b R T H r ( o g ) H p ( o ') . Defining (4.29) OF (4.29) reduces to s = ( M ) ( ~ v , ) R T b (4.31) and relates t he magni t ude of the peak out put vol t age to the i nput power. Wi t h t he definition of ZT(OJ), (4.28), and HT(~)), (4.30), the noise sources given by (4.1 l, 18, 20) can be evaluated. As HT({o ) is defined above, t he vari ous integrals will explicitly depend on the width of the time slot T. Thi s dependence can be fact ored out in a simple manner by defining a normal i zed frequency variable f {~) o) T Y = B = ~2~B - 2r: (4.32) 2~y T With this change of variables we can define t wo new functions, I H ( 2 r c y l ( 4 . 3 3 ) 1 H [2~Y 1 The t ransfer function is then given by H:r(y ) - (4.34) (4.35) HT(O) ) (4.30) Receiver Design for Optical Fiber Communication Systems 101 With these definitions the dar k current , shunt and series noise cont r i but i ons are given by ~ 2 ( "Is ) d a r k . . . . . . t = 2q[lm( M2 ) + I~] R2BI2 ( 4 . 3 6 ) ( l 1 2 ) s h u n t - d(ia)eq R 2 B I , ( 4 . 3 7 ) df ( H 2 ) s e r i c s d(l?2> g 2 = d f ~ 2 B 1 2 ( 4 . 3 8 ) g i n d ,, ~2,, + ( 2 X C T ) Z ~ R Z B 3 1 3 , aj where the 12 and 13 are definite integrals given by co 12 = .[ [H'T(y)12dy ( 4 . 3 9 ) 0 13 = IH'w(y)[2 y2 d y . (4.40) 0 Because of the nor mal i zat i on of H:r(y) the values of these integrals depend onl y on the relative shapes of the i nput and out put pulses: the dependence on the time slot width T - B t is cont ai ned in t he B and B 3 coefficients in (4.36 38). The noise ill the out put at time t due t o the incident optical signal P(t ), is f ound by subst i t ut i ng (4.7) i nt o (4.11), performi ng t he requi red Four i er t ransforms, and using the definitions of the vari ous quant i t i es to give e x p E O 2 = y / r l ( , - k r ) l H ; ( y ) E n ( y ) . u . O l d ? } . (4.41) Thi s noise source is pr opor t i onal t o B as are the ot her shunt cur r ent noise terms, and it is seen to depend in a detailed manner not onl y on the value of the received pulse within t he time slot of interest but it cont ai ns cont r i but i ons f i om all ot her time slots weighted by the t erm in braces. It is conveni ent to separat e the summat i on over k i nt o t wo part s : t he first part being the cont r i but i on from the signal in the time slot in quest i on, and the second part the cont r i but i on due to t he signal in all ot her time slots. The first term, eval uat ed at t =0, the poi nt of maxi mum signal, is [ nq~.M ' ,,O 2B , h < 1 1 2 ( 0 ) > = ./-,q / ] ~ V ) k, ~ 2 1 % T I 1 v , ( 4 . 4 2 ) 102 R. G. Smith a n d S. D. Personick where b is the value of the received signal, b=b(0) or b(l), and 11 is given by I, = Re j" Hv(y ) [H'T(y )* H'r(y) ] dy, (4.43) 0 where Re indicates t aki ng the real part. This expression is seen to be similar in form to the expression for the dark current noise with the pri mary current being given by (nq/hv)b but with a weighting factor I~. As in the case of the definite integrals 12 and 13, 11 depends onl y on the relative shapes of the input and out put pulses. The remaining term, resulting from the cont ri but i ons from the other time slots, cannot be evaluated wi t hout knowledge of all the b k. The worst case condi t i on can be found if it is assumed t hat in all other time slots the maxi mum value of b, b,,,~ x is present. In this case the worst case noise at t =0 is given by (4.44) where ~ t = l / 2 ~ , H ' p ( k ) [ H ; r ( k ) , H : t , ( k ) ] (4.45) k = - - ,x, and I~ is defined above. In summary, the maxi mum signal and the noise terms at the out put are given by s = ( M ) [ ' l q ] R T b ( 4 . 3 1 ) \ / ~ , , / . f q\ = zq t w] ( M 2) R Bl xb (due to signal in time slot) (4.42) (4.44) (4.36) (4.37) ( qql ( M2 ) R-~Bb . . . . (~., - Ix) < , , ~ ( o ) ) . . . . . , . . . . = 2 q \ h ~ q (due to signals in other time slots) (n2(O)),,;,,-k ........ t = 2q [ I , , ( M 2) + 1,3 R~ BI 2 d(iZ)~q RZBI 2 ('2(O))~h"'-- 4f ( / 1 2 ( O ) ) s e r i c s _ _ d ( c 2 ) 2 BI2 B313 l (![" RT ~ i 2 +(2rtCv)2 (4.38) ~ n ' ( ) ) t o t a l = ~ t 7 2 ( 0 ) ) s i g n a l -~- ( ] ' l s ' ) d a r k . . . . . . t + (n2)~hu,u + (nZ)~ri,~ . ( 4 . 2 1 ) Receiver Design for Optical Fiber Communication Systems 103 Fr om these expressions and the knowl edge of the coefficients of the vari ous noise t erms the per f or mance of the receiver can be evaluated. The signal and noise quant i t i es have been expressed in terms of t hei r values at t he out put of t he linear channel. Since t he signal is an i nput current it is conveni ent to define these quant i t i es at the input r at her t han at the out put . Because I t , 12,13 and ~ depend onl y on the pulse shapes, the onl y par amet er of the linear channel remai ni ng in the above definitions is R T. Since the signal and rms noise are bot h pr opor t i onal to R T it can be fact ored out of all the expressions yielding equi val ent i n p u t signal and noise currents. The equi val ent i nput signal cur r ent is t hen Isle, = ( M ) ~hv] b (4.46) and t he vari ous noise t er ms referred t o t he i nput are given by ( i 2 ( 0 ) ) = 2 q ( 7 q ] ( M 2 > b l , B (due to signal in time slot) \:Tvj (4.47) (due to signals in ot her time slots) (4.48) i" - 2 q [ l m ( M 2 ) + I , ] B I 2 < , , ) , , ~ r ~ . . . . . . , - - (4.49) 2 -- d(i2)eq (1 >.hun, d f B I 2 (4.50) d < e ~ ) [ B I 2 + ( 2 n C T ) 2 B 3 1 3 ] (4.51) f f i z ( 0 ) ) t o t . ] = < i ~ ( 0 ) ) i 2 ~ . + ( s >dark . . . . . . t + (/' >shum + (i2>series (4.52) V a l u e s o f I i , 1 2 , 1 9 , The values of the definite integrals 11, 12, 13 and the sum ~1 have been eval uat ed by P e r s o n i c k [4.26] for several i nput and out put waveforms and these results are summari zed here. In choosi ng the shape of the out put pulse it is desirable to define a funct i on which will have its maxi mum val ue at the cent er of the time slot in quest i on and have zero value at the cent er of all ot her time slots. In this way the out put signal, but not necessarily the noise, at t he cent er of the kth time slot is due onl y to the i nput power in the kth time slot. This is anot her way of saying t hat for the assumed pulse shape t he i nt er symbol interference is zero. 104 R. G. Smith and S. D. Personick : H ~ ) u T ( Y } l / ./3 = O.1 ~.-~ , , ~ . , , . s i n ( ' n ' t / T ) c o s ( ~ r l ~ | / T ) 0 . 3 ~ / 7 I ' ~ . ~ h O O T " ' = / r | , T [ , _ ( 2 / ~ t / T ) 2 ] , . o f [ \ i - ~ = = ' - " ~ r - 2 ~ - - i " I ~ " - z - - " v ~ ' T 3 y " ~ " - ' ~ T 1 h o g T ( t ) 0'30.1 ",.J,', I / 7 / Vi~. 4.4. Frequency domain, time domain and eye " K~ diagram representations of raised cosine family I (Copyright 1973 (1978) AT&T Co. Reprinted by W O R S T - C A S E E Y E D I A G R A M S permission from Bell Syst. Tech.) One class of function which satisfies the above criterion and which also tends to minimize the noise is given by the "raised cosine" family defined by (4.53) for which the normalized fourier transform is given by (1 - f l ) H , , , , , ( y ) = J 0 < lY[ < =1/ 2 1- s i n roy 2~j) l #- , l - # l+# 2 < r y r < ~ - =0 otherwise. (4.54) Tile pulse shapes, normalized Fourier transforms and eye diagrams for this family of pulses with fi=0.1, 0.3 and 1.0 are shown in Fig. 4.4. Three different input pulse shapes are considered, The first is a rectangular pulse which fills a fraction c< of time slot. Such a pulse might be expected when the transmitted pulse has sharp leading and falling edges and the pulse remains undistorted in passing through the transmission medium. Receiver Design for Optical Fiber Communication Systems 105 I l l a : 0 ' 2 (:z=1 I I I I , , - - r T I - ( I T hp ( t ) = T F O R T < I < ' ~ , 0 OTHERWISE /a =0.2 I \ -T hp(t)= 2_~aEXP {_12/[2((~T)2] } T T T 2T , . ( I ~: " Ex~ {-,,loT] } Fig. 4.5. Input pulse families (Copyright 1973 11978) AT&T Co.) The second i nput pulse family consi dered has a gaussian shape. Received pulses appr oxi mat i ng this shape mi ght be expect ed if t here is consi derabl e mode mixing in the fiber medi um [-4.28]. The third family of pulse shapes are exponent i al s which mi ght be charact eri st i c of a t ransmi ssi on medi um which i nt roduces dispersion into the t ransmi t t ed pulse. Typi cal pulse shapes al ong with t he anal yt i cal definition of these pulses are f ound in Fig. 4.5. The par amet er cx is a measure of the fract i on of the time slot occupi ed by the pulse. With these definitions for the assumed pulse shapes the values of I 1, 12, /3 and ~ i are f ound in Fig. 4.6-17. In the case of the rect angul ar family it is seen t hat the values of the four quant i t i es of interest are relatively insensitive to t he fl' action of t he time slot occupi ed by the i nput pulse. The values of / 1 and ~ i are bot h close to 0.5 over the range 0 <c~ =< 1. Thus, the cont r i but i on to the noise from pulses in adj acent time slots, which is pr opor t i onal to ~ - 11 is small, especially f o r / / = 1.0. Since 12~0. 5. the noise associ at ed with the dar k cur r ent and shunt cur r ent gene- rat ors has an effective noise bandwi dt h B~rr~0.5B. The integral 13 varies between 0.03 and 0.08 and is very nearl y but not exactly equal to B 2 r r / 3 which would be the case if HT( ~ ) were a perfect low pass filter. Whereas the vari ous integrals and t he sum Y'n are relatively insensitive t o for rect angul ar pulses, they are seen to increase qui t e rapi dl y in the case of gaussian pulses for c~ >0. 5. When ~=0. 5 the val ue of a given pulse is 0.135 of its maxi mum value at the cent er of the adj acent time slots. In or der to satisfy the r equi r ement of zero i nt crsymbol interference at the out put , the equal i zer plus 106 R. G. Smith and S. D. Personick 0 d. ~ 7 o o d- d 0 ~ o d 0 N N d d " I ~ I 2 i I i i I i i I ~ ~ . ~ o q d o d i I ~ ~ - / I I I i I I i I 0 0 0 0 0 0 ~ N )0 oo o o do d 0 o o o ~ 0~ - o w~ > Receiver Design for Optical Fiber Communic~fion Sys{cms ~07 0,1 o ,o, ~. ~0 c5~ ,~ ~ - - '._ ~ ~ ~. d d d q ~ o d o d " - - o , o ~ ' o ~ ' 4 o d ~ d I o d l I i i I l l I i / 0 0 0 0 0 0 ~ oa ~ LO r ~ k~ co ~o d - - - . ko ~0..d o, H- g o o I I I Lq, ~ 04 O0 0 0 ~. ~ i ~ , . . . ~ o o In, t o~ ~6 . I - P ~< .~c~ 108 R. G. Smith and S. D. Personick 5 0 4 0 30 2 0 1 0 5 "r 2 4 5 2 t 0 5 0. 4 O-~ 0. 2' 0 1(3 I 3 o 0 2 05: 0 0. 0 O)4 0.0.3 0. 02 ( 2 1 2 3 4 5 I I l I I ~ = 0 . 1 0.5,1.0 ~ 1.0 I I I I I 0. 2 0. 4 0. 6 0. 8 1. 0- 1. 2 ( 2 ( 2 1 2 3 4 5 I I I = t B = I . 0 0.1 0.5 ~=1. 0 0.5 o ' . ~ o ' . , o ' . ~ o ' . ~ 1 ' . o 1 . ~ (2 Fig. 4. 16. Exponent i al f ami l y 12 vs c~ and fl ( Copyr i ght 1973 (1978) AT &T Co. ) Fig. 4. 17. Exponent i al f ami l y 13 vs and f l ( Copyr i ghl 1973 (1978) AT &T Co.) filter must reshape the pulse, emphasizing the high frequency components of the input signal, thereby increasing the effective bandwidth and the values of the noise integrals. The implication of the increase in 11, 12, 13, ~ is an increase in the noise of the receiver and a corresponding decrease in sensitivity. Consider the same reasoning from the point of view ot the frequency domain. The Fourier spectrum of the out put pulse is specified by its assumed shape. Whcn the input pulse is narrow in time, i.e., ~ 0 , its spectrum is broad 0 >, ~ - 2 0 - 30 - 4 0 0 GAU $SIAN Cl~ 0 ~20 110 10o 90 80 7O 6O 5O 40 3 Q 2O 10 0 - 20 - 5 0 - 40 R E C T A N ~ I I I t I I I ' I I ~ I 0, t 0. 2 0.3 0. 4 0. 5 0. 6 0. 7 0.8 0.9 1.0 Y Fi g. 4. 18. No r ma l i z e d t r a ns f e r f unc t i on f or t he r e c t a n g u l a r f ami l y. [~ ~ 1.0, :~ = 0.5, 1.0 Receiver Design for Optical Fiber Communication Systems 109 I r I I I I I I I 0.1 0.2 05 04 0.5 06 0.7 0.8 0.9 Y Fig. 4.19. Normalized mmsfer function to for the gaussian family, fl = 1.0, ~ ~ 0.5, 1.0 and the shape of H'TO') appr oaches t hat of H'ou,0'), permi t t i ng the rnaxi mum amount of filtering and hence mi ni mum noise. When the i nput pulse br oadens in time its spect rum narrows, limiting t he amount of filtering t han can be empl oyed, resulting in increased noise. The case of t he ideal gaussian pulse t ends to exaggerat e this effect since for a given pulse width in time, t he fi' equency spect rum tends to be relatively nar r ow and falls off rapi dl y with frequency. Figures 4.18, 19 show pl ot s of H~r(y ) for the rect angul ar and gaussian families with [ t = 1.0 and e =0 . 5 and 1.0. Whereas HT(Y) for the rect angul ar family falls off smoot hl y with frequency, the t ransfer function for the gaussian 110 R. G. Smith and S. D. Personick ( a ] IE(~)I~ / <~'> {b) Vou 2 Iin (c) (e} to / IHTi ( f ) <e a > I H T L 2 ~ o (g) - < e i > IH T et (d) (~ (h ) ,.~ Fig. 4.20a-h. Frequency dependence of signal and noise in fi'ont end circuit: (a) system response at inpul of equalizer; (b) response of equalizer; (e) system response at output of equalizer: (d) system response at output of fiher and output noise spectrum of shunt current generator; (e) spectrum of series noise generator at oul pul of amplifier; (f) spectrum of series noise contribution at output of equalizer; (g) spectrum of series noise contribution at output of filter with equalization ratio of 2; (h) spectrum of series noise contribution at output of filter with equalization ratio of 10 f ami l y s hows a rise in t he r es pons e r esul t i ng f r om t he f a l l of f of H' p( y) - whi ch is r equi r ed t o mi ni mi ze t he i nt e r s ymbol i nt erference. I n a pr act i cal si t uat i on, recei ved pul ses whi ch ma y a p p e a r gaus s i an will in all l i kel i hood not r equi r e t he a mo u n t of equal i zat i on pr edi ct ed by t hi s i deal i zed case. The val ues for t he var i ous quant i t i es f or t he . exponent i al fami l y, s hown in Fi gur es 4. 2- 14 t h r o u g h 4. 2- 17 s how a de pe nde nc e on c~ whi ch is less d r a ma t i c t han f or a gaus s i an but mor e r api d t ha n f or t he r e c t a ngul a r fami l y. Bit Rat e Dependence of t he Noi s e In t he pr ecedi ng t he o u t p u t noi se has been f or mal l y eval uat ed in t er ms of t he spect r al densi t i es and t he wei ght f unct i ons of I t , 12, 13 a nd ~, l t he var i ous noi se sour ces. Fr o m t hese resul t s it is seen t ha t all t he s hunt noi se t er ms c ont r i but e t o t he o u t p u t noi se in di r ect p r o p o r t i o n t o t he si gnal i ng r at e or b a n d wi d t h B. Thi s resul t s f r om t he fact t hat t he s hunt noi se s our ces see t he s ame t r ans f er f unct i on as t he si gnal cur r ent and t he b a n d wi d t h o f t he t r ansf er f unct i on is p r o p o r t i o n a l t o t he si gnal i ng rate. On t he ot her hand, t he series noi se ge ne r a t or does not see t he s ame t r ans f er f unct i on as t he s hunt cur r ent ge ne r a t or s and cont r i but es a t er m t o t he noi se pr opor t i ona l t o t he cube of B. Thi s de pe nde nc e can be expl ai ned as follows. Let t he i nput a dmi t t a nc e t o t he pr eampl i f i er be a si mpl e RC ci r cui t in whi ch case t he i nput vol t age due t o t he si gnal a nd s hunt noi se sour ces will fall off b e y o n d m = ( RC) - 1 due t o t he s hunt i ng effect of t he i nput c a pa c i t a nc e as s hown in Fig. 4.20a. Ch o o s i n g t he equal i zer t o cor r ect f or this fal l -off in r esponse, t he Receiver Design for Optical Fiber Communication Systems 111 t ransfer funct i on of the equal i zer is shown in Fig. 4.20b, and the t ransfer funct i on of t he system at the out put of t he equal i zer is given in Fig. 4.20c. The filter response and hence the overall system response H T t aken from Fig. 4.4, is shown in Fig. 4.20d. The out put noise spect rum of the shunt noise sources is given by the pr oduct of (c) and (d) assumi ng t he spect r um of t he noise cur r ent is flat - and hence pr opor t i onal to the integral of t he t ransfer funct i on which is pr opor t i onal to B. In the case of the series noise gener at or the noise vol t age at the amplifier i nput will be i ndependent of frequency, Fig. 4.20e, assumi ng the i nput admi t - tance t o the ideal amplifier is zero. The effect of the equal i zer is t hen t o accent uat e the high frequency component s of the noise as shown in Fig. 4.20t". The spect rum of this noise t erm at the out put of the filter has the shape shown in Figs. 4.20 g, h for the cases when the cor ner frequency of the i nput admi t t ance and hence the equal i zer is equal to 0. 5B and 0.1 B, respectively. Assuming the same i nput noise generat or, the out put noise, pr opor t i onal t o t he areas under these curves, is clearly seen to depend upon the degree of equal i zat i on empl oyed. The integral is easily seen t o cont ai n a t erm pr opor t i onal t o B a resulting from t he .1.2 dependence of t he equal i zer response. Since the magni- t ude of this t erm depends quadr at i cal l y t he l ocat i on of the zero of t he i nput admi t t ance it t hus depends on the squar e of the i nput capaci t ance, given anal yt i cal l y by (4.38). Thi s noise t erm is often referred to as the capacitive noise, since it depends on the i nput capacitance. Physically this noise does not result from the capaci t or, which is a noiseless element, but r at her from the equal i zat i on requi red t o over come the shunt i ng of t he signal by this capaci t ance. Noise Corner Frequency The i nput or out put noise expressions cont ai n t erms pr opor t i onal to B and B 3. Wi t hout knowi ng the magni t udes of the t erms involved it is not clear which t erm will domi nat e. A conveni ent means for discussing the i nput noise is t hr ough the concept of the noise cor ner frequency defined below. Consi der the noise t erms which are i ndependent of the signal. The shunt noise cur r ent densi t y is given by d ( i 2 ( ( o ) ) e q / d f and the series vol t age gener at or is equi val ent to a shunt cur r ent gener at or given by d(e2(~o))/df[Yi,(~,o)l 2. Assuming d ( i 2 ( o ) ) ) c q / d f and d(e2,({o))/t!/" t o be i ndependent of frequency the t ot al equi val ent i nput noise cur r ent is then of the form d (i2((O))toia i =Ao +A2c,)2 ' d f where A _ d . , 1 d , o- 4f + a7 <e:(.,)> 112 R. G. Smith and S. D. Persouick and A ~ = C " ~t < d ( , o ) > " d r ' " The noise cor ner f r equency. ] ~ is defined as t hat frequency for which the val ue of the noise is twice the low frequency value, and is given by 1 d ( i 2 ( u ) ) ) c q / d f 1 ] 1''2 I L - 2 n C . , I ~ / ( d ( ~ o ) ) / a f + ~ 1 " ( 4 . 5 5 ) Under most ci r cumst ances the first t er m in the radi cal domi nat es si mpl i fyi ng the expressi on somewhat . The quest i on of whet her the B or B a t erm in the noise domi nat es is det er mi ned by the rel at i on bet ween the noise cor ner frequency and t he si gnal i ng rat e B. When .1~ > B the noise spect ral densi t y is nearl y const ant over the r ange of HT(~o) and the noise is pr opor t i onal to B. When.li,~ <~ B, the f 2 t erm in (4.20) domi nat es the noise which is t hen pr opor t i onal t o B s. When .f.~ < B bot h t er ms cont r i but e nearl y equal l y to the t ot al noise. The quest i on of whet her ./',~ <, = or >B can be si mpl y answered f r om the knowl edge of the noise sources and the definition of.f,~ given by (4.55). We shall now t urn to the eval uat i on of the noise sources for FET and bi pol ar amplifiers. 4 . 3 No i s e o f F E T and Bi pol ar Ampl i f i e r s In tile precedi ng section the noise of the syst em, referred bot h to the out put , (4.21, 36 38, 42, 44), and to the i nput of the l i near channel , (4.47 52), has been present ed in t er ms of the spect ral densities of the noise sources, the bit r at e B, and the weight funct i ons 11, 12, 13 and ~ . The noise will now be eval uat ed for the specific cases of a c ommon sour ce FET and a c ommon emi t t er bi pol ar, the most c ommonl y used confi gurat i ons. 4.3.1 FET Front End The c ommon source conf i gur at i on for the i nput ampl i fi er when using an FET is shown in Fig. 4.21, where the resi st or R e represent s the parallel combi nat i on of resistors used to bi as the gat e of the t r ansi st or and to pr ovi de a dc ret urn pat h for the det ect or current . The pri nci pal sources of noise in this circuit are the t her mal noise of the equi val ent resi st or R L, the shot noise associ at ed with gat e v i i Receiver Design for Optical Fiber Communication Systems 113 R t . VDD Fi g . 4 . 2 1 . Schematic representation of FET fi'ont end i l l c o m r n o n s o u r c e C O l l f i g t l r a l i o n l e a k a g e c u r r e n t a n d t he noi s e a s s o c i a t e d wi t h t he c ha nne l c o n d u c t a n c e 4. Th e e q u i v a l e n t i nput s hunt c u r r e n t d e n s i t y is gi ven by [4. 29, 30-1 d 2 4 k T d f ( i )~hu,,t= -R-~ +2ql g~, ~, (4.56) whe r e I~,~c is t he ga t e l e a k a g e c ur r e nt . Th e c h a n n e l c o n d u c t a n c e c o n t r i b u t e s a noi s e c u r r e n t at t he o u t p u t of t he F E T wi t h a s pe c t r a l d e n s i t y d ( i a ) , , u m, t / d f =4 k T Fg m whe r e gm is t he t r a n s c o n d u c t a n c e of t he device, F is a n u me r i c a l f a c t or ~ 0 . 7 f or Si devi ces a n d F. . ~I . 1 f or Ga As F E T s [ 4. 31] . Re f e r r i ng t hi s o u t p u t c u r r e n t s our c e t o t he i nput of t he a mpl i f i e r gi ves a ser i es vol t a ge s our c e wi t h a s pe c t r a l d e n s i t y s d z 4 k T F (4.57) dJ " < e " ) = g i l l Un d e r mo s t c i r c u ms t a n c e s t he i n p u t a d mi t t a n c e of t he F E T wi l l be c a pa c i t i ve , hence Y,, =j ~, ) C, , (4.58) Ri. , = RL, (4.59) a n d C T = C d + C~ + C, , (4.60) whe r e C,, is t he s um of t he g a t e - s o u r c e a n d g a t e - d r a i n c a p a c i t a n c e s . 4 Some devices exhibit l / f noise which is not explicitly considered here. See [.4.29-1. 5 The use of this equation in (4.51) gives the correct expression for the series noise contribution, (4.61), where C 1, is the actual capacitance at the input. The Miller effect, which, increases the effective input capacitance, when explicitly taken into account, modifies (4.51, 57) and (4.71), below, in such a way that the Miller effect cancels out, yielding correct expressions with its neglect. l l 4 R. G. Smith and S. D. Personick T h e n o i s e d u e t o t he a mp l i f i e r a n d b i a s r e s i s t o r , r e f e r r e d t o t he i n p u t , f o u n d f r o m (4. 50, 51) a n d ( 4. 56- 59) is g i v e n b y (i2)circuit \ ~ - L +2q/ g; , , 12B{~h~,,tl 4 k r r I l a B ] + .q-~LR~ + ( 2 lZCT)2I~B ( ~ , . (4. 61) R e a r r a n g i n g t h e t e r ms a c c o r d i n g t o t h e i r bi t r a t e d e p e n d e n c e (4. 61) b e c o me s 2 r i k T ( , r < ' + 4k TF (2~CT)2 13 B3 . (4. 62) ,l~ ] I I 1 Mi ni mum FET Noi s e F o r mo s t we l l - d e s i g n e d c i r c u i t s gmRL> 1 a n d f o r Si d e v i c e s / g a t e is s ma l l a n d ma k e s a n e g l i g i b l e c o n t r i b u t i o n t o t h e n o i s e 6. Us i n g t h e s e a p p r o x i m a t i o n s ()4kT (2~CT)213B3"gm (4. 63) ( i 2) ~i r ~ui t ~ RL 12B+4kTF As (4. 63) is wr i t t e n , t h e f i r st t e r m is t h e n o i s e c o n t r i b u t e d b y t h e b i a s c i r c u i t wh i l e t h e s e c o n d t e r m is t he c o n t r i b u t i o n o f t he a c t i v e de vi c e . T h e s e c o n d t e r m t h u s gi ve s t h e a b s o l u t e m i n i m u m n o i s e wh i c h t he f r o n t e n d c a n a c h i e v e i n t h e l i mi t R c - ~ oo, (i2)circuit. mini . . . . . = 4kTF (2rt Cv)2 13 B 3 . (4. 64) gm T h e m i n i m u m v a l u e is o f t e n u s e d t o c a l c u l a t e t h e l i mi t i n g s e n s i t i v i t y o f F E T r e c e i ve r s . F r o m (4. 64) it is s e e n t h a t t he l i mi t i n g c i r c u i t n o i s e f or a n F E T v a r i e s a s B ~ a n d a l s o a s C~/,O,,,. A f i g u r e o f me r i t f o r a n F E T f r o n t e n d is t h u s g i v e n b y gm ( F E T ) . (4. 65) F i g u r e o f Me r i t = C ~ 6 Gate leakage is not negligible in GaAs FETs at this time. A discussion of the effect of leakage current is given below. Receiver Design for Optical Fiber Communication Systems 115 Since C. r cont ai ns cont r i but i ons f r om bot h the det ect or (plus parasi t i cs) and the t ransi st or, and since the t r ans conduct ance and capaci t ance of an FET are bot h appr oxi mat el y pr opor t i onal to the gat e width, an FET desi gned t o opt i mi ze the figure of meri t woul d have C a =( C a +C0 . Example The mi ni mum circuit noise for an FET, (4.64) cal cul at ed as a funct i on of the bit rat e B for the fol l owi ng assumed val ues of the circuit par amet er s, assumi ng a 2N4416 j unct i on FET, C,, = 4.5 pF Ca = l pF C s = 0.5 p F gm= 6 mS = 6 mmh o F =0. 7 k T = 4 . 1 4 x 10 - 2 1 J 13 = 0 . 0 3 (Rectangular Pulse) is given by 2 =8. 24 x 10 -23 B3[ Mbi t - s - 1] [ A2] . (1)ci rcui t , minimum In the above expressi on the bit rat e B is expressed in Mbi t - s 1 and the i nput noise current densi t y is expressed in A 2. The i nput noise is pl ot t ed in Fig. 4.22 as a funct i on of the bit rat e B. Effects of Thermal and Shot Noise Equat i on (4.64) gives the mi ni mum syst em noise assumi ng t hat t her mal noise of t he resi st or and shot noise due to gat e l eakage ar e negligible. We now exami ne the limits on R c and / g a t e such t hat t hei r cont r i but i ons to the noise are negligible. The cont r i but i ons t o the circuit noise f r om R c and Ig,, are given by 4k T + 2ql~ate) I z B (4.66) and are seen t o var y l i nearl y with bit rate. The t her mal noise cont r i but i on of a resi st or R and the shot noise cont r i but i on of a cur r ent 1 are equal when 4 k T - 2ql R 116 R. G. Smith and S. D. Personick , o -,~ . _ o ~ 4 " Y / / . % 2o _ / " ~o ~ ~/ / _ / 8, o v _ / / jy.~*/__ / / " m / 2.."/ / ,o V o.' o' . I / / , . %/ / . : o.o I 0 2 5 ~ 0 I $ ~0 100 BI T RATE 6 [ M b l s } Fig. 4.22. Effective i nput noi se cur r ent vs bit rat e for an FET front end wi t h pa r a m- et ers i ndi cat ed. Also shown is effective i nput noi se cur r ent for shunt resi st or or l eakage current . Tot al noi se is gi ven by t he sum of alI cont r i but i ons or 2k T I R- - 5 . 2 x 10 -2 [ V] . (4.67) Thi s relation, often referred to as the 50 mV rule, st at e t hat the equi val ent noise resi st ance of a cur r ent I is appr oxi mat el y 50mV/ I . Thus a cur r ent of 1 laA cont r i but es the same noise as a 50 kf l resistor. Taki ng 12 =0. 5 ( Rect angul ar Pulse), the noise as a funct i on of bit rat e for vari ous val ues of R L (and l~,t~) are shown as dashed lined in Fig. 4.22. The difference in bit rat e dependence bet ween t her mal or shot noise and mi ni mum FET noise is clearly seen, The effects of R L and Ig,,e may be easily seen f r om the figure. Assume for exampl e t hat a systern is to be oper at ed at a dat a rat e of I Mbi t . s - 1, for which the mi ni mum FET noise is appr oxi mat el y 10 - 2z A 2. At this poi nt the noise is equal to t hat of a resi st or of 108 Q or a l eakage cur r ent of 500pA. Thus to reach the mi ni mum achi evabl e noise level, limited by the channel noise of the FET, it is necessary the Re>> 108~ and I~t~ < 500 pA. In many cases achi evi ng or exceedi ng these limits may be i mpossi bl e in which case the act ual circuit noise will be domi nat ed by R L or l~,t~. It shoul d al so be not ed t hat nonmul t i pl i ed dar k cur r ent of the det ect or cont r i but es noise with the same weight funct i on and bit rat e dependence as the gat e l eakage cur r ent and t hus the same const r ai nt s appl y. At hi gher bit rat es the rest ri ct i ons on R E and l eakage current s become less severe due to the differing bit rat e dependences of the noise sources. A general V b i o s _k T Recei ver Desi gn for Opt i cal Fi ber Co mmu n i c a t i o n Sys t ems Vcc 117 b rbJb b t i s C d + C ~ RI_ C a r b t e 0 F i g . 4. 23. Schemat i c r epr es ent a- t i on of bi pol ar f r ont end in c om- mon emi t t cr conf i gur at i on rel at i on t hat must be satisfied for resist, or noise to be less t han t hat of t he FET is g ml 2 1 Re> F(2~CT)213 B2. (4.68) The cor r espondi ng rel at i on for l eakage cur r ent is I < 2kTF(2~ZCT)213 - B 2 . (4.69) q,qmI 2 4.3.2 Bipolar Front End The common emi t t er confi gurat i on for the i nput amplifier is shown in Fig. 4.23, where the resistor R e represents the parallel combi nat i on of resistors used to bias the base of t he t ransi st or as well as the phot odet ect or . For the bi pol ar t ransi st or t he domi nant sources of noise are t he shot noise associated with the base and col l ect or current s and t he t her mal noise of t he base resistance, rh. b 7. In t he common emi t t er confi gurat i on the shot noise of the base cur r ent is model ed by an equi val ent shunt cur r ent gener at or with a noise cur r ent spect ral density given by d d~ (i2)shun' = 2qlu' (4.70) (base current ) 7 Conf i gur a t i ons ot he r t ha n c o mmo n s our ce a nd c o mmo n emi t t er ar e di scussed in [4. 32]. 118 R. G. Smith and S. D. Personick where I b is the qui escent base current . The shot noise of the col l ect or cur r ent appear s as a current source at the out put with a spectral densi t y d 2 ~ : f ( i )ou~vu, = 2 q I ~ (collector current ) where I~ is the quiescent col l ect or current . Referring this noise gener at or t o the i nput gives an equi val ent series noise gener at or with a spectral densi t y given by d 2 2 q l ~ d f ( e a ) = 2 (4.71) gm where gm is the t r ansconduct ance which in t urn is given by q l (4.72) g i n - - k T " The i nput admi t t ance of the t ransi st or at the i nt ernal base cont act (lead and package capaci t ance are consi dered to be included in the st ray capaci t ance) is given by 1 Y,, = - - + j w C , , (4.73) F b '~ where C~ includes base-emi t t er and base-col l ect or capaci t ances and rb, ~ is the dynami c base-emi t t er resistance, rh'~ = f i g (4.74) fl is the small signal bet a of the t ransi st or and G is the Shockl ey emi t t er resistance 1 k T G - - (4.75) gm q l ~ " The t ot al i nput admi t t ance is t hen given by Y i n = " ~ +Ja)C' r (4.76) and C T = C O + C~ + C. , (4.77) Receiver Design for Optical Fiber Communication Systems 119 where now the stray capacitance C'~ includes t hat port i on of the t ransi st or input capacitance associated with leads and package. Using st rai ght forward circuit analysis, the base resistance noise, which appears as a noise voltage generat or connected between the internal and external base leads, can be model ed as an equivalent series voltage generat or given by d ,2 d Yiy~ 2 ~ f ( % ) = ~ f ( e 2 ) , (4.78) where 1 ! Yi, = w- +fin(Ca + C;) tx L (4.79) and d d ~-=g < eb ) 2 = 4k ' ] ? ' b ' b " (4.80) The effective i nput noise of the amplifier may be found from (4.50, 51) using (4.70, 71, 78). The equivalent i nput noise current is given by < i 2 ) c i r c u i t - = 2 q l t , Bl 2 (base current) (4.81) 2ql [ ( 1 + 1 ]2BI2+(2gCT)ZBaI3] (col l ect orcurrent ) + [kK "b'o/ +4kTJ'b,bl~/L22 +(2g)2(CdWC~)2B313]. (base resistance) Opt i mum Col l ect or Current Exami ni ng (4.81) it is seen t hat the first and second terms depend explicitly upon the operat i ng point, i.e., collector current of the transistor, the base resistance t erm is essentially i ndependent of I c except as rb, b might vary with specific biasing conditions. The base current noise, being proport i onal to I b is t hus proport i onal to the collector current. On the ot her hand, since the t ransconduct ance, gin, is proport i onal to the collector current, the collector current noise is inversely proport i onal to I c. Thus there exists all opt i mum collector current which minimizes the sum of the collector and base current cont ri but i ons to the noise. By differentiating (4.81) with respect to lc and neglecting the variation of C a 1 2 0 R . G. S m i t h a n d S . D. P e r s o n i c k with 1~, which does not significantly alter the results, the opt i mum col l ect or cur r ent is found to be _ k T . 2 x C T f l U Z B ( I 3 ] 1 /2 [ I . . . . . . . . . . . . q \ I ~ / 1 12/13 ] 1 / 2 + ( 2 x ~ 0 2 j (4 .8 2 ) where I c f l O = l b = h F E " ( 4 . 8 3 ) The opt i mum col l ect or current t hus varies di rect l y as the bit rate, directly as the t ot al i nput capaci t ance and as the square r oot of the bet a of t he transistor. Minimum Bipolar Noise At this opt i mum col l ect or cur r ent the base and col l ect or cur r ent noises are essentially equal and are given by 2 k T ( 2 r C C T ) ( I 2 1 3 ) l / 2 [ I 2 / 1 3 ] < i 2 ) h a s e = f l l o / 2 1 + ( 2 r c B C T R L ) 2 i 2 . ( )Ct)III2CtI)F = (i2). . . ~ 1 / 2 ( 4 . 8 4 ) In general the value of tile t erm in bracket s is close to uni t y and for most ci rcumst ances may be set equal to one. At the opt i mum bias poi nt the t ot al circuit noise referred to the i nput is given by 2 - ( 8 x k T ) " ( 1 2 1 3 ) 1 / 2 B 2 1 + ( 2 1 r B C T R L ) 2 ] < l > c i r c u i l , o p t i m u m - - \ 1 " 0 / + 4 k ~ 'b ,b I-~ //- + ( 2 7 ~ ) 2 ( C 3 q - C s ) 2 I d B 3 ] . (4.85) In the limit t hat RL--* oo, the mi ni mum circuit noise for the bi pol ar t ransi st or is given by < i 2 > ~ , r , : , i , , = ( 8 ~ k T ) ( ~ l ( I 2 1 3 ) t ' 2 B 2 mini . . . . \ r i O / / + 4k'/?'wh(2'~) 2 ( C , j + C ~ ) 2 1 3 B 3 . ( 4 . 8 6 ) Receiver Design for Optical Fiber Communi cat i on Systems 121 Excl udi ng t he cont r i but i on f r om the base resistance, which in most, but not all, instances will be small, t he mi ni mum noise is seen t o var y as the square of the bit rat e in cont r ast t o a cubi c dependence for t he FET, and linearly with capaci t ance compar ed to a quadr at i c dependence for the uni pol ar device. An appr opr i at e figure of meri t for the bi pol ar t ransi st or is Fi gure of Meri t = [3~'2 (Bipolar). (4.87) Cv The fact t hat the bit rat e dependence of the noise of t he bi pol ar front end is slower t han for the FET results in the bi pol ar device pr oduci ng lower noise circuits at hi gher bit rates, whereas the FET is superi or at low bit rates. The principal reason for the superi ori t y of the bi pol ar device at the hi gher frequencies derives from the ability to increase the t r ansconduct ance of the device by increasing t he col l ect or current , t hereby reduci ng t he cont r i but i on of the series noise gener at or which varies as B 3. B a s e R e s i s t a n c e N o i s e Since the noise cont r i but i on of t he base resistor is i ndependent of the col l ect or cur r ent it is not subject to the above opt i mi zat i on. It is thus of interest to exami ne under what ci rcumst ances base resistance noise will pl ay a maj or role in det ermi ni ng receiver noise. Equat i ng the base resistance noise to the sum of the base and col l ect or cur r ent noises in the limit of large R L gives c , ( 1 2 1 ,,,2 rb'bB---- 2 n ( C d + ~ s ~ I'O \ I 3 ] t,-.p ]2 [~ 1/2 ( 4. 88) Using [2 =0. 5 and 13 =0. 03 (the rat i o is r easonabl y const ant for varyi ng pulse shapes), and expressing the capaci t ances in pF, (4.86) results in t he appr oxi mat e formul a rb, bB=6. 5 X 1011 CT /3~/2(Co + C, ) 2 . (4.89) As an exampl e choosi ng CT=6 p F , Ca + C' s = 4 p F and /3o= 100, the poi nt at which the base resistance noise equals the ot her noise sources satisfies rb, bB=2. 4 X 101 Thus at a bit rat e of 100 Mb i t . s - 1 the effect of base resistance will begin to domi nat e an opt i mal bi pol ar receiver when r b, u=240~. At a bit rat e of 300Mbi t . s -a the cor r espondi ng val ue is 80f~. It shoul d be not ed in passing t hat for a nonopt i mal receiver t he effect of base resistance is cor r espondi ngl y less i mport ant . 122 ~6 s oc 10 - 5 o t66 R. G, Smith and S, D. Personick I I C T = 6 P F / ;::7:o I 2 = 0 5 / i I I I I t i I I I I I I I 10 t 0 0 1000 BIT RATE B IMDIs) Fig. 4.24. Opt i mum col l ect or cur r ent for bi pol ar f r ont end Example As an exampl e the opt i mum bias cur r ent and the mi ni mum circuit noise for a bi pol ar t r ansi st or in the limit RL~ oo are pl ot t ed in Figs. 4.24, 25 usi ng the following par amet er s C T = 6 pF C a + C~ = 4 pF #o = 1 O0 r w, = 100f~ I 2 =0. 5 13 =0. 03 for which case the opt i mum bias cur r ent is given by I .............. = 2 . 4 x 10-6 B [ A] and the mi ni mum noise by <i 2)ci rcui , , mini . . . . . . ---7.65 X ] 0 - 2 1 B e + 3 . 1 4 x 1 0 - 2 3 B 3 (base resistance) [ A: ] , where B is expressed in Mbi t . s - L Also shown as a dot t ed line in Fig. 4.25 is the limiting behavi or when rwb =0. The effect of the base resi st ance noise is seen to Recei ver Desi gn for Opt i cal Fi ber Co mmu n i c a t i o q Sys t ems 123 I o - 1 4 1(5 t 5 t o - I ~ == 1517 u z t0-18 t 0 - 1 9 ~ o - 2 o o -"- J / / ~ ~'( CT=6P F Cci+C~=4p F / ~ o - i O o ,C,OO "r 2=0.5 , "r3:0.03 | 6 " " ~ t " l o l O O ~ o 0 o BIT RATE B (Mb/s) Fig. 4.25. Effective i npul noi se cur r ent vs bi t r at e for an opt i mi zed bi pol ar f r ont end wi t h pa r a me t e r s i ndi cat ed. Effective i npul noi se cur r cnl is al so s hown. Tot a l noi se is gi ven by t he SUITI of all c ont r i bul i ons become significant for B~ 3 0 0 Mb i t . s -1 in agr eement with the pr ecedi ng ar gument . For reasons of compl et eness the noise associ at ed shunt resi st ance or l eakage cur r ent s ot her t han base bi as cur r ent al r eady specifically included, is shown in the figure. As wi t h the FET, the differing dependences on bit rat e of t he noise sources is seen. 4. 3. 3 F ET vs Bi po l a r Front Ends The mi ni mum circuit noise for the FET and the bi pol ar front ends are given by (4.64, 86). The noise in the case of t he FET vari es as B 3 whereas the dependence is B 2 for t he bi pol ar when l'b'b is small and B 3 when base resi st ance noise domi nat es. Hence at low bit rat es the FET is super i or whereas at hi gher bit rat es the bi pol ar device pr oduces super i or per f or mance. The cr ossover fre- quency at which poi nt the per f or mances are equal is given by equat i ng t he above expressi ons and for t he t wo exampl es given occurs at 93 Mbi t . s i when rb, b = 0 and 150 Mbi t . s ~ when Yb'b = 100 ~'~. Dependi ng on the specific as s ump- t i ons r egar di ng par amet er val ues the cr ossover occurs ar ound 5 0 Mb i t . s t within a fact or of 2. For silicon FET' s high frequency oper at i on is al so limited by the finite gai n of the first st age (fixed 9,,) whi ch i ncreases the cont r i but i on of subsequent st ages to the t ot al noise. 124 R. G. S mi t h and S. D. P e r s o n i c k 4. 4 Front End Designs In the previ ous sections the basic noise propert i es of front ends have been discussed in detail. In this section several al t ernat i ves for front end designs are discussed and their features cont rast ed. The designs discussed are the high i mpedance or i nt egrat i ng front end and the t r ansi mpedance or cur r ent - t o- voltage convert er. The noise charact eri st i cs as well as ot her propert i es of these designs are compar ed with a t hi rd al t ernat i ve which uses an amplifier with an i nput t er mi nat i on chosen to pr ovi de a bandwi dt h great er t han or equal to t he chosen bit rate. 4. 4. 1 St rai ght f orward Termi nat i on The most st rai ght forward - but not necessarily the opt i mum met hod of designing a front end is t o t ermi nat e the i nput to the preamplifier with a l oad resistor, RL, such t hat in conj unct i on with the i nput capaci t ance, C, r, the bandwi dt h of t he i nput admi t t ance is equal to or great er t han the bit rate, B. This requires 1 R L < 2 r c B C , r . (4.90) One appr oach woul d be to use a st andar d 50 f* ternaination pr ovi ded (4.90) was satisfied, and the second to satisfy the equal i t y in the above equat i on. In ei t her case t he i nput signal woul d be passed with good fidelity t hr ough the amplifiers requi ri ng little or no equal i zat i on. The final filter woul d then do the pulse shapi ng and noise filtering with F ( ( o ) ~ - Hv ( w ) . The advant age of this appr oach is its simplicity. The i nput noise associated with t he l oad resistor R L is given by 4 k T 2 - - - . B I 2 . (4.91) ( 1 ) l o a d resi st or-- RL When R L is chosen to satisfy the equal i t y in (4.90) the noise becomes ( i 2 ) l o a d r e s i s t or = 41 T ( 2 n C T ) B 212 (4.92) while for a t ermi nat i on of 5 0 ~ the noise is given by (4.91) with RL=5 0 ~ . In general the noise associated with ei t her of these choices of R L will exceed the mi ni mum value achi evabl e with an FET or an opt i mi zed bipolar. Compar i ng (4.92) with the mi ni mum noise achi evabl e with an FET, (4.64) gives 2 g m l 2 ( l ) l o a d resistor __ (4.93) 2 F(2nCT)I3B ( l ) mi n i . . . . FET and in the limit t hat rb, b is small the equi val ent rel at i on for a bi pol ar at opt i mum col l ect or current , (4.86), is given by i 2 / I \ ~ / 2 ( ),o~ros~s.,r = f l ~ / 2 [ ~ _ } . (4294) (i2)minimum. Bipolar k/ a] Receiver Design for Optical Fiber Communication Systems 125 For t he par amet er values selected as exampl es in t he previ ous sections t he l oad resistor noise is appr oxi mat el y 40 times t hat of the opt i mum bi pol ar at any bit rat e whereas for t he FET it is 103 times great er at 4 Mb i t . s - ~ and 100 times great er at 4 0 Mb i t . s ~. Thus the simplicity of this appr oach is paid for by a consi derabl e increase in the circuit noise over t hat pot ent i al l y achievable. 4.4.2 High Impedance or Integrating Front End The second appr oach to the design of a front end is to reduce all sources of noise t o the absol ut e mi ni mum. Whet her using a bi pol ar or an FET i nput device this is accompl i shed by reduci ng the i nput capaci t ance t hr ough t he selection of low capaci t ance, high frequency devices, by selecting a det ect or with low dar k current s, and by mi ni mi zi ng the t hermal noise cont r i but ed by the biasing resistors [4.26, 29, 30, 32-36]. The l at t er is accompl i shed by maki ng R L large which is the reason this appr oach is referred to as a high i mpedance front end. When the i nput resistance is large the i nput admi t t ance is domi nat ed by the capaci t ance C T and t he signal cur r ent t ends to be i nt egrat ed by this capacitance. Thus, t he high i mpedance front end is also commonl y referred to as an i nt egrat i ng front end. It is i mpor t ant to not e t hat the i nt egrat i on of t he signal is a consequence of havi ng made R L large as a part of the process of minimizing the circuit noise and is not t he reason low noise is achieved. In the high i mpedance front end the equal i zer plays the essential role of rest ori ng the i nput pulse shape which was di st ort ed due to the limited bandwi dt h of the i nput admi t t ance. In many cases the equal i zer takes t he form of a simple di fferent i at or, which at t enuat es t he low frequency component s of the signal, rest ori ng a flat t ransfer funct i on to the system [4.33]. The equal i zer woul d nor mal l y follow the first few stages of ampl i fi cat i on and hence woul d add little noise to t hat al ready present due to t he cont r i but i ons of the front end. A high i mpedance-equal i zed amplifier is t hus capabl e of reduci ng the circuit noise to an absol ut e mi ni mum while ret ai ni ng a t ransfer funct i on t hat preserves t he i nf or mat i on cont ai ned in the signal. It does, however, have several dr awbacks discussed below. The first probl em with this appr oach lies with the need to posi t i on t he zero of the equal i zer to compensat e for the zero of t he i nput admi t t ance. Since the l ocat i on of the i nput zero depends upon the values of C T and R~,, which in t ur n depend on the values of parasi t i c capaci t ances, and in t he case of bipolars on the/3 of the t ransi st or, it will in general vary from unit to unit. Thus, individual amplifiers may need to be i ndi vi dual l y equalized, and this equal i zat i on may have t o be t emper at ur e compensat ed, increasing the compl exi t y and cost of such an appr oach. The exactness with which equal i zat i on must be achieved depends upon the amount of the signal spect r um near the cut-in frequency of the equalizer. For bal anced signals, with appr oxi mat el y equal number s of marks and spaces, the low frequency por t i on of t he signal spect rum is small and exact equal i zat i on may not be i mport ant . Scrambl i ng of the signal is one met hod of obt ai ni ng such bal ance ; using bal anced codes is anot her. 126 R. G. Smith and S. D. Personick The second dr awback of the high i mpedance front end is t hat it has r educed dynami c range (ratio of maxi mum to mi ni mum i nput signals) compar ed to ot her appr oaches [4.6, 37, 38]. The loss of dynami c range occurs because the charge on the i nput capaci t ance, and hence the i nput voltage, builds up over the course of a number of i nput time slots, greatly exceeding t he charge and vol t age associ at ed with a single i nput pulse 8. Al t hough the equal i zer subsequent l y restores the pulse shape, the bui l dup of t he low frequency component s within the circuit pri or to the equal i zer can lead to pr emat ur e sat ur at i on of the amplifiers at high i nput signal levels. The reduct i on of dynami c range depends upon the amount of i nt egrat i on and subsequent equal i zat i on empl oyed. In the case of the FET f r ont end if t he resisto? R L is chosen such t hat its noise is equal to the mi ni mum FET noise, t her eby doubl i ng t he t ot al noise, the amount of equal i zat i on requi red, defined as the rat i o of the bit rat e B to the 3 dB frequency of the equal i zer (or the i nput circuit) is given by (4.93). Thus at a bit rat e of 4 Mbi t . s - 1 t he i nput admi t t ance woul d have a zero at 4 kHz requi ri ng an equal i zat i on of 1000:1. The amount of equal i zat i on woul d be 100:1 at 40 Mbi t - s 1. I f on the ot her hand it was desired to make the noise cont r i but i on of R L equal to 10% of the FET t hen the t ot al noise is reduced by 45% and t he degree of equal i zat i on requi red woul d be increased tenfold. In t he case of a bi pol ar front end increasing R e reduces its cont r i but i on to the noise while R~, asympt ot i cal l y reaches t he dynami c i mpedance of t he j unct i on, rh, . In this case the noise appr oaches the mi ni mum val ue for a bi pol ar front end while the equal i zat i on rat i o takes on a value given by (4.94). The equal i zat i on rat i o is thus i ndependent of the bit rate but does depend specifically on t he [ / o f the transistor. As a general rule, for fixed bias voltages the dynami c range will be smaller the great er the amount of equal i zat i on empl oyed. The high i mpedance front end thus provi des the means for achi evi ng the ul t i mat e in low noise front etads but with reduced dynami c range and t he possible requi rement of i ndi vi dual l y equal i zed amplifiers. In appl i cat i ons requi ri ng the ul t i mat e in sensitivity it does achieve the best performance. 4.4.3 Transimpedance Amplifier The t ransi mpedance or shunt feedback amplifier is the most commonl y empl oyed front end design used in optical fiber appl i cat i ons I-4.6, 37-41]. Shown schematically in Fig. 4.26 it is basically a cur r ent - t o- vol t age convert er. In the limit of large l oop gain t he relation bet ween the out put vol t age and the i nput cur r ent is given by V,,ut = -- ZFiin, (4.95) 8 It is possible to actively discharge the input capacitance - integrate and dump - bot this adds considerable complexity and will not be considered further. Receiver Design for Optical Fiber Communication Systems 127 ZF o Vo -(D Fig. 4.26, Simplified representation of the traasimpedance amplifier where Z v is the effective feedback i mpedance from the out put to the i nput of the amplifier. The t ransi mpedance amplifier is finding widespread use because it is capable of wide bandwidths, provides a greater dynami c range t han the high i mpedance approach [4.6, 38, 41] and has a noise performance which can approach t hat of the high i mpedance front end, al t hough in many applications nonopt i mi zed designs are used [4.38]. The bandwi dt h, or more properly, the transfer function of the t ransi mped- ance amplifier is det er mi ned by the details of the circuit design and the part i cul ar value of the component s used. Wi t h the simplifying assumpt i on of large l oop gain, where (4.95) applies, the frequency response is governed by the response of Z F, which is usually comprised of a feedback resistor, RF, and a capacitance C v. Since the feedback capacitance will in general be smaller t han the input capacitance, and can be made exceedingly small t hrough the use of a cascode design, the bandwi dt h will be greater t han for a similar amplifier which does not use feedback and has its i nput t ermi nat ed with R L = R v. The use of feedback t hus provides for increased bandwi dt h. The basic sources of noise for this t ype of amplifier consist of cont ri but i ons from det ect or leakage, the i nput t ransi st or and the feedback resistor R F. The effective input noise is given by the relations given in Sect. 4.3 for FET and bipolar transistors where now R L is replaced by R v, the feedback resistance. Thus the noise of a t ransi mpedance amplifier is the same as t hat of a high impedance front end if R v = R L. In practice the noise performance of t ransi mpedance amplifiers is not as good as t hat achieved with the high i mpedance approach. The principal reason for this results from the effect R F has on the frequency response of the amplifier. Equat i on (4.95) is onl y an appr oxi mat i on to the actual response of the t ransi mpedance amplifier in the limit of infinite l oop gain. In practice the gain of the amplifier is finite and at high frequencies may not be much greater t han 100. The actual transfer function is composed of two or more poles and may cont ai n zeros. For a fixed open loop gain it is well known t hat increasing the feedback resistance (for fixed CF) tends to make the pole locations complex, and under some conditions, make the amplifier oscillate. Thus, there is some practical limit to how large R F can be made. This limit depends upon the details of the circuit and t hus all t hat can be said here is t hat since R v cannot be 1 2 8 R. G. Smi t h a n d S. D. Pe r s oni c k increased indefinitely its cont r i but i on to the noise will have a pract i cal l ower limit. Thus, the noise of the t r ansi mpedance amplifier will always exceed, to some extent, t hat of the high i mpedance front end. The dynami c range of the t r ansi mpedance amplifier is great er t han t hat of the high i mpedance front end. Thi s arises as follows: In the high i mpedance front end the low frequency component s which are at t enuat ed by the equaliza- tion process cause sat urat i on of the amplifier pri or t o t he equal i zer at high signal levels. In the t r ansi mpedance amplifier at t enuat i on of t he low frequency component s is accompl i shed via the negative feedback and hence the low frequency component s are amplified by the cl osed-l oop, not the open- l oop gain of the amplifier. For a given amplifier the i mpr ovement in dynami c range will be appr oxi mat el y t he rat i o of the open- and cl osed-l oop gains. 4. 4. 4 Experimental Results The precedi ng material has discussed the noise characteristics of front ends from a t heoret i cal poi nt of view. We shall now t urn to the measurement s of this noi se; t hree procedures will be descri bed: The first pr ocedur e is to directly measure the noise at the out put of t he linear channel with a t rue RMS vol t met er. With a knowl edge of the magni t ude of the t ransfer function of the system this noise can be referred to the i nput of the linear channel as an effective i nput noise cur r ent density. While this t echni que is st rai ght forward it does requi re an accurat e knowl edge of the gain and frequency response of the vari ous amplifiers in the receiver. The second technique, a modi fi cat i on of the first, utilizes a p-i-n det ect or as the light detecting element. The amplifier noise is first measur ed with t he det ect or in the dar k and subsequent l y with t he det ect or i l l umi nat ed with a const ant intensity light source. If I o is the cur r ent at which the t ot al noise of t he system doubl es t hen t he effective i nput noise of the circuit is given by ( i 2 ) c i r c u i t = 2 q l o B e f r c c t i v c , where Beff~tiw is the effective bandwi dt h of the system including the filter. The effective bandwi dt h of the system can in t urn be det er mi ned by measuri ng the response of the system t o a modul at ed light source or by injecting white noise into the i nput (such as with an i l l umi nat ed APD oper at i ng at high gain) [4.38]. The effective bandwi dt h is then given by Hr(f)df Berf - o (4.96) H,r(0) ' Receiver Des i gn for Opt i cal Fi ber Communi cat i on Syst ems 129 Table 4.1. A compari s on o f theoreti cal and experi ment al resul ts o f first stage noi s e for several front end desi gns operating from 3 to 274 Mbi t . s Entry Bit rate Signal k~rmat 12 I~ Recei ver t ype; Cal cul at ed Measured Ref. No. i nput devi ce noi s e first noi se Input Out put st age pul se pul se [Mbit. s - I ] [A 2] [A2] I 3 NRZ NRZ 0.55 0.085 Optimized 1. 34xl 0 - ' 2. 24x10 -2 [4.41] Trans. hnped. Si FET 2 6.3 RZ RZ 0.31) 0.13 High. lmped. 1.59x10 1, 2. 12x10 i~ [4.32,33] Si FET 1.6 xl 0 -19 [1) 3 44.7 NRZ NRZ 0.68 (2) Nonoptimized 1.89x10 '~' 2 x 10 -~t' [4.38] Trans. l mped. Si Bi pol ar 4 50 RZ NRZ 0.40 0.036 High hnped. 1.44x 10 - 17 1.46x 10 - ' v [4.32.35] Si FET 5 274 NRZ NRZ 0.55 0.085 Si Bi pol ar 5.3 x 10 -I~' 2.4 xl 0 ,s [4.32,34] 1.76x 10 15 (1) 6 274 NRZ NRZ 0.55 0.085 Ga As FET 1.3 xl 0 - I s 6.2 xl 0 ,5 [4.32.34] 3.0 xl 0 -I5 (1) (1) Contribution o f first stage. {2) Series noi se cont ri but i on negl i gi bl e for this nonopt i mi z e d desi gn. where HT(0) is t he a s y mpt o t i c val ue o f t he transfer f unc t i on at l o w frequenci es. For a di gi tal s ys t em des i gned t o ope r at e at a bi t rate B, (4. 96) reduces t o Bef f ecl i v e = BI 2 . (4. 97) It s ho ul d be no t e d t hat a l t h o u g h t hi s t e c hni que e mpl o y s a s hunt current generat or t o det ermi ne t he s ys t em noi s e it meas ures bo t h s hunt and series c o nt r i but i o ns t o t he ci rcui t noi s e. The t hi rd t e c hni que e mpl o y s t he me as ur e me nt o f recei ver s ens i t i vi t y ( di scussed i n t he next s ect i on) t o infer t he ci rcui t noi se. Thi s t e c hni que is mos t easi l y i mpl e me nt e d by use o f a p- i - n det ect or, a l t h o u g h an AP D o pe r a t i ng at l o w gai n ma y be used. Thi s t e c hni que has t he di s advant age t hat , be i ng i ndi rect , it c a n n o t separat e o u t ot her effects - s uc h as pi ckup - whi c h mi ght affect sensi t i vi t y. In wel l - engi neered s ys t ems all three me as ur e me nt t echni ques , properl y perf ormed, gi ve ext remel y g o o d agreement . Tabl e 4.1 present s t he resul ts o f meas ured and cal cul at ed noi s e perf ormance o f f ront ends des i gned for di gi t al s ys t ems ope r at i ng at bi t rates f rom 3 Mb i t . s - t o 274 Mb i t . s - t [ 4 . 3 2 - 3 5 , 38, 41] . Al t h o u g h n o t a c ompr e he ns i ve s umma r y of all repeat er des i gns it doe s serve as a basi s o f c o mpa r i s o n o f t he or y and experi ment . I ncl uded i n t he t abl e i n a ddi t i o n t o t he noi s e val ues are t he s i gnal i ng f ormat ( RZ or NRZ) , t he a ppr o x i ma t e val ues o f t he i nt egral s 12 and 130 R. G. Smith and S. D. Personick Vbias ,A/v +5V : OUTPUT Fi g. 4. 27. Ci r cui t di a gr a m of t r a ns i mpe da nc e ampl i f i er used to obt a i n da t a s h o wn in Figs. 4.28, 29 13 , the t ype of i nput device empl oyed and the type of circuit design: high i mpedance, opt i mi zed t r ansi mpedance and nonopt i mi zed t ransi mpedance. It is seen t hat the agreement between the t heoret i cal and experi ment al results is very good, especially at bit rates bet ween 6. 3Mbi t . s -1 and 50Mbi t . s - t . At the lowest bit rate, 3 Mb i t . s -1, the devi at i on is 2. 2dB, possibly due to the cont r i but i on of the feedback resistor, while at the highest bit rat e the di screpancy is somewhat great er (5. 2dB for the bi pol ar and 3. 6dB for the GASFET) due possibly to the neglect of parasitic resistances or the use of an oversimplified model for the transistor. In general, however, the present model does provi de good engi neeri ng number s for all but t he most exacting application. Optimized Transimpedance Amplifier The high i mpedance front end achieves the lowest effective i nput noise as st at ed previously. Tr ansi mpedance amplifier designs do not generally appr oach t he low noise levels achieved with the high i mpedance appr oach and hence it is often implied t hat the t r ansi mpedance amplifier shoul d not be empl oyed when a t rul y low noise design is required. Reference t o entries 3 and 4 in Tabl e 4.1 woul d suggest t hat a t r ansi mpedance amplifier using bi pol ar t ransi st ors (entry 3) is appr oxi mat el y one or der of magni t ude mor e noisy t han a high i mpedance design using an FET when oper at i ng at nearl y the same bit rat e (entry 4). The frequency response and noise propert i es of a t r ansi mpedance amplifier similar t o t hat used in ent r y 3 have been studied as a funct i on of t he biasing condi t i ons of the i nput t ransi st or [4.42]. A circuit di agram of the amplifier is shown in Fig. 4.27. The frequency response of the amplifier is charact eri zed by t wo real poles, S t and S 2 whose l ocat i ons are shown in Fig. 4.28 as a funct i on of the feedback and col l ect or resistors R v and R c. R v was set equal t o R c for conveni ence in these experiments. Bot h pole l ocat i ons are seen to vary inversely as the resistance level. An amplifier with a domi nant pole at 50 MHz coul d be designed, for example, with RF = Rc = 6 k ~ ) whereas for Rv = Rc = 2 5 k ~ t he domi nant pole is at 12.5 MHz with the subsidiary pole l ocat ed at 50 MHz. For such an amplifier a simple, single zero equal i zer can be used to cancel t he effects of the domi nant pole Sj resulting in an amplifier with a frequency response Receiver Design for Optical Fiber Communication Systems 131 4 0 0 2 0 0 I00 e0 ::- 60 o <~ 40 g ._i o ~ ] I I I l - $z - '! I 2 4 6 B t 0 2 0 4 0 6 0 8 0 t 0 0 R F : R c ( k , P . ) Fi g. 4. 28. Po l e h ) c a t i o n s o f trans- impedance a mp l i f i e r vs Rv f or p a r t i c u l a r c h o i c e Rv = Rc Iu I5~ , 1 i ' i ' ' ' r i i t i r ~ \ BASE CURRENT + I "-. FEEDBACK RESISTOR NOISE - - t 6 1 \ C U R R E N T / B_ ~ 4 : r M _ b / s " ~ _ ~ - - - > _ _ / ~ " CT= 3 . 9 pF k~ / " a. ~ I0 I? ~ : 1 4 0 ~j x/ o MEASURED rl:j b= 9011 / .~. z 12 =0.52 / \ / "\ I 3 = 0 . 0 9 / \ / \ / ' B A S E RESISTANCE NOISE \ . / t ,.,, / ,.... / \ i ~ 1 8 ~ i , f i I I h I I k I I t \ . l , 1 0 t 0 0 RF: Rc(k 9,) ' 1 0 0 0 Fi g. 4. 29. Contributions to i n p u t n o i s e from various noise s o u r c e s . Da t a a r e shown f or a n l p l i f i c r wi t h d o mi n a n t p u l e e q u a l i z e d to . / ~ 50 M Hz do mi na t e d by t he s ubs i di ary pol e S 2. Equa l i z a t i o n o f t he s e c ond pol e is al s o possi bl e. The meas ured effective i nput noi s e current for a receiver ope r at i ng at 45 Mb i t . s - 1 is s h o wn i n Fig. 4. 29 a l o ng wi t h t he cal cul at ed noi s e resul t i ng f rom t he base current, f eedback resistor, c ol l e c t or current and base resi st ance 132 R. G. Smith and S. D. Perso~lick cont ri but i ons. For resi st ance val ues gr eat er t han 6 kf L single st age equal i zat i on was empl oyed to equalize the domi nant pol e to a frequency equal t o or gr eat er t han 50 MHz. At low values of R o and hence high col l ect or current s, the cal cul at ed noise is domi nat ed by the shot noise of t he base cur r ent and t her mal noise of t he feedback resistor. At high resi st ance val ues the noise is domi nat ed by shot noise of t he col l ect or current . The mi ni mum noise, occur r i ng for R c = R v = 44 k~, (I c =801aA), has a val ue of 2.5 10 17 A2 which is a fact or of 8(9dB) bel ow the nonopt i mum design r epor t ed in Tabl e 4.1 and a fact or of 1.7 (2.3 dB) great er t han ent ry 4 of t hat same table. At this opt i mum poi nt 50 % of the noise is due to t he col l ect or current , 16% t o the base cur r ent and 34')(, to the feedback resi st or 9. The measur ed noise in these exper i ment s is f ound to be in close agr eement with t heory, except at the highest val ues of col l ect or resi st ance where the dat a fall s omewhat bel ow t heory, due pri nci pal l y to the reduct i on in the noise bandwi dt h due to the effect of the unequal i zed subsi di ar y pole. These dat a serve to demons t r at e t hat usi ng a modes t a mount of equal i zat i on in this case a fact or of four - the noise of the t r ans i mpedance ampl i fi er can closely a ppr oa c h t hat of the high i mpedance front end which, in the case of ent r y 4, used an equal i zat i on rat i o of gr eat er t han 1000:1. The measur ed loss in dynami c range of the t r ans i mpedance ampl i fi er wi t h Rv = Rc = 2 5 k O compar ed to a si mi l ar uni t using 4 kf~ resi st ors is 2 5 / 4 =8 dB. 4.5 Sensitivity of Digital Receivers In the t ype of digital syst em bei ng consi der ed the i nf or mat i on is t r ansmi t t ed as a series of mar ks and spaces occur r i ng at a rat e B. At the receiver the opt i cal signal will be at t enuat ed f r om t hat t r ansmi t t ed and may al so be di st ort ed. It is the j ob of the digital r egener at or t o det er mi ne whet her a mar k or a space was t r ansmi t t ed and to regenerat e the i nf or mat i on wi t h a mi ni mum pr obabi l i t y of error. The er r or pr obabi l i t y is often measur ed in t erms of a bit er r or rat e (BER) which is given by the r at i o of bits i ncorrect l y identified to the t ot al number of bits t ransmi t t ed. Typi cal bit er r or rat e r equi r ement s are in the r ange of 10- o to 10-15 dependi ng upon the appl i cat i on. The sensitivity of a receiver is defined in t er ms of thc received opt i cal power requi red t o achi eve t he desired BER. Thi s power is often meas ur ed in dBm where 0 dB m cor r es ponds to 1 mW of opt i cal power at the receiver. One of the goal s of receiver design is to mi ni mi ze the a mount of power requi red t o achi eve the desired objective, t hat is, to maxi mi ze the sensitivity. Receiver sensitivity depends upon t he t ot al noise of the f r o , t end, i ncl udi ng the noise associ at ed with the signal, and on the t ype of det ect or empl oyed and 9 When RI;=R c the thermal noise of R F is not negligible and (4.83) for the optimum collector current and (4.84) for the tolal noise must be modified. Recei ver Desi gn for Opt i cal Fi ber Communi cat i on Syst ems 133 Sll) s( o) 2 G I 2% D E C I S I O N L E V E L PNOI~ABIL FTY Fig. 4.30. Pr obabi l i t y di st r i but i on func- t i on for two level digital signal. E. t is pr obabi l i t y of falsely i dent i fyi ng a space and El o is pr obabi l i t y of falsely identi- fying a nlark, o- o and a, ar e t he rms noi se levels associ at ed wi t h t he signal levels its properties. Di st or t i on of the received pulse will also affect the receiver sensitivity ei t her t hr ough increased noise associated with pulse equal i zat i on (see Sect. 4.2), or t hr ough i nt ersymbol interference resulting from imperfect equal i zat i on. In this section the sensitivity of a receiver will be eval uat ed for the case of an undi st ort ed pulse with the effects of di st or t i on consi dered in mor e detail in a separat e section. 4.5.1 Gaussian Approximation Al t hough the assumed t ransmi t t ed signal consists of t wo well-defined light levels, the signal plus the noise at the receiver will not be as well defined and may be charact eri zed in terms of a pr obabi l i t y densi t y funct i on (PDF). This probabi l i t y density funct i on describes the pr obabi l i t y t hat t he out put vol t age You , (or i nput cur r ent il, ) has a value v (i) within the i ncrement al range d v ( d i ) . Figure 4.30 indicates graphi cal l y the form such a PDF mi ght take. The details of the PDF depend upon the statistics of the signal, the statistics of t he det ect or 134 R. G. Smith and S. D. Personick especially for an APD - and the statistics of t he non-si gnal -dependent noise sources. The t hermal and shot noise sources of the amplifier are well descri bed by gaussian statistics where the noise is charact eri zed by an rms noise vol t age (current). On the ot her hand t he statistics of t he optical signal are Poi sson in nat ur e while t hose of the aval anche process are charact eri zed by a PDF which has been anal yzed by several aut hor s but which is not easily expressed in a simple anal yt i cal form [4.18, 27, 43 51]. In or der to handl e the sensitivity cal cul at i on exactly it is necessary to use the vari ous statistics in a consi st ent manner. To dat e such an appr oach has proven difficult and has requi red digital comput at i on to i mpl ement it [4.46]. On the ot her hand, the use of gaussian statistics is relatively st rai ght forward and yields simple anal yt i cal solutions. It has been shown, however, t hat the results obt ai ned by assumi ng all statistics are gaussian yield predi ct i ons in close agreement with t he exact calculation, especially for the range of par amet er s encount er ed in optical fiber systems [4.46]. Because of t he comput at i onal simplicity and the close agreement with bot h the exact t r eat ment and experi ment al results, gaussian statistics will be used here. The receiver sensitivities cal cul at ed will generally be within 1 dB of t hose cal cul at ed by ot her met hods; significant devi at i ons bet ween comput at i onal t echni ques occur pri nci pal l y in t he areas of the det er mi nat i on of opt i mum aval anche gain and in predicting the opt i mum l ocat i on of the decision level of the decider circuit [4.46]. These latter t wo par amet er s are easily subject to experi ment al det er mi nat i on in the design of an actual system and, when opt i mi zed, lead to experi ment al l y det er mi ned sensitivities in ext remel y close agr eement - usual l y within 1 dB - with those cal cul at ed by the simplified t heory. Ret urni ng to Fig. 4.30 let t he expect ed values of the signals in the t wo t ransmi t t ed states by s(0) and s(1) and assume the PDFs of the t wo states to be 2 and a 2. Since the noise in each signal state cont ai ns gaussian with variances a o cont r i but i ons pr opor t i onal to the signal, a o and a~ will not in general be equal. With a decision level D, located as shown in the figure, the probabi l i t y t hat a signal t ransmi t t ed as a space is falsely identified as a mar k is pr opor t i onal t o the area E0~ in the figure. Similarly the probabi l i t y t hat a mar k is i ncorrect l y identified as a space is pr opor t i onal to the area E~o. The t ot al pr obabi l i t y of er r or is t hen given by p( E) = p(O)E o, + p(l )E ~ o, (4.98) where p(0) and p(l) are the probabi l i t i es the space and mark are t ransmi t t ed and Eol and El o are the probabi l i t i es t hey are misidentified. When t he statistics are assumed to be gaussian t he probabi l i t y a mar k or space is i ncorrect l y identified, P(E) is given by P( E) = 1 7 (x2/2) ~2~_ Q e - d x , (4.99) 1 0 - 5 I 0 - 6 1 0 - 7 10 . 8 tl_ I 0 - 9 w 10 - I 0 k - I d i 0 - 1 1 Q:: o . i 0 - 1 2 IO-L3 10- 14 I l I Recei ver Desi gn for Opt i cal Fi ber Co mmu n i c a t i o n Sys t ems I I l I I 135 1 0 - 1 5 I 4 5 5 5 5 6 6 . 5 T 7 . 5 8 O Fig. 4.31. Pr obabi l i t y of er r or vs Q where Q = Io - sll (4.100) O" i and s i and o- i are the expect ed val ue and st andar d devi at i on of the ith signal level. To an excellent appr oxi mat i on l e - ( g2/ 2) P( E) = ; (4.101) Q Fig. 4.31 shows P(E) vs Q. Fr om this figure it is seen t hat for Q =6, P( E) ~ 10- 9 and P( E) = 10-15 cor r esponds to Q = 7.9. Because of the rapi d vari at i on of P(E) with Q, a val ue of Q=6 , yielding a BER of 10 - 9 will be used in performi ng numeri cal examples. When the statistics are gaussian the value of the variance, o -2, is equal to the mean square noise associated with the signal level. Thus t he val ue of 0 "2 will be given by (4.21) when the signal and noise voltages are referred to the out put and by (4.52) when t hey are consi dered as i nput currents. Before proceedi ng to tile cal cul at i on of receiver sensitivities it is of interest to consi der what limits exist for t he ul t i mat e sensitivity of a digital receiver. The 136 R. G. Smith and S. D. Personick l i mi t i n g c a s e will o c c u r wh e n t he n o i s e o f t h e a mp l i f i e r is r e d u c e d t o z e r o a n d t h e t r a n s mi t t e r s e n d s s o m e f i ni t e p o we r wh e n t h e s i g n a l is a m a r k a n d z e r o p o we r wh e n t he s i g n a l is a s p a c e . Si nc e i n t hi s i d e a l c a s e t h e c i r c u i t n o i s e o f t h e r e c e i v e r is i d e n t i c a l l y z e r o it is o n l y n e c e s s a r y t o d e t e r mi n e wh e t h e r a t l e a s t o n e p h o t o e l e c t r o n is g e n e r a t e d a t t h e d e t e c t o r i n o r d e r t o d e c i d e i f a m a r k h a s b e e n s ent . I f o n t he a v e r a g e N p h o t o e l e c t r o n s a r e g e n e r a t e d wh e n a m a r k is r e c e i v e d , t h e n t he p r o b a b i l i t y t h a t z e r o p h o t o e l e c t r o n s a r e g e n e r a t e d i n a g i v e n i n t e r v a l c h a r a c t e r i z e d b y a m a r k is e x p ( - N) . F o r a p r o b a b i l i t y o f e r r o r e q u a l t o 1 0 - 9, 21 p h o t o e l e c t r o n s , on t h e a v e r a g e , mu s t b e g e n e r a t e d i n t h e d e t e c t o r f o r e a c h m a r k t r a n s mi t t e d . As s u mi n g a n e q u a l p r o b a b i l i t y o f m a r k s a n d s p a c e s t h e a v e r a g e r e c e i v e d p o w e r is g i v e n b y r//5 = 21 hvB/ 2 wh e r e hv is t h e e n e r g y o f a p h o t o n . Th i s m i n i m u m p o w e r is r e f e r r e d t o a s t h e q u a n t u m l i mi t . U n d e r a n y ot he, " c i r c u ms t a n c e s , s u c h a s n o n z e r o c i r c u i t n o i s e o r f i ni t e p o w e r t r a n s mi t t e d f or a s p a c e , t h e r e q u i r e d o p t i c a l p o w e r i s g r e a t e r t h a n t hi s va l ue . 4 . 5 . 2 p - i - n D e t e c t o r Wh e n t he p h o t o d c t e c t o r i s a p - i - n p h o t o d i o d e < M ) = 1 a n d ( M 2) = 1. F o r t hi s c a s e t h e e x p e c t e d v a l u e s o f t h e s i g n a l i n t h e t wo s t a t e s a r e (qq)b(O) (4. 102) Isig ( 0 ) = ( r / q) b ( l ) # ~(1)= hv a n d t h e me a n s q u a r e n o i s e c u r r e n t s in t he t wo s t a t e s a r e d e f i n e d a s <i2>o a n d <i2)l. T h e s e n o i s e c u r r e n t s , f o u n d f r o m (4. 52), i n c l u d e c o n t r i b u t i o n s f r o m t h e a mp l i f i e r a n d b i a s c i r c u i t pl us n o i s e a s s o c i a t e d wi t h t he d a r k a n d s i g n a l c u r r e n t s wi t h i n t h e d e t e c t o r . C h o o s i n g t h e d e c i s i o n l evel D t o yi e l d e q u a l p r o b a b i l i t y o f e r r o r in t h e t wo s t a t e s , a s wo u l d b e a p p r o p r i a t e wh e n p ( 0 ) = p(1), (4. 100) b e c o me s (~lql b(O)=Q< i2> ~ i2 (4. 103) D - ~hv) wh e r e ( 4 . 1 0 2 ) h a s b e e n us ed. F o r p r a c t i c a l digital r e c e i v e r s , s o l u t i o n o f (4. 103) s h o ws t h a t t h e n o i s e a s s o c i a t e d wi t h t he s i g n a l is n e g l i g i b l e c o m p a r e d t o t he c i r c ui t noi s e. He n c e i t is a g o o d a p p r o x i m a t i o n t o n e g l e c t s i g n a l - d e p e n d e n t c o n t r i b u t i o n t o t h e n o i s e i n wh i c h c a s e < i 2 > 0 = < i 2 > l = < i 2 > c wh e r e t h e l a t t e r q u a n t i t y is t he c i r c u i t n o i s e I o. 10This will nol necessarily be true for analog systems which require larger S/N ratios. See Sect. 4.7. Receiver Design for Optical Fiber Communication Systems 137 Using this appr oxi mat i on (4.103) can be solved to yield D = l / 2 ( ~ ) [ b ( O ) + b ( 1 ) ] (~Tqv) [ b ( l ) - b ( O ) ] = 2 Q ( i2> J /2 . (4.104) (4.105) The decision level is t hus l ocat ed mi dway bet ween the t wo signal levels. Using (4.24) and defining b(O) r = - - (4.106) the average opt i cal power requi red to achi eve an er r or rate, defined by the par amet er Q, is given by 1 +r I / h v\ . 2 1/2 ,P= (4.107) In this expressi on the first factor, l + r / 1 - r , describes the penal t y associ at ed with t ransmi t t i ng a nonzer o power level for a space; the fact or h v / q is the phot on energy in el ect ron volts and the ot her paramet ers have been defined above. In the limit r =O this expression reduces to bY / i2 1/2 ' I p = \ q / Q < >c (4.108) At a wavelength 2 = 825 nm, h v / q = 1.5 eV and for an er r or rat e of 10 9, Q = 6, yielding the simple rel at i on for a p-i-n det ect or oper at i ng at 825 nm *l f i =9<i 2>~ 12 (p-i-n, 2 =8 2 5 nm). (4.109) At a wavel engt h of 1. 3gm where t here is consi derabl e current interest the cor r espondi ng expressi on is t i p = 5 . 7 ( i e > ~ / 2 (p-i-n, 2 = 1.3 gin). (4.110) The decreased val ue of t he coefficient results from the reduced energy of t he phot on at t he l onger wavelength. In the above expressions the sensitivity is expressed in t erms of the pr oduct of the average received opt i cal power and the quant mn efficiency. This quant i t y is used as it is directly rel at ed to the average cur r ent generat ed in the phot odet ect or , a quant i t y which is easily measured. The power sensitivity of the 138 R.G. Smith and S. D. Personick receiver is si mpl y rel at ed to this quant i t y t hr ough a separ at e det er mi nat i on of the qua nt um efficiency of the det ect or. Typi cal qua nt um efficiencies for ant i refl ect i on coat ed silicon devices are 80 %. Referri ng to the expressi ons for the sensitivity of receivers using a p-i-n det ect or it is seen t hat t he opt i cal power requi red to achi eve a given sensitivity is pr opor t i onal to the r oot mean squar e noise. Thus a reduct i on in the t ot al circuit noise by a fact or of 10( 10dB) will result in an i mpr ove- ment of 5dB in receiver sensitivity. Conver sel y if it is desired to i mpr ove a receiver sensitivity by 10 dB of opt i cal power it is necessary t o reduce the ampl i fi er noise by 20dB. For a front end whose noise is domi nat ed by l eakage current s or l oad resi st or t her mal noise this cor r esponds to a 100-fold reduct i on in the current or a si mi l ar i ncrease in t he val ue of R L. For a front end with t he mi ni mum achi evabl e noise, a 10 dB i mpr ovement in opt i cal sensitivity cor r esponds to a 100-fold increase in the figure of meri t of the i nput circuit, (4.65, 87). Bit Rate Dependence When the i nput device is a bi pol ar t r ansi st or t he mi ni mum circuit noise is pr opor t i onal to B 2, (4.86) when base resi st ance can be neglected. One f act or of B results f r om the bandwi dt h of the filter; the ot her f r om the fact t hat t he spect ral densi t y of the noise source is al so pr opor t i onal t o B, (4.82). In this case the sensitivity of an opt i mum bi pol ar recei ver vari es as the first power of B, cor r espondi ng to a fixed energy per bit. The mi ni mum front end noise of an FET front end vari es a s B 3 , (4.64). One f act or of B results f r om the filter bandwi dt h. The r emai ni ng B 2 dependence comes f r om the effective spect ral densi t y of the series noise t er m which t akes this f or m due to the equalizer. The bit rat e dependence of a receiver empl oyi ng an opt i mi zed FET t hen varies as B 1'5. The third limiting case occurs when the noise is domi nat ed by a l eakage current or a l oad resi st or which does not var y wi t h bit rate. In this case the equi val ent i nput noise spect ral densi t y is const ant and the t ot al effective i nput noise is pr opor t i onal to the filter bandwi dt h, and hence to B. In such a case recei ver sensitivity will be pr opor t i onal to B '5 The above dependences assume t hat t he par t i cul ar receiver is desi gned for a given bit rat e B. In the case of the bi pol ar f r ont end this assumes t hat t he col l ect or current is adj ust ed to its opt i mum val ue for t he given bit rate, (4.82). In all cases it is assumed t hat the overal l filter response is adj ust ed to give the desired eye response, (4.53) or (4.54). If on the ot her hand a receiver desi gned for a given bit rat e B is used in a syst em oper at i ng at s ome lower bit rat e B' wi t hout changi ng the el ement values or the filter r esponse funct i on t hen the sensitivity is unchanged f r om its val ue at t he bit rat e B. Thi s results f r om the fact t hat the i nput noise is unchanged, and hence the aver age opt i cal power, (4.108), r emai ns the same. If an opt i mi zed receiver is used at a hioher bit rat e tLe sensitivity will decrease due to i nt er symbol interference resul t i ng f r om the i mpr oper filtering. -20 -30 -40 -50 ,% -60 -70 -80 Receiver Design for Optical Fiber Communication Systems 139 I I J I ] r I I i I I J PIN ~ _ " / I ~ UA NT UM LIMIT .I I L / I I I I r J I I i i L I0 I00 I000 BIT RATE ( M b/s] Fi g. 4 . 3 2 . Ca l c u l a l e d s e n s i l i - vi t y o[" r e c e i ve r u s i n g a p- i - n d e t e c l o r a nd o p t i mi z e d FET or b i p o l a r f ront e nd a s a f unc l i o n o f l he bi t rat e. Quart - t a m l i mi t c o r r e s p o n d s t o 21 p h o t o n s per mar k. D a t a p o i n l s show s o me r e por t e d e xpe r i - me n i a l r e s u h s Experi mental Resul ts Cal cul at ed sensi ti vi ti es of opt i cal receivers, expressed in terms o f ~IP, for syst ems operat i ng at 2 = 825 nm are s ho wn in Fig. 4. 32 for the mi ni mum noi s e levels s ho wn in Figs. 4.22, 25. Al s o s hown on the figure are reported sensi ti vi ti es o f several receivers operat i ng at vari ous bit rates [4. 6, 7, 32 35, 38, 39, 52]. Reas onabl e agreement bet ween experi ment s and the s ome what i deal i zed cal cul at i ons is good. 4. 5. 3 Aval anche Phot odet ect or Through its internal gain mechani s m the aval anche phot ode t e c t or ( APD) mul t i pl i es the primary phot ocurrent o n the average by a factor <M). For a gi ven i nput si gnal current t o the front end, the opt i cal powe r is corres pondi ngl y reduced by (M). Thus t o the extent that the system noi se is domi nat ed by the circuit noi se the receiver sensi ti vi ty is i mproved by the same factor ( M) . The APD is used t o i mprove receiver sensi t i vi t y t hrough its mul t i pl i cat i on o f the primary phot ocurrent . Aval anche gain al s o i ncreases the noi s e o f the system. Fl uct uat i ons in the si gnal current (signal shot noi se) are mul t i pl i ed by the aval anche process whi ch in turn adds its o wn noi se due t o t he r andom nature o f its gai n mechani sm. The noi se associ at ed wi t h currents mul t i pl i ed by the aval anche process is pro- port i onal t o ( M 2) whi ch varies more rapidly than ( M ) z [ 4. 27] . As the aval anche gain is i ncreased the noi se as s oci at ed wi t h the si gnal current i ncreases t o a poi nt where it be c ome s comparabl e to, and ul ti matel y exceeds, 140 R. G. Smith and S, D. Personick the signal i ndependent circuit noise. These characteristics have several con- sequences: First of all, the fact t hat the signal-dependent noise becomes significant results in different noise levels for the mark and the space. As a consequence the opt i mum decision level is not located mi dway between the two signal levels. The second principal effect is t hat there exists an opt i mum value of the aval anche gain beyond which the signal-dependent noise increases more rapidly t han the signal power. With an APD detector the expected values of the signal in the two states are given by I~g(O)= ~ (M)b(O) ( r l q ) ( M ) b ( l ) Isig ( 1 ) = ~ (4.111) The noise in the two states can be divided into two port i ons: one which does not depend upon the aval anche gain and a second which in proport i onal to ( M2) . In the former are cont ri but i ons from the amplifier and bias circuit and leakage currents in the det ect or which do not undergo avalanche multipli- cation. The avalanche gain dependent terms include detector leakage currents t hat are multiplied as well as the shot noise associated with tile signal-generated phot ocurrent . These various terms are included in the expressions for the equivalent input noise, (4.46 52). In order to simplify the following analysis several assumpt i ons will be made. The first is t hat the component of the dark current which undergoes multipli- cation is small. It will also be assumed t hat the signal-dependent noise in a given time slot is due only to the signal within t hat time slot, i.e., ~1 =I ~. Finally it will be assumed t hat the space corresponds to zero t ransmi t t ed power, i.e., b(0)=0. These const i t ut e ideal condi t i ons for a system. Following det ermi nat i on of the sensitivity of the receiver under these ideal condi t i ons the effect of relaxing the various assumptions will be considered separately and the impact on the sensitivity will be evaluated. For the above assumpt i ons the noise current in the space or zero state is given by (i2)c and the noise in the mark or "1" state is given by ( i z) t = ( i2)c + 2q (T~qv) ( M 2 ) B l t b(1 ) . (4.112) With b(O)=O the equat i ons anal ogous to (4.103) for the p-i-n are given by rlq) ( M >b(1)- D = Q ( i 2 ) J / 2 (4.113) Receiver Design for Optical Fiber Communication Systems 141 Solving these equat i ons for b(1) yields b(1)=(hv)~[2<i2)~/2 ( M e ) 1 2qBl~ Q <M~- J (4.114) + The average optical power, assumi ng equal probability of marks and spaces, is given by [hv\ {(ie)J 'z ( Me ) ) q P = ~ q ) Q t ~ ) +qBItQ ( M ~ " (4.115) The first term on the ri ght -hand side of this expression results from the circuit noise ; the second term is due to the noise in the signal. In the limit ( M) = 1 the second term is negligible and (4.115) reduces to (4.108) for the p-i-n detector. Equat i on (4.115) is an explicit function of the average aval anche gain ( M) and of ( M 2) where (Me> = (M>2F(<M)). (4.116) The quant i t y F((M)) is referred to as the excess noise factor and describes the factor by which the aval anche process increases the noise over t hat of a perfect, noiseless gain mechanism. The expression for F((M)) is [4.27] F( <M) ) =<M) [ I - ( 1 - k ) ( <M) - I ) 2 ] \ <M) ' (4.117) where k is the ratio of the ionization coefficients of the holes and electrons in the detector j unct i on. For silicon k~0. 02 to 0.04 in well-designed devices and k~0. 5 for germani um. Equat i on (4.117) also assumcs t hat the aval anche is initiated by the most highly ionizing carrier type. The expression for F((M)) is often approxi mat ed by the expression F = ( M) " with x~0. 3 to 0.5 for Si and x = l for Ge devices. Equat i on (4.117) is a more exact expression, fits the experimental dat a well, and is equally easy to use as an analytical expression as is the approxi mat e exponential form. Substituting (4.116) into (4.115) gives an alternative expression for II/5 O[ <P):- +qBItQF(<M>) I. (4.118) \ q / X[ <M> The first term varies inversely as the gain ( M) while the second t erm cont ai ns a term directly proport i onal to (M). There t hus exists an opt i mum value of ( M) for which ~l/5 is mi ni mum. Subst i t ut i ng (4.117) for F(<M)) and differentiating with respect to ( M) gives the opt i mum gain 1 { < i e e , , 2 ) , , 2 < M > < , , , = k , , e + k - 1 (4.119) 142 R. G. Smith a n d S. D. Personick and t he mi ni mum optical power requi red to achieve the er r or rat e charact er- ized by Q rll~ = 2 h v B l I Q 2 ( k ( M) o p t + 1 - - k ) . (4.120) To a good appr oxi mat i on t he first t erm in the parent hesi s in (4.119) domi nat es the expressi on in which case the opt i mum gain expressi on becomes { ( i 2>J / 2 / 1/2 ( M ) " ' " ~ ~ k q B I , Q ] " ( 4 . 1 2 1 ) The opt i mum gain t hus varies directly as t he f our t h r oot of the mean square system noise and inversely as the square r oot of t he k val ue of t he det ect or. When the front end uses a bi pol ar t ransi st or the noise ( i 2 ) c varies as B z and hence the opt i mum gain is i ndependent of the bit rate. When an opt i mi zed FET is empl oyed the noise varies as B 3 and hence the opt i mum gain varies as B '25. Using (4.121) in (4.120) t he expression for the opt i cal sensitivity is given by q f i , ~ 2 h v B I 1 Q Z [ \ q B i ~ Q ] + l - k . (4.122) Consi der the dependence of this expression on the vari ous par amet er s in- volved : When an opt i mi zed bi pol ar t ransi st or is used the t erm in the radical is i ndepcndent of B and hence the overall sensitivity varies as B 1' j ust as was the case with a p-i-n det ect or. Wi t h an FET t he sensitivity varies bet ween B 1 and B 12s dependi ng upon whet her t he radical is less t han or much great er t han unity. For a given bit rat e and k value, the dependence on amplifier noise is relatively weak, varyi ng at most as the fourt h r oot of ( i z ) and mor e generally as the 0.2 to 0.15 power of the noise. Thus the use of an APD can reduce to a large ext ent the dependence of the sensitivity on amplifier noise. Fi nal l y for ot her par amet er s bei ng fixed the dependence of the sensitivity on the k val ue of the det ect or is generally less rapi d t han k s . Equat i on (4.122) can be present ed in an al t ernat i ve form using the definition of <M),,pt, (4.121), and t he sensitivity of the receiver using a p-i-n diode, (4.108). The resulting expression is (2) q /I~ A I'D = ~ M ~ o p t q f i p -i-" + ( 1 - k ) ( 2 B l l ) h v ( Q 2 ) " (4.123) For well-designed silicon devices 1 - k , ~ 1 and for rect angul ar pulses 11 ~0. 5 (see Fig. 4.6) hence _ ( 2 ) ~'/PAI'D~'~' ~ l"/PP-i-n +( h v B) ( Q2 ) . (4.124) -20 -30 -40 -50 -60 -70 -8 Recei ver Desi gn for Opt i cal Fi ber Communi cat i on Syst ems ] i i ] i 1 i i [ I i i T 143 A P D k = O . O 3 X = 8 2 5 n m B E R : I O - 9 $ l e o / I 1 1 " ~ . T o M .,,.,1 - / FET t / I I f ' I I I I I I I I I I I I IO IOO IOOO B I T R A T E ( M b / s ) Fig. 4.33. Cal cul at ed sensi t i vi t y of recei ver usi ng an AP D wi t h k=O. 03, as a funct i on of t he bit rate. Dat a poi nt s show s ome r epor t ed exper i ment al results Thus the sensitivity of a front end using an APD is equal t o 2/<M)opt times t he sensitivity of t he same receiver using a p-i-n plus Q2 phot ons per bit per second. If the opt i mum gain is 100 the i rnprovcment in sensitivity of an APD over a p-i-n det ect or can be nearl y 50 or 17 dB. This represent s a significant fraction of the avai l abl e margi n between the t ransmi t t i ng and receiving ends of a digital system ( ~50dB) . A similar i mpr ovement in receiver sensitivity using p-i-n det ect or woul d requi re a reduct i on in the noise level of t he amplifier by 34 dB. Experimental Results Fi gure 4.33 shows a plot of the receiver sensitivities comput ed for an APD with k=0. 03 as a funct i on of bit rate for the assumed noise levels previ ousl y discussed. Also shown are some report ed results. Of interest to not e is t hat t he difference in comput ed sensitivity bet ween bi pol ar and FET front ends is much smaller with the APD t han with the p-i-n. The experi ment al dat a are general l y somewhat worse t han those cal cul at ed but in some instances the experi ment al dat a are within 1 to 2dB of t he cal cul at i on indicating t hat pr oper receiver design can lead t o excellent agreement with t heory. Effect of Finite Extinction Ratio When the t ransmi t t ed power cor r espondi ng to a space is nonzero, (b(0) +0), t he noise in the signal in this st at e is nonzer o and is in fact a funct i on of tile 144 R. G. Smith and S. D. Personick - 4 4 - 4 6 - 4 8 "~ - 5 0 ,% - 5 2 - 5 4 - 5 6 - 5 8 ' ' ' ' l I I t i , I l l I I , X k = 0. 0~5 I I I I I I i I i I i l l l I I i I 0 2 0 4 0 6 0 8 0 I 0 0 2 0 0 5 0 0 < M> Fig. 4.34. Calculated sensitivity of a bipolar receiver as a flmction of the average avalanche gain ( M) , for two values of the exline- tion ralio of the source, r. Data point shows the average sensitivity and standard deviation of 53 receivers measured avalanche gain ( M) . The noise levels in the two states are given by ( i2)o = ( i2)~ + 2q (qhq) ( M 2) Bl ,b(O) 4i2), = (i2)c + 2q(~v,)(M2)Blah(1). ( 4 . 1 2 5 ) (4.112) Using these equat i ons al ong with the definitions of the signal levels, (4.111), and subst i t ut i ng into (4.100) yields the following expression for the average optical power required to achieve an error rate characterized by Q rlP=(]q') {I+r~I(I+r)Q2qBI1F(<M))~I - r } [ 1 - r I ( & q B I , F ( <M >) ) =. 4 , . + & < / ~ > , ~ ] ' " = ' l + t \ I - , / < ~ T J ] (4.126) where r is defined by (4.106). This equat i on is seen to reduce to (4.118) in the limit r--+0. Al t hough this expression can, in principle, be differentiated to find the opt i mum gain and the correspondi ng sensitivity, the mat hemat i cs is cumbersome and does not lead to any simple analytical expressions yielding physical insight. Equat i on (4.126) is most easily numerically computed. An example of such a comput at i on is shown in Fig. 4.34 where the receiver sensitivity is comput ed as a function of the APD gain ( M ) for a receiver operating at B=45 Mbi t - s - i. The amplifier noise is (i2)c= 2 x 10 t6 A 2, and for the APD k=0. 035. Two curves are presented: one for r = 0 and one for Receiver Design for Optical Fiber Communication Systems 145 r = 0.05, a value typical 'of what mi ght be achi eved with a laser. Two features are evident : the first is t hat the opt i mum gain is l ower for r + 0 due to the increased noise in the "zer o" state, and secondl y t hat the sensitivity is poor er due to t he l ower opt i mum gain. Also shown in the figure is the average of t he measur ed sensitivity of 53 receivers for which the opt i mum gain was appr oxi mat el y 80 and the mean sensitivity ~7/5 = - 55.7 dB m [4.38]. The agr eement bet ween t he predi ct ed sensitivity and the experi ment al results is qui t e good, demonst r at i ng the fact t hat use of t he gaussian appr oxi mat i on yields accept abl e predi ct i ons of receiver sensitivity. The measured opt i mum gain is l ower t han t hat predi ct ed and t he opt i mum posi t i on of the decision level is appr oxi mat el y 43 % of the peak eye height, closer to the mi dpoi nt t han predi ct ed by the gaussian t heory. For these same par amet er s the effect of ext i nct i on rat i o on t he sensitivity was comput ed as a funct i on of r. An empirical rel at i on was f ound for this case to be A sensitivity [ dB] = 18r. (4.127) Thus for r =0. 1 the opt i cal power penal t y is appr oxi mat el y 1.8dB, etc. Ext ensi on of this relation t o ot her bit rates, amplifier noise and det ect or propert i es shoul d yield a similar rel at i on with a slightly different numeri cal value. Multiplied Dark Current The effects of multiplied dar k cur r ent are similar in nat ure to t hose resulting from t ransmi ssi on of a finite opt i cal power in t he "zer o" state. The presence of the cur r ent limits t he amount of APD gain t hat can be used, t hereby limiting t he sensitivity. The effects of mul t i pl i ed dar k cur r ent can be eval uat ed by addi ng a t erm 2ql m<M2>Bl 2 to the circuit noise. The results agai n are most easily eval uat ed by numeri cal comput at i on. On the ot her hand an i ndi cat i on of the al l owabl e level of dar k cur r ent can be obt ai ned by equat i ng t he noise of the dar k cur r ent at opt i mum gain to t hat of t he circuit in t he absence of dar k cur r ent effects. Thi s leads to the appr oxi mat e result lm ~ <i2)~/ZkQ (4.128) 2F(<M>) " Assuming k=0. 03, Q= 6 and F( <M>) ~ 5 at <M>= 100 the above expressi on becomes i mP 2 x 10-2 ;2\1/2 <, /~ . (4.129) For the above exampl e at B = 45 Mbi t . s -1 the dar k current (before multipli- cat i on) will have an i mpact when it appr oaches 3 x 10 I A. Fr om this appr oxi mat e rel at i on it is seen t hat t he permissible dar k cur r ent varies with bit rat e as B 1' to B ~5 dependi ng on the t ype of front end empl oyed. 146 R. G. Smith and S. D. Personick lntersymbol Interference The discussion of i nt ersymbol i nt erference will be limited to a fcw remarks. It is clear t hat when i nt ersymbol i nt erference occurs, ei t her t hr ough pulse dispersion within the fiber medi um or due to noni deal filtering in the receiver, the signal level as well as the noise in the time slots nomi nal l y occupi ed by spaces will be dependent on the gain of the APD. In addi t i on t o eye cl osure the i nt ersymbol i nt erference will result in increased noise in t he space, which has the effect of reduci ng t he opt i mum gain, and hence the sensitivity. The exact amount of sensitivity loss depends in a detailed manner on how the decision level is det ermi ned in the presence of i nt erference but the power penal t y due t o reduct i on of the opt i mum gain will be at least as great as t hat given by (4.127). 4.6 Effects of Transmitter and Channel Parameters on System Performance To first or der the individual pieces of a fiber communi cat i on system call be opt i mi zed i ndependent l y. The most powerful t ransmi t t ers are used al ong with the lowest loss fibers, and t he most sensitive receivers. Invari abl y i nt eract i ons between these pieces occur which lead to the necessity to make compr omi ses in or der to opt i mi ze t he overall system performance. For example, it was seen above t hat in or der to accommodat e a wide range in optical signal levels (due perhaps to unknown fiber losses and lengths) it may be necessary t o t rade off sensitivity (the fundament al receiver figure of merit) against dynami c range. The sensitivity of a fiber opt i c system receiver is dependent upon the shapes of the individual light pulses which arri ve in each time slot. If the received pulses tend to overl ap (i nt ersymbol interference) they will interfere in the decision maki ng process at the linear channel out put (i.e., deciding if a mar k or space is present). The overl ap can be r emoved by equal i zat i on, but this has t wo di sadvant ages : First, in general, equal i zat i on requires a knowl edge of the pulse shape- which may vary from one si t uat i on to anot her. Second, equal i zat i on, or high frequency enhancement , increases the noise at the linear channel out put which may degrade the receiver sensitivity mor e t han allowing the pulses to interfere by overlapping. The amount of over l ap in the received pulse sequence is a funct i on of the pulse br oadeni ng (dispersion) propert i es of the fiber. The fiber designer l ooks to the receiver designer for gui dance as to the al l owabl e amount of dispersion and how to quant i fy the t r adeof f between dispersion and receiver sensitivity. For a given average t ransmi t t ed power, large ampl i t ude low dut y cycle pulses are preferred in or der to maxi mi ze the receiver sensitivity. However, all sources are to some ext ent speed and peak power limited, and ext ernal modul at or s have peak cur r ent and speed const rai nt s as well. Thus the t ransmi t t er designer l ooks for gui dance in picking the pr oper dut y cycle and, in addi t i on, t r ansmi t t er rise and fall times, pulse to pulse width variations, pulse to pulse ampl i t ude variations, etc., in t hat t hey affect receiver performance. Receiver Design for Optical FiBer Communication Systems 147 4.6.1 Dispersion and Equalization In a typical fiber opt i c link, as shown in Fig. 4.35, t he received pulse shape h v ( t ) is dependent upon the t ransmi t t ed pulse shape and the fiber impulse response: h t r ( t ) and h f ( t ) , respectively. h v ( t ) = htr(t), hr(t ) (4.130) The t ransmi t t ed pulse shape can oft en be appr oxi mat ed by a t rapezoi d as shown in Fig. 4.36. The impulse response of the fiber is mor e difficult to characterize. It can be shown t hat for fibers with significant mode mixing (at splices or due to mi crobendi ng) the fiber impulse response appr oaches a gaussian shape in bot h the time domai n and in t he frequency domai n ( Four i er t ransform) [4.28]. For shor t fibers or fibers wi t hout much mode mixing the fiber impulse response can be ahnost anyt hi ng positive. It can be mul t i modal (more t han one local maxi mum) or uni modal . It need not be a mi ni mum phase f unct i on, etc. It is useful to define one par amet er which charact eri zes the received pulse shape h p ( t ) in a quant i t at i ve way to pr ovi de some insight i nt o the relative sensitivities of a receiver oper at i ng with t hat pulse shape and one worki ng with an ideal nar r ow received pulse. One such par amet er is the rms pulse wi dt h defined as follows [ [ h ( t ) t 2 d t ( r = [ ( h ( t ) d t I h ( , ) , J , l = l ; h ( t ) a t / 1 = " where T= ~ h ( t ) t d t . ( h ( t ) d t " (4.131) Thi s rms wi dt h is the square r oot of the moment of i nert i a of the pulse about its cent er of gravity. It has cert ai n i nt erest i ng propert i es. For exampl e for a n y positive pulses h i ( t ) and h 2 ( t ) we have the following rel at i onshi p: If h3(t ) =h l ( t ) , h z ( t ) (convol ut i on) t hen ~r = (cr~ +az2) '/2 . (4.132) Thus referring t o (4.130) it follows t hat the rms wi dt h of the received pulse is the square r oot of t he sum of the squares of the rms widths of t he t ransmi t t ed pulse and the fiber impulse response. 148 R. G. Smith and S. D. Personick 1 TRANSMITTER j _/-L h t r ( t ) h f ( t } h p ( t ) = h l r ( t ) * hf (~) Htr(OJ) H f (w) Hp (~e)= Htr(W) Hf ((~) ~ J RECEIVER - L NEAR CHANNEL hT(t) h~jt(t) = hp (t ] * h T (t) HT ( ~) Houl (o~)= Hp (c~) H l- (oJ) Fig. 4. 35. Block d i a g r a m of fiber s y s t e m s h o wi n g t r ans mi t t e d a n d recei ved op- tical purse s ha pe s a nd el ect ri cal s i gnal at ti l e o u t p u t o f t he l i near channel " I T , t" T 2 " t T ~ I " Fig. 4, 36. Ap p r o x i ma t i o n to t r ans - mi t t ed pul s e s ha pe 20 m 1 6 z 1 2 > - , , = , ~ 4 i i 01 6 8 I0 ~ L G i 0. 2 0. 4 0 . 6 0 B 1.0 2 4 O-IT Fig. 4. 37. Loss in recei ver s ens i t i vi t y as a f unct i on of tile rms pul s e wi dt h wi t h an a v a l a nc he de t e c t or at o pt i - mu m gain. ( Copyr i ght 1973 (1978) AT & T Co. ) Figure 4.37 s hows the d!fference in sensi ti vi ty (referenced to an i deal narrow pulse) of a typical receiver for vari ous recei ved pul se shapes. It is as s umed that the receiver can equal i ze the recei ved pul se t o a nomi nal out put pulse, and therefore that the received pul se shape is known. It is seen that roughl y speaki ng, there is less than 1 dB penal ty compared t o the i deal i f the rms wi dt h of the recei ved pul se is less than about 0.25 o f the pul se spaci ng expressed in terms of the wi dt h o f the ti me sl ot, T. For larger rms wi dt hs, the penal ty is shape dependent . Receiver Design for Optical Fiber Communication Systems 149 This can be expl ai ned as follows. If t he Four i er t r ansf or m of tile received pulse is calculated, the following expressi on obt ai ns / Hp((o) = ~ h p ( t ) e J " ' d t = Hp(0) [1 +j~o~ where ~ = o -2 + (~-)2, i.e., ? , = ~t " h v ( t ) d t [ h ~ ( t ) d t " '02-? ) . . . (4.133) If now t he pulse is cent ered in time about its cent er of gravity, }-, to r emove an i rrel evant delay, H p ( ) ) = H p ( O ) ( 1 ~ 0 2 f f 2 ) 2 . . . +hi ghe r or der t erms ill (0 . (4.134) Since the receiver per f or mance penal t y depends upon how qui ckl y t he pulse rolls off in the frequency domai n, it is now clear why the rms width charact eri zes the effects of di spersi on for small rolloffs. It is also easy to show t hat no positive pulse shape can have a Four i er t r ansf or m whose magni t ude at some par t i cul ar frequency f t exceeds the dc value H(0) 11. Thus the most ideal pulse shape (smallest rolloff) is a const ant H ( f ) , cor r espondi ng t o a nar r ow received pulse. That is IHp(cO)l<Hp(0), for positive h p ( t ) IHo&o)lid~. l=Hp(O)--,hv(t) is a nar r ow pulse (4.135) Thus far the discussion has concent r at ed on equalizing receivers to pr oduce t he ideal pulse shape at the out put which i nt roduces a high frequency enhancement in the linear channel to compensat e for high frequency rolloffs ill the received pulses. An al t ernat e appr oach is to al l ow t he out put pulses from the linear channel to interfere at the sampl i ng time, r at her t han to equalize. It can be shown t hat for modest amount s of equal i zat i on (less t han 1 dB or so of dispersion penalty) this appr oach can be as good or bet t er t han equal i zat i on [4.53]. The nice thing about this appr oach is t hat it is not necessary to know the received pulse shape, since equal i zat i on will not be empl oyed. Ther e is a slight pr obl em in pract i cal receivers with this appr oach. Pract i cal receivers t ypi cal l y use the peak- t o- peak ampl i t ude of the linear channel out put to provi de a reference for aut omat i c gain cont rol . I nt er symbol i nt erference on the linear channel out put signal due to i nput pulse overl ap can cause the aut omat i c gain cont r ol system to set the aval anche or linear channel amplifier gain t oo t I A smaller value of H p(6O ) increases the value of the integrals 11, I z , 13 and the stun ~t- 150 R. G. Smith and S. D. Personick A/ Fi g. 4.38. Repr es ent at i on of a pul se aft er passi ng t hr ough t he fiber showi ng tails oll t he recei ved signal low (compared to optimal) causing an increased dispersion penal t y in receiver sensitivity. Anot her propagat i on effect of interest in receiver design is the presence of tails on the received pulse stream. This is illustrated in Fig. 4.38. These tails t end to increase the intersymbol interference in digital systems by an amount which depends upon the shape of the tail and the type of line codi ng being used. Tails which are j ust a few time slots long are relatively the worst since t hey add the most uncert ai ni t y to the neighboring pulse ampl i t udes at their sampling times. Tails which are many time slots long are less i mpor t ant since they !end to add to and average with ot her tails, to add a dc level to the received signal which is blocked by an ac coupled amplifier. For a given area in the tail, long low tails are much less i mport ant t han shorter higher ampl i t ude ones. 4. 6. 2 Transmi tter Parameters There are a number of t ransmi t t er parameters which affect the receiver design and sensitivity. The t ransmi t t ed pulse shape is a t radeoff between average t ransmi t t ed power and pulse width. For a given received average power, the received pulse should ideally be as narrow as possible. However, since t ransmi t t ers are peak power limited, the width of the t ransmi t t ed pulse must be chosen as a compromi se between the above requirements. It can be shown t hat except for very dispersive fibers, the opt i mal pulse width is close to full dut y cycle. Since t ransmi t t er response speed is always a consideration, it is typical to make use of this result by choosi ng full dut y cycle signaling in high speed systems. Effects like t ransmi t t er rise and fall times, extinction ratio, pulse-to-pulse width and ampl i t ude variations, pat t ern effects, etc., must be quantified to provide specifications to the t ransmi t t er designer (see Chap. 5). Typically a certain fraction of a dB of receiver sensitivity penalty is allocated to each effect with the allocations adj ust ed as a negot i at i on process takes place. A detailed analysis of the above is beyond the scope of this book. Receiver Design for Optical Fiber Communication Systems 151 Li ne codi ng is al so an i mpor t ant consi der at i on in digital recei ver desi gn [4.54]. I f t he t r ansmi t t ed signal is scr ambl ed, t hen the recei ver can be desi gned to be ac coupl ed wi t h a l ow frequency cut off of 0.01 1% of the bit rat e dependi ng upon t he scr ambl er and the al l ocat ed penal t y in recei ver sensi t i vi t y for basel i ne wander 12. Thi s is permi ssi bl e because scr ambl i ng r emoves ma ny low frequency component s in t he pulse s t r eam which woul d cause excessive basel i ne wander in an ac coupl ed receiver. If scr ambl i ng is not empl oyed, a r edundant line code which wast es bandwi dt h, or some f or m of de r est or at i on like cl ampi ng or quant i zed feedback mus t be empl oyed in the receiver. When desi gni ng t he line code, it is al so i mpor t ant to guar ant ee sufficient t r ansi t i ons for t i mi ng recovery. Scr ambl i ng is al so sui t abl e for this purpose. 4.7 Analog Systems Recei ver per f or mance ill t he case of anal og syst ems is nor mal l y specified in t er ms of a si gnal -t o-noi se rat i o or a car r i er - t o- noi se ratio. Let the t r ansmi t t ed opt i cal power be modul at ed by a si nusoi dal wavef or m in such a ma nne r t hat P(t) = Pt(1 + m sino~t), (4.136) where P t is the aver age t r ansmi t t ed power ; m is t he modul at i on i ndex ; and (o is t he modul at i on frequency. Assumi ng the t r ansmi ssi on naedium is nondi spersi ve, tile received opt i cal power will have the same f or m as (4.136) but wi t h an aver age received opt i cal power Pr. The phot ocur r ent gener at ed in t he det ect or will be given by l =l o ( M > (1 + m sincot), (4.137) where Io = ~ Pr (4.138) and <M> is the aver age aval anche gain. When a p-i-n det ect or is used, <M> = 1. Negl ect i ng the c ompone nt of the dar k cur r ent of the det ect or which is mul t i pl i ed, and i ncl udi ng the nonmul t i pl i ed c ompone nt in the circuit noise, the t ot al aver age noise in the syst em, i ncl udi ng the shot noise associ at ed with t he signal is given by <i2> = < iz)~ + 2qlo< M> 2 F( < M> )B ~rf , (4.139) 12 Baseline wander refers to the time varying random sum of the negative tails of individual ac coupled pulses. This averages close to a constant for adequately scrambled signals. 152 R. G. Smi t h and S. D. Personi ck where Bef f is the effective noise bandwi dt h of the system. The si gnal -t o-noi se rat i o, S/ N, defined in t erms of the rat i o of the mean squar e signal cur r ent to the mean squar e noise current is given by S 1 m2(M)212o N = 2 ( i 2 ) + 2 q l o ( M) 2 F ( ( M) ) B f f ( APD) (4.140) A si gnal -t o-noi se rat i o defined in t erms of the peak- t o- peak signal power is a fact or of 8, or 9dB gr eat er t han the above expressi on. When a p-i-n det ect or is used (4.140) reduces to S 1 m2 I 2 (p-i-n) (4.141) N=2 " ( I-)~ + 2qloB~f f 4 . 7 . 1 p- i - n De t e c t o r Consi der first the case where a p-i-n det ect or is empl oyed. Gi ven t he modu- l at i on index m, the effective bandwi dt h B rr and the circuit noise i2)~, the S/ N rat i o can be det er mi ned f r om (4.141) as a funct i on of the aver age signal cur r ent I o. Two distinct signal dependences are obser ved : When I o is small the circuit noise t er m domi nat es the t ot al noise in which case S 1 m2Io 2 N ~ 2 (i2) (Circuit Noi se Li mi t ). (4.142) In this case, referred t o as the circuit noise limit, the si gnal -t o-noi se rat i o will be pr opor t i onal t o the squar e of the signal cur r ent and i nversel y pr opor t i onal to the circuit noise. Since this limit occurs for small val ues of I 0 it cor r esponds to small S/ N ratios. The ot her limit occurs for large values of I o in which case the noise is domi nat ed by the shot noise associ at ed with the signal. Thi s limiting case, referred to as the shot noise lirnit, has S m2Io - - ~ (Shot Noi se Li mi t ) . (4.143) N 4qBf f The si gnal -t o-noi se rat i o depends l i nearl y on the signal cur r ent a n d is i ndependent of the circuit noise. Since this expressi on is i ndependent of the circuit noise it represent s the f undament al or qua nt um limit for t he sensitivity of an anal og syst em and is the count er par t of the 21 phot oel ect r ons per t i me slot qua nt um limit for a digital system. The qua nt um limit for anal og syst ems is general l y associ at ed with high S/ N ratios. Thus when large S/ N rat i os are requi red of a syst em it is general l y oper at i ng at or near the shot noise limit ; on t he ot her hand when a small S/ N rat i o is requi red the syst em is general l y circuit Receiver Design for Optical Fiber Communication Systems 153 noise limited. Di gi t al syst ems requi re onl y qui t e smal l S/N ratios, i.e., (12) 2 for a p-i-n det ect or, and hence will be circuit noise limited (when a p-i-n det ect or is used). 4.7.2 Avalanche Detector Ret ur n (4.140) for the case of an APD det ect or and consi der the case of a syst em which was initially circuit noise limited with a p-i-n det ect or. As the aval anche gai n is i ncreased f r om ( M) = 1 for a fixed val ue of I o, the S/ N rat i o initially i ncreases as ( M ) z as l ong as t he circuit noise domi nat es. As the APD gai n is furt her i ncreased the S/ N r at i o i ncreases until the shot noise t er m becomes compar abl e to the circuit noise, and for furt her increases in ( M) , t he S/ N rat i o decreases as F ( ( M ) ) 1. Ther e t hus exists an opt i mum aval anche gai n yielding a ma xi mum S/ N ratio. On the ot her hand consi der a syst em which is shot noise limited wi t h a p-i-n det ect or. In this case, wi t h an APD the noise i ncreases mor e r api dl y wi t h ( M) t han the signal and t he S/ N rat i o is r educed for any val ue o f ( M) > 1. Hence an APD pr oduces no i mpr ovement for syst ems oper at i ng in the shot noise limit, or at l arge S/ N ratios. Thus when l arge S/ N rat i os ar e requi red a p-i-n will be the preferred det ect or and when small S/ N rat i os are called for, as in digital syst ems, an APD yields bet t er per f or mance. 4.7.3 Example As an exampl e o f a S/ N cal cul at i on consi der the case of an anal og recei ver with a bandwi dt h adequat e t o car r y a single TV channel . Let the effective noise bandwi dt h be Bet r = 6.25 MHz, choose m = 0.707, ). = 825 nm, and consi der t wo circuit noise levels ( i 2) c = 10- 19 A 2 and 10 17 A z charact eri st i c of an opt i mum recei ver and one with t wo orders of magni t ude gr eat er noise. Usi ng a p-i -n det ect or the S/ N r at i o as a funct i on of t he aver age received power, deri ved f r om (4.138, 141), is pl ot t ed in Fig. 4.39. Fr om this figure it is seen t hat when S/ N rat i os less t han 30 dB are requi red, t he difference in sensitivity bet ween t he t wo circuits is appr oxi mat el y 10dB as bot h are oper at i ng in t he circuit noise limit. Between 30 and 60 dB S/ N rat i os, this difference di mi ni shes and for S/ N rat i os gr eat er t han 60 dB the opt i cal power requi red is essentially the same for t he t wo ampl i fi ers (shot noise limit). It is al so i nt erest i ng to not e t hat for a S/ N rat i os of 80 dB t he opt i cal power requi red is gr eat er t han 0 dB. m, a level compar abl e to t hat avai l abl e f r om cur r ent s emi conduct or lasers and in excess of t hat avai l abl e f r om LEDs. Fi gure 4.40 shows the rel at i on bet ween the S/ N r at i o and the received opt i cal power for the same ampl i fi er noise levels when an APD is used as t he det ect or. The as ympt ot i c values of the circuit noise limit for the t wo ampl i fi er noise levels (using a p-i-n) as well as the shot noise limit are shown by dashed 154 R. G. Smith and S. D. Personick 0 -I0 -8o -3o , % -40 -5O - 60 Io - I I N I I I I / I / P Bel l =625 MHz 2 : -I? ~ ' - f ~ ~ S HOT NOISE UMIT ~ I O I S E LIMIT ,..~, a >c--io-'9 I 1 I I I f 20 50 40 50 60 70 80 SIGNAL- TO-NOISE RATIO (dB) Fig. 4.39. Sensitivity as a function of signal-to-noise ratio for an analog system using a p-i-n detector and for two amplifier noise levels. Asymptotes show regions of circuit noise limi! and shot noise (quantum) limit I ~ 0 I I I t I I / / " APD k= 0.035 / . / _ Beff= 6.25 MHz / / / r- X= 825 nrn / m=07o7 / - 2 o F . / : | ' I . , , , " / " - 30 CIRCUi T NOISE LIMIT ~ , ' ~ S HOT NOISE LIMIT ~ - 4 0 " " " _ o L : Z / F ' " lo-le -70 ~ I0 20 30 40 50 60 70 80 SIGNAL-TO-NOISE RATIO (dB) Fig. 4.40. Sensilivity as a func- tion of signal-to-noise ratio using an APD with same as- sumptions as in Fig. 4.39 lines. The sol i d lines r epr esent t he r el at i on bet ween t he S/ N r at i o and t he opt i cal powe r f or t he o p t i mu m a va l a nc he gai n. Val ues of o p t i mu m gai n ar e s hown al ongs i de t he cur ves f or each decade c ha nge in S/ N rat i o. F r o m t hese cur ves several f eat ur es ar e obs er ved : Fi r st o f all at l ow S/ N r at i os t he o p t i mu m a va l a nc he gai n is r e a s ona bl y l arge, decr eas i ng as t he r equi r ed S/ N r at i o i ncreases, ul t i mat el y r e a c hi ng a val ue of 1 (p-i-n) at a poi nt wher e t he ci r cui t noi se and s hot noi se a s ympt ot e s i nt ersect . The s e c ond poi nt is t hat t he over al l sensi t i vi t y cl osel y a ppr oa c he s t he s hot noi se l i mi t f or all val~_tes of S/ N rat i o. Thus t he AP D can pr ovi de si gni f i cant r e duc t i ons in r equi r ed opt i cal powe r when l ow S/ N r at i os ar e r equi r ed but is i neffect i ve f or l ar ge S/ N rat i os. Receiver Design for Optical Fiber Commuqication Systems 155 Anot her quest i on concerns t he ext ensi on of the concl usi ons reached in the above exampl e t o ot her modul at i on indices and bandwi dt hs. Consi der first the bandwi dt h dependence" When the circuit noise varies as B z , as for a bi pol ar front etad, bot h the circuit noise limit and the shot noise limit are consi st ent with the optical power varyi ng as B for any S/ N ratio. Similarly when an APD is used t he opt i cal power scales di rect l y with bandwi dt h as long as t he circuit noise varies as B 2. When the circuit noise varies as B 3 (FET) the opt i cal power varies as B t'5 in t he circuit noise limit and as B 1' in t he shot noise limit. Thus at low frequencies, where the circuit noise is small, the system is likely to be shot noise limited in which case t he optical power dependence is B 1. At high frequencies where the noise woul d t end to be larger, an API ) coul d be used in which case the shot noise limit is appr oached and a B ~' dependence is expected. Thus to a good appr oxi mat i on the optical power requi red to achi eve a given S/ N rat i o will scale directly as t he circuit bandwi dt h when the opt i mum det ect or is empl oyed. When t he modul at i on index is reduced t he signal current is reduced while the average value of the phot ocur r ent , and hence the shot noise, is unchanged. Thus when the system is oper at i ng in the circuit noise limit m/5 is a const ant indicating t hat the requi red optical power varies as m- ~. On t he ot her hand, if the system is in the shot noise limit m2/5 is a const ant and the requi red opt i cal power varies as m- 2 4.8 Remaining Work Earl y analyses of receiver sensitivity deal t with fundament al limits i mposed by device and circuit par amet er s [ 4. 13, 26, 32] . Experi ment al i mpl ement at i on verified these predi ct i ons yielding receiver sensitivities in many cases in excellent agreement with t heor y [4.33 35, 41]. Fur t her wor k went i nt o t he devel opment of so-called pract i cal receivers with somewhat reduced sensitivity but with great er dynami c range and reduced manuf act ur i ng t ol erances [4.7, 9, 37, 38, 40]. Recently, compl et e receivers including a l l stages of ampl i fi cat i on and decision ci rcui t ry have been fabri cat ed on a single silicon i nt egrat ed circuit [4.I I, 123. At the present time practical, manuf act ur abl e receivers oper at i ng at vari ous bit rates (digital) and bandwi dt hs (analog) are r eady for i nt r oduct i on i nt o fiber opt i c systems. Ther e are, however, a number of areas in which bot h t heoret i cal and experi ment al work as well as device design remains to be done, Among the i nt erest i ng probl ems is the effect of fiber dispersion on receiver per f or mance: specifically in det ermi ni ng the opt i mum equal i zat i on strategy, t aki ng i nt o account the effects of increased amplifier noise and i nt ersymbol interference. Thi s pr obl em can be expect ed to become ext remel y i mpor t ant as t he t rans- mission medi um is pushed to the limit of its frequency response. Some form of adapt i ve equal i zat i on may have to be devised to compensat e for t he vari at i on in dispersive propert i es among fibers. 156 R. G. Smi t h and S. D. Pe r s oni c k Anot her area of interest is in obt ai ni ng a simple model for receiver sensitivity which i ncorporat es the nongaussi an statistics of the aval anche phot odet ect or permi t t i ng efficient analysis of overall system performance and the effects various degradat i ons have on the performance. In the field of devices t echnol ogy several areas are part i cul arl y interesting. GaAs FETs have larger figures of merit t han Si FETs and hence hol d promise for i mproved receiver sensitivities. At the present time, however, gate leakage currents are excessive in most devices and an underst andi ng and solution of this problem is required before full advant age can be taken of this new technology. Also new circuit designs operat i ng at the necessarily high impedance levels must be devised. Consi derat i on has been given in this review to receivers using either p-i-n or aval anche phot odi ode detectors. Recent work on new device structures includ- ing phot o- FETs I-4.55], phot ot ransi st ors [4.56], and emi t t er-det ect or com- binations [4.57, 58] shows promise, and one or more of these devices may be practical for some applications. Work remains in eval uat i ng and i mprovi ng the performance of these devices and in developing models for their noise properties. List of Symbols Ao~Z 2 A 1 ( o J ) A2(~o) b bk B Bc f f C. Cd C~ c~ CR D C a d 4i" <e,~(.,)> d E(,,)) Elo El l 1 f . o F((M)) F ( . ) ) g m h Coefficients in power series expansion of input noise spectral density Preamplifier gain Postamplifier gain Value of input signal (general) Value of input signal in kth time slot Bit rate Effective bandwidth Input capaeilance of amplifier (neglecting Miller effect) Capacitance of photodetector Stray capacitance Stray capacitance including lead and package capacitance of input transistor Total input capacitance (neglecting Miller effect) Decision level in digital system Series noise generator of input amplifier Spectral density of series noise generator Spectral density of noise associated with base resistance Equalizer response Probability a mark is misinterpreted as a space Probability a space is misinterpreted as a mark Frequency [Hz] Noise corner frequency Excess noise factor of avalanche photodetector Filter response Transconductance of input transistor Planck' s constant / I f { l ) h o . , ( l ) h p ( l ) h . ( t ) l I F E hT(t) H ,((,)) H'oo,(y) Ho(.~) H'p(y) H T ( ( D ) HT(y) ia /l i o L(O d d r ( i2 ) d d . ~ f < i ~ > o , , y j . < i ~ ( , o ) > o n d (g, <i2> . . . . . . <i2(o)> ( i 2 > c l r c u i , , ( / 2 ) c ( i 2 ) s l . . . . . <i?(0)> <q(o)> . . . . . . . . . . io i~(l) <i2> o, ( i 2 > 1 [b l c I coplm~.m / g a l e I m I n J s i g I ~ ( ( o ) - / ~ i g ( O ) , / ~ i g ( 1 ) I o , 1o((O) 11, 12, I 3 k I l l M < M > (M),,p, <M ~> < " ~ ( 0 > Receiver Design for Optical Fiber Communication Systems 157 Impulse response of the fiber Output pulse shape Input pulse shape Transmitted pulse shape Transistor hybrid parameter hnpulse response of linear channel Fourier transform of output pulse Normalized fourier transform of output pulse Fourier transform of input pulse Normalized fourier transform of input pulse Frequency response of linear channel Normalized frequency response of linear channel Shunt noise current generator of input amplifier Noise current generator of diode Noise current generator of load resistor Photocurrent Signal current (time dependent) Spectral density of shunt noise generator Spectral density of effective input noise current Spectral density of current generator Total mean square input noise current Mean square input noise current Mean square input noise current due to detector dark curreut Mean square input noise current due to load resistor Mean square input noise current due to series noise source Mean square input noise current due to shunt current generator Mean square input noise current due to signal in the time slot Mean square input noise current due to signals in other time slots Photocurrent Signal current (time dependent) Mean square noise current when space (mark) is received referred to input of amplifier Base current of bipolar transistor Collector current of bipolar transistor Optimum collector current Gate leakage current of FET Component of detector dark current that is multiplied Component of detector dark current that is not multiplied Input signal current Fourier transform of input signal current Signal current when space (mark) is received Diode leakage current Definite integrals involved in expressions for circuit noise Boltzmann's constant ; Summation index ; Ratio of ionization coefficients of holes and electrons Modulation index Carrier multiplication of avalanche photodiode (a random variable) Average or expectation value of avalanche gain Optimum avalanche gain Mean square avalanche gain Mean square output noise voltage due to total photodetector current 158 R . G, S m i t h and S . D . P e r s o n i c k ( n . ~ ( 0 ) ) ( J 7 2 ) s e r ie s Q t 2 ~ h , , . , ( n 2 ( 0 ) , . , . ~ p ( 0 ) , p ( 1 ) P . P , . P ( t ) p. P ( . ) ) P ( E ) q Q r r h , h f ' b ' V e R v Ri+l R i . R o RT s s k s(O), s ( l ) S ~ , $ 2 t , t ' T v o,. v d t ) O ~ Y ) ) ( v ~ l + ) ) V.(.~) X Y Y~,. Y , ( ( o ) Y + . , ~ , , ( . ) ) ~ ' . z r : ( , . ) Z ~ , ( t ) Z . d o 2 t F ,5((.I i t V tTO, ~1 * (superscript) Mean square out put noise of interest Mean square out put noise Mean square out put noise Mean square out put noise Mean square out put noise voltage due to the signal current in the time slot voltage due to phot odet ect or dark current voltage due to series noise sources voltage due to shunt noise sources voltage due to phot odet ect or signal current Total mean square out put noise, all sources Probabi l i t y a space (mark) is t ransmi t t ed Received optical power Tratlsmitted optical power Average received optical power Fouri er t ransform of received optical signal Probabi l i t y of error Electronic charge Paramet er det ermi ni ng error rate in digital systems Ratio of t ransmi t t ed power in space to t hat in mark Base resistance Base-emitter resistance Shockley emitter resistance Feedback resistance of t ransi mpedance amplifier Effective resistance of input circuit Load resistance Responsivity of phot odet ect or Effective transfer resistance of linear channel Peak value of out put voltage Peak value of out put voltage in kth time slot Value of out put voltage when space (mark) is received Location of poles of transimpedance amplifier Time Time slot width; Absolute t emperat ure Out put signal voltage Expectation value of out put voltage Expectalion value of square of out put voltage Fourier transform of out put voltage Exponent in power law approxi mat i on to excess noise factor Normalized frequency variable Input admi t t ance of amplifier Total admi t t ance of input circuit Admittance of i nput circuit minus effect of transistor j unct i on capacitances Feedback impedance of transimpedance amplifier Impulse response of linear channel Response of linear channel Paramet ers used in describing input pulse shape Srnall signal bet a of t ransi st or Quiescent beta of transistor Fact or appeari ng in spectral density of series noise generat or of FET Delta function Quant um efficiency of the phot odet ect or Wavelength of light Frequency of light St andard deviation of noise in the space (mark) Mean square noise in space (mark) Summat i on appeari ng in noise expressions Convol ut i on Complex conjugate Receiver Design for Optical Fiber Communication Systems 159 References 4.1 l.Jacobs: Bell Syst. Tcch. J. 57, 1717 (1978) 4.2 E.E.Basch, R.A. Beaudette, H.A.Carnes: IEEE Trans. C-26, 1007 (1978) 4.3 R.W. Berry, D.J.Brace, l.A.Ravenscroft: IEEE Trans. C-26, 1021 (1978) 4.4 J.I.Yamada, M.Sarvwatari, K.Asatani, H.Tsuchiya, A.Kawana, K.Sugiyama, T.Kimut:a: IEEE J. QE-14, 791 (1978) 4.5 T.L.Maione, D.D.SelI: IEEE Trans. C-25, 517 (1977) 4.6 Y.Ueno, Y.Ohgushi, A.Abe: Proc. 1st Europ. Conf. on Opt. Fiber Commun., IEEE Conf. Pub. No. 132, 147 (1975) 4.7 J. Yamagata, S.Senmoto, Y.[namura, H.Kaneko, T.Takahashi: Proc. 1st Europ. Conf. on Opt. Fiber Commun., IEEE Conf. Pub. No. 132, 144 (1975) 4.8 T.L.Maione, D.D.SelI, D.H.Wolaver: Bell Syst. Tech. J. 57, 1837 (1978) 4.9 J.R, Biard, EL.Stewart: IEEE Electro. Comp. Syrup. Rec,, IEEE 74CH0803-7 EMC, 12 (1974) 4.10 W.H. Hackett, J r., J.R.Jones, C. A. Brackett, L.C. Dombrowski, L. E. Howarth, P. W. Shulnate, R.G.Smith, A.W.Warner, R.S.Riggs: Optical Fiber Transmission I I (Technical Digest, Williamsburg, VA 1977) 4.11 W.C. Brown, D.C.Hanson, T. Hornak, S.Garvey: IEEE Trails. C-26, 976 (1978) 4.12 D.C. Hanson, W. W. Brown, S. E. Garvey, G.F.Girot, E. Heldt : l EEE Trans. C-26, 1068 (1978) 4.13 S.D.Personick: Proc. 1EEE 65, 1670 (1977) 4.14 M.K.Barnoski: Fundamentals of Optical Fiber Communications (Academic Press, New York 1976) Chap. 6 4.15 S.E.Miller, T. Li, E.A.J.Marcatili: Proc. IEEE 61, 1703 (1973) 4.16 T. Li: IEEE Trans. C-26, 946 (1978) 4.17 J.Conradi, F.P. Kapron, J.C. Dyment : IEEE Trans. ED-25, 180 (978) 4.18 P.P.Webb, R.J.Mclntyre, J.Conradi: RCA Rev. 35, 234 (1974) 4.19 H.Melchior, A, R. Hartman, D.P.Schinke, T. E.Seidel : Bell Syst. Tech..I. 57, 1791 (1978) 4.20 H. Ando, H. Kanbe, T. Kimura, T. Yamaoka, T. Kaneda : I EEE J. QE-14, 804 (1978) 4.21 L.R.Tomasetta, H.D.Law, R.C. Eden, l.Deyhimy, K.Nakano: IEEE J. QE-14, 800 (1978) 4.22 H.D. Law, L.R.Tomasetta, K.Nakano, J.S. Harris : Appl. Phys. Lett. 33, 416 (1978) 4.23 M. [to, T. Kaneda, K.Nakajima, Y.Toyoma, T. Yamoako, T. Kotani : Electron. Lett. 14, 418 (1978) 4.24 T.P.Pearsall, M.Pauchon: Appl. Phys. Lett. 33, 640 (1978) 4.25 K.Ahmad, A.W.Mabbitt: Tech. Dig., Inter. Elect. Dev. Meeting, (Washington, D.C. 1978) pp. 646 4.26 S.D.Personick : Bell Sysl. Tech. J. 52, 843 (1973) 4.27 R.J.Mclntyre: IEEE Trans. ED-13, 164 (1966) 4.28 S.D.Personick: Bell Syst. Tech. J. 50, 843 (1976) 4.29 T.Witkowicz: IEEE J. SSC-13, 195 (1978) 4.30 T.Witkowicz: IEEE J. SSC-13, 722 (1978) 4.31 W.Baechtold: IEEE Trans. ED-19, 674 (1972) 4.32 J.E.Goell: Bell Syst. Tech. J. 53, 1771 (1974) 4.33 J.E.Goell: Bell Syst. Tech. J. 53, 629 (1974) 4.34 J.E.Goell: Proc. IEEE 61, 1504 (1973) 4.35 P.K. Runge: IEEE Trans. C-2,:1, 413 (1976) 4.36 J.L.Hu[[ett, T.V.Muoi: IEEE Trans. C-23, 1518 (1975) 4.37 J.L.l,iullett, T.V.Muoi: IEEE Trans. C-24, 1180 (1976) 4.38 R.G.Smith, C. A. Braekett, H.W. Reinbold : Bell Syst. Tech. J. 57, 1809 (1978) 4.39 R.C.Eden: Final Rpt. NASA Cont. NAS5-23333 (1975) 4.40 T.A. Eppes, J.M.Holland : Electro-Optics Syst. Des. 9, 41 (1977) 4.41 W.M.Muska: Bell Syst. Tech. J. 56, 65 (1977) 4.42 P.W.Dorman: Unpublished results 4.43 R.J.McIntyre: IEEE Trans. ED-19, 703 (1972) 160 R. G. Smith and S. D. Persouick 4.44 J.Conradi: IEEE Trans. ED-19, 713 (1972) 4.45 S.D.Personick: Bell Sysl. Tech. J. 50, 3075 (1971) 4.46 S.D.Personick, P. Balaban, J. H. Bobsin, P. R. Kumar : 1EEE Trans. C-25, 541 (1977) 4.47 P.Balaban: Bell Syst. Tech. J. 55, 745 (1976) 4.48 D.G. Messerschmitt: IEEE Trans. C-26, 1110 (1978) 4.49 W.Hauk, F.Bross, M.Otlka: 1EEE Trans. C-26, 1119 (1978) 4.50 J.E.Maxo, J.Salz: Bell Syst. Tech. J. 55, 347 (1976) 4.51 J.C.Cartledge: IEEE Trans. C-26, 1103 (1978) 4.52 See for example papers in: [) Proc. of the 1st, 2nd and 3rd Europ. Conf. on Opt. Comm. ; 1I) Tech. Dig., Top. Meeting on Opt. Fiber Transmission II, Williamsburg, Va. (1977); Ill) Tech. Dig., Intern. Conf. on Integ, Opt. and Opt. Fiber Comm., Tokyo (1977); IV) Special issue on : Fiber Opt,, IEEE Trans. C-26 (1978) V) Tech. Dig., Conf. on Laser and Electro-Opt. Syst. San Diego, CA (1979); VI) Special issue on: Quant. Electr. Dev. for Optical-Fiber Comm., IEEE J. QE-14 (1978) 4.53 S.D.Personick: Bell Syst. Tech. J. 55, 957 (1976) 4.54 Y.Takasaki, M.Tanaka, N.Maeda, K. Yamashita, K.Nagano: IEEE Trans. C-24, 404 (1976) 4.55 C.Baak, G. Elze, G.Walf: Electron. Lett. 13, 193 (1977) 4.56 R.A.Milano, T.ll.Windhorn, E.R. Aaderson, G.E.Stillman, R.D.Dupuis, P.D.Dapkus: Tech. Dig., lnter. Electr. Dev. Meeting, (Washington, D.C. 1978) pp. 650 4.57 T.Ozeki, Y. Ucmatsu, T.lto, M. Yamamoto, Y. Unno: Opt. Lelt. 2, 103 (1978) 4.58 J.A.Copeland, A.G.Dentai, T.P. Lee: IEEE J. QE-14, 810 (1978) 5. Lightwave Transmitters P. W. Shmnat e and M. Di Domeni co, Jr. With 32 Figures Li ght -emi t t i ng di odes (LEDs) and injection-laser di odes (1LDs) have cert ai n uni que characteristics which make t hem at t ract i ve as lightwave communi cat i on sources. Thi s chapt er provi des a discussion of the engi neeri ng consi derat i ons one must make to use these semi conduct or sources in lightwave t ransmi t t ers. Among the advant ages, LEDs and lasers are di mensi onal l y compat i bl e with opt i cal fibers, resulting in high coupl i ng efficiencies; t hey emi t at wavel engt hs coi nci di ng with regions of low opt i cal absor pt i on in fibers; t hey are easy to modul at e; and they offer solid-state reliability, with lifetimes now exceeding 106 h. Nevertheless, t here are many engi neeri ng tradeoffs to be dealt with. This chapt er will first compar e and discuss these t radeoffs in view of differing appl i cat i ons. Then, ci rcui t ry for digital and anal og appl i cat i ons will be exami ned. 5. 1 Charact eri s t i cs Al t hough LEDs and lasers exhibit a number of similarities, striking differences exist which must be consi dered in t erms of t he appl i cat i on and t r ansmi t t er design. One maj or difference is, of course, the spatial and t empor al coher ence of laser light which results in a relatively nar r ow beam of light and nar r ow spectral width. In cont rast , most LED sources are Lamber t i an and have a moder at el y large spont aneous spectral width. These fact ors govern, respec- tively, t he amount of opt i cal power which can be coupl ed i nt o a fiber and the influence of mat eri al (or chromat i c) dispersion on the dispersive propert i es of the fiber medium. A second difference is in speed. The st i mul at ed emission from lasers results in intrinsically faster opt i cal rise and fall times in response to changes in drive current t han rise and fall times obt ai nabl e with LEDs. Thi rdl y, EEDs generat e light al most linearly pr opor t i onal to t he cur r ent passing t hr ough the device. In cont rast , lasers are t hreshol d devices, and t he lasing out put is pr opor t i onal to drive cur r ent onl y above threshold. Compl i cat i ng the pi ct ure is the fact t hat the laser' s t hreshol d cur r ent is not const ant but a funct i on of device t emper at ur e and age. Thi s has a significant i mpact on drive ci rcui t ry for lasers. The effect of t emper at ur e changes on LED out put can be handl ed far mor e simply, and in many cases is not even a probl em. Lastly, differences exist in device reliability. At present, LEDs have subst ant i al l y l onger oper at i ng lifetirnes t han lasers. We will now discuss these compar i sons in mor e specific detail, summari zi ng the par amet er s ill Tabl e 5.1 as we proceed. 162 P. W. S h u m a t e and M. Di Do me n i c o , J r . Table 5, I. Representative electrical and optical characteristics lightwave communications for light-emitting devices suitable for Low-radiance High-radiance Injection LEDs LEDs lasers Voltage drop [V] 1.5 to 2.5 1.5 to 2.5 1.5 to 2 Forward current [mA] 5(1 to 300 50 to 300 10 to 300 Threshold current [mA] NA NA 5 to 250 Output power [mW] 1 to 3 I to 10 1 to 10 Launched power [mW] 0.001 to 0.1 0.05 to 0.5 0.5 to 5 Spectral width [nm, at 800nm] 35 to 50 35 to 50 2 to 3 [nm, at 1300nm] 70 to 100 70 to 100 3 to 5 AInP/AT [ %/ ' C ] - 1 - 1 N A A1 n / - i / A T [ %/<'C] N A N A + 1 Brightness [W/cm2/sr] 1 to 10 10 to 1000 ~ 105 Rise time [10% to 90%, ns] 5 to 50 2 to 20 < 1 Frequency response [ 3dB, MHz] 7 to 70 18 to 175 350 to 1000 Nonlinearity [%] 0.03 to 1 0.3 to 3 0.3 Io 30 Feedback stabilization required? No No Yes MTTF (h, at 25 ' C heatsink) 10 ~, to 109 l04 to 107 3 x 103 to 10 <, Operating temperature [C] - 55 to + 150 - 40 to +90 - 55 to +70 (;hip processing complexity Lowest Low to Medium High Packaging complexity Low Low High Cost Lowest Low High N o t e : Spcctral width is full width at half maximum (FWHM). approximately equal to 1.18 x rms (2o-I width. 5.1.1 Power Li ght -emi t t i ng diodes typically oper at e at cur r ent levels of 50 to 300 mA and with vol t age dr ops across the t ermi nal s of 1.5 to 2.5 V. Dependi ng on quant um efficiency, st ruct ure and series resistance, such electrical power levels t ransl at e i nt o the or der of 1 to 10roW of optical out put when the device is oper at ed cont i nuousl y. Of this out put power, from less t han 1% to less t han 10% is t ypi cal l y coupl ed into a single mul t i mode fiber lightguide [5.1]. Coupl i ng efficiencies on the or der of 10% or mor e shoul d be at t ai nabl e if the light- emi t t i ng area is smaller t han t he area of the fiber core and lenses or ot her optical elements are empl oyed [5. 2-5]. Edge-emi t t i ng LEDs, by virtue of their non- Lambcr t i an radi at i on pat t ern also achi eve ~ 10% coupl i ng [5.6]. These consi derat i ons t ransl at e into power levels from a few laW on one ext reme to several hundr ed I.tW on the ot her typically being l aunched i nt o single mul- t i modc fibers. The specific value depends st rongl y on di ode structure, fiber di amet er, and fiber numeri cal apert ure. Injection lasers oper at e at similar levels of electrical i nput (10 to 300 m A, 1.5 to 2 V), but are t hreshol d devices as illustrated by the l i ght -current ( L - I ) t ransfer ~o Lightwave Transmitters 163 8 E 6 , , = , o a_ 4 d 2 0 0 250 20C 50C THRESHOLD A T 5 0 ( / I 50 100 t50 200 FORWARD CURRENT (mA) F i g . 5.1. The light-power output vs forward current, or L - - l . transfer charac- teristic for a typical injection laser diode. Curves for two typical heatsink tempera- tures are shown charact eri st i c in Fig. 5.1. Below t hreshol d the laser' s out put is spont aneous LED light, whereas above t hreshol d the emission is coherent , of nar r ow spectral wi dt h compar ed with an LED, and generat ed at a high differential quant um efficiency ( d L / d l ) . Cont i nuous opt i cal power levels of several mW per mi r r or are typical of t oday' s laser diodes. Al t hough this out put level is numeri cal l y similar to t hat of an LED, mor e of it is effectively coupl ed i nt o a fiber because the beam is di rect i onal (spatially coherent). Fur t her mor e, a lens can readily be used to advant age since t he emi t t i ng area of a laser is much smaller t han most fiber cores, Wi t h no lens, coupl i ng efficiencies near 30 % are obt ai ned by placing a fiber close to a laser mi rror. By mel t i ng a "lens" on the end of this ot herwi se flat-ended fiber, or by using ot her lens schemes, coupl i ng efficiencies appr oachi ng 100 % are realizable [5. 7-9]. (The very highest efficien- cies are not necessarily practical, however, because of t hei r sensitivity to mechani cal mot i on, assembl y tolerances, etc.) For lasers, then, power levels of appr oxi mat el y 0. 5r ow to several mW are l aunched, an advant age of 10 to 20dB when compar ed with most LEDs. Thi s 10 to 20dB power advant age t ransl at es into subst ant i al l y longer fiber lengths being al l owed between source and det ect or, or higher si gnal -t o-noi se ratios (SNRs) for fixed lengths of fiber in anal og appl i cat i ons, or addi t i onal margi n to accommodat e passive taps or coupl ers for signal di st ri but i on in a dat a net work. 5. 1. 2 Spectral Width The i ncoherent emission of an LED oper at i ng in the 800 900 nm range (e.g., a GaAI As device) usually displays a spect rum whose full wi dt h at hal f maxi mum ( FWHM) is 35 to 50nm. Chr omat i c or mat eri al dispersion in vari ous high- silica fibers results in rms i mpul se br oadeni ng of 1.6 to 3.0 ns/ km at 800 nm (1.0 to 2. 0ns/ km at 900nm) for this range of spect ral wi dt hs [5. 10-12]. Thi s di spersi on in t urn places a 3dB-frequency x length l i mi t at i on on an LED system of appr oxi mat el y 60 to 120 MHz . k m at 800 nm ( appr oxi mat el y 95 to 164 P. w. Shumate and M. DiDomenico, Jr. 190 MHz . k m at 900nm) ~. Thi s l i mi t at i on can be ci r cumvent ed ei t her by oper at i ng at a wavel engt h where chr omat i c di sper si on is less, or by usi ng a sour ce wi t h reduced spect ral width. Chr omat i c di spersi on in high-silica fibers appr oaches zero near 1.3p.m [5.12, 13] 2. Ther ef or e if a "l ong- wavel engt h" LED is used (one whose emi ssi on peaks near 1.3 lam), even t hough its F WHM is as l arge as 100nm [-5.14, 15], the frequency x length pr oduct is i ncreased at least an or der of magni t ude to the vicinity of 1 GHz . k m over what is obt ai nabl e in the 800 to 900nm region. ( Anot her benefit of oper at i ng near 1.3 ~m is the very low at t enuat i on glass fibers di spl ay at t hat wavel engt h. ) The spect ral width of the source is nar r ower if one uses a laser r at her t han an LED. The coher ent lasing out put is far mor e monoc hr oma t i c t han the s pont aneous cmi ssi on of an LED, by a fact or of at least 20. Thus even at 800nm, the frequency x length pr oduct rises to a val ue near 1 GHz . k m, and consi der abl y hi gher at 1.3 t, tm. Ther ef or e the frequency (or bit rate) x l engt h r equi r ement s of a par t i cul ar syst em may di ct at e the choi ce of source or wavel engt h. 5. | . 3 Speed The effective mi nor i t y- car r i er lifetime in a heavi l y doped 0018 to 1019/cm3) di r ect - gap mat er i al such as Ga As is t ypi cal l y about 1 t o 10ns [5.16]. The 10 to 90 % rise t i me of an LED made f r om GaAs, reci procal l y rel at ed to the frequency response of t he device, is at least twice this value, and frequent l y much l onger due to j unct i on (space-charge plus diffusion) capaci t ance in combi nat i on wi t h any st ray cont r i but i ons [5.17]. Rise times for LEDs cur r ent l y avai l abl e range from 2 to about 50 ns, cor r espondi ng to 3 dB frequencies of 7 to 175 MHz. By pr opcr choi ce of dri ve circuitry, malay of the sl ower di odes can be dri ven to show rise times of 10ns or less. Thi s will be discussed in the circuits section of the chapt er. The t i me const ant charact eri st i c of the st i mul at ed- emi ssi on process in lasers is much short er, and mos t lasers di spl ay rise times of 1 ns or less. As wi t h the LED, pr oper choi ce of dri ve ci rcui t ry can mi ni mi ze this value. To achi eve the highest speeds with lasers, and mi ni mum t ur n- on delays, lasers are usually bi ased ei t her j ust bel ow or j ust above t hreshol d and a hi gh-speed dri ve current on the or der of a few mA to tens o f mA is added to this bias. Lasers can be used beyond 1Gb/ s N RZ ( nonr et ur n- t o- zer o) or f r om a bout 500 t o 1000MHz, as compar ed with pr ac t i c al limits of abot , t 100 Mb/ s NRZ (or about 50 MHz) for most LEDs. (Li ght -emi t t i ng di odes have been oper at ed at frequencies much hi gher t han this in vari ous exper i ment al envi r onment s I-5.18-20-1.) 1 This limitation always scales linearly with distance. 2 The second derivative of refractive index dZn/d22, to which chromatic or laaterial dispersion is proportional, goes to zero at an inflection point in n(2). This inflection point is connected with the dipole absorption in glass at longer wavclcngths responsible for classical "anomalous dispersion." Lightwave Transmitters 165 5. 1. 4 Li neari t y Al t hough the out put f r om an LED is appr oxi mat el y pr opor t i onal to dri ve current , mos t LEDs di spl ay some degree of nonl i near i t y in t hei r L- I charact eri st i cs. In the l i ght -generat i ng regi on above t hreshol d, lasers can di spl ay nonl i neari t i es in t hei r L- - I charact eri st i c, ari si ng f r om mode in- stabilities. Gr os s nonl i neari t i es ari si ng f r om j unct i on heat i ng in an LED or ki nks in a l aser can be cl earl y seen f r om a quasi st at i c pl ot of t he L - I charact eri st i c on a r ecor der or oscilloscope. The mor e subt l e di st or t i ons ari si ng f r om cur r ent cr owdi ng or the j unct i on equat i on for LEDs can be measur ed in t erms of har moni c- di s t or t i on levels, two-, three-, or mul t i f r equency (noise l oadi ng) i nt er modul at i on component s , or di fferent i al -gai n and differential- phases responses to a st andar d compos i t e signal. These t ypes of meas ur ement s can be rel at ed t o one anot her as descri bed in [Ref. 5.21, Chap. 10]. As a resul t of source nonlinearities; mos t at t ent i on has been di rect ed to oper at i ng light- wave t r ansmi t t er s, par t i cul ar l y t hose usi ng lasers, in a 2-level (bi nary) digital mode which is fairly insensitive to these di st ort i ons. (Digital pr obl ems ari si ng f r om nonl i neari t i es will be di scussed later.) Cert ai n LEDs, however, di spl ay high l i neari t y where di st ort i on pr oduct s ( har moni c and i nt er modul at i on) lie 50dB [5.22], or even 60dB or mor e [5.23] bel ow the modul at i ng signal level. Hi gh- r adi ance devices such as etched-well ("Burrus") LEDs consi st ent l y dis- pl ay di st or t i on pr oduct s 35 to 45 dB bel ow the signal level [5. 24-26]. (These di st or t i ons have been anal yzed and predi ct ed f r om t heoret i cal consi der at i ons of LED r ecombi nat i on kinetics and di ode st r uct ur e [5.27, 28].) And very rel evant to the pi ct ure, cert ai n new si ngl e- mode laser st r uct ur es are al so showi ng l i neari t y in this r ange ( ~ 50 dB) [5.29]. These levels of di st or t i on are adequat e for ma ny anal og appl i cat i ons. I f not adequat e, t here are circuit designs for i mpr ovi ng t he l i neari t y of ei t her t ype of source, the mos t obvi ous bei ng the use of negat i ve feedback. 5. 1. 5 Thermal Behavi or Nei t her LEDs nor lasers mai nt ai n a const ant opt i cal out put if the device t emper at ur e (more specifically, the j unct i on t emper at ur e) is changed. It is welt known t hat mos t LEDs have a negat i ve t emper at ur e coefficient of a bout - 1 %/ C [5.30] 3. Thi s arises f r om an i nt er pl ay bet ween the t emper at ur e dependence of the band det ai l s on one hand ( compet i t i on bet ween r adi at i ve and nonr adi at i ve processes), and changes in car r i er - gener at i on efficiency on the ot her. A t emper at ur e dependence of this magni t ude can result in a change of 2 dB (a f act or of 1.6) or mor e in opt i cal power over the t emper at ur e range 0 to 70 ~C. For t unat el y, if this present s a pr obl em, t he out put can be stabilized very 3 More accurately, the quantum efficiency closely follows an exponential activation energy behavior near room temperature, with an activation energy in the range 40 to 70meV. 166 P. W. Shumate and M. DiDomenico, Jr. sat i sfact ori l y by desi gni ng a circuit to adj ust t he dri ve cur r ent wi t h t emper at ur e and cancel the predi ct abl e change in LED efficiency. Lasers present far mor e of a pr obl em. The t hr eshol d current s of t ypi cal lasers change by appr oxi mat el y + 1%/C. The 1%/'~C is far mor e di sast r ous in t er ms of out put st abi l i t y t han an LED' s coefficient of si mi l ar magni t ude because its effect is magni fi ed by the high differential qua nt um efficiency beyond t hreshol d. In a t ypi cal laser (Fig. 5.1), a change of onl y 30~'C, which results in a t hr eshol d i ncrease of 3 0 mA in the figure, results in a compl et e loss of lasing out put for levels bel ow 5 mW (al t hough the s pont aneous out put remai ns) 4. Cl earl y this must be deal t wi t h at the t r ansmi t t er - desi gn level and furt her discussion must be deferred until l at er in the chapt er. 5.1.6 Reliability Mean- t i me- t o- f ai l ur es ( MTTFs ) gr eat er t han 107 h at a j unct i on t emper at ur e of 70"C have recently been r epor t ed for cert ai n LEDs [5.31]. Typi cal MTTFs of commer ci al LEDs, bot h high- and l ow- r adi ance devices, are in the 10 "s h r ange [5.32]. For LEDs, it is expect ed t hat l ow- cur r ent - densi t y di odes (J < 1 kA/ c m 2) will have the l ongest MTTFs , the reduced-j unct i on (J ~ 3 to 6 kA/ cm 2) Burrus- t ype device the short est , and t hat lifetimes for all devices will const ant l y show i mpr ovement s as the t echnol ogy mat ures. Impressi ve laser MTTFs of ~ 1 0 6 h at a heat si nk t emper at ur e of 22C (average j unct i on t emper at ur e appr oxi mat el y 32"C) have recently been r epor t - ed [5.33]. Inj ect i on lasers are expect ed to show subst ant i al i mpr ovement s in lifetime as this t echnol ogy mat ures. Typi cal MTTFs of commer ci al l y avai l abl e lasers range f r om 3 x 103 to 105 h at 25 C [5.32] 5. Lasers have mor e pot ent i al da ma ge mechani sms t han LEDs (e.g., mi r r or damage, nonr adi at i ve r ecom- bi nat i on in the act i ve region, defects in the act i ve-regi on geomet r y, etc.) Thus it is r easonabl e to expect that, given equal oper at i ng t emper at ur es and cur r ent densities, lasers will al ways trail LEDs in reliability. The reliability requi re- ment s for any appl i cat i on must be eval uat ed very specifically in each case. Temper at ur e ext remes and the details of the l og- nor mal failure statistics must be consi dered. Power - suppl y l i mi t at i ons may al so come i nt o the picture, because in s ome appl i cat i ons, t her moel ect r i c cool i ng may be used pr of i t abl y to enhance device lifetime. The onl y gener al i zat i on one can make is t hat LEDs are far mor e reliable at higher t emper at ur es t han lasers. Therefore, if oper at i on in an unpr edi ct abl e envi r onment (e.g., out door s) is cont empl at ed, LEDs present l y have a subst ant i al edge over lasers. 4 Lasers with substantially reduced thresholds, however, will be affected to a much lesser extent. 5 For comparison with LEDs specified at 70 "C, laser lifetimes at 25 "C must ~e reduced by a factor of approximately 20 to 70. The exact ratio depends on junction tempcrature, activation energy, thermal impedance and average operating current, which may change widely for a laser during its lifetime. Li ght wave Tr a ns mi t t e r s 167 5.1.7 Other Factors Aside from these technical tradeoffs, pract i cal consi derat i ons include device cost (lasers are mor e expensive t o fabri cat e and to package t han LEDs) and circuit compl exi t y. Presently, cost is a rapi dl y movi ng t arget and must be consi dered, again, in each specific appl i cat i on. Ci rcui t compl exi t y affects cost, bot h in part s count and assembly and set up time. 5. 2 Modul at i on Format s We t ur n now t o the modul at i on format s useable with LEDs and lasers. Bi nary digital met hods, including pul se-code modul at i on (PCM) and adapt i ve delta modul at i on (ADM), are frequent l y preferred in lightwave appl i cat i ons because of t he high signal-to-noise ratios or low bit er r or rates at t ai nabl e for even modest carri er-t o-noi se ratios [5.21, 34]. That is, a bi nary signal t ol erat es large amount s of at t enuat i on or signal degr adat i on arising from dispersion, noise, etc., wi t hout degradi ng the i nf or mat i on cont ent since onl y a two-level (pulse/no pulse) decision must be made. Thi s usually t ransl at es into increased repeat er spaci ng (or increased length of nonr epeat er ed links) when compar i ng digital with i nt ensi t y- modul at ed transmission. Fur t her mor e, bi nar y digital modu- lation is highly compat i bl e with a source like an LED or laser t hat can be switched rapidly, and nonlinearities in the L - I charact eri st i c of t he source do not usually affect its suitability for two-level signaling. Digital transmission, however, acqui res these advant ages at the expense of bandwi dt h expansion. For PCM, an anal og signal must be sampled at twice its highest frequency component (i.e., the Nyqui st rate) in the anal og-t o-di gi t al conversi on process. Thus a vi deo signal of bandwi dt h of 4.25 MHz must be sampl ed at least as oft en as 8.5 x 106 samples/s. In fact, col or vi deo must be sampl ed at the next highest multiple of the 3.58 MHz NTSC chr oma subcarri er frequency 6 in or der to prevent undesi rabl e col or effects in the pi ct ure [5.35, 36]. Therefore, a col or vi deo signal t hat coul d be sent in a 4. 25MHz anal og baseband f or mat woul d instead be sent at 86 Mb/ s if 8-bit codi ng is desired t o preserve a high signal-to-noise rat i o (3 x 3.58 x 10 6 samples/s x 8 bits/sample). Clearly, a PCM signal at this rat e requires st at e-of-t he-art sources (t r < 5 ns), graded-i ndex fibers if l ong t ransmi ssi on lengths are desired (BW > 500 MHz - km) and possibly l ong-wavel engt h LEDs or narrow-l i ncwi dt h sources such as lasers to avoi d mat eri al -di spersi on constraints. In spite of t he advant ages of digital signaling, anal og appl i cat i ons such as di rect i nt ensi t y modul at i on (IM) or subcarri er FM ~ offer simplicity and low 6 Na t i ona l Tel evi si on Sys t em Commi t t e e s t a nda r d for 525-1ine br oadcas t i ng. 7 Subcar r i er FM is used to denot e t he fact t hat t he f r equency of t he l i ght sour ce itself is not bei ng modul a t e d to t r a ns mi t t he i nf or mat i on (direct FM) , but r at her t hat t he f r equency of an rf electrical s ubcar r i er is modul a t e d. Thi s si gnal , in t ur n, i nt ensi t y modul a t e s t he light sour ce. The s ubcar r i er can be a si ngl e cent r al f r equency or sever al i ndependent l y modul a t e d s ubcar r i er s ( mul t i - subcar r i er ) , mul t i pl exed in t he electrical f r equency domai n. 168 P. W. Shumat e and M. DiDomenico, Jr. cost at the terminal ends (11o CODECs) and may frequent l y deserve con- si derat i on 8. IM and mul t i subcarri er FM are realizable using direct i nt ensi t y modul at i on of many LED sources and some lasers because of the adequat e linearity these devices possess, requi ri ng no feedback or ot her schemes to reduce di st ort i on. This is especially t rue for t hose appl i cat i ons where second- or der t erms lie out - of - band [-5.37, 39, 40, 42, 43] and onl y the smaller t hi rd- or der i nt er modul at i on t erms such as 2o) t - ~) 2 and e) t +co2- o~ 3 are i n-band and need be considered. Of course, the devices can be linearized furt her using feedback or ot her techniques, as will be discussed. A probl em peculiar to lasers t hat was not discussed earlier but t hat may limit their appl i cabi l i t y in cert ai n IM links is excess noise. Near t hreshol d, the light out put of injection lasers has super i mposed on it wi deband noise of an ampl i t ude great er t han or di nar y shot noise referred ei t her to the laser cur r ent or to the det ect or phot ocur r ent [5.44 47]. Thus, t he signal-to-noise rat i o (for IM or subcarri er FM) will be reduced. At levels about 10 % above t hreshol d, the excess noise of many lasers dr ops back close to the shot -noi se limit, but for some lasers it may remai n qui t e large (total noi se/ shot noi se=excess noise ~ 10 dB). Thi s is part i cul arl y common near L - I nonlinearities (kinks). ( LEDs are Mso known to exhibit some excess noise, but it is o f a 1/ f nat ur e and smaller in ampl i t ude, t herefore pot ent i al l y less t r oubl esome [5.48].) Some lasers also show sustained pul sat i ons at frequencies rangi ng fi'om hundr eds of MHz to several GHz, and the frequency of such pul sat i ons may decrease with aging. The modul at i on index of such pul sat i ons can be very large, nearly 100%, and t herefore pot ent i al l y di sast rous in cert ai n IM applications. Pul sat i ons can also limit the per f or mance of digital systems and may be a fact or in det ermi ni ng ul t i mat e device reliability in cert ai n applications. At present, until nonlinearities, excess noise, and pul sat i ons are mor e clearly under st ood and eliminated, LEDs woul d appear to be safer choices in IM applications. In publ i shed papers, at t ent i on has been given t o modul at i on and codi ng schemes ot her t han IM, FM, and PCM. For example, pul se-frequency modu- lation, pulse-position modul at i on, and pul se-ampl i t ude modul at i on have been discussed, as well as encodi ng schemes for bi nary dat a such as bi-phase (Manchester), del ay modul at i on (Miller), multilevel, and ot hers [5.49 59]. The compl ex subject of the relative merits of each scheme is beyond the scope of this chapt er but the reader is referred elsewhere [5.21, 34, 60] as well as to t he references j ust ment i oned. It shoul d be recognized t hat whi chever anal og scheme mi ght be di ct at ed by a part i cul ar appl i cat i on, the anal og drivers to be present ed may be suitable as t he interface bet ween a user-provi ded modul at or and t he opt i cal source. Likewise, the digital circuits to be present ed may be used between a digital encoder circuit and the source. 8 Di rect i nt ensi t y modul at i on has received much al t ent i on recent l y for mul t i channel vi deo appl i cat i ons wher e l he bit rat e of a digital composi t e woul d be excessively hi gh [5.37 41] . Lightwave Transmitters 169 - 1 0 ~" - 2 0 - 3 0 d . > bn - 4 0 - I u~ - 50 o 0,_ - 6 0 - 7 0 - 8 0 - 9 0 ..../....A P O FET ~ BI POLAR I I I I I :5 10 30 1 0 0 3 0 0 1 0 0 0 BI T RATE ( Mb / s , NRZ) Fig. 5.2. Typical output power levels vs operating frequency for comnlon lightwave sources, as coupled into single fibers (core diameter ~ 50 70 pro, N.A. ~0. 2 to 0.36). Also shown are receiver sensitivity curves for silicon p-i-n photodiodes and for silicon awdanche photodiodes operated at opt i mum gain. The break near 50 Mb/s denotes a switch from silicon FETs to silicon bipolar transistors to optimize noise performance 5.3 Source Comparisons To summari ze and concl ude tile first par t of this chapt er, the i mpact of many of t he per f or mance par amet er s j ust discussed will be present ed in t wo sets of curves appl i cabl e to digital systems. Fi gure 5.2 is useful in compar i ng the speed-power t radeoffs among the sources. This figm'e also includes receiver sensitivities for 10- 9 bit er r or rate as a funct i on of bit rate. These wel l -known curves for p-i-n phot odet ect or s and APDs charact eri ze opt i mum receiver designs opt i mmn filtering at each bit rate, st at e-of-t he-art semi conduct or and circuit design, and opt i mal aval anche gain in the case of the APD [5.61] (see Chap. 4). Pract i cal designs can achieve per f or mance within 1 to 2 dB of these curves. The curves at the t op of the figure represent typical average power levels l aunched by each source. The ranges of power levels and pract i cal frequency l i mi t at i ons are also indicated. At any bit rate, the difference between the source power and the receiver sensitivity for the chosen phot odet ect or represent s a margin t hat can be di st ri but ed among fiber, connect or and splice losses, t emper at ur e and end-of- life allowances, addi t i onal margi n for l ower bit er r or rates, l ess-t han-opt i mal receiver designs, etc. The al l ocat i on of these quant i t i es in assembling a t ransmi ssi on loss budget has been discussed elsewhere [5.62 65]. 170 P. W. Shumate and M. DiDomenico, Jr. m lOOK E 1oK Z 3 ~E o I00 ( J ~" 10 { E s LASER-APD f--FIBER MODAL HIGH- RADIANCE LED-APD / DISPERSION / / L E D C H R O M A T , C - - - - - - - - - - - - / D . S E R S , O . - ~ ~ ~ X= B25nm .- L E D & THERMAL ~ DRIVER NOISE LIMITED ~- LIMITED i r i i I s s r l t l I t I I t l l , ~ i , J J 10 100 1000 BIT RATE ( Mb/ s) Fig. 5.3. Representative t ransmi ssi on di st ances as a function of b i t rate for various combi nat i ons of sources and detectors. For these curves, 2=825 nm, phol odi odes and bi pol ar t ransi sl ors arc silicon, and fiber loss is 5 dB/km. Fiber core di amet er is 55 gm and N. A. =0. 23 As an exampl e one finds f r om Fig. 5.2 a p-i -n recei ver sensi t i vi t y of - 44 dB m at 30 Mb/ s (again, NRZ f or mat and BER= 10-9). If one selects the surface- emi t t i ng br oad- ar ea LED wi t h an aver age out put power of - 2 0 dB m (10 laW) i nt o a fiber, t here r emai ns a t ot al mar gi n of 24 dB to al l ocat e a mo n g loss and syst em mar gi n cont r i but i ons. For a fiber wi t h a cabl ed loss of 5 dB/ km and a 2 dB connect or on each end, one coul d assembl e a 4 km link if t emper at ur e, end-of-life, and ot her syst em mar gi ns were negl ect ed, Wi t h these addi t i onal consi der at i ons included, somet hi ng on t he or der of 3 km is mor e reasonabl e. Fi gure 5.3 uses the dat a of Fig. 5.2 al ong wi t h some r epr esent at i ve assumpt i ons about mar gi n consi der at i ons and present s length vs bi t - r at e plots. Again, one mus t realize t hat these curves por t r ay the general charact eri st i cs of devi ces; a specific source f i ber - det ect or combi nat i on used with specific ci rcui t ry may differ by ma ny dB f r om Fig. 5.3. Nevert hel ess, Fig. 5.3 can usefully be used t o i ndi cat e t ypi cal combi nat i ons of sources and det ect or s and their l i nk-l engt h capabi l i t i es at different bi t rates. Al t ernat i vel y, Fig. 5.3 can be regenerat ed by the r eader t o compar e specific devices, fibers, etc. 5.4 Digital Drive Circuitry Havi ng selected a l i ght -emi t t i ng device based oll t radeoffs rel evant to the par t i cul ar appl i cat i on, we now t urn t o t he circuit aspect s of oper at i ng t he device. Fi rst we will discuss digital dr i ver desi gns for LEDs since LEDs ar e mor e Li ght wave Tr a ns mi t t e r s 171 CI R1 +Vc LED@. < ~ R 2 - - - - - - 7 [ I f I Fig. 5. 4. A c o mmo n - e mi t t e r s a t ur a t i ng swi t ch. Res i s t or R 2 l i mi t s t he cur r ent t hr ough t he LED: R~ a nd ('~ ar e set by t he i nput levels wi t h C~ pr ovi di ng t he desi r ed "speed- up' " or over dr i ve to i ni t i at e t ur n- on st rai ght forward to operate. All of the circuits to be discussed are rat her el ement ary so t hat no lengthy analyses are needed or supplied. No at t empt is made to recommend a best choice for a driver : the applications are t oo varied, and the best choice for 300 Mb/s, for example, mi ght not be the best choice for 10 Mb/s. Instead, the operat i ng performance of each circuit described is given where details have been published. Our purpose is mai nl y to review the various workabl e approaches reported in the literature. 5 . 5 L E D D r i v e r s For bi nary (2-1evel) digital applications, a current in the range 50-300 mA must be switched on and of f at high speed t hr ough an LED ill response to a low-level dat a-i nput signal. A small dc forward bias ( ~ 1 mA) added to the switched current may prove advant ageous in high-bit-rate applications by mai nt ai ni ng charge on the diode' s capacitance. The t radi t i onal circuit for cont rol l i ng such currents is a bipolar t ransi st or switch operat ed in the common-emi t t er confi gurat i on (Fig. 5.4). This circuit offers current gain (the base current is smaller by a factor of [3 t han the base drive current), a small voltage drop across the switch (in sat urat i on, with the collector-base j unct i on forward-biased, the collector-to-emitter voltage VCE<~,t ) is approxi mat el y 0.3 V), and ease of provi di ng i ndependent bias to the LED. In Fig. 5.4 such a bias is provided t hr ough R 3. The speed of the common- emi t t er switch is limited, however, compared with ot her confi gurat i ons to be discussed. Bandwi dt h is t raded off for current gain, al t hough this can be compensat ed somewhat by preemphasizing (over- driving) the base current duri ng the t urn-on period. In Fig. 5.4, preemphasis is accomplished by using speed-up capaci t or C~. For certain LEDs, speed can be limited by the fact t hat the LED drive appears to come from a current source (the reverse-biased collector-base junction). Thus, if the LED has a high 172 P. W. Shumate a n d M. DiDomenico, Jr. + 5 V 1 / 2 7 4 S t 4 0 I I ] R1 t , 4 1 1 f 3 8 0 t r , ~ _ ~ V c c + 5 V c T Y - , - J l t - - - 4 o 2 %+ , . " b /47-- 07 Fig. 5. 5. (a) A s i mp l e , l o g i c - l e v e l - c o mp a t i b l e L E D d r i v e r ut i l i z i ng a c o m m e r c i a l l i n e - d r i v e r i m e g r a t e d c i r c u i t ; (hi t h e i n t e r n a l d e t a i l s o f t h e l i ne d r i v e r capaci t ance, dri ve cur r ent is initially di vert ed into char gi ng this capaci t ance, reduci ng the cur r ent avai l abl e for light generat i on. Thi s can be offset in t wo ways : a speed- up capaci t or can be used as descri bed above, or the LED can be f or war d- bi ased near its cut-in voltage. Al t ernat i vel y, one must go to a low- i mpedance dri ver such as an emi t t er-fol l ower circuit to pr ovi de a vol t age st ep r at her t han a cur r ent step. Finally, speed of the common- emi t t er dri ver can be limited by the t i me requi red to r emove mi nor i t y- car r i er charge st ored at (principally) the col l ect or- base j unct i on dur i ng sat ur at i on. Thi s effect can be mi ni mi zed by using a Schot t ky di ode cl amp bet ween the col l ect or and base of the drive t ransi st or. Negat i ve feedback t hus pr ovi ded limits the ext ent of sat ur at i on to a bout 0.3 V while the di ode itself adds no addi t i onal mi nor i t y- car r i er st ored charge. (A Schot t ky di ode is a maj or i t y - or hot - car r i er - device.) Such cl ampi ng is aut omat i cal l y provi ded if one uses cert ai n commer ci al T2L i nt egrat ed circuits. For exampl e, Schot t ky- cl amped 74S04 i nvert ers and 74S140 line dri vers are capabl e of driving LEDs at useful current levels. Each of the t wo sect i ons of a 74S140 can drive an LED at 60 mA, and the sections call be paral l el ed for 120mA if one follows or di nar y current -equal i zi ng pract i ces (splitting the l oad resi st ance a mong the out put s) or assures t hat the paral l el ed gat es al ways switch at preci sel y the same time. Bot h t he 74S04 and the 74S140 al so pr ovi de compact size, T Z L signal-level compat i bi l i t y, and T2L noise mar gi ns (0.3 V rain when used with 74S-series logic, or 0.4 V mi n when used wi t h 74-series logic). A 74S140 LED dr i ver is shown in Fig. 5.5a. Wi t h a 50f2 resi st or to limit the LED cur r ent to 60 mA and with a 47 pF speedup capaci t or , opt i cal rise times of 5 ns have readi l y been at t ai ned usi ng LEDs wi t h 150-200 pF of capaci t ance. Resi st or Rl is selected to dr aw a const ant small cur r ent of about I mA t hr ough the LED as di scussed earlier. A few obser vat i ons can be made here r egar di ng the use of T2L logic gat es as drivers. In Fig. 5.5b showi ng the i nt ernal circuit of a 74S140 [5.66], the series connect i on of Q1 and Q2 is called a t ot em pole. Because TZL t ot em poles are Lightwave Transmitters 173 usually requi red to sink mor e cur r ent t hr ough Q2 t han t hey source t hr ough Q1, Q2 is a physically large, hi gh-current t ransi st or. Ill an or di nar y gate it can sink 16- 20mA. On the ot her hand Q1, a smaller device, can source onl y 1 mA or less. (The 74S 140 is one of a few except i ons : it can sink 60 mA or source 40 m A.) In general, one shoul d connect the out put as shown in the figure, r at her t han reversing the di ode and gr oundi ng its n cont act (in which case Q1 now serves as an emi t t er follower). Connect ed pr oper l y as in Fig. 5.5, t he t ot em pole has one pot ent i al di sadvant age. The presence of Q1 adds an addi t i onal const rai nt to even the simplest schelnes for forward biasing the LED in the off state. For exampl e, the current drawn t hr ough Q1 by R~ must not exceed Ion, regardless of the val ue of the LED bias. Finally, one shoul d be aware of t he short crossover i nt erval duri ng switching when bot h Q1 and Q2 may be on. Hi gh-frequency noise t ransi ent s (glitches) are easily generat ed, appear i ng on the power suppl y rails unless adequat e suppressi on measures are taken. Fi gure 5.6 shows a discrete t ot em- pol e dri ver combi ni ng hi gh-current capabi l i t y with commerci al emi t t er-coupl ed-l ogi c (ECL) compat i bi l i t y. Opt i cal rise times of 6- 7ns have been obt ai ned using hi gh-capaci t ance LEDs ( 200- 500pF) with no off-state bias. The l ower t ot em- pol e t ransi st or acts duri ng t ur nof f to remove charge st ored in t he LED. An i mpor t ant vari at i on of the col l ect or-l oaded common- emi t t er switch is the emi t t er-coupl ed or cur r ent - r out i ng circuit shown in Fig. 5.7. Al t hough this circuit resembles a linear differential amplifier, it is oper at ed out si de the nar r ow range of linearity in t he switching mode of operat i on. The circuit funct i ons as if it were st rongl y overdri ven at the i nput while at the same time remai ni ng out of sat urat i on. This mode of oper at i on follows fi'om t he use of the cur r ent source Q3 and selection of appr opr i at e i nput drive levels, as discussed in [5.67]. The effective overdri ve results in switching speeds faster t han a col nmon- emi t t er amplifier woul d exhibit were it driven from a COlnparable st ep-i nput signal. Emi t t er-coupl ed drivers have been used at speeds beyond 300 Mb/s. With the const ant - cur r ent source as shown, the LED can be placed arbi t rari l y in ei t her col l ect or lead. The cur r ent source shoul d be const ruct ed with a fast t ransi st or havi ng a low out put capaci t ance Cob SO t hat the cur r ent supplied is not affected by the high-speed switching oper at i ons via col l ect or-t o- base coupling. If, for simplicity, however, a resistor is used in place of the current source, one shoul d be aware t hat the LED out put may r epr oduce ripple and noise present on the i nput signal if placed in the collector circuit of Q~. The emi t t er-coupl ed dri ver is readily compat i bl e with commerci al emi t t er- coupl ed logic and, being a const ant - cur r ent driver, no significant t ransi ent s are placed on t he power suppl y rails dur i ng switching. The compl et e dri ver shown in Fig. 5.8 has level-shifting t ransi st ors to make the circuit fully level com- patible with ECL (high = - 0 . 8 V, l o w= - 1.8 V). This shifting is necessary if one wishes to have a gr ounded light emi t t er since a typical LED requires a vol t age dr op of about 1.5 to 2.5 V : it is not possible ot herwi se to gr ound the p cont act of the LED and still switch the circuit with ECL-level signals. 174 P. W. Shumate and M. DiDomenico, Jr. t ~ I / 3 MC102t0 M M 4 0 4 9 k ~ Fi g. 5.6. An ECk- compat i bl e t ot em- pol e 4 7 ~ _ . _ ~ t o o o dr i vcr V I N < Q1 ( ) Q2 "+Vec V B B R ~ + Vcc R3 D~ D:~ Fig. 5.7. A basi c emi t t er - coupl ed switch wherein the LED dr i ve cur r ent is det er mi ned by the const ant - cur r ent source Q~ The bi asi ng pr obl em can be el i mi nat ed by connect i ng the p cont act of the LED t o + 1 V or more, r at her t han to gr ound, but this may be i nconveni ent . If one is willing to suppl y this posi t i ve bias, the circuit shown in Fig. 5.9 usi ng a commer ci al ECL i nt egrat ed circuit is wor t h consi deri ng [5.68]. The 10210 i nt egrat ed circuit cont ai ns uncommi t t ed npn t r ansi st or s which can be con- nect ed in parallel and used as shown in the figure. To at t ai n t he highest speed f r om an LED, one must resort to a low- i mpedance dri ver - one t hat supplies a l ow- i mpedance vol t age step, wi t h or Li ght wave Tr a ns mi t t e r s 175 A; L 121 ilO I/gMCIOItE 1.88K 3OO 196 - -5.2V Fig. 5.8. A compl et e ECL- c ompa t i bl e emi t t er - coupl ed dri ver, useful to at l east 50 Mb - s -1. A change t o a fast er ECL logic f ami l y a nd 2 GHz t r a ns i s t or s will ext end t he ope r a t i on of t hi s ci rcui t to 300 Mb . s wi t hout pulse shaping (speed-up) to the device. The space-charge and diffusion capacitance are thus charged as rapidly as possible. The simplest low-impedance driver is of the emitter-follower (or common- collector) type shown in Fig. 5.10. This circuit is known to be capable of extracting 2.5 ns optical rise times from 180pF LEDs, clearly adequat e for 100 Mb/ s operat i on [5.69]. The Darl i ngt on confi gurat i on of QL and Q2 (to reduce source impedance) and the combi nat i on of R~ and C 1 (to compensat e the pole arising from R 1 and CLE o while permitting the i nt roduct i on of R~) bot h serve to optimize the speed of the circuit. Since electrical rise times are often in the fi' equency region where /3 is complex (i.e., near the cut-off frequency in current gain, or near f ) , the out put emitter load, if not purely resistive, can induce oscillation in this circuit. One should be aware of this potential problem and the steps needed to prevent it [5.70, 71]. The other type of low-impedance driver we shall discuss is the shunt confi gurat i on shown in Fig. 11. Wi t h the switching element now in parallel with the LED, a low-impedance path is provided for t urni ng off the LED by shunt i ng current ar ound it. The LED can be slightly reverse-biased fit t urn-off to aid in removi ng stored charge as shown in the figure by having VEE below ground. In this case, the Schot t ky clamp limits the extent of the reverse bias while at the same time not 176 P. W. Shumate and M, DiDomenico, Jr. o ~Z %
? ~ h YSI ~Y h YEs J ] " ) o ~ ~, 2.~. .~. o , & 0 > Lr~ + w J <i Y-?) :~-~vv~ + Y C ) ~ Y Z Y ' ~ Y C ) Ill z : ~ i.Z Lightwave Transmitters 177 i nt r oduci ng any mi nor i t y- car r i er st ored charge of its own. Alternatively, if no reverse bias is desired, the t ransi st or emi t t er can be gr ounded, in which case the Schot t ky di ode is no l onger required. Resistor R 1 and the LED capaci t ance det ermi ne the t ur n- on per f or mance of this circuit. Fi gure 5.12 shows a simple i mpl ement at i on of t he shunt dri ver using a st andar d TZL i nt egrat ed circuit. The shunt dri ver offers ot her advant ages. By replacing R 1 in Fig. 5.11 with a const ant - cur r ent source, the design of which is now made easier by the increased vol t age available (the switch is no longer in series with the LED), power-suppl y noise can be reduced. In shunt -dri ver circuits, charge can be mai nt ai ned on the LED capaci t ance by placing a resistor between t he LED and the shunt-switch col l ect or (e.g., at poi nt A in Fig. 5.12). Any pole i nt r oduced by this resistor can be compensat ed in the usual manner. Finally, an ECL- compat i bl e shunt driver, again maki ng use of the 10210 i nt egrat ed circuit, is shown in Fig. 5.13. Thi s circuit appears in [5.68]. Not e t hat it cont ai ns a const ant - cur r ent source, and the cur r ent l oad on bot h the + 5 and - 5. 2 V power supplies is nearly const ant . The magni t ude of cur r ent fluc- t uat i ons on the - 5 . 2 V rail is det er mi ned by the selection of R1: the emi t t er cur r ent set by R1 shoul d be slightly higher t han the cur r ent from the current source. Thus, the shunt (left-hand-side) t ransi st or will at t empt to sat urat e, the ext ent of sat ur at i on limited by the Schot t ky diode. At this point, the digital LED drivers so far discussed have been for 2-level, on- of f signaling. If one is able at t he receiver to di scri mi nat e among mor e t han t wo levels, mor e i nf or mat i on can be sent in a single bit interval. For example, four discrete levels (off, 1/ 3on, 2/ 3on, full on) describe all four possible combi nat i ons of t wo bi nary bits. Therefore, if these four levels can reliably be decoded duri ng one signaling interval, the bit rate associ at ed with this interval is doubl ed. Similarly, 8 levels are requi red to triple the bit rate, 16 levels for quadrupl i ng, etc. On occasion, 3-level signaling can be used not to increase the i nf or mat i on rate of a channel but to code a signal in such a way to make the receiver mor e stable to widely varyi ng dat a patterns. For example, the frequency spect r um of NRZ dat a cont ai ns a dc component cor r espondi ng to indefinitely l ong strings of I s or 0 s. To respond to these pat t erns a receiver must ampl i fy at dc or very low frequencies maki ng it susceptible to dc drift and/ or l ow-frequency noise. Long time const ant s also may cause dynami c- r ange and i nt ersymbol - interference problems. Wi t hout using 3-level codi ng such probl ems are usually solved by t ransmi t t i ng or di nar y 2-level opt i cal pulses aft er scrambling, bi -phase (Manchest er) encodi ng, Miller encodi ng, etc. While also aiding in clock recovery at the receiver when needed, these schemes nevertheless requi re addi t i onal synchronous circuitry, which can restrict t he applicability of a system. Fi gure 5.14a shows a por t i on of an ar bi t r ar y NRZ dat a pat t ern. Figure 5.14b shows the same dat a pat t ern coded in a 3-level f or mat [5.72, 73]. Not i ce t hat this encodi ng scheme t ransmi t s onl y chan.qes in the NRZ pat t ern as 178 P. HI. Shumate and M. DiDomenico, Jr. ) . . . . . . +5V i N ~ ~ . ~ " ~ R t MECL 10210 - 5 2 V Fig, 5.13. An ECL-compatible shunt driver with a constant-current source for the LED [-5.68] I I I (a) o F-~~ o (b) I-L- l j ~ ( c ) i- - 1 J - - I I - t l I I I i I I I 0 1 1 1 0 J " / ' - - ~ 0 n i _ _ i - ! . - i r _ _ l Y t i U 1 F1 Fig. 5.14a~t. Several coding schemes discussed in the text. (a) unencoded NRZ: (b) 3-level edge encoding: (e) 3-level alternate-mark-inversion encoding: (d) bi-phase or Manchester encoding, commonly used in tape recording and lightwave applications t r ans i t i ons f r om hal f powe r t o full powe r or f r om t he hal f - power t o t he of f level in an RZ f or mat . A dc hal f - power level is t hus ma i nt a i ne d c ont i nuous l y for possi bl e AGC use at t he recei ver. But t he ac ampl i f i cat i on r equi r ement s of t he recei ver ar e na r r owl y defi ned by t he wi dt h of t he t r ansi t i ons, opt i ma l l y equal t o hal f t he NR Z bit i nt erval . Al so s hown in t he s a me fi gure ar e a 3-level a l t e r na t e - ma r k- i nve r s i on ( AMI ) f or ma t and Ma nc he s t e r or bi - phas e encodi ng. The AMI i or ma t (Fig. 5.14c) however , a l t hough i l l ust rat i ve of a 3-level code, has c l oc k- r e c ove r y probl ems. . Thes e pr obl e ms ar e c i r c umve nt e d by i nser t i ng ext r a t r ans i t i ons ( not r ecogni zed as Is) in l ong st ri ngs of 0 s [5, 21]. V$ N + VCC R DELAY k / ONE- SHOT MV Li ght wave Tr a ns mi t t e r s 179 F i g . 5.15. Re pr e s c nt a t i on of how mul t i l evcl s i gnal i ng can be a c c ompl i s he d wi t h a n LED (or laser) R BIAS (OPTIONAL) 56 1 5 0 Fig. 5.16, A c ompl e t e T2L- compt l t i bl e 3-level edge- encoder ci rcui t +5V 56 ( ~ L E D 0-- 54/74S00 ~ 54/74S04 Since ma ny LEDs are r easonabl y l i near sources, or at least t hei r non- l i neari t y is pr edi ct abl e and st abl e with aging, a dri ver circuit such as shown in Fig. 5.15 can be used to generat e mul t i l evel signaling. Since the vari ous open- col l ect or "swi t ches" do not i nt eract , the swi t chi ng on of any one of t hem results in a cur r ent of appr oxi mat el y (Vcc-VLW D - V c E ( s , t ) ) / R . Cl earl y if n of these switches are on si mul t aneousl y, the cur r ent is n times larger. Thi s circuit is useful for gener at i ng t hree or four di scret e dr i ve- cur r ent levels. For mor e t han four levels, wi t hout r esor t i ng to t emper at ur e cont r ol of the LED and/ or feedback st abi l i zat i on of some sort, the errors in the levels will resul t in unsat i sf act or y bi t - er r or - r at e per f or mance at the receiver. Fi gur e 5.16 shows a compl et e T2L- compat i bl e circuit for encodi ng the leading and t rai l i ng 180 P. W. Shumate and M. DiDomenico, Jr. edges of NRZ dat a accor di ng to the scheme depi ct ed in Fig. 5.14b. Monost abl e mul t i vi brat ors (one-shots) provi de the pulses in response to positive and negative edges in the input, and ext ra gates have been added to equalize t he pr opagat i on del ay for each type of edge. To close this section oil digital LED drive circuitry, let us ret urn to the quest i on of t hermal behavior. As ment i oned previously, typical LEDs display a change of up to 2dB in optical out put over a nor mal range of operat i ng t emperat ure. One call allow for this change by subt ract i ng 2 dB from a system loss budget cal cul at ed for nomi nal oper at i ng condi t i ons, but this is oft en 2 dB out of onl y 15- 30dB to st art with. Alternatively, one call compensat e for this t emper at ur e vari at i on within the circuit design. In many cases, the l at t er is preferable. Obvi ousl y the i nt ent for such a circuit will be to raise LED cur r ent with rising ambi ent t emperat ure. For series drivers (Figs. 5.4, 5) or shunt drivers (Figs. 5.11, 12) this can be accompl i shed by condi t i oni ng the vol t age appl i ed t o the LED, as a funct i on of t emperat ure. A circuit which raises the vol t age supplied to the LED can be qui t e simply built using the t emper at ur e vari at i on of the cut-in voltage of a silicon j unct i on ( - 2.5 mV/<>C). Side benefits of this appr oach include increased isolation from glitches, reduced LED drive i mped- ance (low Z voltage source and lower series resistance), and i mmuni t y of the LED dri ve-current level to power-suppl y changes. Temper at ur e compensat i on call also be added to a shunt driver or to an emi t t er-coupl ed dri ver (Figs. 5.7 9) by using a t emper at ur e- var i abl e const ant - current source such as that shown in Fig. 5.17. Again t he t emper at ur e dependence of silicon j unct i ons is used t o increase cur r ent with rising t empera- ture. The base voltage of Q l is 1000 VB~ i450(Vcc- 3VBE), where VBE~0. 7 V near r oom t emperat ure. This analysis assumes bot h Q1 and Q2 have reasonabl e [:Is, ~5 0 or higher. The LED drive, equal to t he col l ect or current of Q2, is given by 100 2VBE I LED=I c ' ~I E= 1 4 ~ ( Vcc- 3 VBE) 10 which is 60mA if Vcc = +5 V. The t emper at ur e vari at i on AI E/ AT is trivially f ound to be + 1 mA/ C, again if Vcc = + 5 V. This coefficient was found to correct for one part i cul ar GaAs LED. Ot her LEDs may requi re modi fi cat i ons to this circuit, but such modi fi cat i ons are easily wor ked out. A l ow-i mpedance driver such as an emi t t er follower (Fig. 5.10) presents somewhat mor e of a probl em. It is amenabl e to nei t her the const ant - cur r ent - source appr oach nor t he vari at i on of Vco al t hough the n cont act of t he LED coul d be ret urned to a negative VEE source pr ogr ammed as j ust discussed. VIN 1K VBE VBE VBE 450 +5V 10 Q2 16 1/2 7545'1 i L ~ ] , r . . . . . . . . . ~ _ _ ~ ,o o e _ _ ,(~ ~ ~ o Fig. 5.17. Temperature compensation using lhc coeffi- cient of silicon pn junctions to adjust a current source Lightwave Transmitters 181 -t-Vcc v o i LED VBE VBE VBE Fig. 5.18. Temperature compensation of an LED using the temperature- dependent series voltage drop in multiple silicon pn junctions to adjust the current through the LED, assum- ing the total voltage drop relnains constant [5.74, 75] Co n t r o l of t he base vol t age of Q2 is r i sky becaus e [3 of ma n y t r ans i s t or s is itself sensi t i ve t o t e mpe r a t ur e - and f l ( T ) is not pr edi ct abl e e n o u g h t o use in pl a nni ng a c o mp e n s a t i o n scheme. On e possi bl e a p p r o a c h is t he use o f a c o mb i n a t i o n of r esi st or s and t her mi s t or s in pl ace of R~ in Fig. 5.10. A d r a wb a c k of this me t hod, unl i ke t he ot he r a ppr oa c he s pr evi ous l y di scussed, is t he i mpossi bi l i t y of i nt egr at i ng t her mi s t or s on a si l i con I C s houl d i nt egr at i on be c ome desi rabl e. An a l o g o u s t o t he use o f t her mi s t or s , however , is t he c onne c t i on of several si l i con j unc t i ons in series wi t h t he L E D in such a way t ha t t he negat i ve t e mpe r a t ur e coeffi ci ent s of t he si l i con a nd Ga As j unc t i on vol t ages ar e expl oi t ed t o rai se t he cur r ent [5. 74, 75]. Fo r t he ci r cui t in Fig. 5.18 V o - 3 VB E - VLE D /LED = R a nd t he t e mpe r a t ur e coeffi ci ent is A/LE D - - 4 ( - - 2. 5mV/ "C) + 10mV/ C AT R R ( Not e t hat t he t e mpe r a t ur e coeffi ci ent of t he Ga As L E D j unc t i on has been a s s ume d t o be t he s ame as silicon. Thi s is a ppr oxi ma t e l y cor r ect , but vari es 182 P. w. Shumat e and M. DiDomenico, .It. f r om one LED mat er i al syst em to anot her . ) If R is 10~, we obt ai n 1 mA/ C as before, and if V o is adj ust ed t hr ough bi asi ng of the dri ver circuit to appr oxi - mat el y 4. 2V, we f ur t her mor e have the 6 0 mA r oom- t emper at ur e drive as before. Use of this circuit may requi re Vcc > + 5 V, or the use of a negat i ve bi as (VE~) on the t ransi st or. Fur t her mor e, cert ai n di odes ma y sl ow the t ur n- on response of this circuit. One scheme not ment i oned is use of feedback cont r ol t o mai nt ai n the light out put const ant at all t emper at ur es. At low bit rat es ( < 1 Mb/ s) it is fairly easy to det ect the LED light with a phot odi ode and use this signal in a negat i ve- feedback scheme t o cl amp the light level to some reference "on" value. At hi gher bit rates, however, where mos t i nt erest lies, the phase shifts and avai l abl e gai n become a pr obl em and st abl e designs are difficult to obt ai n. Si mpl er appr oaches are avai l abl e for LED cont rol , and furt her discussion of feedback r egul at or s will be reserved for laser dri vers where such cont r ol is far mor e i mpor t ant . 5.6 Laser Drivers Unl i ke LEDs, which emi t light appr oxi mat el y in pr opor t i on to t he t ot al device cur r ent (aside f r om the small cut-in region), lasers are t hreshol d devices as de- scribed earlier. As such, light out put is pr opor t i onal t o t he i ncrement al cur r ent above t hreshol d. Several of the circuits discussed for LED appl i cat i ons can be used as laser dri vers wi t h onl y mi nor modi fi cat i ons. These changes are di rect ed t owar d suppl yi ng the laser wi t h a subst ant i al bi as (often called prebi as) in t he off state. It is desi rabl e t hat this bias be j ust bel ow t hreshol d in the off st at e for several reasons. First, by keepi ng the laser close t o t hreshol d, t ur n- on del ay and l eadi ng-edge over s hoot of t he opt i cal out put are mi ni mi zed 9. Second, a subt hr eshol d bias can easily be cont r ol l ed t o compens at e for changes in t he ambi ent t emper at ur e and, in mos t cases, device agi ng as will be discussed short l y. Thi rd, dat a- pat t er n- dependent j unct i on heat i ng and its effect on t he opt i cal out put are reduced since t he on c u r r e n t (/bias-l-/drive) is close to the off c u r r e n t {Ibi.s) for mos t lasers. A subt hr eshol d bias, in spite of causi ng s pont aneous light emi ssi on in the off state, ext ract s little penal t y for mos t digital appl i cat i ons since the on- t o- of f rat i o of light levels is usual l y gr eat er t han 10 : 1. The si mpl est dri ver t h a t has often been used to expl oi t the hi gh- speed capabi l i t y of lasers is a modi fi ed shunt driver. By using Ga As MESFETs in the circuit of Fig. 5.19 [5.79], lasers have been modul at ed at rat es exceedi ng l Gb/ s [ 5 . 8 0 ] . Resi st or R 1 and compensat i ng capaci t or C1 are used to add sufficient vol t age in series wi t h the laser to bi as Q~ well i nt o its act i ve or pi nch- of f region. 9 These aspects of laser dynamics are discussed broadly in [5.76-78], and the delay aspect is further considered in [5.17]. Lightwave Transmitters 183 Therefore, for a specific i nput vol t age V~n (or VGS), a specific amount of the t ot al current flowing t hr ough R 2 is di vert ed ar ound t he laser. That cur r ent cont i nu- ing to flow t hr ough R 1 and the laser is the bias, or off-state, value. For maxi mum ext i nct i on ratio, ?, the off-state bias woul d be below but close to t hreshol d. In this way, ext i nct i on ratios of 10-30 are commonl y at t ai ned. For maxi mum speed, however, it may be desi rabl e to bias the laser slightly above t hreshol d. Now all modul at i on is per f or med on st i mul at ed emi ssi on: the time const ant is very short ( < 1 ns) and the t ur n- on del ay is minimized. The penal t y one pays for the l ower ext i nct i on rat i o may be uni mpor t ant . Personick has discussed the power penal t y referred to receiver sensitivity for different 7's. For example, in lowering V f r om 16 to onl y 4, one loses at worst onl y 3 dB in receiver sensitivity [-5.81]. The circuit of Fig. 5.19 woul d pr obabl y not be useful in a system appl i cat i on, however. Recall t hat t he lasing t hreshol d is a sensitive funct i on of t emper at ur e, changi ng about + 1%/C. It was shown earlier how a rise of 30 C coul d r ender such a circuit i noperat i ve. Thr eshol d is not onl y a funct i on of t emperat ure, but usually increases as the device ages, br ought about by an increase in i nt ernal losses. This has been observed to be as large as 1 mA/ kh for lasers with a mean time to failure of l 0 s h. Cl earl y AI/At depends on the reliability (i.e., MTTF) of t he par t i cul ar device. If t emper at ur e changes were the principal concern, the laser t emper at ur e coul d be regulated. In fact, if cool i ng were placed under cl osed-l oop cont r ol using the laser' s opt i cal out put as the cont r ol l ed paramet er, t hen this cool i ng coul d cor r ect for aging effects as well as ambi ent t emper at ur e changes - a laser woul d be increasingly cool ed as it aged t o keep its t hreshol d const ant , i ndependent of the t emper at ur e aspect. This scheme woul d qui ckl y become i mpract i cal and inefficient, however, if subst ant i al amount s of cool i ng were requi red t o offset a large t hreshol d increase. It is qui t e easy to get ar ound this pr obl em by cont rol l i ng the laser bias instead. Since the bias is a quasi-static current , it can be cont rol l ed using tow- speed feedback ci rcui t ry and raised or lowered as necessary. It has usually been found desirable to mai nt ai n some aspect of the light out put const ant in such a loop. (Predictive cont rol of the bias, similar to what was done for an LED, woul d provi de onl y poor t emper at ur e compensat i on and no compensat i on at all for device aging.) In the emi t t er-coupl ed driver shown in Fig. 5.20 [5.82], the laser cur r ent is the sum of the col l ect or current s of Q2 (the drive current ) and Q3 (the bias). Since Qt and Q2 form a nonsat ur at i ng cur r ent - r out i ng switch, Q2 is always in its active region and its switching act i on is unaffected by the presence of Q3. Similarly, t he col l ect or cur r ent of Q3 is unaffected by the switching oper at i on of Q2 if Qa has a small out put capaci t ance (Cob) so t hat high-speed voltage fl uct uat i ons due to switching are not coupl ed into Q3's base circuit. (These vol t age fl uct uat i ons are nor mal l y very small, on the or der of 0. 1-0. 2V due to laser series resistance.) In oper at i on, the emi t t er-coupl ed switch adds a const ant drive cur r ent sufficient to exceed t hreshol d and reach a poi nt subst ant i al l y up the lasing 184 P. W. Shumate and M. DiDmnenico, .h'. v~N + 3C R2 ? Fig. 5.19. A laser shunt-driver circuit. Bias is adjusted through the combination of R~, R 2 and the conductivity of Q~ [5.79] Fi g. 5 . 2 0 . A feedback-stabilized laser driver circuit. Laser output, taken from the mirror not coupled to the output fiber or from a tap in the output fiber is compared with the data pattern and a fixed reference to control laser bias [5.82] + 5V +PHOTODIODE LASER~ ~ / OUTPUT MONITOR I - J - , C O S T A N , DATA M~NITOR VREF r egi on of t he L - 1 cur ve whenever l i ght o u t p u t is cal l ed for, The bi as is adj us t ed by t he f eedback l oop c ont a i ni ng ampl i f i er A 1 and t he p h o t o d i o d e so t hat t he aver age p h o t o c u r r e n t is hel d c ons t a nt wi t h r espect t o t he aver age of Vi,, t he da t a pat t er n. Co mp a r i s o n wi t h t he da t a pat t er n gi ves cont r ol of bot h t he on and off st at es i ndependent l y of t he pat t er n st at i st i cs ; i.e., t he da t a need not be s cr ambl ed t o assur e a 50% dut y cycl e, etc. Thi s scheme - c ons t a nt dr i ve pl us cont r ol l ed bi as - is an effective means t o st abi l i ze l aser o u t p u t whenever t he sl ope effi ci ency is expect ed t o r emai n rel at i vel y cons t ant . I t has been s hown capabl e of pr ovi di ng o u t p u t st abi l i t y wi t hi n 1% over t he t e mpe r a t ur e r ange 20 t o 50 "C [-5.82] a nd wi t hi n 10 % dur i ng 2 0 k h o f ope r a t i on for n u me r o u s lasers [5. 83]. Thi s has been achi eved f or t r ans mi t t er s ope r a t i ng at bi t r at es of 44.7 Mb/ s and 274 Mb/ s . I n t he first ver si ons of t r ans mi t t er s as s embl ed usi ng t hi s ci rcui t , t he " ba c k" mi r r or of t he l aser was mo n i t o r e d by t he p h o t o d i o d e f or cont r ol whi l e t he " f r ont " mi r r or was coupl ed t o a fiber. Thi s s cheme wor ks well, as descr i bed Lightwave Transmitters 185 above, onl y if t he t wo mi rrors t rack in relative power changes. Laser nonlinearities evidenced by ki nks in the L - I curve can arise however, due to the appear ance of hi gher - or der t ransverse modes m the active regions of many lasers. Appear i ng with agi ng or t emper at ur e changes, these nonlinearities are somet i mes accompani ed by f r ont - t o- back mi r r or mi st racki ng leading t o poor regul at i on of the f r ont - mi r r or out put even t hough t he back mi r r or is itself stabilized I-5.83]. In l at er versions of the t ransmi t t ers descri bed above, the opt i cal out put was stabilized by moni t or i ng t he act ual fiber power using opt i cal -fi ber taps depi ct ed in Fig. 5.21 [5.84]. Basically, a beam splitter is assembled using angle-polished fibers aligned in a precision vee groove, and t wo 4 % (glass-to-air) reflections are directed t owar d a l arge-area phot odi ode as seen in Fig. 5.22. Thus, appr oxi - mat el y 8% of t he fiber power is conver t ed i nt o a phot ocur r ent and the r emai nder is t ransmi t t ed. Opt i cal - out put regul at i on of _ 2 % has been achieved from 0-50~-'C even in the presence of l aser-mi rror mi st racki ng [5.83]. This represent s mor e t han an or der - of - magni t ude i mpr ovement over t he back- mi r r or - moni t or i ng scheme used in the earlier design, for cases where laser mi st racki ng occurred. Thr ough addi t i onal processing of t he phot odi ode out put in Fig. 5.20 enough i nf or mat i on can be derived to cont rol the drive cur r ent in addi t i on to the bias. Thus changes in laser slope efficiency can be compensat ed as well. An i mpor t ant advant age of such a circuit, however, lies in the fact t hat the circuit woul d possibly not requi re adj ust ment to different lasers dur i ng manufact ure, saving the time and cost of addi t i onal steps if all lasers do not have the same values of t hreshol d and slope efficiency. Phot ocur r ent from the out put moni t or carries with it, in its ac and dc component s, all the i nf or mat i on needed to deduce the on, off, and average light levels. Circuits acqui ri ng such i nf or mat i on fl-om the ac signal have been descri bed [5.85 88]. Fi gure 5.23 por t r ays one of these circuits, based on [5.85]. The most positive excursi on of t he moni t or phot ocur r ent cor r espondi ng to the peak light level is st ored in a peak det ect or aft er being buffered and amplified by a fast ac amplifier (A2). Because A 2 is ac coupl ed, one-hal f of this st ored level cor r esponds to the mi dpoi nt of the modul at i on dept h which differs from the average light level by the amount of light in the off or zero state. (The peak det ect or is assumed to be perfect.) Amplifier A 3 subt ract s one- hal f the modul at ed power level from the average, deri ved t hr ough slow dc amplifier A~, and uses t he resulting anal og of the zero light level, Po, to cont r ol the bias t hr ough compar at or A 4 and pass t ransi st or Q3. The drive level is adj ust ed by compar i ng the peak signal or modul at i on dept h with a desired level: the out put of compar at or A s pr ogr ams the cur r ent source in the emitters of Qt and Q2. By use of this scheme, no reference to the dat a pat t ern need be made. However, now the statistics of the dat a must be such t hat t ransi t i ons occur frequent l y to keep the peak det ect or charged. If the dat a di sappear for a pr ol onged interval (e.g., an idle line), the drive level coul d be adj ust ed t o t oo high a value and the first bits of a new dat a st ream will result in excessive 186 P. W. Shumate and M. DiDomenico, .lr. ~ 1 Fig. 5.21. Configuration of a beam- splitter optical-fiber tap using a vee P - I - N P H O T O D I O D E ~ i i l i - , . . ~ PHOTOCURRENT groove to align two angle-polishcd fibers. The reflected output is convert- ed directly to an electrical signal using the p-i-n photodiode ALIGNMENT TAP OUT P - , ; N P H O T O O ,a O E . " = ~ - " , I II INPUT F IB E R ~ J O L T P U T F IB E R ~ 2OFm )OUT Fig. 5.22. Details of the reflections taking place in tile optical-fiber tap [5.84] o p t i c a l o u t p u t . Cl e a r l y a d a t a - p a t t e r n r e f e r e n c e c o u l d b e a d d e d i n t o t h e c i r c u i t o f Fi g. 5. 23 t o p r e v e n t t hi s ef f ect . A s i mi l a r c i r c ui t , d e s c r i b e d i n d e t a i l b y Gr u b e r et al. [ 5. 88] wa s s h o wn t o r e d u c e l a r g e c h a n g e s me a s u r e d wi t h o u t a n y f e e d b a c k c o n t r o l t o l ess t h a n ___ 3 % v a r i a t i o n s o v e r a c h a n g e i n t e m p e r a t u r e f r o m 10 t o 40 ~C. O v e r t hi s r a n g e , b i a s c h a n g e d , f o r e x a mp l e , f r o m 190 t o 227 mA, a n d d r i v e c h a n g e d f r o m 50 t o 58 mA. S i mi l a r p e r f o r ma n c e h a s b e e n r e p o r t e d f o r o t h e r p e a k - p o w e r / m e a n - p o w e r c o n t r o l s c h e me s b y S a h e r e t al . [ 5. 86] a n d b y Br o s i o e t al. [-5.87]. A n o t h e r i n t e r e s t i n g l a s e r c o n t r o l c i r c ui t is s h o wn i n Fi g. 5. 24 [ 5. 89] . Th i s " a u t o m a t i c b i a s c o n t r o l " us e s t h e f a c t t h a t t h e j u n c t i o n v o l t a g e o f a l a s e r Li ghl wave Tr ans mi ner s +V F L . M O N I T O R P H O T O D I O D E +Vcc (+) t "-:V D R I V E ~ I I I C U R R E N T L ~ ) . I - - - - - <~ I I m r)7 . . . . . . . . . . l I / 2 P M O D 187 i P E A K D E T E C T O R ( P M o D ) D R I V E R E F E R E N C E /77T Fig. 5.23. A circuit concepl using bot h t he ac and dc i nf or mal i on in tile laser out put So cont r ol laser dri ve and bi as i ndependent l y. ( Used wi t h per mi ssi on of St andar d Tel ecommuni cat i on Labor at or i es, Ltd. (c) 1976.) [5.85] becomes pi nned at the bandgap val ue with the onset of st i mul at ed emission [5.90]. By raising t he bias cur r ent of the laser until the j unct i on vol t age is sensed t o be clamped, we have an all-electronic confi gurat i on (i.e., 11o optical par amet er in the feedback loop) t hat changes the bias in response to t empera- t ure or aging. The degree of sat ur at i on of the j unct i on vol t age is sensed as i l l ust rat ed in Fig. 5.25. The laser with its i nt ernal series resistance R s is placed in a bridge circuit, as shown. Resistors R2, and e2b are adj ust ed so t hat , if the laser' s j unct i on voltage Vj were const ant (pinned at t he bandgap), 11o change in bal ance woul d arise if drive current I d were vari ed; (i.e., the dc i nput s to the difference amplifier may be unequal at all times, but the magni t ude and sign of t hei r difference stays const ant in response to a dat a input signal.) If, on the ot her hand, Vj changes with I a due to lack of pinning, the difference represent ed as A- B in the figure changes. Obvi ousl y, if the amplifier is ac coupl ed, t he out put of the amplifier goes from zero to some nonzer o value rel at ed in magni t ude to the change in Vj. Thi s represents an er r or signal which, 188 P. W. Shumate a n d M. DiDomenico, Jr. vi N 0 !! R 2 LASER ~ . ~ ~ DI / - ) 7 I'I8 ~ M ~ ~ C 3 +v f . ~ Q2 / q , Fig. 5.24. An " a u t o ma t i c bi as c o n t r o l " c i r c ui t t hat s e n s e s l a s e r t h r e s h o l d e l e c t r i c a l l y [5.89-1 Id A N RI / l ~ \ ~ A-B o-&/. V J x / ! 1 ' I Fig. 5,2N. I l l us t r a t i o n o f t he br i dge pr i nc i pl e us e d in t he rJ7 ""automatic bi a s c o n t r o l " ci rcui t , , n e a s u r i n g t he e x t e nt t o wh i c h / the laser junction voltage ~ is saturated t hr o ug h addi t i onal ci rcui try, is mi ni mi z e d in t he ordi nary c l o s e d- l o o p, negat i ve- f eedback sense. Re t ur ni ng n o w t o Fig. 5.24, we find t he bri dge i mpl e me nt e d as R 1 (for Rt ) and R z (for R2. and Rzb ). The laser series resi st ance is no l onger l abel ed. Capac i t or s C 1 and C 2 r e move t he dc c o mp o n e n t s o ampl i fi er A~ wi l l r e s pond o nl y t o c hange s in bal ance. Resi st ors R a a nd R 4 provi de c l o s e d - l o o p dc s t abi l i z at i on o f A~ as wel l as adj us t abl e offset for t he out put . The t rans i ent o ut put s o f A ~ (in r e s pons e t o t he si gnal i nput V~,,) are st ored in a peak de t e c t or (D 1 and C3), t he o u t p u t o f whi c h c ont r ol s t he bi as l u t hr ough t ransi st ors Q, and Q2. The bi as is raised s o as t o mi ni mi ze t he error si gnal , but t he gai n o f t he ci rcui t is kept l o w e n o u g h s o t hat an error si gnal al ways persists (is no t reduced t o t he noi s e level). Hence, t he laser is al ways bi ased j us t be l ow t hres hol d. Lightwave Transmi t t ers 189 5.7 Analog Drive Circuitry In cont r ast to the precedi ng digital circuits which generat e onl y t wo or t hree distinct out put levels, anal og drivers must cause the light out put L ( t ) from a source t o follow a t i me-varyi ng i nput vol t age waveform V ( t ) as accur at el y in ampl i t ude and phase as possible. If LED or laser sources responded linearly to vol t age or cur r ent drive, our task woul d be simplified. Unf or t unat el y this is not the case, as was discussed earlier in t he chapter. Most commonl y available sources (hi gh-radi ance LEDs and mode-st abl e lasers) display di st ort i on pro- duct s 30-50 d B below the signal. Thus the source itself limits t he per f or mance in most anal og signaling appl i cat i ons since phot odi odes show much hi gher linearity [5.91] and t ransi st or dri ver circuits can be designed to have, in many cases, l ower di st ort i on levels. If one has a reasonabl y linear, high-speed source and if the appl i cat i on does not requi re ext remel y low di st ort i on levels (i.e., ~ 1 ~,;), very st rai ght forward dri ver circuits can be used. A simple and often adequat e high-speed dri ver is the common- emi t t er t r ansconduct ance amplifier shown in Fig. 5.26 which convert s an i nput base vol t age to a col l ect or current . Base bias is adj ust ed in this circuit for class A oper at i on: the quiescent col l ect or cur r ent is about half t he peak value and bot h the t ransi st or and t he LED (or laser) are biased well up into t hei r linear oper at i ng region. For this circuit dri vi ng a resistive l oad (no LED or laser) and dependi ng on t ransi st or selection, response to over 100MHz with har moni c di st ort i on levels on t he or der of 45 dB or mor e below signal can be at t ai ned at 80~,, modul at i on index. Since the circuit as shown provi des onl y 3 0 _ 2 4 mA (i.c., 80 % modul at i on), addi t i onal bias must be supplied if a laser with a hi gher t hreshol d is used. A similar t r ansconduct ance driver, but utilizing a Dar l i ngt on pair for Q 1 to reduce source i mpedance, has been used at 70 MHz to drive high- radi ance LEDs [5.92]. Laser bias under feedback cont rol to compensat e for t emper at ur e or aging effects has been used qui t e effectively with the circuit of Fig. 5.26, also at 70 MHz [5.93]. The emittel;-coupled dri ver of Fig. 5.20 was replaced with the common-enai t t er circuit of Fig. 5.26. Since the laser average power must be regul at ed i ndependent l y of t he i nput waveform, the dat a reference lead in Fig. 5.20 was removed. The l ow-frequency response of the resulting circuit is now det ermi ned by the time const ant of the feedback amplifier A 1 and the i nput coupl i ng capaci t or. By increasing these values, response down t o a few Hz can be obt ai ncd if requi red for a part i cul ar appl i cat i on. If one has a source whose linearity is higher t han t hat of a simple drive circuit, the dri ver itself can be linearized. It is well known t hat even- har moni c di st ort i on can be reduced t hr ough use of a push-pul l confi gurat i on, or electrical negative feedback can be used to force the drive cur r ent to follow the i nput voltage. Alternatively, compl ement ar y di st ort i on can be used to minimize di sl ort i on pr oduct s generat ed in the amplifier [5.94]. The feedback appr oach 190 P. IV.. Shumat e a nd M. Di Domeni co, Jr. VIN 1/J.F I I 5 0 2K ~SK / ; +SV 7 LEDORLASER 50: iO M O N I T O R Fig. 5. 26. A s i mpl e t r a ns c onduc t a nc e dr i ver f o r i nt ensi t y modul a t i on of a l aser or LED D I LED R Fig, 5. 27. Us e of a di ode net wor k to pr ovi de s ome degr ee of compl e- me nt a r y di s t or t i on for l i near i zi ng an LED [5. 97] becomes difficult at vi deo frequencies because of the high gai n- bandwi dt h requi rement s and short pr opagat i on delay times needed for stability. At frequencies below about 1 MHz, however, feedback l i neari zat i on can be used mor e reliably. The third met hod, compl ement ar y di st ort i on to linearize tile dri ver alone, is pr obabl y not wor t h t he effort consi deri ng t he l i mi t at i ons present ed by the source itself, t hus i nt r oduced after the driver. The real payoffs come when the opt i cal source is included in t he lineariza- tion scheme. For negat i ve-feedback circuits the opt i cal out put can be det ect ed and compar ed with the i nput waveform. This sampl i ng can be done using a tee coupl er in the out put fiber, or a phot odi ode can be packaged al ong with the source to pick up st ray emission. The har moni c di st ort i on of an LED plus dri ver has been measur ed at 70 dB bel ow the f undament al f r om dc to 100 kHz in an opt oi sol at or i mpl ement at i on of the ext r a- det ect or scheme [5.95]. Clearly, by using hi gher speed circuitry, an LED plus dri ver coul d be linearized to a compar abl e degree in a hi gher speed fi ber-opt i c appl i cat i on. It has been poi nt ed out, however, t hat negative feedback, part i cul arl y for vi deo signals, may be poor in reduci ng t hi r d- or der di st ort i ons since these cont ai n compl ex cr oss- modul at i on t erms as well as t hi r d- har moni c com- ponent s [5.96]. Anot her scheme for obt ai ni ng a signal for use in the closed l oop woul d be to drive in series with the pr i mar y emi t t er a secondary, reference emi t t er coupl ed to a phot odet ect or . The reference channel woul d be placed in the feedback l oop and the pr i mar y emi t t er woul d be coupl ed to a fiber. Thi s assures maxi mum power coupl ed i nt o the fiber and may simplify obt ai ni ng a st r ong signal for feedback. It is interesting to not e t hat if the reference-channel phot odet ect or is of the same type as t hat used at the far-end receiver, phot odet ect or non- linearities will be compensat ed out to within the capabilities of t he feedback circuit. Lightwave Transmitters 191 At vi deo frequencies, mor e stable correct i ve schemes are needed. Vari ous predi st ort i on t echni ques appear to be the most popul ar solution. Pr edi st or t i on or compl ement ar y di st ort i on is commonl y used in el ect roni cs to compensat e for peculiarities in a t ransmi ssi on or recordi ng medi um, or in devices at the t ermi nal ends. It is vari ousl y called preemphasi s, equal i zat i on, speedup, pr edi st or t i on or, loosely, feedforward dependi ng on the appl i cat i on. The scheme is to al t er the signal at some poi nt so t hat the "pecul i ari t i es" i nt r oduced l at er on are as nearl y as possible cancelled out leaving the signal as if nei t her act i on had t aken place. The result, of course, is t hat the t ransmi ssi on pat h l ooks compl et el y t r anspar ent with regard to di st ort i on. One of the simplest pr edi st or t i on schemes is to al t er the LED or laser cur r ent in response to t he signal ampl i t ude by connect i ng a net wor k ar ound the device. In Fig. 5.27, the use of a ger mani um di ode and resistor in parallel with a Burrus-t ype LED was found to ext end t he range of drive cur r ent for a given level of out put linearity by nearl y 50% [5.97]. The net wor k t ends to shunt a l arger fract i on of the t ot al cur r ent ar ound the LED at low drive levels and vice versa at high drive levels. Thus at high dri ve levels, the negative cur vat ur e of the LED charact eri st i c arising from j unct i on heat i ng is compensat ed by great er drive. Since j unct i on- heat i ng effects are time dependent , it has also been found useful to deemphasi ze t he l ow-frequency component s in the signal at the t r ansmi t t er (and make a cor r ect i on at the receiver) [5.98]. Such l ow-frequency rol l off combi ned with a di ode pr edi st or t i on scheme was f ound in this last reference to reduce differential gain and phase to <1 % and <0. 75 at 70% modul at i on index. The t echni que of predi st ort i on using di odes in net works coupl ed to the source dri ver circuit are exemplified in [5.96, 99, 100]. Differential gain and phase cor r ect i on circuits, shown in Figs. 5.28, 29, respectively, are placed ahead of the source driver. Compl ement ar y di st or t i on is i nt r oduced in different por t i ons of the response of each circuit by the i ndependent l y biased diode- resistor net works which affect the gain in Fig. 5.28 and t he phase response in Fig. 5.29. The bias voltages V i det er mi ne where in the amplifier' s t ransfer charact eri st i c the associated resistors R i affect t he gain (Fig. 5.28) or phase (Fig. 5.29) response. Thus, in a pi ecewi se-cont i nuous manner, source (and circuit) di st ort i ons can be t r i mmed out by adj ust i ng the bias V~ and resistance Ri for each diode. In principle, di st ort i on coul d be reduced to any ar bi t r ar y level by increasing the number of diodes, and hence bias variables, in each net work. The aut hor s of this work achi eved reduct i ons in differential gain and phase to less t han 1% and 1 , respectively, over reasonabl e changes in t emper at ur e [5.96]. Closely rel at ed t o the pr edi st or t i on t echni ques discussed above is "quasi - f eedf or war d" which has been r epor t ed effective in r educi ng di st or t i on for wi deband appl i cat i ons [5. 101, 102]. Inst ead of predi st ort i ng the electrical drive t o t he source using passive component s (diode characteristics, R's, C's), the drive signal is al t ered with an "er r or signal" obt ai ned by first passing the 192 P. W. Shumate and M. DiDometlico, Jr. V,rN - - + V c c R 5 - - O U T P U T TO DRIVER OR DIFF, P H A S E CIRCUIT ) R I ~ R2 i - V ~ - - V 5 Fig. 5.28. Differential gain correction circuitry [5.96] + V c c R4 v l r", v2 R t I~ _1 %%,% R2 R3 OUTPUT TO DRIVER OR DIFF. GAIN CIRCUIT Ct Fig, 5,29. Differential phase correction circuitry [5.96] i nformat i on t hrough a reference light-emitting channel. The mixing of this compl ement ary di st ort i on with the signal at the electrical level differentiates quasi-feedforward from feedforward, where the mixing would t ake place in the optical regime. The latter is also possible, but the aut hor s are aware of no published reports of such techniques. In Fig. 5.30, LED 1 is modul at ed directly by the i nput signal, detected immediately by phot odi ode D 1 and compared with the input signal which has been delayed to preserve the correct phase relationship. The result of this compari son is an error signal cont ai ni ng onl y the di st ort i on component s added by the source and detector. Next the error signal is subt ract ed from the input which has been delayed still further to remain in phase with the error signal. Lightwave Transmitters 193 VZN +Vcc +V t , , ' 7 1 7 COMPARISON SUM Fig. 5.30. Quasi-feedforward lincarization principle [5.101] +%c LED 2 (OUTPUT) Fi nal l y the resul t i ng pr edi st or t ed i nput modul at es the pr i mar y sour ce LED 2. If phase and ampl i t ude have been pr eser ved careful l y and t he sources are mat ched, t hen the di st ort i on and compl ement ar y pr edi st or t i on component s cancel and the r esponse is linear. Use of this scheme has reduced di st or t i on component s 70dB bel ow a 3.1 MHz f undament al . The aut hor s of t he quasi - f eedf or war d t echni que have r epor t ed a not he r novel scheme, a phase- shi f t - modul at i on t echni que for selectively cancel l i ng har moni c component s of a specific or der [5.102, 103]. As seen f r om Fig. 5.31, t wo i dent i cal sources are dri ven out of phase wi t h each ot her and the opt i cal out put s are combi ned in a opt i cal tee coupl er. Fr om a st r ai ght f or war d anal ysi s given in the reference, one finds t hat the det ect ed ac signal at t he recei ver has zero s econd- har moni c c ompone nt if the phase shift is 90 . Al t ernat i vel y, t hi rd- har moni c component s cancel if t he phase shift is 60 c', etc. As furt her discussed in t he reference, schemes can be devised usi ng mor e coupl ers and sources to cancel several har moni c component s si mul t aneousl y. As descri bed in the reference, a t wi n- cont act - st r i pe LED was used to obt ai n mat ched LEDs, and a 25dB decrease in s econd- har moni c di st or t i on to a level near - 6 0 d B was measur ed at 3.1 MHz. All of our di scussi on on l i neari zi ng t echni ques has addr essed LEDs r at her t han lasers. Thi s is because mos t lasers avai l abl e now still have the pot ent i al for (a) devel opi ng ki nks in t hei r L - I charact eri st i c or (b) at least showi ng a change in l i neari t y wi t h agi ng or wi t h t emper at ur e changes. Onl y a scheme t hat measur es the nonl i near i t y and correct s it in real t i me such as feedback, feed f or war d or phase modul at i on woul d be helpful here. Then one is faced, in the l at t er t wo schemes, with mat chi ng t wo lasers in their mode- i nst abi l i t y behavi or 194 90OPHASE VIN SHIFT P. W. Shumate a nd M. DiDomenico, Jr. 4 Vcc Z +Vcc I LED 2 . . _ ~ . ~ F PASSIVE OPTICAL FIBER COUPLER OUTPUT T _ _ J ' F ' B E R Fig. 5.31. Pha s e - s hi f t - modul a t i on t echni que for sel ect i ve h a r mo n i c cancel l at i on when l i nearl y mo d u l a t i n g an LED [5. 103] - a very difficult if not i mpossi bl e task. Ther ef or e we feel t hat for any syst em requi ri ng low di st ort i on, wi t hout the chance of degr adat i on with aging, mode- st abl e lasers are absol ut el y necessary. As with the vari ous digital drivers, we will make no absol ut e compar i s ons a mong the anal og drivers. Far t oo little has been publ i shed for gui dance in maki ng a compar i son. If a si mpl e dri ver such as t hat in Fig. 5.26 can be used, or if negat i ve feedback can be appl i ed, t hen clearly these circuits are the choices because of their simplicity. All of the ot her t echni ques so far publ i shed for i mpr ovi ng source l i neari t y requi re careful t r i mmi ng, or precisely mat ched t i me del ays (especially at hi gh frequencies such as 100--300MHz), or mat ched sources or hi gh- qual i t y tee coupl ers. The endur ance of these t echni ques to large changes in t emper at ur e or to agi ng changes in t he sources is yet to be demons t r at ed for some of these techniques. Ther ef or e it r emai ns to be seen which are the mos t useful schemes if low- di st ort i on l i ght wave dri vers are to be realized for syst ems appl i cat i ons. 5.8 Subsystem Considerations Up t o this poi nt we have discussed device t radeoffs and circuit consi derat i ons, not t ryi ng to r ecommend best choices for ei t her since choices depend on the Lightwave Transmitt.ers 195 appl i cat i on, but to provi de perspective for maki ng these selections. Ther e remain some consi derat i ons t hat are i mpor t ant from the subsyst em view but often are so diverse and t ransi ent t hat they are i nappr opr i at e for extensive discussion. Device packaging, part i cul arl y for lasers, is such an item. In closing this chapt er, we will ment i on and comment on several of these topics to clarify what we consi der the mor e i mpor t ant facets of each. Source Packaging: A main consi der at i on here is t he met hod for transfer- ring the light out of the package, which is preferabl y hermet i c for achi evi ng highest reliability. Present l y t here are t hree sour ce- packagi ng scenarios : the first is a lensed oi" wi ndowed package. Here the bur den of coupl i ng the source' s opt i cal out put i nt o the fiber in an efficient, stable way lies with the user. A package of this sort is the hardest to use. The second package style brings the light out via an opt i cal fiber per- manent l y at t ached to the package and aligned with the source. The fiber, often rei nforced as a mini-cable, is called a pigtail. The out put end of the pigtail may have a st andar d opt i cal -fi ber connect or installed, ot herwi se the user provi des the connect i on. A pigtailed package solves the device coupl i ng pr obl em for the user. In addi t i on, the flexible pigtail, often between 10 and 30cm in length, allows the user much freedom in l ocat i ng the par t on a circuit boar d and in strain-relieving the connect or on the end. On the ot her hand, a pigtail is mor e costly and takes up ci rcui t -board area. The third style of package has the connect or rigidly at t ached to the package wall, or has the light source embedded in t he connect or itself. A compact packagi ng scheme such as this is desirable for densely packed appl i cat i ons where a pigtail woul d be in the way or subject to damage. A possible di sadvant age, however, coul d be provi di ng adequat e strain relief, or l ocat i ng the part itself, if an opt i cal -fi ber cable is to be connect ed and br ought off a circuit boar d cont ai ni ng the package. For all of these package designs, it is desirable to have the smallest pract i cal dimensions, with the package also cont ai ni ng the drive ci rcui t ry where possible (i.e., an i nt egrat ed-ci rcui t chip). Package design shoul d also provi de low t hermal i mpedance between the source j unct i on and the ambi ent , since source lifetime is t hermal l y activated. This is very i mpor t ant for lasers, where the junction t emper at ur e shoul d be mai nt ai ned as low ( 25- 50"C) as reasonabl e. For act i vat i on energies EA of ~0. 7 cV, a difference of 20 "C i mpact s lifetime by a f act or of five- to tenfold (degradat i on rat e is pr opor t i onal to e x p ( - EA/kT). Ther ef or e a package design i ncor por at i ng a stud for heat r emoval is highly desirable. Coolin~i : Al ong these same lines, for some appl i cat i ons it may be necessary to cool a laser using a t hermoel ect ri c device. The cost of provi di ng AT' s on the or der of 20 ~'C is modest and t he power di ssi pat i on low, on the or der of a few watts. Thus a laser coul d be rendered mor e practical for use in out si de-pl ant appar at us or hot equi pment frames. LEDs, havi ng a less sensitive internal st ruct ure and higher reliabilities, pr obabl y do not war r ant cooling. 196 P. W. Shumate and M. DiDomenico, Jr. i ..--...~ OPTO I SOL AT OR Fig. 5.32a, b. Ci rcui t s of laser L I si mul at ors. (a) Basic circuit pr ovi di ng t hr eshol d and sl ope- efficiency adj us t ment s : (b) circuit l hat pr ovi des addi t i onal feat ure of ki nks [5.104] R. C = 0. 5/ . ~F R= t O K ~ , Laser Prot ect i on : Al t hough not addressed specifically in the circuit par t of this chapt er, it shoul d be recogni zed t hat lasers are cat ast rophi cal l y damaged by overdrive. This is part i cul arl y i mpor t ant duri ng the power - up sequence. Most laser circuits shoul d have time const ant s of milliseconds to seconds placed on strategic nodes so t hat bias, for example, rises smoot hl y f r om zero and is under feedback cont r ol by the time it reaches full value [5.82]. Cr owbar s and clamps somet i mes offer i nadequat e pr ot ect i on because t hey do not act fast enough. We have ci rcumvent ed the l aser-prot ect i on pr obl em when experi ment i ng with new laser circuitry (part i cul arl y of the t ype provi di ng i ndependent bias and drive cont rol ) by using a laser simulator in place of the laser. The circuit of Fig. 5.32a behaves electrically similar to a laser at the i nput port s [5.104, 105]. By connect i ng the shunt/series net wor k ar ound the LED in an opt oi sol at or and maki ng use of the LED' s light-emitting cut-in behavi or near 1.1 V, the phot ocur r ent at the out put port s is found to display an " L" - I t ransfer charact eri st i c very close to t hat of Fig. 5.1. The out put phot ocur r ent , presumed to be a measure of " L" , is ret urned to the feedback cont rol circuit in place of the laser moni t or . Adj ust ment s of R 1 and R 2 al t er the "t hr eshol d cur r ent " and "slope efficiency", respectively. We have used the augment ed version of this circuit in Fig. 5.32b one which provi des "ki nks" to test the stability of cl osed-l oop bias cont rol in the presence of such kinks [5.104]. (These ki nks can cause instabilities in certain feedback circuits, possibly resulting in laser damage were an act ual laser to be used for testing purposes.) Li ght wave Tr ansmi t t er s 197 T~'ansmitter Subsystem Evaluation : To charact eri ze a compl et ed design, one must eval uat e bot h its electrical and opt i cal performance. Electrical charact eri - zation includes measurement s of i nput current s for all possible combi nat i ons of i nput voltages over the t emper at ur e range for which the design is i nt ended to operat e. Dynami cal l y, this must also include worst-case dat a or signal-input patterns. Opt i cal per f or mance includes measur ement s of out put power, on/ of f (ex- tinction) ratio, opt i cal rise time and fall time, delay, j i t t er, over shoot , dr oop, frequency response, di st ort i on, etc. Power measur ement s are easily made using cal i brat ed phot odet ect or s or commerci al power meters. Pulse propert i es are eval uat ed by exami ni ng the optical out put in response to a known input, using a fact phot odet ect or and oscilloscope. For digital t ransmi t t ers, a square-wave or ps eudor andom- wor d i nput is useful. The behavi or of the optical out put in the presence of strings of 0s or 1 s in a pseudor andom pat t ern is a good test for heat i ng effects or laser bias-circuit effectiveness. Ps eudor andom- wor d gener at or s are easily built using i nt egrat ed- circuit shift registors: two or mor e bits in the register are summed (exclusive OR) and ret urned t o the i nput of the register [5. 106-109]. Of the many wel l -known met hods for measuri ng di st ort i on [5.21, 110 112], one of t he best is t he t wo- t one i nt er modul at i on test. Two sinusoidal frequencies are added in a wi deband hybri d coupl er (mixer) and t hen used t o modul at e t he source. By l ooki ng with a spect r um anal yzer at Ji , f z, f ~ -+f2 and t erms such as 2ft-+J2, second- and t hi r d- or der t erms are readily measured. Har moni c- di st ort i on figures can be cal cul at ed from i nt er modul at i on measur ement s [5.21]. The advant age of mul t i t one IM testing is t hat harmoni cs present in the test signals themselves are n o t present in t he sum or difference frequencies. The same spect rum anal yzer also facilitates measuri ng the frequency response of a t ransmi t t er, part i cul arl y if a swept -frequency measur ement is made. Ackmm,ledqements. We wish t o t hank F.S. Chen, P. W. Dor nr an, W.J. Cl emet son, and C. A. Bracket for useful di scussi ons on circuits and per mi ssi on to use some of t hei r dat a and desi gn ideas. List of Symbols L I ?l J tr BW f BER VCE(s;II} R Li ght - power level [ mW or ].tW] Devi ce cur r ent [ mA or A] l ndex of refract i on Wavel engt h of light [ mn or / . t m] Cur r ent densi t y [ kA/ cm 2] Ri set i me [ ns ] Bandwi dt h [ MHz or GHz ] Fr equency [ Hz] Bit er r or rate I s - J] Tr ansi st or cur r ent gain /h; E Col l ect or - t o- emi t t er vol t age of a sat ur at ed bi pol ar t r ansi st or [ V] Resi st ance lEVI 198 P. W. Shumate and M. DiDomenico, Jr. C Capacitance [ pF or laF] 1 "2 L Transi st or-t ransi st or logic (family) / o, Out put current of a gate in the high slate [ mA] ECL Emitter-coupled logic (family) C,, h Out put capacitance, common-base configuration [ pF] E'I~H~ LED capacitance, equal to space-charge plus diffusion plus stray (e.g., lead) component s [ pF] Vcc Colleclor supply voltage IV] VI~ E Emitter supply voltage [V] AGC Automatic gain control I/i.~,:~ ~ Voltage across terminals of LED when emitting light [V] VBE Forward voltage of a p-n (e.g., base-emi t t er)j unct i on [V] I c Collector current [ mA] ll~ Emitter current [ mA] 7' Temperat ure [ K] Vi~ Input voltage [V] VGs Gate-to-source w)ltage IV] Extinction (on/off~ ratio t Time [h] Vj Junct i on voltage [V] I d Drive current [ mA] E A Activation energy [eV] k Boltzmann const ant [8.625 10- "~ eV/ K] N.A. Numerical aperture References 5.1 M.L. Dakss: Laser Focus II, 31 (Dec. 1975) 5.2 B.S.Kawasaki, D. C. Johnson: Opt. and Quant . Electr. 7, 281 (1975) 5.3 K. Thyagarajan, A.Sharma, A. K. Ghat ak: Appl. Opt. 17, 2416 (1978) 5.4 S. Iloriuichi, K. lkeda, T. Tanaka, W. Susaki: IEEE Trans. ED-24, 986 (1977) 5.5 M.Abc, l. Umebu, O. l l asegawa, S. Yamakoshi, T. Yamaoka, T. Kot ani , H. Okada, H. Takanashi : IEEE Trans. ED-24, 990 (1977) 5.6 M. Ettenberg, H.Kressel, J. l ' . Wi l t ke: IEEE J. QE-12, 360 (1976) 5.7 C.A. Brackett : .1. Appl. Phys. 45, 2636 (1974) 5.8 C. C. Ti mmer mann: Appl. Opt. 15, 2432 (1976) 5.9 M. Maeda, I. lkushima, K. Nagano, M. Tanaka, H. Naskshima, R. l t oh: Appl. Opt. 16, 1966 (1977) 5.10 D. Gloge: I EEETr ans. MTT-23, 106 (1975) 5.11 S.E. Miller, E. AJ. Marcat i l i , T. I , i : Pr oc. I EEE61, 1703 (1973) 5.12 J.W. Fleming: Electron. Lclt. 14, 326 (1978) 5. 13 D.N.Payne, W. A. Gambl i ng: Electron. Letl. 11, 176 11975) 5.14 A. G. Dent ai , T.P. Lee, C. A. Burrus: Electron. Left. 13, 484 (1977) 5.15 l. Umebu, O. Hasegawa, K. Akita: Electron. Lett. 14, 499 (1978) 5.16 R.J.Nelson, R. G. Sobers: J. Appl. I'hys. 49, 6103 (1978) 5.17 T.P. Lee: Bell Sysl. Tcch. J. 54, 53 (1975) 5.18 J.Heinen, W. Har t h: El ect r on. Lelt. l l , 512 (1975) 5.19 R.C.Goodfellow, A.W. Mabbi t l : Elcctron. Left. 12, 50 (1976) 5.20 J.1teinen, W. Huber, W. Har t h: Electron. Lett. 12, 553 (1976) 5.21 Transmission Systems f or Communications, 4th ed., rev. (Bell Tel. Lab. 1971) Chap. 10 5.22 F. D. King, J.Strauss, O.I.Szentesi, A.J.Spri,agthorpe: Proc. lEE 123, 619 (1976) 5.23 L.R. Dawson, V.G. Keramidas, C. L.Zipfel: In prcparation. (LEDs are the same type as t'hose reported in [5.31]) 5.24 T.Ozeki, E. H. Hara: Electron. Lett. 12, 80 (1976) 5.25 R.W. Dawson: IEEE Trans. ED-25, 550 (1978) Lightwave Transmi t t ers 199 5.26 J . St r aus s : I EEE J. QE-14, 813 (1978) 5.27 T. P. Lee: Prec. IEEE 65, 1408 (1977) 5.28 K. Asatani, T. Ki mur a: IEEE Trans. ED-25, 199 (1978) 5.29 K.Saito, N.Shige, T. Kajirnura, T. Tsukada, M. Maeda, R. It o: Tech. Digest, Intern. Con['. Integrated Optics and Opt. Fiber Commun. , Tokyo (1977) p. 65 5.30 A.A.Bergh, P. J. Dean: Prec. IEEE 60, 156 (see Sect. VI.A) (1972) 5.31 L. R. Dawson, V. G. Keramidas, C.L.Zipfel: Extended Abstracts Eleclrochem. Soc. Meeting 7?-2, 855 (1977) C.L.Zipfel, V. G. Kerami das, R. H. SauI: Device Res. Conf., Boulder, Co. (1979), paper TA-B2 5.32 Fiber Optics Handbook and Market Guide (Information Gatekeepers, Brookline, Mass. 1978) Sect. IV 5.33 R.L. Hart man, N. E. Schumaker, R. W. Di xon: Appl. Plays. Left. 31, 756 (1977) 5.34 P. F. Pant er : Modulation, Noise, and Spectral Amdysis (McGraw-Hill, New York 1965) 5.35 A. A. Gol dberg: Electronics 49, 94 (Febr. 5, 1976) 5.36 J. O. Li mb: IEEE Trans. COM-25, 1349 (1977) 5.37 E. H. Hara, T. Ozeki: IEEE Trans. CATV-2, 18 (1977) 5. 38 B.S.Helliwell: IEEE Intercon. Tech. Papers, Session 12, paper 5 (1974) 5.39 E. H. Har a: I EEETr ans . CATV-2, 120 (1977) 5.40 S.Akiyama, T.Otsuka, M.Sekiguchi, K. Fukushi ma, T. Hamanaka, K. Ni shi mura: Tech. Digest, Intern. Conf. Integrated Optic~ and Opt. Fiber Comrnun., Tokyo (1977) p. 481 5.41 F. W. Dabby, R.B.Chesler, J. Arbut hofl ; M. Canest ri : Intern. Conf. Integrated Optics and Opt. Fiber Commun. , Tokyo (1977) postdeadline paper P-13 5.42 J.Shekel: Prec. IEEE 58, 1024 (1970) 5.43 E. Rot hol z: Electron. Lett. 14, 584 (1978) 5.44 T. L. Paol i : Appl. Phys. Lett. 24, 187 (1974) 5.45 T. L. Paol i : I EEEJ . QE- I I , 276(1975) 5.46 G. Guekos, H.Jackel, K. F. Scbmi d: Electron. Lett. 12, 64 (1976) 5.47 H. Jaskel: Electron. Lett. 12, 289 (1976) 5.48 T. Lec, C.A.Burrus, Jr.: IEEE J. QE-8, 370 (19721 5.49 W.M. Hubbar d: Bell Syst. Tech. J. 52, 731 (1973) 5.50 C. R. Pat i saul : SPIE-Guided Opt. Commun. 63, 76 (1975) 5.51 C. Game, A.Jessop: 1EEE Conf. Pub. Opt . Fi ber Comm. 132, 171 (1975) 5.52 M. Rousseau: IEEE Conf. Pub. Opt . Fi ber Comm. 132, 174 (1975) 5. 53 Y.Takasaki, M. Tanaka: Prec. 1EEE 63, 1081 (1975) 5.54 Y.Takasaki, M. Tanaka, N. Maeda, K.Yamashita, K. Nagano: IEEE Trans. COM-24, 404 (1976) 5.55 C. C. Ti mmerman : IEEE Trans. BC-23, 12 (1977) 5.56 M. Rousseau: Electron. Lett. 12, 478 (1976) 5.57 R.Petrovic: Electron. Lett. 14, 541 (1978) 5.58 K. Nakagawa, E. Yoneda : Prec. 4th Europ. Conf. Opt. Fiber Commun. , Genoa (1978) p. 574 5.59 D. Ka t o: I EEE J. QE-14, 343 (1978) 5.60 Reference Data fi)r Radio Engineers, 5th ed. (Howard W.Sams & Co., New York 1973) Chap. 21 5.61 M. K. Bar noski : Fundamentals of Optical Fiber Communications (Academic Press, New York 1976) Chap. 6 5.62 F. R. McDevi t t : SPIE-Guided Opt. Commun. 63, 31 (1975) 5.63 J.E.Goell, T.A.Eppes, C. Kao: SPl E-Gui ded Opt. Commun. 63, 50 (1975) 5.64 C. K. Kao, J. E. Goel l : Electronics 49, 113 (Sept. 16, 1976) 5.65 Optical Fiber Communi cat i ons - Technical Note RI (ITT, Electro-Optical Products Div. 1977) Roanoke, VA 5.66 The TTL Data Book ./or Design Engineers, 1st ed. (Texas Instruments, Dallas, Tx. 1973) 5.67 D. Koehl er: IEEE Spectrum (Nov. 1965) 5.68 B. Jarret t : Electron. Des. 22, 96 (July 19, 1974) 5.69 G. White, C. A. Burrus: Int. J. Electron. 35, 751 (1973) 5.70 M. Chessman, N. Sokal : Electron. Des. 24, 110 (1976) 5.71 J.L. Kos i kows ki : l EEETr ans . CT-I1, 129 (1964) 200 P. W. Shumate and M. DiDomenico, Jr. 5.72 W.W. Brown, D.C.Hanson, T.Hornak: U.S.Patent 4,1)27, 152(1977) 5.73 W. Brown, D. Hanson, T. Hornak: IEEE Intern. Solid State Circuits Conf. Digest of Tech. Papers XXI, 18 (1978) 5.74 R.A. Deters : Digital Des. 8, 70 (May, 1978) 5.75 Johnson Space Center: Mach. Des. 50, 40 (June 8, 1978) 5.76 P.R.Selway: Proc. IEEE 123, 609 (1976) 5.77 El.Kressel, J.K. Butler: Semiconductor Lasers and Heterc~junction LEDs (Academic Press, New York 1977) Chaps. 8, 17 5.78 l-l.C.Casey, Jr., M.B.Panish: Heterostructure Lasers Part B: Materials and Operating Characteristics (Academic Press, New York 1978) Chap. 7 5.79 V.Ostoich, P.Jeppesen: Electron. Lett. 11,515 (1975) 5.80 S.M.Abbott, W.M.Muska, T.P. Lee, A.G. Dentai, C.A. Burrus: Electron. Lett. 14, 349 (1978) 5.81 S.D.Personick: Bell Syst. Tech. J. 52, 875 (1973) 5.82 P.W.Shumate, Jr., F.S.Chen, P.W.Dorman: Bell Syst. Tech. J. 57, 1823 (1978) 5.83 M.A. Karr, F.S.Chen, P.W.Shumate: "'Output Power Stability of GaAIAs Laser Transmitters Using an Optical Tap for Feedback Control". Appl. Opt. 18, 1262 (1979) 5.84 M.A. Kcrr, T.C. Rich, M.DiDomenico, Jr.: AppI. Opt. 17, 2215 (1978) 5.85 R.E. Epworth: Proc. 2nd Europ. Conf. Opt. Fiber Commun., Paris (1976) p. 377 5.86 S.R.Salter, D.R. Smith, B.R.White, R.P. Webb: Proc. 3rd Europ. Conf. Opt. Fiber Commun., Munich (1977) p. 208 5.87 A. Brosio, P.L.Carni, A.Moncalvo, V.Seano: Proc. 4th Conf. Opt. Fiber Commun., Genoa (1978) p. 438 5.88 J.Gruber, P.Marten, R.Petschacher, P. Russer: IEEE Trans. COM-26, 1088 (1978) 5.89 A.Albanese: Bell Syst. Tech. J. 57, 1533 (1978) 5.90 T.L.Paoli, P.A. Barnes: Appl. Phys. Lett. 28~ 714 (1976) 5.9l T.Ozeki, E.H.Hara: Electron. Left. 12, 80 (1976} 5.92 A. Albanese, H.F. Lenzing:J. SMPTE 87, 821 [1978) 5.93 F.S.Chen, M.A.Karr, P.W.Shumate: Bell Sysl. Tech. J. 58, 1617 (1979) 5.94 A.Prochazka, P. Laucaster, R.Neumann: IEEE Trans. CATV-I, 31 (1976) 5.95 S.Waaben : IEEE Intern. Solid State Circuits Conf. Digest of Tech. Papers CVIII, 30 (1975) 5.96 K. Asatani, T. Ki mura: IEEEJ. SSC-13, 133 (1978) 5.97 l,',.W. Dawson: Electronics 50, 119 (Nov. 10, 1977) 5.98 I.Umebu, M. Abe, T. Yamaoka, T. Kotani, N.l-lanano, K.Iguchi, T. Yoshibayashi: Tech. Digest, Intern. Conf. Integrated Optics and Opt. Fiber Commun., Tokyo (1977) p. 109 5.99 K. Asatani, T. Kimura: Electron. Lett. 13, 162 (1977) 5.100 K. Asatani, T.Kimura: Tech. Digest, Intern. Conf. Integrated Optics and Opt. Fiber Commun., Tokyo (1977) p. 11)5 5.101 J.Strauss, O.I.Szentesi: Electron. Lett. 13, 158 (1977) 5.102 J. Strauss, O. I. Szen test : Proc. 1977 Intern. Symp. Circuits and Systems, Phoenix, Ariz. ( 1977) p. 288 5.103 J.Strauss, A.J.Springthorpe, O.I.Szentesi: Electron. Lett. 13, 149 (1977) 5.104 G. Moy: I EEEJ. QE-14, 865 (1978) 5.105 C.C. Timmermann : Nachrichtentech. Z. 3 I, 579 (1978) 5.106 F.J.MacWilliams, N.J.A.Sloane: Th~ Theory of Error-Correctim! Codes-Part II (North- ltolland, Amsterdam 1977) Chap. 14 5.107 M. Damashek : Electronics 49, 107 (May 27, 1976) 5.108 J.T. Harvey : Electronics 4g, 104 (Nov. 27, 1975) 5.109 H.Pangratz: Electron. Des. 26, 134 (May 19, 1978) 5.110 T~,leeommunications 7i'ansmission Engineerin~i, Vol. 1 (American Telephone and Telegraph Co. 1974) Sect. 4 5.111 L.E. Weaver: Television Video 7~'ansmission Measurements, 2nd cd. (Lbenezer Baylis and Son, London 1977) 5.112 See numerous Application Notes available from test-equipment manufacturers, e.g., Hewlett-Packard Application Note 150-11, "Spectrum Analysis-Distortion Measurements" (19761 6. Fiber Couplers M. K. Barnoski W i t h I1 F i g u r e s To use opt i cal fibers in communi cat i ons systems requires component s for coupl i ng light-emitting semi conduct or devices to the fibers and for i nt ercon- necting separat e lengths of fiber. This chapt er briefly treats the pr obl em of coupl i ng light sources to fibers by bot h direct but t j oi nt coupl i ng and coupl i ng using i nt ermedi at e optics. Tr eat ment of the coupl er probl em is followed by a discussion of the at t enuat i on losses t hat result from mechani cal mi sal i gnment of t wo i nt erconnect ed fibers. 6.1 Source-to-Fiber Coupling Thi s section reviews vari ous t echni ques used for coupl i ng power from LEDs and laser di odes into single st rands of mul t i mode optical fibers. Source-t o-fi ber coupl i ng t echni ques include direct but t coupling, mi cro- lenses, and the use of t apered or bul b-ended fiber pigtails. The simplest of these is di rect but t coupl er shown schemat i cal l y in Fig. 6.1. For a source with a cylindrically symmet ri c radi at i on pat t ern, it is a simple exercise [6.1] to show t hat the power coupl ed into t he fiber is, to a good appr oxi mat i on, given by s i n ~ (N.A.} S B(O)sinOdO P t ' i b e r = o P ....... = ~l~ P . . . . . . .[ B(O)sinOdO o (6.1) where B(O) is the source brightness, P . . . . . . is tlae t ot al power emi t t ed by the source, and N.A. is the fiber numeri cal apert ure. It is assumed t hat t he emission area of the source is less t han or equal t o t hat of t he fiber core. For a source such as a surface-emi t t i ng LED, the r adi at i on pat t er n of which is Lambert i an, the coupl i ng efficiency ~L=(N. A. ) z. A plot of coupl i ng efficiency vs the numeri cal aper t ur e of a step-index fiber is shown in Fig. 6.2. ( For an edge- emi t t i ng LED, the radi at i on pat t er n is asymmet ri cal as discussed in Sect. 2.6, where fiber coupl i ng is discussed for this case.) For a step-index waveguide, t he number of bound modes is appr oxi mat el y equal to N = V 2 / 2 , [2.66] where V 2 = ( k( d) 2 ( H 2 _ 1 1 2 ) = ( k a ) 2 ( N , A.) 2 . ( 6 . 2 ) 202 M. K. BurRo,ski 9 - - 8 7 6 5 5 ,7 ~ 4 ~ a 2 1 0 ,(cos 8)n.9- ANGUL AR DI STRI BUTI ON OF OPTI CAL POWER EMI TTED BY SOURCE / \ n 2 n] OPTOE LECTRONI C SI NGLE FI BER SOURCE STRAND f _~_ I ~ I I 0.10 0.20 0.30 NUMERICAL APERTURE Fig. 6. 1. Sc he ma t i c i l l ust r at i on of di rect but t c oupl i ng Fig. 6.2. Pl ot of coupl i ng efficiency vs numer i cal a pe r t ur e of st ep- i ndex fiber for a La mbe r t i a n s our ce The efficiency with which power is coupl ed i nt o a fiber is t herefore di rect l y pr opor t i onal to the number of bound modes. In the above expressi on, k and a are the free space wave number (2~/2) and fiber core radius, respectively. Her e 2 is the free space opt i cal wavel engt h. For a gr aded- i ndex fiber, t he refract i ve index varies with core radi us as l 6 3 , Fiber Couplers 203 OPTI CAL SOURCE OPTICAk ~ ~ SOURCE ~ & . SIZE ds t - f - ANGUL AR DI STRI BUTI ON OF OPTICAL POWER EMI TTED BY SOURCE . . . . . q l CLAD - ! OPTI CAL COUPLI NG SYSTEM l Fig. 6,3. l[lustration of optical source to fiber coupler q_-- where n 1 = n(0) is t he refract i ve index at the cent er of the fiber core, ~ specifies the shape of the refract i ve index profile, and A=( n ~- n z ) / 2 n ~ wi t h 1 1 2 = n ( a ) bei ng t he refi' active index at the edge of the fiber core. The numbe r of bound modes of a gr aded- i ndex fiber is given by [-2.66], M= ( / ~ a ) ~ ( n ~ - n ~ ) = ~ + 2 ' (6.4) Since the number of bound modes of a gr aded- i ndex fiber wi t h a given n 2 and n 1 is smal l er by a fact or of [~z/(:+2)] t han the numbe r for a st ep-i ndex fiber wi t h core and cl addi ng refract i ve indices of n 1 and 172 the coupl i ng efficiency is less for a gr aded- i ndex fiber t han for a st ep-i ndex fiber. For a fiber with a par abol i c profile (e = 2), t he coupl i ng efficiency is reduced by a fact or of 2. The coupl i ng efficiency bet ween a l aser di ode or LED and an opt i cal fiber can be enhanced by i nt r oduci ng an i nt erveni ng opt i cal syst em (i l l ust rat ed in Fig. 6.3), such as a sphere lens or a t aper ed section of fiber, onl y if the pr oduct of the opt i cal emi ssi on area times the emi ssi on solid angl e is less t han the pr oduct of the cross-sect i onal ar ea of the fiber core times the accept ance solid angl e of t he fiber. The opt i cal el ement serves to magni f y the sour ce ar ea ill the same pr opor t i on t hat it demagni fi es (collimates) t he solid angul ar di st ri but i on of the radi at i on, so t hat the r esul t ant r adi at i on pat t er n is bet t er mat ched to the fiber N. A. The ma x i mu m coupl i ng efficiency I-6.2] t hat can be realized usi ng a Lamber t i an sour ce is ~1~ ( maxi mum) = ( N . A . ) 2 AF As ' (6.5) where A F is the cross-sect i onal ar ea of the fiber core, and A s is the l i ght -emi t t i ng ar ea of t he source. The efficiency of the i nput opt i cal coupl i ng i mpr oves as the 204 M. K. Barnoski Fig, 6.4, Tapered launcher input coupler 2a emission area of t he source decreases and the fiber core cross-sectional area increases. The t ot al optical power from a source with a uni form r adi at i on di st ri but i on across its emission area is appr oxi mat el y Ps ~ 2 x A s B 5.1(O)dO (6.6) when cylindrical symmet r y is assumed. The source brightness B(O) has been expressed as B(O) = Bf (O), where If(0)[ < 1, and B is the maxi mum value of B(O). The t ot al optical power emi t t ed by the source is directly pr opor t i onal to the pr oduct of the emission area and brightness. If t he emission area is decreased to enhance i nput coupl i ng efficiency, the source brightness must be increased if the t ot al power emi t t ed by the source is t o remai n const ant . Since t he power coupl ed into the fiber is di rect l y pr opor t i onal to the optical source brightness, it is advant ageous to make the semi conduct or source as bri ght as is consi st ent with the restraints placed on the size of t he device by semi conduct or mat eri al and device processing t echnol ogy. An optical system t hat efficiently coupl es light from a semi conduct or source i nt o an opt i cal fiber is the t aper l auncher [6.3, 4] shown in Fig. 6.4. This optical coupl i ng system consists of a short section of fiber heat ed and pulled into the shape of a taper. Equat i on (6.5) predicts that, for a Lamber t i an emi t t er with emission di amet er <2 a l , the i mpr ovement in coupl i ng efficiency over t hat of di rect but t coupl i ng is R z = (a/ a 1)2. Experi ment al results [6.4] obt ai ned using a pl anar, Lamber t i an LED with a di amet er of 22 gm are shown in Fig. 6.5. The figure shows a pl ot of the i mpr ovement fact or ( a / a l ) 2 vs t aper rat i o ( a/ al ) with the measured dat a points also shown. The tapers used in the experi ment were fabricated using Cor ni ng silica fibers with N. A. = 0.18. The experi ment al results obt ai ned are in excellent agreement with t hose predicted by t heory. The t aper l auncher has also been used as an optical coupl i ng system with a het er oj unct i on [6.4] LED and a doubl c- het cr oj unct i on laser [6,3]. Nei t her of these devices is a Lamber t i an emitter. Both have mor e di rect i onal radi at i on pat t erns, a consequence of which is hi gher coupl i ng efficiency for bot h direct but t coupl i ng and coupl i ng with an opt i cal element. Coupl i ng efficiencies as 10 / , , / , i / ~ N A 0 t 8 I / 3 4 5 6 T A P E R R A ] I O H Fi ber Coupl e r s 205 Fig. 6. 5. l ) e pc nde nc e of t he power coupl i ng efficiency on t he t aper r al i o for a La mbe r l i a n s our ce [6. 4] 600 500 400 300 200 i lO0 50 / / . / - / / / / / - e / / / 7 ' / / ! / '_ / / / I I I / NA : 0.26 / / NA : 0.18 I [ I I I I I I I I I 2 3 4 5 6 7 8 TAPER RATIO, R I - 10 I 9 10 lOO 50 >- 0 z t.t0 r , O t . t . z o Fig. 6.6. De pe nde nc e of power coupl i ng bet ween edge- emi t t i ng doubl e - he t e r oj unc t i on LED and l aper ended fiber on t be t aper r at i o [6.4] 206 M. K. Barnoski SPHERE LENS (n = 2.0),Xx ~ ( a ~ A I R (n = 1.0) EPOXY % ~ L I G H T SOURCE M E T A L SPHERE LENS / . Z n D I F F U S E D \ EPOXY ~ / / / n - G a A I A s / I f SUBSTRATE I.. . . . . . ~ . . . . . . ~ M E T A L I " H E A T S I N K " ~ ' S O L D E R b FiR. 6.7. (a) A si mpl e geomet r i cal con f i gur at i on of t he LE D wi t h a s pher e l ens [6, 5] ; (b) a s chemal i c cr oss- sect i onal dr a wi ng of t he Ga As - Ga AI As he t e r os t r uc t ur e I , ED wi l h a sel f-al i gned s pher e l ens high as 97 % were observed using a DH laser and a t aper l auncher fabri cat ed with Cor ni ng fiber (N.A. =0.18). The length of the t aper was 4.3 mm with a t aper rat i o 3.5. The direct but t j oi nt coupl i ng efficiency measured using the same laser di ode is 30 %. The t apered l auncher t herefore provi ded an i mprove- ment rat i o of 3.2 in this case. Maxi mum experi ment al coupl i ng efficiencies obt ai ned [6.4] using an edge-emi t t i ng doubl e- het er oj unct i on LED and t apered fibers with numeri cal apert ures of 0.18 and 0.26 are 53 and 83 %, respectively. The plot shown in Fig. 6.6 displays the dependence of the optical power coupl ed into the fiber on the t aper rat i o using an LED t hat emits 650gW at 100mA drive current. The measured coupl i ng efficiencies for direct but t coupl i ng to t he 0.18 and 0.26 numeri cal aper t ur e fibers are 11% and 17%, respectively. Microlenses can also be used as the optical coupl i ng system. Self-aligned [6.5] sphere lenses have been used to increase the power coupl ed from an LED i nt o the fiber waveguide. The geomet ri cal confi gurat i on used is shown in Fig. 6.7. In this confi gurat i on, the sphere is aut omat i cal l y cent ered in the et ched hole, which is filled with clear epoxy. Since the surface-emi t t i ng doubl e- het er oj unct i on di ode has a Lamber t i an radi at i on pat t ern, the maxi mum coupl i ng efficiency obt ai nabl e using the ball lens is given by [6.5]. The experi ment al l y det ermi ned efficiency of coupl i ng i nt o a 80gm core di amet er fiber with an N.A. of 0.14 is shown as a funct i on of LED emi t t i ng area di amet er in Fig. 6.8. The ball lens di amet er was appr oxi mat el y 100pm. The solid curve was comput ed using [6.5]. The solid dot is the experi ment al l y measured coupl i ng efficiency using a 35!am di amet er LED. The open circles are the efficiencies est i mat ed based on experi ment al det er mi nat i on of t he power within a numeri cal aper t ur e of 0.14 using LEDs with emi t t i ng di amet ers of 25, 30, 35, and 50 pm. The results agree closely with t hat predi ct ed theoretically. For an emi t t i ng-area di amet er of 35 gin, a coupl i ng efficiency of 9 % was observed. This is an i mpr ovement of appr oxi mat el y 4.5 above t hat expect ed for direct but t coupl i ng i nt o a fiber with N.A. =0. 14. Fi ber Coupl er s 207 > . - LU < '2- u. Lu z o 100 5 0 2 0 1 0 5 0 2 0 10 I I I I 1 A M 8 0 p . m 10 20 30 4 0 5 0 6 0 70 E M I T T I N G A R E A D I A M E T E R d s , H - m Fi g. 6.8. Coupl i ng efficiency vs sur f ace- emi ni ng LED emi t t i ng- ar ea di amet er . A solid dot and curve r epr esent t he exper i ment al result and t heoret i cal limil, re- spectively. Open dot s shrew l hc coupl i ng efficiencies expect ed from t he exper i ment al results of t hc i mpr ovement fact ors 1"6.5] Anot her optical system for enhanci ng coupling efficiency which has been experimentally investigated is the spherical, or bulb-ended, fiber [6.6 8]. As with the tapered launcher and ball lens, the experimental results obt ai ned for this type of coupler are in good agreement with theory. High-brightness semi conduct or light-emitting devices can be efficiently coupled to communi cat i on-grade optical fibers using the optical elements discussed above. The semi conduct or source, coupling optics, and a short section of fiber can be packaged as a uni t with i nt ercounect i on to the fiber cable provided for by a fiber connect or or permanent splice (Sect. 2.4.2). 6.2 Fiber-to-Fiber Coupling Practical i mpl ement at i on of fiber systems requires interconnecting fibers with maxi mum transfer efficiency of the optical signal. There are two i nt e r c on- nection techniques. The first, the fiber splice, is a permanent i nt erconnect i on of two fibers. The second, tile fiber connector, is designed for many repeated connect i ons of two fibers. Both types of interconnection may have at t enuat i on resulting from variation in the intrinsic parameters of the two fibers to be connected. The intrinsic fiber paramet ers are core di amet er and shape, cl addi ng di amet er and shape, numerical aperture, refractive index profile, and the concentricity of the core and cladding. In addi t i on to at t enuat i on resulting from vari at i ons in these intrinsic fiber parameters, interconnection losses can also result from mechanical mi sal i gnment of the fiber ends, i mproperl y prepared and cleaned fiber end surfaces, and Fresnel reflections. 208 M. K. Barnoski a POWE R SUPPLY 100x 20x MODE ~ _ . _ . . LENS L E N S STRIPPER (-~ r-"a f } LAUNCHING MIC R OPOSITION E R S ~ .,,,~,~r~ ~ FIBER ENDS MODE SOLAR CELL / STRIPPER J CHART L _ _ _ ] D I G I T A L ~ R E C O R D E R ) I b . POWER ] METER _ 1 DIGITAL MICROPOSITIONER --_ T T ~ LAUNCHINGIMULTIMETERI LED FIBER DRIVER OBJECTIVES x 25 ~ 4 IN. S @ ~ 4 1 N . SPOOL ROTATING TABLE POWE R METER IN. SPOOL DIGITAL MULT METER Fig. 6.9. (a) Fiber and misalignment measurement setup in the first experiment; (b) fiber and misalignmenl measurement setup in the second experiment [6.9] Th e de s i gn o f fi ber s pl i c i ng t e c hni que s a nd fi ber c o n n e c t o r s requi res a de t ai l e d k n o wl e d g e o f t he s e ns i t i vi t y o f t he i n t e r c o n n e c t i o n i ns e r t i o n l o s s o n me c h a n i c a l a l i g n me n t o f t he f i ber e nds . Ex p e r i me n t a l s e t ups us ed t o me a s u r e [ 6 . 9 ] t he effect o f me c h a n i c a l mi s a l i g n me n t o n c o u p l i n g l o s s are s h o w n in Fi g. 6.9. In b o t h a r r a ng e me nt s , t he f i bers we r e e x c i t e d by s ur f a c e - e mi t t i ng L E D s Fi ber Coupl er s 209 NORMALIZED ANGULAR MISALIGNMENT, ~'/sin - 1 NA 0.1 0.2 0.3 0.4 0.5 0.6 4 I I I l I FIRST ,,~ ; - ~ OFFSET ALONE EXPERIMENT 3 I / s E c o N D ~ 2 - - i / E X P E R I M E N T Q / / ANGULAR MISALIGNMENT I I ATSIa = 1 / / / / ' / / / / j SEPARAT,O [ / / ~ ~ ALONE - ..........---~. o 1 L I I 0 1 2 3 4 5 6 NORMALIZED OFFSET, d/a OR SEPARATION,S/a Fig. 6.10. Loss in dB vs nor mal i zed offset din, separ at i on S/a, and angul ar mi sal i gnment #/ si n - 1N.A. [6.9] with a 50 gm emi t t i ng di amet er. The experi ment al ar r angement shown in Fig. 6.9a was used to measure the dependence of coupl i ng loss on the lateral offset of the t wo fiber cent ers and on t he l ongi t udi nal (on-axis) separat i on of t he t wo fiber ends. A 1.83 m length of graded-i ndex fiber with 50 gm out si de di amet er was used. The ar r angement shown in Fig. 6.9b was used to measure the effects of lateral, longitudinal, and angul ar misalignments. A 20m length of graded- i ndex fiber with 55 lain cor e di amet er and 110 gm out si de di amet er was used in this case. In bot h cases, the on-axis numeri cal aper t ur e is N.A. =0. 2. The lateral offset (defined as d) and the l ongi t udi nal separat i on (defined as S) can be normal i zed to the fiber core radius a (t hat is, d/a and S/a). The angul ar mi sal i gnment (defined as #) can be normal i zed to the axial numeri cal aper t ur e N. A. = ~ - n 2 (6.7) [ t hat is, #/ s i n- ~ (N.A.)]. The measured coupl i ng loss in dB as a funct i on of t hree mechani cal mi sal i gnment s is shown in Fig. 6.10. The first experi ment refers to the ar r angement of Fig. 6.9a; the second t o t he ar r angement of Fig. 6.9b. The results clearly reveal t hat the coupl i ng loss is most sensitive t o lateral misalignment. I ndependent measur ement s made of the coupl i ng loss de- pendence on one of the t hree normal i zed par amet er s with the ot her t wo held const ant yield dat a t hat can be used t o pl ot const ant loss curves as caused by vari ous kinds of mi sal i gnment (Fig. 6.11). A coupl i ng loss of 0.5 dB can result 210 M . K . B a r n o s k i Z O < < 03 t~ uJ N -J < n- O Z 0. 5r i b LOSS o ~-~/sin - 1 NA : 0 . . . . . . 1. 0dB LOSS A : 0. 087 . . . . 1. 5dB LOSS (3 = 0. 174 = 0.261 \ " ' , \ , ' , " , \ . !I I \ \ " ' " " " " i " I \ \ \ \ - - , ~ \ \ , \ I \ \ " . , \ ",.' ,. \ 0 0.1 0. 2 0. 3 0. 4 NORMALI ZED OFFSET, d/ a 0.5 F i g . 6. 11. Cons l al l l l oss l i nes as t he r esul t of fiber end offset d / a , s e pa r a t i on S / a , and a ngul a r mi s a l i gnme nt f l / s i n i N. A. [6. 9] f r om a nor mal i zed offset of 0.2 alone, a nor mal i zed separ at i on of 2 al one, or a nor mal i zed offset of 0.1 and a nor mal i zed separ at i on of l. Since coupl i ng losses of 0.5 dB or less are very desirable, the const ant loss curves shown in Fig. 6.11 pr ovi de a val uabl e guide to tile connect or designer. The losses shown in the figure are, of course, mi ni mum losses since t hey include onl y the effects of mechani cal mi sal i gnment and not fiber intrinsic losses or the effects of i mpr oper l y pr epar ed or cl eaned fiber ends and Fresnel reflection loss. Fresnel reflection losses, which are appr oxi mat el y 0. 2dB per fiber face, can be mi ni mi zed by using index mat chi ng mat eri al . El i mi nat i on of the effect of end pr epar at i on and cleanliness are, of course, i nt i mat el y rel at ed to the design and assembl y of the connect or. List of Symbols I11 H2 N.A. / )fi ber P s o u r c e R(0) N k (J O~ Refract i ve index of fiber core Refract i ve index of fiber clad Fi ber numeri cal aper t ur e I nput coupl i ng efficiency Opt i cal power coupl ed i nt o fiber Opt i cal power emi t t ed f r om source Source bri ght ness Nu mb e r of bound modes of st ep-i ndex fiber Free space wave number Fi ber core radi us Profile shape fact or Fiber Couplers 211 M Av As (l 1 R d S d~ Numbe r of bound modes of gr aded- i ndex fiber Cross-sect i onal ar ea of fiber core Li ght - emi t t i ng ar ea of the sour ce Mi ni mum radi us of t aper ed fiber Taper r at i o Lat eral offset Longi t udi nal offset Angul ar mi sal i gnment Di amet er of emi t t i ng ar ea of source References 6.1 M.K. Barnoski: "'Couplim.! Components f or Optical Fiber Waveyuides", in Fundamentals ()[" Optical Fiber Communications, ed. by M. K. Barnoski (Academic Press, New York 1976) Chap. 3 6.2 M.C. Hudson: Appl. Opt. 13, 1029 (1974) 6.3 T.Ozcki, B. S. Kawasaki : Electron. Lett. 12, 607 (1976) 6.4 Y. LJematsu, T.Ozeki: "E[ficient Power Coupling Between a M H LED and a Mul t i mode Fiber with Tapered Lamwher", Tech. Dig. of 1977 Intern. Conf. Integrated Optics and Optical Fiber Commun., Tokyo, Japan (1977) p. 371 6.5 S. I toriuchi, K. Ikeda, T.Tanaka, W. Sasuki : [EEE Trans. ED-24, 986 (1977) 6.6 D. Kato: J. Appl. Plays. 44, 2756 (1973) 6.7 B.S.Kawasaki, D.C.Johnson: Opt. Quantum Electron. 7, 281 (1975) 6.8 M.Abe, l.Umebu, O.Hasegawa, S.Yamakoshi, T. Yamaoka, T. Kotani, H.Okada, l t.Takanashi: [EEF Trans. ED-24, 990 (1977) 6.9 T.C.Chu, A.R.McCorrnick: Bell Syst. Tech. J. 57, 595 (1978} 7. Modul at i on of Laser Di odes * G. Arnol d, P. Russer, and K. Pet er mann Wi l h 25 Fi gures GaAs doubl e- het er ost r uct ur e semi conduct or injection lasers which now exhi bi t mor e t han 25,000 h cw r oom t emper at ur e lifetime are of great interest for fut ure use as directly modul at ed t r ansmi t t er for high bit-rate fiber optical com- muni cat i ons. The effects limiting this appl i cat i on are modul at i on di st ort i ons, spectral width and addi t i onal spectral br oadeni ng in the case of modul at i on and self-pulsations of the out put power, The dynami c and spectral behavi or of injection lasers, the met hods of high bi t -rat e modul at i on and the i mpr ovement of the high bit-rate modul at i on capabi l i t y by coupl i ng t wo lasers are discussed. 7 . 1 B a c k g r o u n d Semi conduct or injection lasers are of great interest as t ransmi t t ers for high bit- rate fiber opt i cal communi cat i on systems [7.2]. The main adwmt ages of semi conduct or injection lasers are simple const ruct i on, small dimensions, high efficiency and direct modul at i on capabi l i t y up to the GHz range. Semi conduct or injection lasers yield a good coupl i ng efficiency also i nt o monomode fibers [7.3] and for t hei r nar r ow emission spect ra onl y a low pulse br oadeni ng due to fiber, dispersion when monomode fibers are used. At the wavel engt h of GaAs injection lasers the dispersion of usual monomode fibers is 1 ns k m - ~ for 1% relative opt i cal bandwi dt h [7.4, 5]. Consi derabl e effort has been under t aken in recent years to devel op cw lasers with long lifetimes at r oom t emperat ure. Today lifetimes great er t han 25,000 h have been achi eved [7.6, 7]. Li fet i me measur ement s at elevated- t em- perat ures yield ext r apol at ed r oom t emper at ur e lifetimes in excess of 100 years [7.8]. Several review papers on semi conduct or injection lasers have been written [7. 9-13]. In this paper we shall discuss the modul at i on and spectral propert i es of doubl e- het er ost r uct ur e (DHS) st ri pe-geomet ry GaAs/ Gal _xAl . , As i nj ect i on lasers for cw r oom t emper at ur e oper at i on. The results on these devices may be consi dered to be . representative also for semi conduct or lasers based on ot her materials. Especially injection lasers for l onger wavelengths based on l nGaAs P/ I nP are very at t ract i ve [7.14 17]. The dynami c behavi or of these devices is very similar to t hat of GaAs lasers, as report ed in [7.18, 19]. " Expanded and updat ed t r eat ment based upon a paper publ i shed in Appl. Phys. [7.13, 214 G . A r n o l d et al . For high bit-rate communi cat i on from several 100 Mbi t / s up into the Gbi t / s range, lasers must have the following properties: i) No modul at i on distortions (pattern effects); ii) A narrow spectral bandwi dt h; iii) No high spectral broadeni ng due to direct modul at i on; iv) No self-pulsations. The rate equat i ons which describe the dynami c behavior of injection lasers are discussed in detail in Sect. 7.2. In the following sections we review and discuss the modul at i on, spectral and self-pulsation behavi or of injection lasers, and the met hods for direct modul at i on at high bit rates. 7.2 The Rate Equations Semi conduct or injection lasers exhibit a very complex dynami c behavior. Until now, not all experimentally observed effects can be explained satisfactorily ; but for the main properties, a good theoretical underst andi ng has been achieved. The dynami cs of a semi conduct or injection laser is governed by rate equat i ons [ 7. 2027] . The quant um-mechani cal rate equat i ons for the electron density and polarization operat ors and the phot on ampl i t ude operat ors give i nformat i on about the time devel opment of the phot on amplitude, frequency and phase and also about the statistical properties due to quant um fl uct uat i ons [7.20, 23, 24, 26]. In many cases, where onl y the time dependence of the phot on number mean value is of interest and an interaction of modes with a very narrow wavelength spacing does not occur, the analysis can be performed by the much simpler classical rate equat i ons for the electron density in the active layer and the phot on numbers in the modes [7.21, 22, 25, 27]. Due to the i nt raband scattering processes spectral hole burni ng is uni mpor t ant in injection lasers and the cause for the mainly observed mul t i mode operat i on of injection lasers is assumed to be spatial hole burni ng [7.28-31]. We give mul t i mode rate equat i ons for the electron density and phot on numbers. In the mul t i l node case the i nhomogeneous di st ri but i on of electron and phot on densities must be taken into account. For an active layer, parallel to the x - y plane with a thickness d, these rate equat i ons are ( ' ) n ( x , y , t ) J ( x , y , t ) l - eod ~ R ~ p ( n ) + D V 2 n ( x , y , t) (7.1) at 1 F i _ / ' i ~?S,(1) S i ( t ) + ~ ( l t p i ( x , y ) l Z r s p ( E i , n ) d x d y (7.2) r")l Z ph + q ) ( E i ) F i + ~ S~(t) ,f Iq~i(x, y)l 2 r, , ( E i, n) d x d y . Mo d u l a t i o n o f L a s e r Di o d e s 215 The el ect ron densi t y n depends on x and y and is assumed t o be confi ned within the active l ayer and const ant therein in z direction. J is the injection current density, e o the absol ut e value of the el ect ron charge. S i is the phot on number in the ith mode, ~p~(x,y) is the normal i zed scalar compl ex phot on ampl i t ude function. The compl ex ampl i t ude funct i on is normal i zed so t hat ]q~(x,y)] 2 i nt egrat ed over the whol e x - y pl ane yields 1. F~ is the phot on confi nement fact or which gives the rat i o of phot on energy concent r at ed within the acti'~e l ayer of vol ume V to the t ot al phot on energy, bot h for the ith mode [7.13]. The stimulated and spont aneous emission rates per unit of vol ume and unit of phot on energy r~, and r~p have been calculated by Las her and St e r n [7.32] for GaAs at r oom t emperat ure, assumi ng t ransi t i ons between par abol i c bands with k selection rule for pur e mat eri al and wi t hout k selection rule for highly doped material. For parabol i c bands wi t hout k selection rule, and addi t i onal Gaussi an i mpuri t y band tails cal cul at i ons have been performed by Case); and St er n [7.33]. For t ransi t i ons between parabol i c bands wi t hout I< selection rule an appr oxi mat e expressi on for r~t has been given by Mar i nel l i [7.34]. R~p is t he t ot al spont aneous emission rat e per unit of volume, and is the i nt ernal quant um efficiency. 4~(E~) is the number of modes per unit of vol ume and unit of energy and is given by --~--2 2 ( E i ) - - ni ni Ei ( 7 . 3 ) g 2 h 3 c 3 ' where ~i is the index of refract i on for the ith mode, and the effective index of refract i on ii' i consi deri ng the dispersion is given by ~ i i = / / ~ ( 1 - f i ~ d2~]" ( 7 . 4 ) For a vacuum wavel engt h of 8500 ,~ we obt ai n E; = 1.46 eV and with ~ = 3.6 and ~i'~ 5 I-7.35] the number of modes per unit of vol ume and unit of energy is q~( Ei ) =l , 817xl 0~2me V- ~c m -3. D is the diffusion const ant and can be calculated from D = i ~, kT/ e o (where/~n is t he el ect ron mobility, k the Bol t zmann const ant and T the absol ut e t emper at ur e) or from D = L~/Z~p (where r.~p is the spont aneous el ect ron lifetime and L n the el ect ron diffusion length). Fr om / ~o =3 0 0 0 c m2 V- l s -~ for GaAs and k T / e o = 2 6 m V the el ect ron diffusion const ant D, = 78 cm 2 s - ~ for p- doped GaAs is calculated. This value coincides well with the result calculated from L, = 5pm for Ge- doped p-t ype GaAs layers [7.36] and Z~p=3ns. In highly doped GaAs, the mi nor i t y carri er diffusion const ant can be smaller by mor e t han an or der of magni t ude [-7.28, 30]. The phot on lifetime I"ph i ill the ith mode is given by [7.25] ph, = ~i ~'i lnRi , (7.5) 216 G. Arnohl et al. where ~i is the i nt ernal opt i cal loss per unit length, R~ the reflectivity of the end mi rrors, bot h for the ith mode and L is the laser length. For sol ut i on gr own j unct i ons at r oom t emper at ur e the phot on lifetimes are bet ween 1 and 2ps [7.37]. The comput at i onal anal ysi s of the si ngl e- mode laser in ma ny cases yields consi derabl e insight i nt o the laser dynami cs. To some ext ent the results are al so appl i cabl e to the physi cal l y mor e realistic mul t i mode case. If onl y one mode oscillates, hol e bur ni ng effects can be neglected and the spat i al di st ri but i on of the el ect ron and phot on densities can be consi dered uni f or m within the act i ve region. In this case the s ummat i on in (7.1) over i has to be deleted and d - l J ( x , y ) [ 2 has to be subst i t ut ed by V 1 in (7.2) the i nt egral s over [ q) ( x, y) [ 2 have to be replaced by l. A furt her si mpl i fi cat i on of the mo n o mo d e rat e equat i ons is possi bl e by the following appr oxi mat i ons l Rsv(n)' ~ n/'c, p , (7.6) F c~ 1~7 q , ( E, ) r ~ v ( E~ , n ) , , ~ . (7.7) Ts p Measur ed s pont aneous el ect ron lifetimes r~p of DHS lasers at r oom t emper a- t ure are bet ween 2 and 8 ns [7.18, 38, 39]. The coefficient c~, defined by (7.6, 7) gives the rat i o of the s pont aneous emi ssi on rat e i nt o the osci l l at i ng mode to the t ot al s pont aneous el ect ron r ecombi nat i on rate. To give a r ough est i mat e of the ma xi mum val ue of e we set the i nt ernal qua nt um efficiency = 1 and R~v(n ) equal r~v(E~, n) times t he s pont aneous emi ssi on linewidth. For V / F = 10 -'~ cm 3, a s pont aneous emi ssi on linewidth of 300/~, and for the above cal cul at ed 4 ) ( E) = l . 8 1 7 x 1 0 1 2 me V - l c m - 3 we obt ai n c ~ = l . 0 7 x l 0 .s. On the basis of Mari nel l i ' s appr oxi mat i on formul a, A d a m s has pr oposed a si mi l ar appr oxi ma- t i on for the st i mul at ed emi ssi on coefficient F V4~(E~) r~t(E~, n ) = O n ' , (7.8) where l = 3 for Ga As doubl e- het er ost r uct ur e lasers at r oom t emper at ur e [7.27]. These appr oxi mat i ons yield the rat e equat i ons dn J n g S n I , (7.9) d t e o d r~p d S S n - + ~ V - - + V, q S n I . (7.10) dt "cph z~p Modulation of Laser Diodes 217 For 7 = 0 and J > = J t h the explicit sol ut i on of the st eady-st at e rate equat i ons is , , , , , "c~p \ V o z r , l-~ / ( J / J , h - 1), (7.12) where n m is the t hreshol d el ect ron densi t y at which the gain in the active region compensat es the r esonat or losses. The t hreshol d current densi t y Jth is given by J t h = e o d n t h / ' r ~v " (7.13) Fi gure 7.1 shows the experi ment al light out put vs cur r ent characteristics of t wo different injection lasers. Laser 1 shows a linear slope above threshold. The slope below t hreshol d results from the spont aneous elnission with a br oad spectral distribution. The nonl i neari t y in curve 2 wilt be discussed later. 7. 3 Di rect Modul ati on of Injection Lasers If a step cur r ent pulse of ampl i t ude I (the injection current 1 is given by the pr oduct of the injection current densi t y J and the j unct i on area) is appl i ed to t he laser, an initial del ay time t d passes until the onset of the laser oscillations and t hen the coherent emission starts with rel axat i on oscillations in the out put power [7.40, 41]. The t ransi ent sol ut i on of t he rate equat i ons has been given by several aut hor s [7.27, 42-45]. Fi gure 7.2a shows the t ransi ent response of a monomode injection laser to a step current pulse. We use the normal i zed el ect ron density z = n / n , , , the normal i zed phot on number x = S r , r , / V n t h Z r , h and the normal i zed injection current r / = l / I t h = J / J t h " The laser par amet er s are z~p/rp~, = 1 0 3, 1= 3, C~=2 X 10 -5. The normal i zed step cur r ent pulse ampl i t ude is 11= 1.1. If a step cur r ent pulse of ampl i t ude 1 is appl i ed t o an initially unbiased injection lascr, the el ect ron density in the active l ayer increases. As long as n is well bel ow nth no consi derabl e ampl i fi cat i on of the spont aneousl y emi t t ed phot ons takes place. After the initial delay time [7.40, 41] ta = z,p In [ I / ( I - Ith)] (7.14) t he el ect ron density in t he active layer reaches its t hreshol d value and the phot on number rises fast. As long as the phot on number is below its st at i onar y value, the el ect ron densi t y furt her increases above n,w When S passes its st at i onar y value, due to the rapi dl y increasing st i mul at ed r ecombi nat i on processes, the el ect ron density qui ckl y decreases but the phot on number furt her 218 G. Arnold et al. 10 E 0~ T o o 2 100 200 300 l, Injection current [mA] Fig. 7.1. Experimental light output vs cur- rent characteristics for two different injection lasers F i g . 7.2a-c. Transient response of an in- jection laser to a step current pulse for different spont aneous emission contributions 0.8 1 a) i 0 , L 0,2 04-0 x ~ ct = 2.10 -s b) o 8 1 U t 0 , 2 0 0 c) N Z ~=10 3 t o . t o 2 t OJ-O ' ' 0 ~ r 0 5 t00 150 Fig. 7.2 -- t / ~ increases until n passes again n,,. The fast decrease of the electron density cont i nues until the phot on number falls below its st at i onary value. If the ampl i t ude of the first phot on spike is much higher t han the st at i onary phot on number, a considerable decrease of n below n~h duri ng the decrease of the light pulse takes place. Since the electron densi t y is slowly raised again the phot on number may decrease by several orders of magni t ude until n has again reached nth. Afterwards the whole process is repeated, but since the phot on number in the considered mode now is higher t han at ~he beginning of the process when n passed n,,, the phot on number reaches its equilibrium value at a shorter time. Therefore the overshoot in the electron density and the following Modulation of Laser Diodes 219 re" W~ WLJJ r r r y ~LU LU klJ e_ 212 tD 2ns o) 5ns 2ns b) C) Fig. 7.3. Current pulse (upper trace) and corresponding light pulse (lower trace) ofa GaAs/GaAIAs DHS stripe-geometry laser operated with double pulses with different pulse spacing (a) 2 ns, (h) 5 ns, and (c) 2 ns with pulses of forward and reverse swings [7.48] phot on number over s hoot are smal l er t han before. The process is repeat ed until the st at i onar y st at e is reached. For pulse modul at i on appl i cat i ons of injection lasers the del ay t i me can easily be reduced by pr ebi asi ng t he l aser wi t h a dc c ur r e nt / o [7.44, 46]. When a cur r ent pulse with ampl i t ude l p is super i mposed the del ay t i me is t d = T~v I n [ I v / ( l P - I m + I o) ] . (7.15) I f the laser is biased up to t hreshol d, t d vanishes. I f the laser is unbi ased or bi ased bel ow t hr eshol d and modul at ed with t wo subsequent pulses the del ay t i me for the second pulse is reduced [7.40] since the el ect ron densi t y aft er the first current pulse is hi gher t han before. In the case of di rect pulse code modul at i on this woul d cause a pat t er n effect. Oz e k i and I t o suggest ed the modul at i on of the injection laser by an addi t i onal compens at i on pulse before each modul at i on pulse which is preceded by a logical zero [7.47]. Thi s compens at i on pulse is t oo small to generat e a light pulse but raises the el ect ron densi t y up to the same level as a f or egoi ng modul at i on pulse woul d have done. Two si mi l ar met hods for pat t er n effect reduct i on have been shown by L e e and De r o s i e r [7.48]. In the first case, the modul at i on pulse ampl i t ude is dependent on whet her a modul at i on pulse preceded or not. In the second case the modul at i on signal consists of doubl e pulses with a f or war d and a reverse swing (Fig. 7.3). The first f or war d swing causes the light pulse, whereas the second negat i ve swing r emoves the excess char ge in the act i ve region. For prebi asi ng near t hr eshol d a pulse spaci ng of 2 ns is achi eved wi t hout pat t er n effect. 220 G. Arnold et al. Wi t h increasing cont r i but i on of spont aneous emission into t he oscillating modes, the dampi ng of the rel axat i on oscillations is raised since the initial phot on number in t hat case is higher, the st at i onar y phot on number is reached earlier and the over shoot consequent l y is smaller [7.43, 45, 49]. In the mul t i mode case the relative cont r i but i on of spont aneous emission i nt o the oscillating modes is pr opor t i onal to the number of oscillating modes when the monomode cal cul at i on is t aken as represent at i ve for the phot on number in all oscillating modes. Fi gure 7.2b, c shows cal cul at i ons of t he t ransi ent behavi or wi t h ~ = 1 0 - 4 and c~=10 -3. Angerst ei n and Si emsen deduced from measure- ment s on DHS st ri pe-geomet ry injection lasers ~ as high as 5 10 - 4 [7.39]. Values of 10 3 <: ~<1 0 - 2 fitted to measurement s on buried het er ost r uct ur e lasers cannot be expl ai ned by such a st rong spont aneous emission [7.50]. Pulse code modul at i on of injection lasers has been achieved up to mor e t han 2 Gbi t / s [7. 51-56]. Figure 7.4 shows the direct modul at i on of an injection laser at 2.3 Gbi t / s. Fi gure 7.5 shows the result of a 280 Mbi t / s modul at i on experi ment with a low mesa-stripe DHS injection laser [7.56]. When the laser is biased below t hreshol d the light out put shows a st rong pat t ern effect. If the laser is biased 5% above t hreshol d the pat t ern effect vanishes but the laser out put exhibits a st rong ringing, since every modul at i on pulse causes rel axat i on oscillations. We have seen t hat in the case of biasing below t hreshol d pat t er n effects arise from the el ect ron density dependence on a precedi ng pulse. In the case of biasing the laser above t hreshol d the optical out put pulse not onl y depends on the initial electron densi t y in the active l ayer but also on the initial phot on number in the oscillating modes. Dani el sen has suggested avoi di ng the pat t ern effects in Gbi t / s PCM by biasing the laser appr oxi mat el y to t hreshol d and giving the height and dur at i on of the appl i ed current pulses such values t hat the laser onl y emits the first spike of the rel axat i on oscillations and the el ect ron and phot on densities at t he end of the cur r ent pulse ret urn t o t hei r initial values [7.57]. In many experiments, we have seen t hat in the case of pulse code modul at i on above 250 Mbit/s, when the laser is biased near t hreshol d an exact adj ust ment of the bias cur r ent and the modul at i on ampl i t ude is necessary. This can be well expl ai ned by the 1,heory of Danielsen. Sinusoidal modul at i on of injection lasers above t hreshol d is a powerful tool for investigating the dynami c propert i es of injection lasers. Small-signal analysis yields a resonance in the modul at i on dept hs vs modul at i on frequency curve [7.27, 58 60]. If I~ is the compl ex modul at i on current anaplitude and S~ t he compl ex phot on number ampl i t ude, bot h at the angul ar frequency to, and if the laser is biased to I 0 and S 0, respectively, we obt ai n by small-signal analysis of the rate equat i ons (7.9, 10) for ~ = 0 11/lo S , / S o = (l/z,pZph) e) 2 +j o o~1-- ~O 2 (7.16) with (D2/Tsp"f ph = l ( l / l t h - 1) (7.17) Modulation of Laser Diodes 221 < 2 6' 1 >, ~a 1,25-" 1 . 0 0 ' J i I I I I I I I r I I , o , b o , ~ , o 6 , I ' , o o i I I I i i I 0 I 2 3 4 5 6 TIME (ns) - o ) Light @ 1o=1,05 Ith 11=45mA It =45mA ~ , [0 : O,g5 [ t h I~ = 45mA d) 2_SyLZ u n L L I I k 2ns/div. Ith=2OOmA Fig. 7.4. Direct modulation of an injection laser at 2.3 Gbit/s Fig. 7.5a~1. Direct modulation of an injection laser at 280 Mbit/s and different bias currents and f l ~ p = I ( l / l , h - 1) + 1. (7.18) Small-signal calculations including the effect of spontaneous emission have been performed by H a r t h and S i e m s e n [7.49]. Figure 7.6 shows the dependence 222 G. Arnold el al. ( z : 2 . I 0 ~ - % . , w . j I 0 - ~ 1 0 l o l L / [ t 0 , 0 1 I I X X ~ . I 1 0 . 2 1 0 - 1 1 1 0 I . - LO ' ~ ' ~ ' ~ s Fig. 7.6. Smal l -si gnal modul at i on dept hs dependence on modul at i on frequency for bi asi ng 10% ( cont i nuous curve) and 20% (broken curve) above t hr eshol d for z ~ , / z p h - l 0 B and di fferent :~ of modul at i on dept hs on the modul at i on frequency for biasing 10 and 20%, respectively, above t hreshol d calculated from (7.9, 10) for l = 3 and e = 2 x 10- 5, 10 4, 10 B, 10-2. The last t wo values o r e are possible onl y in the case of many excited modes. The results of the small-signal analysis suggest t hat a biasing of the injection laser much above t hreshol d woul d be preferabl e for high- frequency modul at i on. Unf or t unat el y, the laser t here exhibits i rregul ar pheno- mena which we shall discuss in Sects. 7.5,6. Fur t her mor e, for opt i cal com- muni cat i on appl i cat i ons a high dc light power level woul d also raise the shot noise in t he receiver. The use of mul t i mode lasers with a hi gher e is in cont r adi ct i on with the nar r ow bandwi dt h requi rement s for low fiber dispersion. When the laser is modul at ed by large sinusoidal current s the resonance frequency call be reduced consi derabl y and the light out put signal becomes di st ort ed [7.61, 62]. With l = l and ~ = 0 in (7.9, 10) accordi ng t o [7.62] the anal yt i cal expression for the time dependence of the phot on density is So S( t ) = ~ exp [a cos(sot + 0)], (7.19) where i o(a) is the modified Bessel function of the first ki nd and of or der zero, and a is an ampl i t ude fact or depicted in Fig. 7.7. At the large-signal resonance frequency where a >> 1 from (7.19) a pulse-like shape of the optical out put signal arises (Fig. 7.8). We emphasi ze t hat this st rong nonl i neari t y occurs al t hough a compl et el y linear dc light out put vs current charact eri st i c is assumed. This imposes restrictions on the appl i cat i on of ampl i t ude modul at i on for high 10 cl Modul a t i on of Las er Di ode s 223 ["~ Fi g. 7. 7. Ampl i t ude fact or a as a f unct i on of t he nor ma l - I I I 5 4 ~2 1 o L 0 # 2 r t out I uJ/ uJ 0 I I " " ~ I 100 80 c" -~ 6o c . o ~ 40 o ~r ~ - - - - I 2o 3 ,rr 4~q Fig. 7,8. Nor ma l i z e d phot on dens i t y as a f unc- t i on of t i me I-7.62] I ! [ m A t 10 ~ 2 0 / . 0 O L l t 0 , 9 1.0 1.1 1.2 10/ l ~h F i g . 7 . 9 . M o d u l a t i o n d e p t h as a f u n c t i o n o f t he bi as c ur r e nt 1~ for 280 MHz si nusoi da] mo d u l a t i o n at di fferent mo d u l a t i o n ampl i - t udes It bandwi dt h fiber communi cat i on systems. The ampl i t ude fact or a has a maxi mum at t hreshol d for which it yields a maxi mum modul at i on depth. Our experi ment al investigations of t he large-signal modul at i on dept h as a funct i on of the normal i zed bias current l o/ l t h confi rm these t heoret i cal results (Fig. 7.9). By generat i ng spikes with a small sinusoidal cur r ent at t he modul at i on resonance frequency and removi ng one or mor e light pulses by a shor t l oweri ng of t he bias, S c h i c k e t a n z has demonst r at ed pulse code modul at i on at 650 Mbi t / s [7.54]. Fur t her investigations of the nonl i near rate equat i ons have shown t hat by modul at i on at the doubl e r esonance frequency also subhar moni cs can be excited [7.63] and the small-signal modul at i on sensitivity can be increased by an addi t i onal large sinusoidal modul at i on cur r ent [7.64, 65]. 224 G. A r n o l d e t al. Z n - d i f f u s i o n o x i d e Z n - d 4 f u s i o n ~ mp l o n t e d r e g i o n s Z n - d d l u s i o n .................. p : G u A I As " . . . . . . . . . . . . . p - G a A t A s ~ p - G a A I A s " \ . . . . . .................... " ~ - - ' ~ . ? . ?.C,a.~,.I A s n - Go A l A s n - Go A l As n - Go A l A s n - Go A I A s n ~ Go A s n - G a A s n G a A s n - G o A s e l b) c ) d l Zn - d i f l u s ~ o n Z n - d i f f u s i o n ~ a A t A s p-GoAIAs . ~ z ~ p_C, aAta s ;/,J//,J)J//,.~>A P - ~xtlx n - G o A I A s ............ ~ / ; ) . . . . G A I A s ~ _ . _ ~ n G o A t A s / ........... n - S a A I A s n - O o A I A s n - G o A I A s n - G o A s n - O a A s n - Ga A s el fl g l h l ~- a c t i v e f o y e r Fi g. 7. 10a h. Se ve r a l l a s e r s t r u c t u r e s . {a) Ox i d e - s t r i p e l a s e r ; (b) Di f f u s e d - s t r i p e l a s e r ; (c) P r o t o n - i mp l a n t e d l a s e r : (d) V- g r o o v e l a s e r ; (e) L o w- me s a - s t r i p e l a s e r ; (f) C h a n n e l l e d - s u b s t r a t e - p l a n a r ( CSP) l a s e r ; (g) B u r i e d - h c t e r o s t r u c t u r e ( BH) l a s e r ; (h) T r a n s v c r s e - i u n c t i o n - s t r i p e ( TJ S) l a s e r 7. 4 Modul at i on Behavi or of Speci f i c Laser St ruct ures 111 the preceding section, the modul at i on behavior of injection lasers has been discussed in general wi t hout referring to special laser structures. In recent years a large number of proposals for specific laser structures has been made. We shall now direct our at t ent i on to the correlations between geomet ry and the dynami c properties of injection lasers. Some of these structures are shown in Fig. 7.10. It is convenient to relate the large variety of laser structures to the following subgroups which differ mai nl y in the mode-gui di ng mechanism and the mode vol ume of the lasing modes: i) Injection lasers with no built-in index waveguide. The lasing mode in these lasers is guided only by the gain profile due to the current injection from the stripe contact. Especially the oxide-stripe laser (Fig. 7.10a), the diffused- stripe laser (Fig. 7.10b) with shallow diffusion, the prot on-i mpl ant ed laser (Fig. 7.10c) and the V-groove laser [--7.66] (Fig. 7.10d) belong to t hat group. The gain- guiding mechani sm also predomi nat es in the low-mesa-stripe laser [7.67] (Fig. 7.10e), as long as a broad mesa structure with a stripe width in excess of about 15 lam is used. ii) Injection lasers with a built-in index waveguide. Improved laser charac- teristics are expected if the lasing mode is guided by a stable built-in waveguide. In order to have efficient waygui di ng the built-in waveguide should be sufficiently narrow so t hat a gain-guided mode with self-focusing properties [7.68] cannot develop. The CSP laser [--7.693 (Fig. 7.100 and the diffused-stripe Modulation of Laser Diodes 225 laser with deep diffusion [7.70] are examples, bel ongi ng to the second group. In addi t i on, the bur i ed- het er ost r uct ur e (BH) laser [7.71] (Fig. 7.10g) also exhibits a stable built-in waveguide. iii) Lasers with very small t ransverse di mensi ons of the active layer. Exampl es of this third gr oup are represent ed by t he BH laser as ment i oned above and by the t ransverse-j unct i on-st ri pe laser [7.72] (Fig. 7.10h). The laser groups ment i oned above differ in their modul at i on behavior. These differences are i nt r oduced on t he one hand by t he different rat i os ~ of t he spont aneous emission i nt o the lasing modes, as discussed in t he precedi ng section. On the ot her hand, diffusion processes within the active l ayer and dynami c i nt eract i ons with hi gher or der modes also play an i mpor t ant role in the modul at i on behavi or [7.30, 73, 74]. Let us begin the discussion by consi deri ng pl anar-st ri pe lasers accordi ng to the laser subgr oup (i). Since the wavegui de for the lasing modes is accompl i shed here mai nl y by the gain di st ri but i on as i nt r oduced by the injected carriers, any dynami c change of the carri er di st ri but i on also i nt roduces a change in the wavegui di ng properties. An accurat e descri pt i on of the modul at i on behavi or of pl anar-st ri pe lasers t herefore requires numeri cal cal cul at i ons [7.73]. If t he wi dt h of the injected carriers in such lasers exceeds the width of the fundament al mode field, def or mat i ons of the spatial gain profile (hole burning) may occur so t hat event ual l y a fi rst -order mode is creat ed duri ng the modul at i on. An exampl e of t hat kind will be discussed in the next section. I mpr oved modul at i on characteristics are t herefore obt ai ned, if t ransverse spatial hol e burni ng is avoi ded. This can be done by lowering the stripe wi dt h of lasers down to the or der of 2 to 3 lain [7.75]. Such a small stripe width can also be easily achieved when using the V-groove st ruct ure [7.66] accor di ng to Fig. 7.10d. The wi dt h of the fundament al lasing mode is then ira t he same or der as the width of the injected current or even larger. Ther ef or e hol e- bur ni ng effects are less pr obabl e to occur. In addi t i on, the diffusion length comes into the same or der of magni t ude as t he modal width. Diffusion effects then yield a reduct i on of t he rel axat i on oscillations as cal cul at ed in [7.74]. Experi ment al l y, very nar r ow stripe lasers t herefore show no rel axat i on oscillations [7.75], which is also caused by the larger value of t he spont aneous emission coefficient due to the small active vol ume accor di ng to (7.7). Diffusion processes yield a significant reduct i on of rel axat i on oscillations for lasers of the subgr oup (ii) if the width of the built-in wavegui de is in the or der of the diffusion length or even smaller. A reduct i on of rel axat i on oscillations especially occurs if the t ransverse cur r ent injection ext ends beyond the built-in wavegui de since t hen carriers from out si de may diffuse into the region of the oscillating mode yielding a reduct i on of rel axat i on oscillations [7.73]. Such a behavi or has been t heoret i cal l y cal cul at ed and is found t o be in good agreement with experi ment al observat i ons on CSP lasers [7.73]. In t he laser subgr oup (iii) the t ransverse wi dt h of t he active l ayer is reduced down to ! to 21am. For the case of lateral carri er confi nement and an active l ayer wi dt h consi derabl y smaller t han the diffusion length, the shape of the gain 226 G. Arnold et al. a) / 799 800 801 802 X[ nml 799 800 801 802 ;~ [ a m ] Fig. 7. 11a, b. Emi s s i on s pe c t r um of a l ow- mes a- st r i pe Ga As DHS i nj ect i on l aser 3 % a bove t hr es hol d {a) a nd cl ose to t hr e s hol d (b) profile is onl y weakl y influenced by diffusion processes. Suematsu et al. [7.74], and Chinone et al. [7.73] have cal cul at ed t hat , t herefore, no significant di ffusi on-i nduced reduct i on of the r esonance peak of the r el axat i on osci l l at i ons occurs in this case. The case of lateral carri er conf i nement in connect i on wi t h low st ri pe wi dt h is realized by the BH laser [7.71]. Due t o the small vol ume V of the act i ve l ayer (7.7) yields a large s pont aneous emi ssi on coefficient yielding a st r onger reduct i on of the rel axat i on oscillations. A nearl y flat frequency response for frequencies up to 2 GHz has been r epor t ed for BH lasers [7.71]. TJS lasers have al so a small vol ume of the act i ve l ayer but differ f r om the BH laser in t hat the carri ers are not strictly confined. TJS lasers w.ere shown to exhi bi t onl y a small pat t er n effect when modul at ed with a 400 Mbi t / s PCM signal [-7.76]. 7. 5 Effect of Modulation on Spectrum and Near Field The dc spect ral [7.10, 13, 35, 77] and near field [7.13, 78-80] behavi or of injection lasers have been widely discussed. We have i nvest i gat ed the influence of modul at i on on the emi ssi on pr oper t i es of injection lasers. Far bel ow t hreshol d, injection lasers exhibit a br oad s pont aneous emi ssi on spect r um ( appr oxi mat el y 300 ~) which is nar r owed wi t h i ncreasi ng , , urrent and exhi bi t s a mode st r uct ur e at and above t hreshol d (Fig. 7.11). The laser can oscillate in a single o1" in a number of t r ansver se modes and al so in one or mor e filaments. Modulation of Laser Diodes 227 The t endency to mor e filaments or to hi gher-order transverse modes increases with the stripe wi dt h of the active region and with the pumpi ng above threshold. The latter increase is due to t ransverse spatial hole bur ni ng I-7.30]. Each filament or t ransverse mode exhibits a l ongi t udi nal mode group. Since the gain spect rum of injection lasers has a br oad maxi mum, also very small l ongi t udi nal hol e burni ng shoul d yield a number of l ongi t udi nal modes within the same transverse mode family. The envel opes of the l ongi t udi nal mode gr oups can be different in wavel engt h and intensity of the maxi mum. For the l ongi t udi nal modes the wavelength separat i on A2 of adj acent modes can be calculated from [7.35] A2 - )2 (7.20) 2 L n l " For a typical laser length of 200-400 lain the l ongi t udi nal modes are separat ed by 1.5-3/~. The t ransverse modes al ong the j unct i on plane, however, have a separat i on in the or der of 0.1/~ I-7.10]. The spectral posi t i on of these lasing modes is very sensitive to t emper at ur e changes owi ng to t wo effects [7.13]. First, the band gap of a semi conduct or decreases with increasing t emperat ure. As a result the wavelength of the laser emission increases with t emperat ure. This t emper at ur e coefficient is appr oxi mat el y 2.5/~/K. The wavel engt h of an individual spectral mode has a t emper at ur e coefficient of appr oxi mat el y 0.4 A/ K because of the t emper at ur e dependence of the refractive index of the semi conduct or [7.81]. To observe nar r ow spectral lines a careful t emper at ur e cont r ol of the laser is necessary. The t emper at ur e effects can influence the emission spect rum if the t emper at ur e of the j unct i on region of the laser is increased in the course of the modul at i on pulse durat i on. Many st ri pe-geomet ry DHS GaAs / GaAI As lasers show nonlinearities - the so-called kinks in the light out put vs cur r ent charact eri st i cs [7.80, 82, 83]. Laser 2 in Fig. 7.1 shows a typical kink in the light out put charact eri st i c of a l ow-mesa-st ri pe DHS injection laser with stripe wi dt h 25 gin. It has been found t hat these kinks are associated with fi l ament ary st ruct ures of the near-field intensity di st ri but i on parallel to the j unct i on plane, spatial movement of the filaments within the stripe width, and excessive spectral br oadeni ng of the emission spect rum. Mor eover , t he l ocat i on and char act er of the kinks can change duri ng l ongt i me oper at i on of lasers. Recently, Di x o n et al. [7.83] r epor t ed an i mproved linear light out put vs cur r ent charact eri st i c by reduci ng the excited stripe width of the active region of the DHS GaAIAs lasers below 101.tm, whereas al most all lasers with a stripe wi dt h rangi ng from 10- 20gi n show kinks [7.84]. In nar r ow stripe lasers the ki nks move to a hi gher current range and may occur again at higher light out put power. Experi ment al and t heoret i cal investigations show t hat hi gh-frequency di rect modul at i on causes an intensity decrease of domi nant modes while the number of nei ghbori ng l ongi t udi nal modes is increased. Consequent l y, the spectral envel ope is br oadened [7.85 88]. When t he injection cur r ent is 2 2 8 G. A r n o l d e t a l . 0.8 0.7 0,6 i 0.5 0/ , 0.3 0 2 03 0 I iif ~ \ " ~ ~ / ~\ I I / ~ % \1 I I t , , I . . . . . . . . . . . . . . . . . I I I I I I ' , / , , / " / / / / ' / i I 0 20 30 40 I h t "z' V "~ s p ' ~ p h ' 5 0 F i g . 7 . 1 2 . R e s p o n s e t o a m o d u l a t i o n w i t h r e c t a n g u l a r c u r r e n l p u l s e s : ( . . . . ) i n j e c t i o n c u r r e n t ; ( ) p h o t o n d e u s i t y ; ( ) s p e c t r a l h a l f w i d t h modul at ed, the electron density also oscillates with the modul at i on frequency. The electron density modul at i on ampl i t ude increases with the modul at i on current ampl i t ude and also with the rnodulation frequency. If the electron density oscillates there are periods when modes with a higher threshold electron density may have a net gain and the number of oscillating modes is increased. Figure 7.12 shows a theoretical calculation of the phot on response and the t emporal development of the spectral emission halfwidth by use of the mul t i mode rate equations, if the laser is operat ed with rectangular pulses which change the injection current from 3.5% above threshold to 403, above threshold [7.88]. The spectral halfwidth A is normalized with respect to the mode spacing A2. Though the st at i onary spectral width in this specific example at 3.53, above threshold is onl y A= 1.6A2 and decreases with increasing current, the spectral width duri ng modul at i on becomes much broader. The spectral broadeni ng j ust occurs duri ng the time interval of small phot on density. As soon as the phot on density increases the spectral width decreases onl y slowly so t hat a relatively broad spectrum is mai nt ai ned duri ng the light pulses. Modulation of Laser Diodes 229 Figure 7.13 shows the effect of 300 Mbi t / s pulse code modul at i on on the emission spectrum of a GaAs low-mesa-stripe DHS injection laser. The quasi single-mode emission of the laser wi t hout modul at i on changes with increasing modul at i on current to a mul t i mode emission. Besides the increase of the number of l ongi t udi nal modes, a new mode family appears. For the i nt erpret at i on of the spectral change, the modul at i on influenced emission spectrum has been investigated spatially and time resolved. Figure 7.14a shows the integral intensity di st ri but i on of the near field al ong the j unct i on plane and Fig. 7.14b the correspondi ng monochr omat i c near-field distributions of neighboring modes belonging to different longitudinal mode families. The near-field distribution A belongs to the fundament al transverse mode and the intensity di st ri but i on B/C exhibits the first-order transverse mode. The spatially resolved emission spectra at the near-field positions A, B, and C are shown in Fig. 7.15 t oget her with the integral emission spectrum. As can be seen, the fundament al transverse mode (A) yields a quasi single-mode emission spectrum which is essentially the same as the spectrum wi t hout modul at i on (Fig. 7.13). At the positions B and C of the first-order transverse mode an identical spectrum is measured, but it is different from the spectrum at position A. The peak wavelength is 4.5/~ shorter t han in the spectrum of the fundament al transverse mode. The lower wavelength for the first-order mode is in agreement with the measurement s of Buus et al. [7.30] and in cont radi ct i on to their theoretical considerations. The wavelength shift can probabl y bc explained by higher bandfilling caused by the modul at i on current pulses. In order to investigate the dynami c behavi or of the emission of the modul at ed laser, the time resolved light pulses have been measured. In Fig. 7.16 the light pulses of the five figures "1" of the word 1000110001111100 at 300 Mbit/s are shown for the same modul at i on condi t i ons as for the spatially resolved spectra. Besides the light pulses for the overall intensity di st ri but i on also the light pulses for the fundament al (A) and first-order (B, C) transverse modes are measured. The light pulses show spikes caused by relaxation oscillations. The first spike of the relaxation oscillations is lasing in the fundament al mode (A) in agreement with the results of [7.30]. The first-order transverse mode (B,C) is excited later, as can be seen from Fig. 7.16. Measurement s of the time resolved spectra have yielded essentially the quasi single-mode emission spectrum of Fig. 7.13 in the first spike of the light pulse and the modul at i on induced spect rum with the short er peak wavelength exhibited by the first-order mode is observed in the following spikes of the light pulse. The results of these investigations show t hat for certain modul at i on condi t i ons it may be possible to retain the emission spectrum unchanged in spite of direct modul at i on of the laser, provided t hat the modul at i ou pulses are very short or the light pulse is suppressed after the first spike of the relaxation oscillation. By appropri at e choice of pulse ampl i t ude and durat i on direct Gbi t / s modul at i on free of pattern effects and spectral broadeni ng is possible [7.89, 90]. However, the need of an accurat e bias and pulse ampl i t ude control would complicate technical applications of this method. 2 3 0 G. Arm)ld e l a l . J c <:; [ s j ! u n q J D ] /~)~ISUO:IU 4 co ~. ~ . ~ _ ~ ~ o o ~ N Z ~ ~ o . ~ ~ . ~ ~ . ~ .~_~ .= ~ E J I [ S ~ l U rl ~ J D J ~ ! q J D ~ 1 1 5 u a l u l 4 < o o ~ o c - . o ~ ~ j j ~ . - = = ~
2ns/ di v Modulation of Laser Diodes 231
Fig. 7.16. Time resolved light outpul of the laser operated as in Fig. 7.14. Light pulses of the five "'1"" of lhe word 1000110001111100. @ Light pulses in tile fundamental transverse mode, and @, light pulses in the first-order mode 2 t . . . . . . . . Fig. 7.17. Double section injection laser 7.6 Self-Pulsing Phenomena F r o m (7.17, 18) can be seen t hat t he mo d u l a t i o n b a n d wi d t h of i nj ect i on l asers i ncr eases wi t h bi as level. Unf or t una t e l y, i nj ect i on l asers t end t o exhi bi t s t a t i ona r y sel f - pul sat i ons when t hey ar e dc bi ased t o mo r e t ha n a few per cent a b o v e t hr es hol d [7. 85, 91, 93, 95]. The f r equency of t hese sel f - pul sat i ons has a r ange f r om 0.1 t o a few GHz a,~d coi nci des wi t h t he mo d u l a t i o n r es onance f r equency f or si nusoi dal modul a t i on. As t he mo d u l a t i o n dept hs can t ake val ues up t o 80% [ 7. 96] , t hi s effect is a severe l i mi t at i on for hi gh- f r equency di r ect mo d u l a t i o n of i nj ect i on lasers. Several t heor et i cal model s have been pr opos e d f or a physi cal unde r s t a ndi ng of t he sel f - pul sat i ons. B a s o v et al. [7.91, 97] s uppos e d spat i al i nhomoge ne i t i e s t o be t he r eas on f or t he sel f - pul sat i ons. The i nfl uence of spat i al i nhomogenei t i es oil t he l aser dyna mi c s has been t r eat ed by tile si mpl e t heor et i cal mode l of a l aser di vi ded i nt o Sects. 1 and 2 (Fig. 7.17). The gai n in t he t wo sect i ons can be cont r ol l ed 232 G. Arnold el al. separately by the injection currents 1~ and I z. Both active regions are within the same Fabr y- Per oI resonator. To achieve laser action it is sufficient t hat onl y one of the sections exhibits optical gain. In the diode which is onl y biased in the Iossy region a phot on field in the r esonat or raises the electron level, whereas in the diode, which exhibits optical gain, the electron density is lowered by a phot on field. Now if the absorbing section saturates faster t han the amplifying one, there exists a region of phot on number where the net optical amplification in the Fabr y- Per ot resonat or increases with the phot on number and the steady-state solution of the laser rate equat i ons becomes unstable. Ifr~t a and r~, 2 are the stimulated emission coefficients in Sects. 1 and 2, and 7 is the ratio of the volume of Sect. 2 to t hat of Sect. 1, the condi t i on of instability is ~ # ' s l 1 ~ . 0r s t 2 ,.,,, + < o . (7.21) By this model possible pulsations can be explained in all laser structures where the phot on field can couple to lasing semi conduct or regions with a band gap smaller t han the phot on energy. Kobayashi has shown experimentally and theoretically t hat st at i onary pulsations can arise when two parallel lasers are optically coupled I-7.98]. In the same way, self-pulsations could also be caused by the coupling of two filaments or two mode groups with different lransverse mode structures. The t heory of Kobayashi is on the basis of the rate equat i ons and includes no mode locking phenomena, where the pulsation frequency is related to the difference of oscillating frequencies of interacting modes. Or di nar y mode locking would yield spiking frequencies in the order of 101 ~ Hz. Therefore second-order mode locking has been proposed for injection lasers [7.85, 92, 99]. Al t hough [7.100] brings experimental evidence for second-order mode locking, the measured pulsation behavi or of lasers with a narrow emission spectrum [7.94] can rat her be explained by the t heory of Basov. We think t hat the Q switching, proposed by Basov is the main reason for spont aneous self-pulsing phenomena. These phenomena are surely influenced and possibly also enhanced by mode or filament interactions. We have measured the microwave spectrum of the st at i onary optical out put pulsations of dc operat ed low-mesa-stripe geometry DHSGa As injection lasers. Figure 7.18 shows the typical decrease of the rf bandwi dt h of the out put pulsations and the increase of the pulsation frequency and ampl i t ude with increasing injection current. Generally we observed t hat an abrupt change in the self-pulsation frequency is always accompani ed by a change in the filamentary structure of the near-field intensity di st ri but i on [7.95]. An emission spectrum wi t hout and with (Fig. 7.19) self-pulsations is shown for the same laser at different injection levels. Self-pulsations cause a broaden- ing of the spectral envelope of the longitudinal modes and also a broadeni ng of Modul at i on of Laser Di odes 233 b) o) "7 1,C z 0.5 1.0 >, , u l _c T / 1 ~ 1 ~ o 1 ; 5 / , o , , , h . ~ / ~ . . . "~. 1,13 A 1.0g / \ , - __ ,oo, ~ ~ 1.0 3 0,5 1 - - ~ f l O H z l Fig. 7.18. (a) Fr equency s pect r um of t he sel f-pul sat i ons of a cw oper at ed low- mesa- st r i pe GaAs / GaAI As laser for dif- ferent i nj ect i on cur r ent s; (b) fl uct uat i on frequency as a funct i on of t he nor mal i zed i nj ect i on cur r ent l o / I , h c o 35 c 806 808 i o) 810 806 P- X [nm] 808 810 Fi g. 7 . 1 9 a , b. Longi t udi nal - mode spect r a of a cw oper at ed l ow-mesa- st ri pe GaAs/ GaA1As laser: (a) laser oper at i on wi t hout sel f-pul sat i ons (1 o =2 1 4 mA) ; (b) laser oper at i on wi t h sel f-pul sat i ons (Io = 283 mA) 234 G. Arnold et al . the individual modes as can be obvi ousl y seen from tile intersections in Fig. 7.19. The linewidth widens by a fact or of two. The br oadeni ng of the spectral lines can be expl ai ned in the following way. Tile self-pulsations cause an oscillation of the el ect ron density and t herewi t h also an oscillation of the refractive index in the active region and a cor r espond- ing oscillation in the wavelength shift. In the time averaged measur ement this effect shows a spectral broadeni ng. In spont aneousl y pulsating injection lasers the spectral br oadeni ng of the emission spect rum is tile same as in the case of direct modul at ed injection lasers. A significant correl at i on between the occurrence of nonlinearities in the light out put vs cur r ent characteristics (so-called kinks) and the repetitive self- pul sat i on has also been observed [7.95]. A typical exampl e of t hat ki nd is shown in Fig. 7.20. Figure 7.20a shows the light out put vs current charact eri st i c of a laser, where the self-pulsations (Fig. 7.20b) j ust occur at the onset of the "ki nks" (shaded areas in Fig. 7.20a). It is i nt erest i ng to consi der t he related near-field di st ri but i on (Fig. 7.21). As l ong as the near field shifts to the right boundar y a self-pulsation occurs. For larger current s the near-field shift stops and tile stabilized near field also yields a nonpul si ng light out put . Experi ment s of t hat kind suggest t hat the repetitive self-pulsations are related to the t ransverse near-field movement i nt o possibly absor bi ng regions. Ther ef or e the most pr obabl e expl anat i on for these self-pulsations is a repetitive Q-switching process i nduced by sat urabl e absor bi ng regions, as ment i oned above. 7.7 Coupled Laser Structures Opt i cal l y coupl ed injection lasers exhibit very interesting modul at i on and spectral properties. Dependi ng on the st rengt h of coupl i ng quite different effects can be observed. St r ong optical coupl i ng is achieved when bot h laser systems are within the same Bragg r esonat or or very close by aligned end to end [7.21, 97, 101]. We have discussed such a system ill the precedi ng section. For i nhomogeneous exci t at i on there also exist points of oper at i on where such j unct i ons exhibit a bistable switching behavi or. Appl i cat i ons can be seen for pulse shapi ng and opt oel ect r oni c logic AND gates. Anot her promi si ng appl i cat i on of opt i cal l y coupl ed injection lasers is to i mprove the modul at i on performance of injection lasers. It has been shown t heoret i cal l y and experimentally, t hat the modul at i on bandwi dt h of an in- j ect i on laser can be increased by i nj ect i on of a coher ent light signal i nt o one of its oscillating modes [7. 102-109]. In sect i on 7.3, we have shown t hat a high initial phot on number in the oscillating modes causes a st rong reduct i on of the rel axat i on oscillations. By injection of a coher ent light signal into one oscillating mode of an injection laser, this st rong reduct i on occurs wi t hout 6,0 ~,,0 ~t [rnW] 2,0-
2,0 1,5 T I , O - flGHzl 05- I / / , ! / 1 J / / : " i I [ [ r e A l - I i I : ' ~ I I I ] [ m A } F i g . 7 . 2 ( l a , b . L i g h t o u t p u t v s c u r r e n t , a n d m i c r o - wave f r equency vs cur r ent char act er i s t i c of a pr ot on- i mpl a nt e d l aser Modulation of Laser Diodes 235 ~ N e o r f i e [ d I lp n 1 / A X , . z 6 s J / / / \ \~.__ z s 0 ~ / / X~_ - 21s ~ / A ~- - - 2 0 0 - - - '- L ' ~ t F i g . 7 . 2 1 . N e a r - f i e l d d i s t r i b u t i o n a l o n g t he j unc t i on pl ane mul t i mode operation. Figure 7.22 shows the experimentally observed optical response of an injection laser to a step current pulse [7.108]. The comput er si mul at i on of a 2 Gbi t / s direct pulse code modul at i on of a mul t i mode injection laser has shown t hat by coherent light injection not onl y the pattern effects are el i mi nat ed but also the noni rradi at ed modes are suppressed [7.104] (Fig. 7.23). The suppression of the noni rradi at ed modes results from the reduction of the electron density and the associated gain caused by the light injection. Practically, coherent light injection at the center wavelength of a mode is impossible and there is always a det uni ng between the wavelength of the injected radi at i on and the wavelength of the free-running laser mode. The locking range for synchroni zat i on to the wavelength of the injected radi at i on is proport i onal to the ampl i t ude ratio of injected radi at i on and radi at i on produced in the laser and inversely proport i onal to the laser length [7.107 109]. For an ampl i t ude ratio of 10 -2 a locking range of more t han 0.1 A can be achieved [7.108, 109]. If two lasers with different l ongi t udi nal mode spacing are coupled, by appropri at e choice of the laser paramet ers there always exists one or more pairs of modes with sufficiently close wavelengths to insure locking [7.107, 110]. Figure 7.24 shows the small-signal modul at i on depths vs modul at i ol l frequency characteristics with and wi t hout coherent light injection [7.109]. The 236 G. Arnold e t al. a I----4 ins Fi g. 7. 22a, b. Re s p o n s e o f a n i n j e c t i o n l a s e r t o a s t e p c u r r e n t p u l s e wi t h a n a mp l i t u d e a p p r o x i ma t e l y 3% a b o v e t h r e s h o l d wi t h - o u t (a) a n d wi t h (b) c o h e r e n t l i ght i n j e c t i o n 2 0 without irradiation $1 l 0/. 0 I 0.2 $2,2LIS~ wilh i rradi al i on 0 t [ n s l Fi g. 7. 23. N o r ma l i z e d p h o t o n n u m b e r s xi=Siz~p / Vn,Jph a n d n o r ma l i z e d mo d u l a t i o n c u r r e n t q = I / l,h a s a f u n c t i o n o f t i me wi t h o u t a n d wi t h c o h e r e n t l i ght i n j e c t i o n calculations with coherent light injection have been performed for different detuning A2 between the wavelength of the injected radiation and the free- running wavelength of the laser mode. When the detuning is smaller than the locking range, the resonance in the modulation characteristics vanishes. Figure 7.25 shows the large-signal response to a step pulse which changes I / I t h from 0.9 to 1.1 at the time t =0 [7.109]. With light injection the response is strongly damped within the whole locking range and exhibits no spiking response. For a detuning larger than the stationary locking range the frequency locking breaks down and the laser produces strong spiking oscillations. The frequency of these spiking oscillations is identical with the difference between the frequencies of the free-running mode and the incident radiation. This oscillation, induced by coherent light injection has also been experimentally observed I-7.111]. 7.8 Outlook Although stripe-geometry DHS GaAs injection lasers under certain experimen- tal conditions can be modulated up into the GHz region there are still problems I 0 T 0,1 0.01 I I 0 2 ~ w ith o u l . / ~ irradiation ~ ~ A X / A X ~ o ~ = 0 ~ ~ ~Wo.6 0.8 i i I 0 q 1 10 mB W I W 0 Mo d u l a t i o n o f La s e r Di o d e s 237 Fi g. 7. 24. S ma l l - s i g n a l m o d u l a t i o n d e p t h s vs mo d u h l t i o n f r e q u e n c y c h a r a c t e r i s t i c wi t h a n d wi t h o u t c o h e r e n t i r r a d i a t i o n f or di f f e r e nt r a l i o s o f t h e wa v e l e n g t h d e t u n i n g r a n g e A2 t o t h e wa v e l e n g t h l o c k i n g r a n g e A2 . . . . ( z ~ o / "r,,i, - - 10 3, 1 = 3 , I / o = I . l ) Fi g. 7. 25. La r g e - s i g n a l r e s p o n s e l o a s l e p c u r r e n t pul s e wi t h c o h e r e n t i r r a d i a t i o n , f2 is t h e a n g u l a r f r e q u e n c y di f f e r e nc e b e t we e n t h e f r cc r ui a ni ng mo d e a n d t h e i n j e c t i o n r a d i a t i o n . T h e l i g h t i n j e c t i o n i n t e n s i t y c o r r e s p o n d s t o a m a x i m u m l o c k i n g r a n g e Oz o l , =0 . 0 5 (Z~r,/ Tph - - l 0 3, l - - 3) 0 . 1 5 i 0.1 0,05 0 i'o 2 ; 3'o 4'o 5'o 6 o 7 6 8 ; ; o ~do ~+ t / ~ "cnh "c+~ to be solved for practical application. The self-pulsations can possibly be eliminated when laser structures can be developed where the optical field does not reach into absorbing regions with a band gap smaller than the photon energy. Since the multimode emission structure is caused by spatial hole burning and the hole burning is intensified by a low minority carrier mobility, undoped active layers with a very good crystal perfection should be used. Another possibility to obtain a stable longitudinal single-mode output is the distributed feedback injection laser [7.112]. Theoretical investigations yield for distributed feedback injection lasers the same modulation behavior as for Fabry-Perot type monomode lasers [7.113]. Single longitudinal mode emission has also been reported for transverse- junction-stripe lasers [7.114, 72]. 238 G. Arnold et al. Coher ent light injection will result in a consi derabl e i mpr ovement in the hi gh-frequency modul at i on behavi or. Possibly tile fut ure devel opment in i nt egrat ed optics [7.115] will make monol i t hi c i nt egrat ed coupl ed laser struc- tures feasible for technical applications. Injection lasers which show a stable fundament al mode lasing oper at i on are becomi ng mor e and mor e at t ract i ve as anal ogue t ransmi t t ers in opt i cal communi cat i on systems. Lasers exhibiting a good linearity between the light out put and the injection current exhi bi t onl y little hi gher-order har moni c di st ort i on. BH lasers (see Fig. 7.10g) show a second- or der har moni c di st or t i on of as low as about - 5 0 dB, as r epor t ed in [7.116]. St r i pe- geomet r y lasers with very nar r ow stripes as, for exampl e, t he V-groove laser (Fig. 7.10d) may also exhi bi t low har moni c di st ort i ons. In our l abor at or y, pr el i mi nar y mea- surement s showed a second- or der di st or t i on of - 5 0 d B with a modul at i on of 1 mW peak- t o- power light power, coupl ed into a fiber, for modul at i on frequencies up to about 100 MHz. Today Ga As DHS injection lasers with stripe geomet r y are tile most devel oped fiber optical transmitters. Increasing effort is under t aken in devel op- ment of laser materials emitting at wavel engt hs ar ound 1.2 to 1.3 gm since the fiber mat eri al dispersion vanishes t here [7.117] (see Chap. 2). In this case, the spectral emission ba ndwi dt h of the laser does influence the t ransmi ssi on bandwi dt h of the optical fiber channel s onl y by a second- or der effect [7.118]. With l nP/ Gal nAs P injection lasers of 1.1 t o 1. 3pm emission wavel engt h cw lifetimes of 3500hrs have been report ed [-7.14, 15]. Acknowledgment. The a ut hor s ar e i ndebt ed t o F.-J. Berlec for as s i s t ance in per f or mi ng t he me a s u r e me n t s and to H. Go t l s ma n n , C. W/51k, and P. Mar s chal l for t he f abr i cat i on of t he l asers i nvest i gat ed. Thi s wor k has been s pons or e d by t he " Bunde s mi ni s t e r i um fiir Fo r s c h u n g und Technol ogi e". The a ut hor s al one ar e r esponsi bl e for t he c ont e nl of t he paper. List of Symbols 6/ C d D D. ~o Ei f g h I [ t h Ampl i t ude fact or Velocity of light in vacuum Thi ckness of active layer Diffusion const ant [cm z s- 1 ] El ect ron diffusion const ant [cm 2 s - 1] Absolute el ect ron charge Phot on energy of ith mode Fr equency of self-pulsation Gai n of lasing mode Pl anck' s const ant Injection cur r ent Thr eshol d injection current Modulation of Laser Diodes 239 I o I ,A3 lo(a) d dth k L L~ 11 ~t PL I ' s p / ' s t R i R s p S SI t t d T V X Z # F t/ 2 A2 A Pn 2"sp 27ph 4,(E~) O) O dc bias current Pulse current Modified Bessel function Injection current density [A cm-2] Threshold current density [A cm 2] Boltzmann's constant, also wave number Laser cavity length Electron diffusion length Electron density Threshold electron density Index of refraction Effective index of refraction Light-output power per laser mirror Spontaneous emission rate per unit of volume and unit of photon energy E i Stimulated emission rate per unit of volume and unit of photon energy El Reflectivity of end mirrors for the ith mode Total spontaneous emission rate Photon number Photon number in the ith mode Time Inital delay time Absolute temperature Active volume Coordinate; normalized photon number Normalized electron density Spontaneous emission coefficient Internal optical loss per unit length Damping parameter Volume ratio Photon confinement factor Normalized injection current Internal quantum efficiency Emission wavelength Wavelength separation of neighboring longitudinal modes ; wavelength detuning Half-width of the stimulated emission Electron mobility Spontaneous electron lifetime Photon lifetime Normalized complex photon amplitude function Number of modes per unit volume and energy Angular frequency Angular detuning frequency 240 G. Arnold et al. Ref erences 7.1 G. Arnold, P. Russer : Appl. Phys. 14, 255 (1977) 7.2 M. Boerner: Wiss. Ber. AEG-Telefunken 44, 41 (1971) M. Boerner, S.Maslowski: Proc. IEE 123, 627 (1976) 7.3 E.Weidel: Electron. 1,ett. II, 436 (1975) 7.4 D. Gl oge: Appl. Opt. I0, 2442 (1971) 7.5 D. Gloge: Appl. Opt. 13, 249 (1974) 7.6 R.L. Hart man, R. W. l )i xon: Appl. Phys. Lett. 26, 239 (1975) 7.7 H.Kressel: Proc. Opt. Fibre Transm. Conf. II, Williamsburg, Va. (1977) H. Kressel, J.K. Butler: Semicomluctor Lasers and Heterojtowtion LEDs (Academic Press, New York 1977) 7.8 R.L. Hart man, N.E.Schumaker, R.W. Di xon: Appl. Phys. Lett. 31, 756 (1977) 7.9 C. H. Gooch: GaAs Lasers (Wiley Interscience, London 1969) 7.10 L. A. D' As ar o: J . Lumin. 7, 310 (1973) 7. 11 M.B.Panish, l. Hayashi: Appl. Solid State Sci. 4, 236 (1974) 7.12 l . Hayashi : Appl. Phys. 5, 25 (1974) 7.13 M. B. Pani sh: Pr oc. I EEE64, 1512 (1976) 7.14 J. H. Itsieh, J.A. Rossi, J. P. Donel l y: Appl. Phys. Lctt. 28, 709 (1976) 7.15 .1.H. Hsieh: Proc. AGARD Conf., London (1977) p. 35/1 7.16 T. Yamamot o, K.Sakai, S. Aki ba: Jpn. J. Appl. Phys. 16, 1699 (1977) 7.17 K.Oc, S.Ando, K. Sugiyama: Intern. Conf. Integrated Optics and Opt. Fiber Commun. , Tokyo (1977) post deadline paper P5 7.18 T. Yamamot o, K.Sakai, S. Akiba, Y. ltaga, Y. Sucmatsu: Intern. Conf. lntcgraled Optics and Opt. Fiber Commun. , Tokyo (1977) paper B2 3 7.19 K. Wakao, H. Morita, T. Kambagashi , K. Iga: Jpn. J. Appl. Phys. 16, 2075 (1977) 7.20 E. Haken, H. Haken: Z. Plays. 176, 421 (1963) 7.21 G..1.Laser: Solid-State Electron. 7, 707 (1964) 7.22 J.Vilms, L.Wandinger, K. L. Kl ohn: IEEE J. QE-2, 80 (1966) 7. 23 H. Haug: Z. Phys. 194, 482 (1966) 7.24 H. Haug: Z. Plays. 200, 57 (1967) 7.25 M. J. Adams, P.T. Landsbcrg: In Ref. 7.9, p. 5 7.26 H. Haug: Phys. Rev. 184, 338 (1969) 7.27 M..l. Adams: Opto-Electron. 5, 201 (1973) 7.28 H.Statz, C.L.Tang, J.M. Lavine: J. Appl. Phys. 35, 2581 (1964) 7.29 H. Haug: Z. Phys. 195, 74 (1966) 7.30 J. Buus, M.Danielsen, P.Jeppescn, F. Mcngcl, H. Moeskjaer, V.Ostoich: Proc. 2rid Europ. Conf. Opt. Commun. , Paris (1976) p. 231 7.31 W. Streifer, R. l). Burnham, D.R.Scrifres: IEEE J. QE-13, 4(13 (1977) 7.32 G. Lasher, F. Stern: Phys. Rcv. 133, A553 (1964) 7.33 H.C.Casey, Jr., F. St ern: J. Appl. Plays. 47, 631 (1976) 7.34 F. Marinelli: Solid-State Electron. 8, 939 (1965) 7.35 T. H. Zachos, J.E. Ripper: IEEE J. QE-5, 29 (1969) 7.36 lt.C.Casey, Jr., B.I. Miller, E. Pinkas: J. Appl. Phys. 44, 128l (1973) 7.37 S.N. Biswas, N. Kumar : Indian J. Pure Appl. Phys. 11,855 (1973) 7.38 J.C. Dyment, J.E. Ripper, T.P. Lee: J. Appl. Phys. 43, 452 (1972) 7.39 J.Angerstein, D.Siemsen: Arch. Elektron. IJbcrtragungstcch. 30, 477 (1976) 7.40 K. Konnert h, C. Lanza: Appl. Phys. Lett. 4, 120 (1964) 7.41 R. Roldan : Appl. Phys. Lett. I I, 346 (1967) 7.42 D. A. Kl ei nman: Bell Syst. Tech. ,1.43, 1505 (1964) 7.43 D. Ri3ss: Z. Naturforsch. 19a, 1169 (1964) 7.44 T.Ikegami, K. Kobayashi, Y. Suematsu: Electron. Commun. Jpu. 53-B, 82 (1970) 7.45 P.M.Boers, M. T. Vlaardingerbrook, M. Dani el scn: Electron. Lett. 11,206 (1975) 7.46 T.Ozeki, T. It o: IEEE J. QE-9, 388 (1973) 7.47 T.Ozeki, T. l t o: IEEE J. QE-9, 1098 (1973) 7.48 T.P. Lee, R. M. Derosier: Proc. IEEE 62, 1176 (1974) 7.49 W. Hart h, D.Siemsen: Arch. Elektron. Ubertragungsteeh. 30, 343 (1976) Modul at i on of Laser Diodes 241 7.50 T. Kobayashi , S. Takahashi : ,Ipn. J. Appl. Phys. 15, 2025 (19761 7.51 M. Chown, A. R. Goodwin, D.F. Lovelace, G. H. B. Thompson, P.R.Selway: Electron. Lett. 9, 34 (19731 7.52 H,W, Thim, L. R. Dawson, J. V. Di Lorenzo, J. C. Dyment, C.J.Hwang, D.L. Rode: Intern. Solid-State Circuits Conf., Dig. of Tech. Papers (1973) p. 92 7.53 P. Russcr, S.Schultz: Arch. Elektron. ~bert ragungst ech. 27, 193 (1973) 7.54 D. Schicketanz: Siemens Eorsch. Entwicklungsber. 2, 218 (1973) 7. 55 H.Yanai, M. Yano, T. Kami ya: IEEE J. QE-I1, 519 (1975) 7.56 J. Gruber, P. Mart en, R.Petschacher, P. Russer: IEEE Trans. COM-7, 1088 (19781 7.57 M. Dani el s en: l EEE J. QE-12, 657 (19761 7.58 T. lkegami, Y. Suematsu: Electron. Commun. Jpn. 51, 51 (19681 7.59 T.Ikegami, Y. Suematsu: Proc. 1EEE 55, 122 (19671 7.60 F.L.Paoli, J.E. Ripper: Proc. IEEE 58, 1457 (1970) 7.6l T.Ikegami, Y. Suemat su: Electron. Commun. Jpn. 53-B, 69 (19701 7.62 W. Har t h: Electron. Lett. 9, 532 (1973) 7.63 W. Harth, D.Siemscn: Arch. Elektron I]bertragungstech. 28, 391 (19741 7.64 P.Russer, H. Hillbrand, W. Har t h: Electron. Lett. 11, 87 (1975) 7. 65 H. Grot he, W. Hart h, P. Russer: Electron. Lett. 12, 522 (19761 7.66 P. Marschall, E.Schlosser, C.WiSlk: Electron. Lett. 15, 38 (19791 7.67 O. Nakada, N. Chinone, S. Nakamura, H. Nakashi ma, R. It o: Jpn. J. Appl. Phys. 13, 1485 ( 1974) 7.68 P.A. Kirkby, A. R. Goodwi n, G. H. B. Thompson, P.R.Selway: IEEE J. QE-13, 705 (19771 7.69 K. Aiki, M. Nakamura, T. Kuroda, J. Umeda, R. lto, N. Chinone, M. Maeda: IEEE J. QE-14, 89 (1978) 7.70 K. Kobayashi , R.Lang, H. Yonezu, Y. Mat sumot o, T. Shi nohara, l.Sakuma, T.Suzuki, l . Hayashi : IEEE J. QE-13, 659 (1977) 7.7l M. Maeda, K. Nagano, l . l kushi ma, M. Tanaka, K.Saito, R. l t o: Proc. 3rd Europ. Conf. Opt. Commun. , NTG-Fachberi cht e Bd. 59, 120 (1977) 7.72 H. Nami zaki : Trans. Inst. Elcclron. Commun. Eng. Jpn. E-59, 8 (1976) 7.73 N. Chinone, K. Aiki, M. Nakamura, R. l t o: IEEE J. QE-14, 625 (1978) 7.74 Y.Suematsu, T. Hong, K. Furuya: Nachrichtent. Zeitschrift NTZ 31, 127 (1978) K.Euruya, Y.Suematsu, T. Hong: Appl. Opt. 17, 1949 (1978) 7.75 T. Kobayashi, H. Kawaguchi, Y. Furukawa: Jpn. J. Appl. Plays. 16, 601 (1977) 7.76 M. Nagano, K. Ka s a h a r a : l EEEJ . QE-13, 632 (1977) 7.77 S.Iida, Y. Wat anabe: Jpn. J. Appl. Phys. 13, 1249 (1974) 7.78 .l.E. Ripper, F. D. Nunes, N. B. Patel: Appl. Phys. Lett. 27, 328 (19751 7.79 B. W. Hakki : IEEE J. QE- I I , 149 0975) 7.80 T. L. Paol i : I EEEJ. QE-12,77(I(19761 7.81 C. l t . Gooch: b~jection Electroluminescem Devices (Wiley, London 1973) p. 138 7.82 T.L.Paoli, P.A. Barnes: Appl. Phys. Lett. 28, 714 (19761 7.83 R.W. Dixon~ F. R. Nash, R. L. Hart man, R. T. Happl ewhi t e: Appl. Phys. Lett. 29, 372 (1976) 7.84 K. Kobayashi, R. Lang, H. Yonezu, J. Sakuma, I. Hayashi : J pn. J. Appl. Plays. 16, 207 (1977) 7.85 T.L.Paoli, J.E. Ripper: Plays. Rev. Left. 22, 1085 (19691 7.86 T. lkegami: Proc. 1st Europ. Conf. Opt. Commun. London (1975) p. 111 7.87 l).Sicmsen, J. Angerstein: Electron. Lett. 12, 432 (1976) 7.88 K. Pet crmann: Opt. Quant um Electron. 10, 233 (1978) 7.89 P.R.Selway, A. R. Goodwi n: Electron. Lett. 12, 25 (1976) 7.90 W. Freude: Arch, Elektron. lJbertragungstech. 32, 105 (1978) 7. 91 N.G. Basov, V.N. Morozov, V.V.Nikitin, A.S.Semenov: Sov. Phys. Semicond. 1, 1305 (1968) 7.92 T.P. Lee, R. Rot dan: IEEE J. QE-5, 551 (1969) 7.93 T.L.Paoli, J.E. Ri pper: Appl. Phys. Lett. 18, 466 (1971) 7.94 N. Chinone, R. It o: Jpn. J. Appl. Plays. 13, 575 (1974) 7.95 G. Arnold, K. Pcl ermann : Sclf-pulsing phenomena in injection lasers. Opt. Quant . Electron. 10, 311 (1978) 7.96 T. k. Paol i : IEEE J. QE-13, 351(1977) 7.97 N. G. Basov: IEEE J. QE-4, 855 (1968} 7.98 K. Kobayashi : 1EEE J. QE-9, 449 (1973) 7.99 H. -G. W6hrstein, H. Ha ke n: l EEE J. QE-9, 318 (1973) 242 G. Arnold et al. 7.100 J.E. Ri pper, T. L. Paol i : I EEE J. QE-8, 74 (1972) 7.101 A.B. Fowl er : J. Appl. Plays. 35, 2275 (1964) 7.102 P. R usser : Arch. Electron. 1~ ber t r agungsl ech. 29, 231 (1975) 7.103 P. Russer : Laser 75 Opt oel cct r oa. Conf. Proc., Mi i nchcn (1975) p. 161 7.104 H. Hi l l br and, P. Russer: El ect ron. Lett. I I , 372 (19751 7. t05 R. Laug, K. Ko b a y a s h i : l EEE. I . QE- I I , 6 0 D( 1 9 7 5 ) 7.106 K. Kobayashi , R. Lang, K. Mi nemur a : Proc. 1st Europ. Conf. Opt . Commun. , London (1975) p. 138 7.107 R. Lang, K. Kobayashi : I EEE .1. QE-12, 194 (1976) 7.108 P. Russer, G. Arnol d, K. Pe t e r ma nn: Hi gh-speed modul at i on of DI I S lasers in t he case of coher ent light injection. Proc. 3rd Europ. Conf. Opt . Commun. Mi i nchen, 1977 7.1(19 G. Arnol d, K. Pet er mann, P. Russer, F.-J. Berlec : Arch. El ekt ron. 1Jberl ragungst ech. 32, 129 (1978) 7.110 R. Salath6, C. Voumar d, H. Weber : Plays. St at us Solidi (a)23, 675 (1974) 7.111 J.-l. Ni shi zawa, K. Ishi da: [ EEEJ . QE- I I , 515(1975) 7.112 M. Nakamur a, K. Aiki, J. Umeda, A. Yariv: Appl. Phys. Lett. 27, 403 (1975) 7. l 13 S. R. Chi nn : Opt . Commun. 19, 208 (1976) 7.114 H. Na mi z a k i : l E E E J . QE- I I , 427 (1975) 7.115 T. Tami r (ed.): Integrated Optics, 2ud cd., Topi cs in Appl i ed Physics, Vol. 7 (Springer, Berlin, Hei del berg, New York 1979) 7.116 K. Nagano, M. Maeda, K. Sai t o, M. Tanaka, R. I t o : Tr a n s . I E CE o f J p n . E-61, 441 (1978) 7.117 D. N. Payne, W. A. Ga mbl i ng: Electron. Lett. I I , 176 (1975) 7.118 F. P. Ka p r o n : Electron. Lett. 13, 96 (1977) Addi t i onal Re f e r e nc e s wi t h Ti t l e s s. M. Abbot t , W. M. M uska, T.P. Lee, A. G. Dent al , C. A. Burrus : 1.1 Gbi t / s ps e udor a ndom pul se- code modul at i on of 1.27 ~m wavel engt h cw I nGaAs P/ I nP d.h. lasers. Electron. Left. 14, 349 (1978) S. Aki ba, K. Sakai , T. Ya ma mot o: Di rect modul at i on of I nGa As P/ I nP doubl e het er ost r uct ur e lasers. El ect ron. Lett. 14, 197 (1978) C. Baack, G. Elze, B. Enni ng, G. Waft: Modul at i on behavi our in t he gi gabi t range of several GaAIAs lasers. Fr equenz 32, 346 (1978) J . A. Cope l a nd: Semi conduct or - l as er sel f pul si ng due to deep level t raps. El ect ron. Let/. 14, 809 (1978) G. Deut sch, G. Li ndner, K. Ltibke, H. W. Thi m: Hi gh- bi t - r at e pul se r egener at i on and modul at i on of i nj ect i on lasers wi t h a pl anar Gt mn device. El ect ron, Lett. 15, 285 (1979) W. Fr eudc : Monomodc oper at i on of di rect modul at ed GaAI As DHS i nj ect i on lasers from 260 Mbi t /s up to 1.4 Gbi t / s. Arch. El ekt ron. (. Jbert ragungst ech. 32, 105 (1978) O. Hi rot a, Y. Suemat su: Noi se pr oper t i es of i nj ect i on lasers due to reflected waves. 1EEE J. Quant . El ect ron. QE-15, 142 (1979) T. -H. Hong, Y. Suemat su: Ha r moni c di st or t i on of i nj ect i on lasers. Trans. I ECE Jpn. E62, 144 (1979) D. Kat o: Hi gh- qual i t y br oad- band opt i cal communi cat i on by t i medi vi si on mul t i pl exed pul se anal og modul at i on: nonl i near i t y in di ode lasers. I EEE J. QE-14, 343 (1978) T. l t o, K. Nakagawa, K. Ai da, K. Takemot o, K. Sut o: Nonr epeat er ed 50ki n t r ansmi ssi on ex- per i ment usi ng l ow-l oss opt i cal fibres. El ect ron. Lett. 14, 520 (1978) M. Ito, T. Ki mur a: Longi t udi nal mode compet i t i on in a pul se modul at ed AIGaAs DH semi- conduct or laser. I EEE J. QE-15, 542 (1979) W. Langc: Int ensi t y fl uct uat i ons of i nj ect i on lasers oper at ed wi t h high frequency modul at i on. El ect ron. Lett. 14, 7 (1978) K. Nawat a, S. Machi da, T. Ito : An 800 Mbi t / s opt i cal t r ansmi ssi on exper i ment using a si ngl e- mode fiber. I EEE J. QE-14, 98 (1978) M. Sar uwat ar i , K. Asat ani , J.-I. Yamada, 1. Hat akeyama, K. Sugi yama, T. Ki mur a : Low loss fibre t r ansmi ssi on of hi gh speed pul se signals at 1.29/am wavel engt h. El ect ron. Lett. 14, 187 (1978) T. Ya ma mot o, K. Sakai , S. Aki ba: Fast pul se behavi our of l n Ga As / l n P doubl c- het er os l r uct ur e lasers at 1.27 p-m. El ect ron. Left. 13, 142 (1977) 8. The Ef f ect of Junct i on He at i ng on Las er Li neari t y and Har moni c Di s t ort i on J. K . B u t l e r wifl~ i0 Figures The modul at i on charact eri st i cs of a laser di ode ar e frequency dependent (Chap. 7) being affected by r esonance effects, noise associ at ed wi t h ki nks and t ur n- on delays. In this chapt er, we anal yze the anal og modul at i on l i mi t at i ons i mposed by nonl i near power - cur r ent curves when these nonl i neari t i es are t her mal l y induced. In the oper at i ng regi me where ot her effects are negligible (low to moder at e frequencies) and st ar t i ng wi t h a laser with nor mal l y a linear P - I curve (Chap. 2), the t her mal l y i nduced effects discussed here will limit the pr oper t i es of a laser used in anal og modul at i on. We deri ve expressi ons for the t ot al har moni c di st ort i on as a funct i on of device par amet er s and oper at i ng frequency. 8.1 Static Laser Characteristics The rel at i onshi p between out put power P and dri ve cur r ent I for a semi con- duct or laser di ode is given by P = O l h v / q ) ( I - 1,1,), (8. I) where h v is tile l asmg phot on energy, q is the el ect ron charge, I1 is the ext ernal differential qua nt um efficiency, and l t h is the dri ve current at t hreshol d. Below t hreshol d, I < / t h , the lasing power is zero. If l,h is assumed to be const ant t hen t he P - 1 curve is linear. However , it is f ound t hat pract i cal cw lasers have P - I curves t hat are somet i mes nonl i near (sec Chap. 2). These nonl i neari t i es in P - I curves are frequent l y due to instabilities of the opt i cal l asi ng mode. Her e we will limit our discussion to lasers which oper at e in a st abl e wavegui de mode so t hat t he P - I curve will have an absence of kinks. The nonl i neari t i es discussed here resul t f r om j unct i on heat i ng which pr oduces a "fl oat i ng" val ue of [ t h ' Assumi ng t hat I n , o c e x p ( T / T o ) where T o is a charact eri st i c par amet er (Chap. 2), the t hr eshol d current increase for a AT t emper at ur e change is i , . = tl',~ e p ( A T / T o ) . ( 8 . 2 ) AT is pr opor t i onal to the drive cur r ent because of the power di ssi pat i on in the laser. In the absence of heating, Ith = 1% the nomi nal t hreshol d current of tile 244 d. K. Butler device. Experi ment al l y, lh can be det ermi ned using low dut y cycle pulses where heat generat i on is negligible. If V3(~-E/q) is the j unct i on voltage, R~ the series resistance, and I the drive current then the power dissipation Q in cw oper at i on is Q = (1 - , l v ) ( v i i + R j 2 ) , (8.3) where tlp is the power efficiency. Because q, < 10 % in typical devices it will be neglected in the subsequent analysis. Consequent l y, the t emper at ur e rise AT above the heat sink t emper at ur e is assumed to be A T ~ R , . , ( V3I + R J 2 ) , (8.4) where R,m is the t hermal resistance. The laser o u t p u t p o w e r P can be expressed as P = ( q h v / q ) { I - lh exp [ R , , , ( V.il + R J 2 ) / 7 o ] }. (8.5) In cw oper at i on the t hreshol d cur r ent is a sol ut i on of ltu = llJh exp ['R,m(Vjl2,h + R J Z O / T o ] . (8.6) R~ is det ermi ned experi ment al l y from the di ode V- I characteristics, I, , and lob are found from the pulsed and cw P - 1 curves, and the rat i o R , , / T o is from (8.6), I n ( l , , , / I , ) R , , , / G - ~i,~, + R J , ] , ' ( 8 . 7 ) We illustrate P - I characteristics in Fig. 8.1 using q =30'%,, V i = 1.4V, and I~), = t 00mA. In Fig. 8.1a R , m / T o =0. 5 W- ~ and the t wo curves represent values of R~= 1 and 10f2. In Fig. 8.1b R~=2f ~ and R, , / 7 o =0 . 2 and 0. 8W -1. 8. 2 Transient Analysis In pulsed oper at i on heat i ng occurs duri ng t he pulse. For example, let I be the pulse curren! ampl i t ude under the condi t i on I , < I < I , , . This pulse st rengt h causes the laser to reach t hreshol d in the early stages of the pulse but it will turn off in the latter parts. But when I > I,,, the di ode will , mi t for the t ot al pulse durat i on, al t hough light i nt ensi t y at the end will be lower t haq at the beginning. To anal yze this t ransi ent response, we use an electrical anal og to model the di ode heat i ng phenomena as shown in Fig. 8.2. In Fig. 8.2a, the "cur r ent Effcct of.lunction Heating on Laser Linearity and l larmonic Distortion 245 15 i a I0 5 O q I I I I I I I I I I I I l t h 0 - I O O m A / RlmTo = 0 . 5 W - I / / / / Rs = ~ I I I I I I I I I I I I I I [ I I 50 I 00 150 CURRENT I (mA) 200 15 , 0 O o 5 illl O = IOOmA / R s " 2 ohm Rtm = 0.2 W J / " I I I I I 1 I I I I I I I I I I I I 5 0 I 0 0 150 CURRENT I ( mA) Fig. 8.1a, b. Static power-cur- rent curves for l a s e r diodes: (a) series resistance is a pa,am- 2oo eler: (b) thermal resishmce is a paranaetcr source" in the equivalent circuit is Q (t), tile i nst ant aneous heat generat i on in tile active region, and AT=A(t ) is the t emperat ure rise as a function of time, modeled as the voltage rise at the node. The capacitance C is equivalent to the thermal capacity and the conduct ance G is the equivalent of the rcciprocal of the t hermal resistance. Figure 8.2b is the electrical anal og with Q(t) as the "voltage source" and A(t) is the "' current" flowing in the circuit. The units of the various parameters arc Q(t) w (power generation) A(t) K (temperature rise) Rt. , K/ W (thernaal resistance) K h J/ K (heat capacity). 246 d. K. But l er Q(t) & ( t ) l a ) C = Kt m G= l - - Rim L ,, g' y' y" y% Q ( t ) + R R = ~ Rt m ( b ) Fig. 8.2a, b. Electrical analog circuits used to model temperature in the active region O0 A(t) I ( ~( t ] ~ t (a) ~ K h Q Rf m = ~0 Qo ( b ) A ( t ) Qo Rtm Kh=O (c) Fi g. 8 . 3 a - c . Temperature response to a step Ctll'l'enl Fr om Ki r chhof f laws, dA 1 O_(O=Kh eF + ( s . s ) For example, assume t hat Q( t ) =Qou( t ) where u(t) is the unit step function t urni ng on at t = 0 as shown in Fig. 8.3a. In Fig. 8.3b we show t he case where R,,, = c~, i.e., heat cannot be conduct ed from the active region. Solving (8.8), the t emper at ur e rise A( t ) =( Qo/ Kh) t , i.e., the t emper at ur e rises indefinitely. In Fig. 8.3c we show the case where the heat capaci t y K~=0 so t hat (8.8) gives A(t )=QoRt mu(t ), a const ant value for t >0. The heat capaci t y Kf~ plays a maj or role in the t ransi ent as well as the low- frequency response of the laser diode. In our simple model the value of K, can be est i mat ed from the vol ume of the active region and the specific heat capaci t y of the material. Of course, an exact model giving the t ransi ent t emper at ur e response can be obt ai ned onl y by accurat el y model i ng the laser geomet ry. Pract i cal laser diodes are mul t i l ayered devices mount ed on copper heat sinks and accurat e t hermal model i ng woul d include the layers' t hermal capacities and electrical resistances. Ncvertheless, our model can give qual i t at i ve as well as fi rst -order quant i t at i ve dynami c behavi or in the row-frequency regimes. Finally, we not e t hat for har moni c exci t at i on of the driving function Q(t), the t emper at ur e function A(t) lags in time. Thi s phenomenon will be discussed in later sections. Effect of J unc t i on Heat i ng on Laser Li near i t y and Ha r moni c Di s t or t i on 247 Wc n o w a n a l y z e t h e d y n a m i c b e h a v i o r o f t h e l a s e r i n p u l s e d o p e r a t i o n . T o o b t a i n t h e mo s t g e n e r a l a n a l y s i s we a s s u me t h a t t h e d i o d e is d c b i a s e d t o a c u r r e n t v a l u e I 0 wi t h a s wi n g a m p l i t u d e Im. Spe c i f i c a l l y, f o r a p u l s e wi d t h r , i (t ) = 1 o + I mH [(t - - r / 2 ) / r ] , (8.9) wh e r e t h e w i n d o w f u n c t i o n t , l t l < l / 2 / / ( t ) = 0 , I t l > 1 / 2 ' T h e i n s t a n t a n e o u s p o w e r g e n e r a t e d i n t he a c t i v e l a y e r is Q( t ) = Vii o + R . J o + [ Vj l m + R j m ( 2 1 o + l m)] H [ t - - r / 2 ) / z ] . ( 8 . i o ) S u b s t i t u t i n g (8. 10) i n t o (8.8) a n d s o l v i n g t he r e s u l t i n g d i f f e r e n t i a l e q u a t i o n , we o b t a i n t he t e m p e r a t u r e c h a n g e wi t h t i me ~Ao + A 1 [1 - - e x p ( - - t / ' r o ) ] , t < z A( t ) = [ A 0 "-l- A 1 [ 1 - e x p ( - z/ z0) ] e x p [ ( z - t ) / r o ] , t > r , ( 8 . 1 1 ) wh e r e R~ g) Ao =Rt m( V3 I o + I " A 1 =Rt m [ V i m + Rs l , , ( 2 1 0 + l,n)] T O = Kh R t m T h e v a l u e ~0 is t h e t h e r ma l t i me c o n s t a n t . ( T y p i c a l d i o d e s h a v e r 0 ~ a f ew h u n d r e d ns. ) T h e p o we r o u t p u t is I }. p(r) = ( qhv/ q) { l o + I , , n [(t - r / 2 ) / 2 ] - ,h e x p [ A( t ) / To] (8.12) N o t e t h a t wh e n t he d i o d e is b i a s e d a t t he c w t h r e s h o l d c u r r e n t l o = I , e x p ( A o / T o ) . F i g u r c 8.4 s h o ws a s e r i e s o f p l o t s i l l u s t r a t i n g t h e l i ght r e s p o n s e a s a f u n c t i o n o f t i me o v e r o n e t h e r l n a l t i me c o n s t a n t r 0, T = % . T h e v a r i o u s p l o t s r e p r e s e n t d i f f e r e n t b i a s l e ve l s : (a) I o = 0 , (b) lo=l~)h, a n d (c) t o > l , . In Fi g. 8. 4a, t he l i ght o u t p u t is m a x i m m n a t t h e b e g i n n i n g o f t he c u r r e n t p u l s e b u t d r o p s a s t he d i o d e h e a t s . T h e m a x i m u m p o w e r o u t p u t o c c u r s , o f c o u r s e , wh e n t he a c t i v e r e g i o n t e m p e r a t u r e e q u a l s t h a t o f t he h e a t s i nk. E v e n t u a l l y t h e p o w e r d r o p s t o P3 wh i c h is st i l l l a r g e r t h a n t he c w p o w e r l evel a t t he d r i v e c u r r e n t I = I 0 + I m. Wh e n t h e p u l s e wi d t h is s uf f i c i e nt l y e x t e n d e d , h o we v e r , t h e p o we r wi l l r e a c h t h e cw p o we r l evel . As t he t h e r ma l c a p a c i t y K h is r e d u c e d , t h e t i me r e q u i r e d t o r e a c h t he c w p o w e r l evel fal l s. 248 . I . K. Butler h- == I - 0 n,. w P3 I o ' O o I m I t ~ P 2 P3 PI P 4 0 ! 5 1.0 115 ( . ) 2 0 P2 P3 I o " I t h [ b ) am ) 0 O 0 ! 5 1.0 11,5 2.0 P 3 i PI t . . . . . . P4 I ( c ) Io> It~ Im>O NORMALIZED TIME t / t o H E A T S I N K T E M P E R A T U R E S T I < T 2 F , T 2 0 + LOW D U T Y _ _ . . ~ C = - ( = I , ~ + I o I t h Fig. 8.4a c. Laser pulse response for different current bias levels D R I V E C U R R E N T F i g . 8 . 5 . C u r r e n t p u l s e c y c l e o n t h e P l c u r v e s In Fig. 8.4b, tile light power is i l l ust rat ed f or a di ode dc bi ased t o t hr eshol d. Thi s bi as c ur r e m heat s t he act i ve r egi on a bove t he heat si nk t emper at ur e. When t he cur r ent pul se is t ur ned on, t he powe r rises t o P : and t hen dr ops back t o P~ as t he act i ve r egi on t e mpe r a t ur e rises. Fi nal l y, in Fig. 8.4c, not e t he t mde r s hoot of t he light power when t he pul se is t ur ned off'. Thi s case will be i l l ust rat ed by t r aci ng t he event s on t he P - I cur ve (Fig. 8.5) dur i ng t he t i me n e i g h b o r h o o d of t he pulse. At t = 0 , an i nst ant j ust bef or e t he pul se t ur n- on, t he di ode is bi ased t o a val ue / 0 whi ch is a bove t he di ode' s cw t hr e s hol d; t he c or r e s pondi ng light power is P, . As t he cur r ent pul se rises fi ' om I o t o I o + I m t he power i ncreases fi' om P t t o P2. The pr ogr es s i on is t aken al ong t he t angent t o t he power cur ve at the poi nt P = P1 and I = 1o. As t he di ode heat s dur i ng t he Effect of .lunction Heating on Laser Linearity and llarmonic Distortion 249 pulse, the power dr ops fi'om P2 to P3; at this point, the active region t emper at ur e T 2 is higher t han 7"1, the heat sink t emperat ure. Just before the pulse is t urned off, t = r , the power level is P.> At the time t = , + , the light power dr ops to P4 at which time the active region t emper at ur e is 72. As the j unct i on cools the light power begins to rise to its initial value P1, thus compl et i ng the cycle. 8.3 Harmonic Analysis In this section, we discuss tile nonl i near characteristics of a laser di ode and how they affect the di st ort i on of harl noni c signals. We will st udy a laser dc biased to a level above t hreshol d with an al t ernat i ng signal superi mposed on tile bias. The swing level of the al t ernat i ng signal must be small enough so that the laser remains above t hreshol d duri ng the t ot al signal period. A st udy of the har moni c generat i on and the resulting signal di st ort i on is ai med at est i mat i ng the per f or mance of a laser di ode used as a t ransmi t t er in anal og systems. The i nformat i on cont ent of the signal is the al t ernat i ng par t ; t hus , if s ( t ) i s the base band signal, normal i zed to +1, then the current appl i ed to the di ode is i ( t ) = l o + 1 , , s ( t ) where I,, is tile swing about tile bias level. Ordi uari l y, the har moni c generat i on in a nonl i near device can be est i mat ed fi'om the static transfer characteristics, l Iowever, the nat ure of signal di st ort i on cannot al one be del ermi ned from the static characteristics because tile device' s transfe,- function may di st ort the signal. (The transfer function used here is tile rat i o of t he light out put power to the i nput signal current at the baseband fi'equcncy.) In this section, we anal yze the laser diode' s l ow-frequency t ransfer function and the har moni c signal generat i on using out" model of tile previ ous section. The amount of har moni c signal di st ort i on is est i mat ed by i nt roduci ng the t ot al har moni c di st ort i on fact or (THD). This quant i t y is det ermi ned l heoret i cal l y by assumi ng a sinusoidal drive current of strength I m, and t hen calculating the har moni c cont ent in the out put light power. TH D is then defined as the rat i o of power in tile harnl oni c component s 21~ 31~ 4./;... and the power in the fundament al of the out put intensity signal. With the di ode biased to 1 o with the baseband signal s ( t ) = c o s ~ , ) t super- imposed, the cur r ent is i ( t ) = I o + I , , coscot. (8.13) The heat i ng power pr oduced by this current becomes (2(t)= (Vii o + R ~ I o + R J ~ , / 2 ) + Ira( Vj + 2 I o R J costol + R~ I~1 (cos 2 t o t ) , / 2 . ( s . 1 4 ) 250 3 . K . B u t l e r N o t e t h a t t h e r e is a s e c o n d h a r m o n i c c o m p o n e n t i n t he i n s t a n t a n e o u s h e a t i n g p o w e r o wi n g t o t he d i o d e s e r i e s r e s i s t a n c e . S u b s t i t u t i n g Q ( t ) i n t o (8.8) yi e l ds t he i n s t a n t a n e o u s t e m p e r a t u r e r i se o f t h e a c t i v e r e g i o n . T h e s t e a d y - s t a t e s o l u t i o n is A( t ) = A o + A t cos((~)t - ~P) + A 2 c o s 2 ( , ) t - VJ), ( 8 . 1 5 ) wh e r e A o = R, m ( V J o + R f l ~ + R f l ~ , / 2 ) , ( 8. 16a) z] 1 = R , m l , n ( V j .-}- 2 R i l e ) [ 1 + (//.1o) 2] - ' / 2 ( 8. 16b) A2 = ( R , m R f l ~ / 2 ) [ 1 + ( 2 1 7 1 J ] - 112 (8. 16c) IP = t a n ' ( I ' / . I o) . ( 8. 16d) T h e t e r ms A o, A ~, A 2 a r e t h e dc, f u n d a me n t a l a n d s e c o n d h a r m o n i c a mp l i t u d e s . T h e c h a r a c t e r i s t i c f r e q u e n c y .[~, = I / ( 2 ~ R , , , , K h ) . (8.1 7) T h e t e mp e r a t u r e f l u c t u a t i o n s o f t h e a c t i v e r e g i o n l a g t h e i n p u t c u r r e n t b y t he p h a s e t e r m V j. At l ow f r e q u e n c i e s 7 J ~ 0 whi l e a t hi gh f i ' e que nc i e s W~r ~/ 2. N o w t he i n s t a n t a n e o u s l i ght p o w e r c a n b e wr i t t e n a s p ( t ) = Olh v / q ) 1Io + Im COS (91 - - ll~l, c x p [A o/ To + (A l / T0 ) cos(~,)t - tp) + ( A 2 / T o ) c o s 2 i , , > l - ' I ' ) - I } . ( 8 . 1 8 ) Wh e n t he a c s i g n a l is s u p e r i mp o s e d o n t he dc one , it is n o t i c e d t h a t t i l e e f f e c t i ve dc b i a s p o i n t o n t he P - I c u r v e wi l l be s hi f t e d d u e t o t he a d d i t i o n o f t he l as t t e r m in ( 8. 16a) wh i c h is t he a v e r a g e p o w e r o f t h e a c s i gna l . T h e i n s t a n t a n e o u s h e a t i n g t e r ms wh i c h af f ect t he h a r m o n i c s i gna l s t r e n g t h s , d i mi n i s h a s t he f r e q u e n c y o f t he b a s e b a n d s i g n a l i n c r e a s e s . At hi gh f r e q u e n c i e s , A 1 a n d A 2 a r e p r o p o r t i o n a l t o 1[[ T h c h a r m o n i c f r e q u e n c y c o m p o n e n t s o r F o u r i e r s e r i e s e x p a n s i o n o f t h e p o w e r s i g n a l c a n b e d e t e r mi n e d u s i n g t he Be s s e l g e n e r a t i n g e q u a t i o n f o r mo d i f i e d Bes s el f u n c t i o n s . I f l i nt (0) i s t he mo d i f i e d Be s s e l f u n c t i o n , o r d e r I, ( f i ni t e a t 0 = 0 ) t h e n ~J e x p ( x c o s 0 ) = ~ I m , ( x ) e x p ( j l 0 ) . (8.1 9) F.ffect of ,lunction Heating on Laser Linearity and Harmonic Distortion 251 [ For negative indices I m _ t ( x ) = l m l ( x ) . ] Thus, the t wo t erms in (8.18) are wri t t en as exp [(A j / T o ) cos(, )t - ~u) + (A z/To) cos 2 ((or - '/'9] = L I m t ( A ~ / T o ) l m , . ( A 2 / T o ) e x p [ j ( l + 2 n ) ( ( ' ) t - t f ' ) ] " l , t l : c t ~ ( 8 . 2 0 ) The kth har moni c of this signal is obt ai ned when I + 2n = k. Subst i t ut i ng (8.20) i nt o (8.18) and using the kth har moni c term, the light power becomes p ( t ) = ( q h v / q ) [Io + (1,,/2) (e J''' + e - J'"') --lO, em'/r" L Ck eJk( . . . . ~'~]' k = - ~ J a (8.21) where Ck= y~ I m , ( A 2 / T o ) I m k _ z , ( A j / T o ) . (8.22) I t = - - c~, We see t hat (8.21) is the compl ex Four i er series represent at i on of the power, i.e., p( t ) = ~ P k e x p O k o J t ) . (8.23) k= - 9: , - - ) * Since p ( t ) is a real signal, P k - 1 k, the compl ex conj ugat e of Pv Because tile coefficients are generally complex, we write P k = [ P k [ e x p ( 4 ) a ) where 4h, is the phase of the kth har moni c compar ed to the phase of the input drive current . In ter, ns of the device paranaete,' s and the C k terms, the vari ous coefficients are P o = O l h v / q ) ( 1 _ Cot,h e - - - o .,,J.,,T,,~,, P , = Ol h v / q ) ( I r a ~ 2 - - C 1 I[' h c"J""r" e -j r, ), _ _ C l o ,~A,,,"r,, e - j2,v), P , = ( t l h v / ' q ) ( ~ 2 tl~ ~ P k = O l h v / q ) ( - c l o oA,,,,r. , , , i , w " I Vk*[ I I ~ ~ / - (8.24a) (8.24b) (8.24c) (8.24d) Tile ampl i t ude and phase of tile fundament al can be calculated fronl the coefficient P L :tile phase is referenced with respect to tile phase of tile sinusoidal por t i on of the drive current. With tile light intensity written as p(O = L & cos(/.or- G) k=O 252 J . K. Buf f er X ( t ) H ( f l y ( t ) I NPUT OUT P UT SI GNAL S I GNAL Fig. 8.6. Linear system network t he f u n d a me n t a l ' s a mp l i t u d e a n d p h a s e a r e A l = ( t l h v / q ) [ I ~ , + ( 2C 1 I l l 1 c A / T n ) 2 - - 4 C I /m eA""'r" COS ~P] 1 / 2 _ ( C , G e ' s mq ' I </~, = t a n ~ , . . g - ~ - ~ ] . Z X , ~ o s ~ l J l . \ I m / Z - - C I t h ~ ~., / ( 8. 25a) ( 8. 25b) Si nc e v a l u e s o f ~/~ a r e s ma l l a n d p o s i t i v e ( < 10~), t i l e l i ght l a gs t i l e c u r r e n t . T r a n , ~ [ e r F u n c t i o n . I t is c u s t o m a r y t o c h a r a c t e r i z e a s y s t e m b y i t s t r a n s f e r f u n c t i o n I I ( f ) wh i c h wc de f i ne as f ol l ows . Le t x ( t ) be t he i n p u t a n d y ( t ) b e t h e o u t p u t o f t he s y s t e m, a s s h o wn i n Fi g. 8.6. T h e i r F o u r i e r t r a n s f o r ms a r e X ( f ) a n d Y ( f ) , r e s p e c t i v e l y . T h e t r a n s f e r f u n c t i o n is t h e n l l ( f ) = Y ( [ ' ) / X ( f ) . (8. 26) Ti l e o u t p L l t t i me f u n c t i o n b e c o me s / y ( t ) = j' l f ( f ) X ( f ) e - i 2 ' ~ . r ' q f ". , 1 Vo r s i n u s o i d a l i n p u t , t i l e o u t p u t will be s i n u s o i d a l . Le t x ( t ) = A c o s o ) t = A / 2 e i,,,, + A / 2 e + j ' ' . (8.27} (8.28) T h e o u t p u t b e c o me s y (t) = A / 214 ( - f ) e -.i,,,, + A / 2 H (. [') e . i ' ' ' . (8. 29) I n t he a b o v e e x p r e s s i o n s , we us e t he c o mp l e x s i gna l n o t a t i o n s o t h a t b o t h p o s i t i v e a n d n e g a t i v e f i ' e q u e n c y c o m p o n e n t s a r e p r e s e n t in t he s i gna l s . I t is c o n v e n i e n t , h o we v e r , t o us e o n l y t he p o s i t i v e f r e q u e n c y c o m p o n e n t s a n d u s i n g t he e x p r e s s i o n y( t ) 2 Re a l p a r t { A / 2 t t ( f ) e J ' " ' } . (8. 30) Ge n e r a l l y , t he t r a n s f e r f u n c t i o n is c o mp l e x , H ( f ) = I H ( f ) t e j'I', s o t h a t t he o u t p u t d u c t o a s i n u s o i d a l i n p u t is y I t ) = A I H ( f ) l cos ( <, t - 4, ) ( 8. 31t Effect of ,lunction Hcating on Laser Linearity and Iqarmonic Distortion 253 which shows how the phase of tile t ransfer funct i on ent ers i nt o the phase of the out put t i me function. Real syst ems have compl ex t ransfer funct i ons satisfying H* (.[) = H( - . [ ) ( 8. 32) hence (8.29) is identical to (8.31). In the present pr obl em, we nmst address the nonl i neari t i es pr oduci ng har moni cs. To a first appr oxi mat i on, the out put har moni cs ot her t han the f undament al will be di sregarded. Thi s means t hat the i nput cur r ent i ( t ) = ( I m / 2 ) e x p ( j r , ) t ) pr oduces a light response p ( t ) = P 1 exp(j~ot). The t ransfer funct i on becomes H ' ( f ) = 2P1 ( 8. 33) I nl A slight modi fi cat i on can be made by fact ori ng the coefficient q h v / q . H ( f ) = ( q / ~ l h v ) 2 P t / I , , . (8.34) Usi ng (8.24b), t i ( f ) = I - ( 2 / I , . ) C l 1 , , e ' l " " 7 < ' e - j q ' ( 8 . 3 5 ) The magni t ude [H(.[)[ is identical to (8.25a) except for the fact or l , , q h v / q : its phase is identical to (8.25b). To illustrate the use of the expressi ons derived above, we now choose some pract i cal device par amet er values. Consi der a laser with a pulsed t hreshol d current ll~l, = 100 mA, series resi st ance R~ = 1.4 ~ and rat i o of t her mal resi st ance to charact eri st i c t emper at ur e R , , , , / T o =0. 5 W J Fi gure 8.7 shows the ampl i t ude and phase of H(./') as a function of nor mal i zed frequency f / [ o . The curve was cal cul at ed by assumi ng l , , = 4 0 mA and 10mA with no appr eci abl e wi ri at i on fi' om curve to curve and t hus i mpl yi ng a "l i near syst cm". The m inimtun val ue of J t t ( f ) t occurs at . 1= 0 and increases to 1 at high frequencies. The ma xi mum phase shift of 3' occurs at . 1 7 1 ; = 1. H a r m o n i c D i s t o r t i o n . The har moni c di st ort i on of the out put signal is found fi' om the Four i er series represenl at i on. We define tile tolal har moni c di st ort i on ( THD) , T HD= Power in har moni cs Power ill f undameut al " 254 J. K. Butler 1. 00 0 , 9 8 g 0 . 9 6 0 . 9 4 0 . 9 2 0 . 9 0 0 . B e , , , , I , , , , i , , , ~ i , , , , i , , , , 4 ~ E _ R s = 1.4 ohm I , , , , I , , , , I , , , , i 2 3 4 ,5 5 3> m 2 A m NORMALI ZED FREQUENCY ( f / f o ) Fig. 8. 7. Tr a ns f e r f unc t i on a mpl i t ude and pha s e as a f unc t i on of mo d u l a t i o n f r equency Usi ng the series r epr esent at i on (8.23), ~ , P , , P , ~ , THD - , , :2 P1P* (8.36) and in dB, T HDm= 101Ogl oTHD. Fi gure 8.8 shows tile di st ort i on as a funct i on of the nor mal i zed frequency with I m, tile current swing, a par amet er . Ma xi mum di st ort i on occurs at f = 0 , decreasi ng l' rom - 4 5 dB as the modul at i ng cur r ent swing drops. The device par amet er s R~ and R, m/ T o play maj or roles in det er mi ni ng the di st ort i on as seen below. In Fig. 8.9, R, m/ T o is vari ed f r om 0.2 to 0.8 W- ; where I is agai n t reat ed as a par amet er . As expect ed, mi ni mum di st ort i on occurs at low R, , , / T o values where thc device is most "linear". Fi gure 8.10 shows the di st or t i on as a funct i on of R~. Again, mi ni mum di st or t i on occurs at low R~ values and increases monot oni cal l y with resistance. Not e t hat for an illustrative st ruct ure there is a dr amat i c dr op in the di st ort i on when R~<2f~. Hi gh series resi st ance cont r i but es significantly to the di st ort i on because it pr oduces second har moni c heat generat i on in the act i ve regi on of the laser. Effect of J u n c t i o n He a t i n g on La s e r Li n e a r i t y a n d Ha r mo n i c Di s t o r t i o n 255 "1= v Z __0 o z o Ig .J o F- -50 - 6 0 -70 -80 - 9 0 - T 0 = . R a - 1,4 ohm " 50 I m - I O m A ~ . I I , , I , , , , J , , , , I , , i , I , , , , I 2 3 4 5 NORMALIZED FREQUENCY ( f i f o ) Fi g, 8. 8. To t a l h a r mo n i c d i s t o r t i o n vs n o r ma l i z e d f r e q u e n c y wi t h a mp l i t u d e s wi n g as a p: u' a me t e r - 5 0 i , , I ' ' ' ' I ' ~ ' ' I ' ' i , I ' ' ' ' I ' ' ' ' o l t h = I 0 0 m A 1 o = 1 6 0 m A 5 0 ~ . . - - " " " ~ ~ I m = I0 mA ~ I 8o ~ - 9 0 . . . . , . . . . , . . . . , , , , , , . . . . , . . . . - , 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 8 R t m / T o (W "1 } Fi g. 8. 9. To t a l h a r mo n i c d i s t o r t i o n vs t he r a t i o o f t h e r ma l r e s i s t a n c e a n d ma t e r i a l c h a r a c t e r i s t i c t ern pe r a t u r e 256 J. K. t3mh~r - 40 m -50 Z o_ i - r , - o - 60 o_ z O ~E n, . r -70 _J O I - - ' ' ' 5 40 . 8 0 0 = I ~ I , I t I 2 4- 6 8 I0 SERIES RESISTANCE R s (ohm) Fig. 10. Tot al har moni c di st or t i on vs di ode series resi st ance 8. 4 Conclusion The thermal characteristics as modeled ill this chapter give complete infor- mat i on regarding the distortion properties of laser diodes for anal og appli- cations where relaxation oscillations can be neglected. Wc have st udi ed: 1) Response of laser light intensity to a current pulse: 2) The laser transfer characteristics which give t i l e phase and ampl i t ude di st ort i on of the modul at i on ; 3) Development of the harmoni c cont ent of an intensity modul at ed signal from its Fouri er expansion, and 4) Calculation of the total harmoni c di st ort i on of the out put modul at ed signal. Basically, the Fourier analysis of the laser intensity signal contains all of the useful i nformat i on regarding signal distortions. For example, i nt ernl odul at i on di st ort i on can be easily calculated by superi mposi ng tv, o signals of different frequencies and comput i ng the out put signal at the sum and difference frequencies. Comput at i ons of the transfer function show t hat maxi mum ampl i t ude and phase di st ort i on occur at low frequencies. Numerical calculations are made in Effect of ,hmction Heating on Laser Linearity and Harmonic Distortion 257 terms of normal i zed frequency . / ' / f ~ where . / ' = 1/(2~zKhR,,~). Maxi mum phase di st ort i on occurs at . f = . l c . Estimates of the cut off frequency from the t hermal time const ant are in the MHz vicinity. The key device paramet ers affecting the characteristics anal yzed in this chapt er are t he di ode series resistance, and the rat i o R , , , , / T o (t hermal re- si st ance/ charact eri st i c t el nperat ure det ermi ni ng the t emper at ur e dependence of the t hreshol d current density). Both of these par amet er s should be as low as possible and can be cont rol l ed to some extent, as discussed in Chap. 2, within limits i mposed by the t echnol ogy. We have made a series of cal cul at i ons of the t ot al har moni c di st ort i on (TH D) as funct i ons of these paramet ers. The range of par amet er values t aken reflects practical AIGaAs device values. In addi t i on, the st rengt h of the modul at i on cur r ent was varied to det ermi ne the effect of the swing ampl i t ude on di st ort i on. Ackmm, ledgen,ents. R.W. Klopfenstein assisted in computer modeling. Discussions with H. Kressel, M. Enenberg, C.J. Nuesc, and D. O. North are gratefully acknowledged. List of Symbols . f . f o h l I , ] t h I o I m i (t) l m t K h P p(t) P k Q Q.(t) Qo q R~ Rim T To l v j Modul at i on frequency [ s - 1] Charact eri st i c frequency Jo = 1 / 2 r c R t h K h Pl anck' s const ant [ J- s] Static drive cur r ent [ A] Pulsed t hreshol d cur r ent cw Thr eshol d current Bias cur r ent Maxi mum swing current I nst ant aneous current Modified Bessel funct i on of first kind and of or der l Heat capaci t y [ K/ J ] Static light intensity [ W] Inst ant aneous light i nt ensi t y [ W] Four i er coefficient of intensity expansi on Active region power generat i on [ W] I nst ant aneous power generat i on Static power generat i on El ect roni c charge [ C] Series resistance [f2] Thermal resistance [ K/ W] Temper at ur e [ K] Charact eri st i c mat eri al t emper at ur e [ K] Ti me l-s] Junct i on voltage IV] 258 J. K. But l e r 6T, A A(t) Ao, A1,A2 q qp 1' 17( t) TO T tlJ Temperat ure differential [ K] Inst ant aneous t emperat ure differential Coefficients of t emperat ure differential Differential quant um efficiency Power efficiency Light frequency [ s - a] Wi ndow function Characteristic time I-s] zo =RmKh Pulse width is] Phase of t emperat ure differential relative to current 9. An Illustrative Optical Communication System J. H. Mullins With 12 Figures The previous chapters dealt onl y with the devices and associated electronic circuitry which are used in fiber optic transmission systems, more specifically transmitters, receivers, and light coupling methods. Obviously, no such system can exist wi t hout the fibers themselves. But by the same token a complete system will require fiber protection, splicing, connect ors (both optical and electrical), physical hardware of all sorts, ancillary equi pment including terminals and test gear, and last but not least, a justification for the existence of the system. We do not pretend to cover all of these items in great detail in this chapter, but hope to put the role of devices into some perspective by describing a real system, including these ot her component s. At the same time we will at t empt to provide t hat "justification" ment i oned above. This requires delving into some topics which have little to do with optical devices, but a proper expl anat i on of device application seems to require it. In the remai nder of the chapter, we describe fiber optic (lightwave) systems in a fundament al sense, then classify them accordi ng to type of application. We discuss the economics of the systems, and at t empt to explain why the part i cul ar appl i cat i on discussed here, telephone interoffice t runki ng, appears economi - cally attractive. This also involves some expl anat i on of the telephone t runk envi ronment and the economics of compet i ng (metallic) systems. The system described is an experimental one designed by Bell Tel ephone Laboratories, Inc., which was installed in the Illinois Bell Tel ephone plant in Chicago. This digital facility is carryi ng a wide variety commerci al traffic, and is being eval uat ed by the designers. This chapt er describes the system com- ponents, configuration, installation and testing procedures, as well as the overall performance of the system. Al t hough not all of the devices and techniques discussed in the previous chapters are used in the system, enough of them are used to give the reader a good feeling for the practical application of these things. 9.1 Background and Purpose 9.1.1 General Fiber Optic System and Range of Applications Before proceeding to describe a specific application of the principles of fiber optic systems, using many of the systems and devices described in the previous 2 6 0 J. If. Mullins chapt ers, it woul d be well to exami ne t he fundament al nat ur e of the systems mor e generally. 111 its simplest form, a fiber opt i c (or lightwave) system consists of an opt i cal or light source which can be modul at ed, an opt i cal det ect or, and t he light conduct i ng fibers, which we shall often call lightguides. The sources may be LEDs or lasers, the det ect ors may be p- i - n di odes or APD det ect ors, and the lightguides may be mul t i mode (step index or gr aded index) or single mode. Out of this ar r ay of component s, some of which are i ncompat i bl e, one can assemble a wide variety of systems. In addi t i on to the basic component s, t here must be the requi red processing electronics (bot h terminals and line repeat ers or amplifiers), provi si ons for prot ect i ng and splicing the lightguide, and an assor t ment of ot her equi pment , including specialized testing apparat us. However, t here is anot her way of classifying a system, and this is by its use, or appl i cat i on, as descri bed below. One general classification refers to the mode of t ransmi ssi on: digital or analog. Here we make a sharp distinction between pure digital and the vari ous pulse modul at i ons such as PAM, PWM, PPM, PFM (pulse ampl i t ude, width, position, or frequency modul at i on) which are act ual l y anal og systems. In this cat egory, of course, we also place [inear systems such as intensity modul at ed o n e s . Besides the di gi t al -anal og distinction, we can classify systems accor di ng to their range. This is quite i mpor t ant for lightwave systems, as it to a large degree det ermi nes the qual i t y of the lightguide in t erms of at t enuat i on and bandwi dt h. The dividing lines bet ween classes of length are qui t e arbi t rary, of course, and we will cont ent ourselves with a relatively qual i t at i ve division into t wo classes. In the "shor t " class, we place systems t hat are general l y i nt ended for use within a building, in general as a single link bet ween t wo pieces of equi pment . These links are usually subst ant i al l y less t han one km in length. Systems in the "l ong" classification are generally one km in length or much longer. They are i nt ended for interbuilding, intercity, or cr oss- count r y appl i cat i on. Systems carryi ng t el ephone interoffice t runks fall into this cat egory, and we will be addressi ng this appl i cat i on in Sects. 9.2-6. In short, the system we describe is a "long", "digital" system. 9.1.2 Economic Considerations Before any new t echnol ogy can be expect ed to have wi despread appl i cat i on it must demonst r at e some significant advant age over an existing t echnol ogy. Most often, the domi nant effect is the economi c one. Al t hough this book is devot ed pri mari l y to optical devices, it t urns out t hat the economi cs of the system are often domi nat ed by the lightguide. A short discussion of this aspect is t herefore in order. Ther e are essential differences between metallic and opt i cal medi a with respect to their behavi or as a function of frequency, and this leads to some basic An Illustrative Optical Communication System 261 economi c differences. Specifically, in the t ransmi ssi on region, metallic medi a are domi nat ed by skin effect, and thus have a loss (in nepers or dB) which is pr opor t i onal to ]/ / f, where f is the frequency. The loss is also inversely pr opor t i onal to t he t ransverse di mensi on, d, of the conduct ors, so we observe t hat t he loss per uni t length is L ~ ~f / d. (9.1) However, we also notice t hat the cost per unit length is roughl y pr opor t i onal to the area, which is pr opor t i onal to the squar e ol d. Ther ef or e we find the cost per unit length is C ~ d 2 ~. [ / L 2 . (9.2) A measure of the economi cs is the cost per unit bandwi dt h, or per unit of i nformat i on. This is pr opor t i onal to ./'; so we iliad the specific cost of the medi um (per unit length per unit of i nf or mat i on capaci t y) is given by C m, where C m ~ 1/ L 2 . (9.3) On the ot her hand, t he loss per unit length of a fiber is i ndependent of the length of the span, the fiber cross section, and is i ndependent of fi' equency, up to t he poi nt at which dispersion effects set in. Ther ef or e the specific cost of the fiber C r is given simply by Cf ~ 1 / [ . (9.4) The concl usi ons from this are simple. Gi ven the length of a span, the quant i t y L is det ermi ned by si gnal -t o-noi se requi rement s. Then for a fiber with sufficiently low loss to bridge this span and retain the necessary si gnal -t o-noi se ratio, the par amet er most i mpor t ant for economi c prove-in of fibers is t he frequency f. if the fibers can handl e a large enough frequency, they can always pr ove in economi cal l y over metallic media, at least from t he st andpoi nt of i nf or mat i on per unit cost (and if t hat bandwi dt h is needed !). Hi dden here is also anot her equal l y i mpor t ant paramet er. The quant i t y L (the loss per unit length) is generally inversely pr opor t i onal to t he length of the span, because of signal- t o-noi se requi rement s. Thus, all ot her things being equal, the specific cost (per unit length) of metallic medi a t end t o be pr opor t i onal to the square of the length of the span, whereas fibers are i ndependent of it. This pr oper t y is very i mpor t ant for single spans, but less so for repeat ered lines. The latter are i nt errupt ed at frequent intervals by repeaters, so t hat the medi a costs can be exchanged for electronics costs. 262 J. H. Mullins 9. 1. 3 Other Considerations Ther e are some pr oper t i es of lightguide syst ems which may be very i mpor t ant , but which are difficult to quant i fy. These are freedom f r om electrical inter- ference, very low or nonexi st ent crosst al k, very smal l size and wei ght for a given capaci t y, and, for repeat ered lines, general l y much l onger spans bet ween repeat ers. The last t wo propert i es will be par t i cul ar l y evi dent in the syst em descri bed in the following sections. Ther e are, of course, some di sadvant ages. Besides the wel l -known field splicing and handl i ng pr obl ems of fibers, t here are some i nherent pr oper t i es of the sour ce- det ect or process, resul t i ng in a poor er si gnal -t o-noi se rat i o for fibers t han for met al l i c medi a. Specifically, the "zero loss" si gnal -t o-noi se rat i o for opt i cal syst ems (the si gnal -t o-noi se rat i o for a syst em with zero length medi um) is usually some 40 to 50 dB poor er t han for a zero loss met al l i c system. Thi s difference derives bot h from the rel at i vel y small power t hat can be injected i nt o the fiber by the t r ansmi t t er and f r om qua nt um effects in the receiver, l eadi ng to significant shot noise. For br oa dba nd anal og systems, which general l y requi re large si gnal -t o- noise rat i os (40 to 60 dB), met al l i c syst ems may have an advant age. A si mpl e numeri cal exampl e may hel p to clarify this. We assume a 10 MHz bandwi dt h syst em with an i nt ensi t y modul at ed source. We t ake the best of ci rcumst ances, in which the source may accomodat e a l arge (nearl y 100%) modul at i on. Thi s in fact may be difficult for LEDs and ext r emel y difficult for lasers owi ng to l i neari t y pr obl ems, but we i gnore this for the moment . Under these condi t i ons, it may be shown by the results in Chap. 4 or t hose discussed by Personi ck [9.1] t hat the requi red det ect or power for a 50 dB si gnal -t o-noi se rat i o is appr oxi - mat el y - 26 dB m, under t he best of ci r cumst ances. Gr a nt i ng t hat a l aser mi ght inject ar ound one mi l l i wat t i nt o a fiber, or an LED about a fact or of 10 less, we find t hat our avai l abl e t ransmi ssi on loss is f r om 16 to 26 dB. Actual avai l abl e loss will be subst ant i al l y less t han this because of l i neari t y restrictions, connect or and splice losses, and syst em agi ng and t emper at ur e effects. Ther ef or e our syst em length is severely limited with cabled, spliced fibers whose t ypi cal loss may be 4 to 8 dB/ km. A baseband met al l i c system, on the ot her hand, with a 10 MHz bandwi dt h, an ampl i fi er noise figure of 8 dB, and a 100 mW source, will have an avai l abl e loss of over 78 dB at 10 MHz for an overal l 50 dB si gnal -t o-noi se ratio. In this case, the avai l abl e loss is so high t hat the length is usual l y det er mi ned by ot her things, such as difficulty of equal i zat i on. In any case, the met al l i c syst em ma y compet e well for this appl i cat i on. On the ot her hand, if we consi der a digital syst em empl oyi ng the same bandwi dt h, say ar ound 20 Mb/s, we iliad t hat the requi red received opt i cal power is onl y ar ound - 6 0 dB m ([9.1] or Chap. 4). Her e we have an avai l abl e t r ansmi ssi on loss of 50 dB or great er, and the very low loss of opt i cal fibers real l y begins to have an effect, resul t i ng in very l ong r epeat er spans. A met al l i c syst em in t heor y woul d al so pick up this 30 to 40 dB advant age, but in pract i ce An Illustrative Optical Communi cat i on System 263 this becomes ext remel y difficult to use. The equal i zat i on pr obl em posed by the ] / - f loss of the metallic medi um, spuri ous noise or crosstalk, and similar probl ems tend to limit metallic digital systems to a gain usually well bel ow 80 dB. Under these circumstances, lightguide systems t end to have a cl earcut advant age. 9. 2 Di g i t a l I nt e r of f i c e Tr unki ng 9.2.1 The Telephone Trunk Network Since we describe a system designed for interoffice t r unki ng appl i cat i ons, it is helpful to exami ne the condi t i ons under which the system will operat e and the funct i ons it must perform. ' qnt eroffi ce t runki ng", t aken in its literal sense, can cover distances of from less t han I km to t housands of kilometers. In our case, however, we wish to concent r at e on the typical met r opol i t an or ur ban- s ubur ban area, in which the distance bet ween nearest nei ghbor switching offices tends to average ar ound 5 to 10 kin, dependi ng upon the popul at i on density, and the area of greatest common interest or i nt erconnect i vi t y tends to be encompassed within a region ar ound 100 km in di amet er or less. That is, t he length of these systems rarely exceeds 100 kin, and is rarely less t han t wo or three. The voice grade channel s t hat make up the interoffice t runks are i mpl ement - ed in several ways. They can consist of a simple 4 kHz baseband channel on a single wire pair, t hey can be frequency division mul t i pl exed ( FDM) on wire pairs or coaxial cable, or they can be digital. This last is by far the fastest growi ng met hod, and it will be the focus of our at t ent i on for the r emai nder of this chapter. 9.2.2 Conventional Digital Metallic Facilities The first digital system i nt r oduced to the interoffice t r unki ng pl ant was the Bell System T1, first used in the U.S. in 1962. This is still by far the most widely used, al t hough newer systems are now maki ng t hei r appearance. The T1 system t ransmi t s 24 digitally encoded voice channels by pulse code modul at i on (PCM) over st andar d t el ephone exchange plant paired cable (0.6 mm conduct or s 22 gauge). Since each voice channel requires about 64 kb/s, the transmission rat e of TI is 1.544 Mb/s, which is a st andar d level, DSI , in the Nor t h Ameri can digital hierarchy. Since t hat time t hree new 48 channel systems for oper at i on on paired cable have been i nt r oduced into the U.S. t el ephone pl ant : the Bell System' s TI C [9.2], ITT' s T148, and GTE Lenkurt ' s 9148A system. Also some very high capaci t y digital coaxial cable systems have been i nt r oduced [9.3, 4], as well as a 96 channel system, T2, oper at i ng on high qual i t y pairs [9.5]. 264 J. H. Mullins Since it is i mpor t ant for the under st andi ng of one of the significant advant ages of lightwave transmission, it shoul d be not ed t hat all of these systems except T2 have a designed repeat er spacing of about 2 km maxi mum. That is, electrical repeat ers (or digital regenerat ors) are requi red every 2 km or less. Since the switching office spacing is usually subst ant i al l y great er t han this, the maj ori t y of these repeat ers are in under gr ound housings, or manhol es. Al t hough the part i cul ar choi ce of 2 km is largely historical, being the same as the spacing of i nduct i ve l oadi ng coils on long voice pair t runks, it has nonet hel ess become an ingrained feat ure of the systems. These under gr ound electronics are a cont i nual source of mai nt enance headaches, and any system t hat eliminates them will have a distinct advant age. The reason for the success and rapi d growt h of digital t ransmi ssi on r at her t han anal og FDM systems in these areas, however, has to do mor e with terminal costs t han with repeat er electronics. Systems as short as the ones we are describing t end to be heavily weighted by terminal costs. FDM systems must bear the cost of the FDM terminal, and PCM systems must bear the cost of the A/ D conver t er and digital multiplexer. Technol ogy has been favori ng the digital appr oach for some time, and with t he rapid progress in digital circuit i nt egrat i on we are seeing, this is a t rend which is accelerating. Digital terminals are cheaper and are likely to become mor e so in time. If we add to the above the enor mous flexibility and power of digital dat a processing, we see t hat digital t ransmi ssi on systems for anal og i nf or mat i on make a great deal of sense, and are likely to become even mor e domi nant in the future. This is f or t unat e for lightwave systems, for, as we have seen, they funct i on best as digital facilities. 9. 2. 3 The Phys i c al Envi ronment The physical envi r onment in the t el ephone plant is ext remel y hostile. Most trunks, especially in met r opol i t an areas, are carri ed in under gr ound ducts with the cable spliced in manhol es, some of which also house the el ect roni c repeat ers (cont ai ned in wat er pr oof cases). These condui t and manhol e facilities are often shared with ot her services, so t hat the t el ephone cables and equi pment are beset by ext remel y r ough t reat ment , with water, soil, steam, hydr ocar bons, and somet i mes corrosi ve chemicals present, not to ment i on the physical abuse which derives fi'om frequent activity in the manhol es. An unpr ot ect ed optical fiber is a poor mat ch for this envi ronment . Consequent l y, all lightguide cables designed for this use must be prot ect ed by rugged sheaths, which prevent exposure of the fibers to the hostile surroundi ngs. This includes t he splices. On the ot her hand, one cannot over do the prot ect i on, since one of the chi ef advant ages of lightwave systems is the very small cross section for a given i nformat i on capacity. This tends great l y to relieve the ext reme crowdi ng of the under gr ound duct system (and manhol es) in many of our ur ban areas. An Illustrative Optical Communication System 265 A rel at ed physi cal advant age grows out of the happy fact t hat the low specific t r ansmi ssi on loss of lightguides, coupl ed with their use in digital facilities, al l ows very l ong spaci ngs bet ween repeat ers. Thi s reduces great l y the numbe r of r epeat er s t hat raeed to be housed in manhol es, an advant age r emar ked upon ira Sect. 9.2.2. In fact, ira ma ny cases it ma y be possi bl e to house the r epeat er s ent i rel y ira t el ephone offices or ot her above- gr ound st ruct ures. Thi s is an i mmense advant age to the t el ephone compani es f r om t he st andpoi nt of mai nt enance and i nst al l at i on experase, as well as reliability. 9.2.4 Comparative Economics It is difficult to be very quant i t at i ve about the compar at i ve costs of met al l i c and fiber systems, since the costs depend so much on the model used, the met hod of appl i cat i on, and the cost s of the opt i cal el ement s t hemsel ves, which are not yet compl et el y det ermi ned. However , as t he anal ysi s in Sect. 9.1.2 has shown, the domi nant par amet er , at least for r epeat er ed lines, is the i nf or mat i on capaci t y per lightguide. The quest i on is whet her fibers demons t r at e adequat e capaci t y, and whet her this capaci t y is requi red in the par t i cul ar appl i cat i on. A si mpl e exampl e may serve to pin down the r ange we are deal i ng with. A moder n syst em (the Bell Syst em' s T1C), when it makes use of s ome recent l y avai l abl e exchange ar ea pai red cabl e of i mpr oved charact eri st i cs, can put some 3 Mb/ s on each of 1800 pairs. Thi s cabl e is about 9.6 cm in di amet er, which will essentially fill a s t andar d t el ephone duct. Taki ng t wo- way t r ansmi ssi on as our st andar d, we classify this syst em as about 2700 Mb/ s. Si mi l ar number s woul d be obt ai ned for a new coaxi al system, T4M. (The l at t er is a bit mi sl eadi ng, since the cabl e used for t hat syst em coul d pr obabl y easily handl e doubl e the capaci t y of T4M. ) Speaki ng very broadl y, we find t hat a t el ephone duct full of conduct or s has an installed cost of ar ound 30 to 40 dol l ars per met er, al t hough coax ma y run higher. If so, our cabl e above yields a cost in the nei ghbor hood of $13 per t wo- way k m- Mb ( TWKM) . Now if we admi t some ranges of gains (and costs) we mi ght concl ude t hat met al l i c medi a yield us a range of, say, $7 to $25 pe t. TWKM, at least in t el ephone appl i cat i ons where fairly large t ransrni ssi on cross sect i ons ar e empl oyed. How do fibers c ompa r e ? Of course, the installed costs of fibers are not well det er mi ned as yet, but i ndi cat i ons ar e now t hat the costs under high vol ume condi t i ons will range f r om per haps 5 to 20 cents per fiber met er, installed. Taki ng 100Mb/ s as a comf or t abl e wor ki ng upper limit for fiber capaci t y, at least of current gr aded- i ndex design, we concl ude t hat fibers coul d yield costs in the r ange of $1 to $4 per TWKM : cheaper t han met al l i c by a fact or of f r om as little as 2 to as much as 25. It is t hus safe to concl ude t hat fibers shoul d beat out met al l i c conduct or s in "specific" costs. However , the si t uat i on is compl i cat ed by the fact t hat fibers do 266 J. H. Mullins not subdi vi de conveni ent l y as do met al l i c medi a, a fact which can be viewed as bot h an advant age and a di sadvant age. The i nf or mat i on car r yi ng capaci t y of a fiber is essentially i ndependent of its area, and t here is no cr osst al k pr obl em. So since smal l er fibers are cheaper the economi cs push us t owar d fibers as small as possible. Splicing and handl i ng consi der at i ons place a limit to this. The net result is t hat there is an economi c l ower limit to the " qua nt um" of i nf or mat i on rat e for fibers for a given set of syst em c ompone nt costs. For exampl e, digital carri er qual i t y t el ephone exchange pai red cabl e gets installed t ypi cal l y for ar ound 2cent s/ met er or less. Therefore, fibers are not likely, at least in the near future, to compet e well if t hey handl e the same dat a rat e as pairs. Based on the admi t t edl y br oad number s above, we woul d expect to requi re each fiber to car r y several t i mes as much as a pai r of wires, per haps as much as a fact or of 10 or more. The above is highly simplified, of course, and does not t ake i nt o account the l onger fiber r epeat er spaci ngs and smal l er duct occupancy. However , these fiber advant ages t end to be offset by cert ai n pr obl ems, havi ng mai nl y to do with new t echnol ogy and field handling. As a result, at least at the outset, it appear s t hat the t el ephone pl ant woul d be bet t er served by fibers bei ng used at a relatively high dat a rate. Such consi der at i ons have led to the design of the exper i ment al interoffice t r unki ng syst em descri bed in the following sections. The syst em oper at es at 44.736 Mb/ s, which is a st andar d level in the Nor t h Amer i can digital hi er ar chy (see Sect. 9.3.1). Thi s is high enough, it is felt, to make economi c prove-i n very likely, yet not so high, at least in the U.S. t el ephone net work, as to di scour age wi despread use. One mi ght s ummar i ze the requi red charact eri st i cs as fol l ows: a) Hi gh i nf or mat i on capaci t y per fiber ( > 20 Mb/ s) ; b) Long r epeat er spaci ngs ( > 7 kin) ; c) Very small cabl e size ( ~ i cm di amet er). We will see t hat the descri bed syst em comes r easonabl y close to these charact eri st i cs. 9. 3 The Chicago Lightwave Project, System Components 9.3.1 Basic System Parameters The system, as has been not ed in Sect. 9.1.1, is a "l ong", "di gi t al " syst em [-9.6, 7]. It is desi gned to oper at e as a mul t i pl y r epeat er ed line at a digital rat e of appr oxi mat el y 45 Mb/ s. Its pr i mar y use is for t el ephone interoffice t r unki ng, at which t i me it oper at es at 44.736 Mb/ s, the DS3 level in the Nor t h Amer i can digital hi erarchy. However , it can al so be used to car r y vi deo and dat a, and these appl i cat i ons are al so being tested in Chi cago. In or der to t ake ma xi mum advant age of t he l ong r epeat er spaci ng the syst em empl oys APD det ect or s and An Il l ust rat i ve Opt i cal Communi cat i on Syst em 267 I E L E C T R I C A L I O P T I C A L .,I Fig. 9.1. Basic el ement s of u l i ght wave t r ansmi ssi on syst em ! [ I TERMINAL OPTI CAL LIGHTGUIDE LIGHTGUIDE (ELECTRICAL REPEATERS INTERFACE CABLE ONLY) ( TRANSMI T- ( PAT CH ( FI BERS IN RECEIVE) FI ELD) SHEATH) I . R , O I ' I R E a O E R I Fig. 9.2. Typi cal equi pment TRANSMITTING ) , i S C R ' B ' E R V - - ' , R , " M , T T E " t ermi nal RECEIVING io O E - S C R , M B , E . I - I . . . . . . . . I - I . O . I , J - - FRANIt= R* ~ * PLUG-INS USED IN ERROR * OTHER SYSTEMS PROCESSOR GaAI As injection lasers as sources. To handl e this high rate and long spacing it is necessary to use graded-i ndex fibers. For short spans it is possible to subst i t ut e the simpler LED source for t he mor e compl ex laser, and this is being done in one span of t he Chi cago trial. However, for the sake of pl ant commonal i t y it woul d seem desirable always to use the high qual i t y graded- index fibers. Any compl et e t ransmi ssi on system nmst car r y with it a sat i sfact ory signal processi ng system at the terminal, and a usable set of equi pment and pr ocedur es for mai nt enance, per f or mance moni t or i ng, and repair. Much of this l at t er set of functions resides also in the terminal. Li ght wave systems are not except i ons to these requi rement s, and in fact the maj or por t i on of the expense and compl exi t y of these systems may well be with the terminal functions. However, these have very little to do with optical fiber transmission, so we will ment i on t hem onl y in passing. 268 J. H. Mullins Table 9.1. Chief system characteristics Modulation scheme Code format Information rate Sources Receiver Repeater spacing Fiber type Cable type Number of fibers Splicing Single fiber connectors Rearrangement technique Expected bit error rate General application Digital Binary, uonreturn to zero 44.736 Mb/s GaAIAs laser or LED Silicon APD > 7 km (with laser) ~3km (with I.ED) Graded index, NA~0.23 Ribbon structure Up to 144 Array connectors, factory applied Transfer molded conie-biconic Lightguide distribution panel <10 -7 Interoffice trunking, video, data The si mpl est bl ock di a gr a m of t he t er mi nal , opt i cal , i nt erface, a nd cabl e por t i ons of t he syst em is s hown in Fig. 9.1. We will be c onc e nt r a t i ng mos t l y on t he opt i cal por t i ons , but first wi sh t o emphas i ze t he c o mme n t s in t he pr ecedi ng pa r a gr a phs by r ef er ence t o Fig. 9.2. Thi s r epr esent s t he a r r a y of e qui pme nt used in t he t er mi nal t o al l ow pr ope r i nt er f ace t o t he 44. 736 Mb (DS3) si gnal and t o ensur e pe r f or ma nc e moni t or i ng. I f t he s ys t em is t o be used t o t r ans mi t vi deo, a di fferent t er mi nal a r r a n g e me n t is used, a nd t he r at e ma y be sl i ght l y different. I n t he f ol l owi ng five sect i ons, t he c o mp o n e n t s whi ch ar e mor e r el evant t o l i ght wave t r ans mi s s i on will be descr i bed. The chi ef char act er i st i cs of t he s ys t em ar e s umma r i z e d in Tabl e 9.1. 9.3.2 Cable and Splicing The cabl e used for this syst ern was ma d e at Bell Te l e phone La bor a t or i e s at At l ant a, Ge or gi a [9. 8], usi ng fibers ma n u f a c t u r e d at t he Wes t er n El ect r i c facility t here. Thes e fibers were ma de by t he Modi f i ed Chemi cal Va p o r Depos i t i on ( MCVD) pr ocess [9. 9], and ar e pr i mar i l y ge r ma ni a dope d silica. Thes e gr aded- i ndex fibers have a numer i cal aper t ur e of a ppr oxi ma t e l y 0.23, a cor e di ai net er of 55 lam, and an out e r di amet er of l l 0 g m. The y have a t ypi cal l oss at 0 . 8 2 g m of a r o u n d 4 d B/ k m, and b a n d wi d t h f or a si ngl e ki l omet er of fi ber of a r o u n d 600 MHz , t ypi cal l y. The fibers ar e cover ed wi t h a pr ot ect i ve pl ast i c c oa t i ng and as s embl ed in r i bbons of 12 fibers each. These r i bbons ar e t hen as s embl ed in st acks of up t o 12 and t hen ar e hous e d in a pr ot ect i ve s heat h as s hown in Fig. 9.3. The s uppor t i ng or s t r engt h me mbe r s in this a p p r o a c h ar e in t he sheat h, or out er cover i ng. Ot h e r cabl e desi gns [ 9. 10] e mpl oy a s t r engt h me mb e r in t he cent er of t he cabl e, An I l l u s t r a t i v e Op t i c a l C o mmu n i c a t i o n S y s t e m 2 6 9 12ramOD POLYOLEFI N SHEATH SHEATH % TWI NE STRENGTH \ ~ '~ MEMBERS PEJACKET~ ~ ~ ~ ~ ~ ~ ~ i ~ T W O R I B B O N ~.- ~" ~ i PAPER CONNECTOR Fi g. 9. 3. L i g h t g u i d e c a b l e d e t a i l , c o n t a i n i n g t wo r i b b o n s ( " C h i c a g o ' " cable) with t he lightguides sur r oundi ng it. Except for the cent ral st rengt h member, this st ruct ure is similar to t hat of conveut i onal metallic cable. Since this is not i nt ended to be a text on cable and cable designs, we will cont ent ourselves with the observat i on t hat t he cable has been descri bed mor e fully in ot her publ i cat i ons [9.9]. It is sufficient to not e t hat the r i bbon design is i nt ended to facilitate splicing, which is accompl i shed by the use of precision ar r ay connect or s [-9.113. These are i nt ended to be fact ory applied in the usual case, al t hough they can be appl i ed in the field when condi t i ons warrant . They consist of stacks of precision gr ooved chips, in this case made from silicon by use of phot ol i t hogr aphy and preferential et chi ng techniques. The ext remel y high precision t hat results from this process is responsible for the excellent fiber al i gnment and low splice loss achi eved using this technique. At hough the gr ooved chips creat e the precise rect angul ar ar r ay for posi t i oni ng of the lightguides, the met hod places severe demands upon the di mensi onal t ol erances of the fibers themselves, and upon the cent eri ng of the core inside the glass cladding. The connect or splicing met hod is i nherent l y very inexpensive and requires little time and onl y a low level of skill to appl y in the field. Ther e are ot her met hods, such as manual single fiber al i gnment or electric arc fusing of single fibers [9. 12-14] which depend less on t he accur acy of fiber paramet ers, but which call for consi derabl y great er skill and requi re much mor e time per fiber in the field to apply. 9 . 3 . 3 Cabl e Te r mi na t i o ns and Si ngl e Fi ber Co nne c t o r s Fr om the st andpoi nt of pure practicality, one of the most i mpor t ant system requi rement s is t hat of the single fiber connect i on. Whet her we envision lightguide rearrangement s, pat chi ng ar ound a br oken or defective lightguide, or connect i ng up the sources and det ect ors, we see t hat we always requi re some t ype of "single fiber" connect i on t hat is demount abl e. A nondemount abl e 270 J. H. Mullins Biconical Nut / , sl eeve I Plug Snap ring Fig. 9.4. Single fiber connector connect i on coul d in principle be used, but the clumsiness of such an arrange- ment woul d be extreme. Any t rul y practical system will pr obabl y requi re demount abl e connect i ons, and we will see examples in the system described here. Different types of single fiber connect or s have been r epor t ed in the l i t erat ure [-9.15, 16]. One general class which has a number of vari ant s depends upon precision machi ned pieces to insure accur at e al i gnment of the lightguides for low coupl i ng loss [9.16]. Some of these types even r emove the r equi r ement of accur at e cent eri ng of the fiber cor e in the glass cl addi ng by deferring the final machi ni ng or bondi ng until after the fiber is in place in the connect or. One can then use the lighted fiber core to cent er the machi ni ng, grinding, or bondi ng operat i on. Most of the precision piece part connect or s are i nherent l y fairly expensive. The connect or empl oyed in the Chi cago system takes anot her appr oach. Rat her t han dependi ng upon precise pieces, it depends upon a precision transfer mol di ng process [9.15]. The lightguide is mol ded in place into a t apered plug which is designed to but t against an identical second plug inside a mat i ng bi coni c guide, as shown in Fig. 9.4. St abl e mat eri al s are requi red for t he finished parts, so a heavily silica filled epoxy is used. The process depends for its success also on t he accur acy of the dies, accurat e process cont rol , and last but cert ai nl y not least, consi derabl e accuracy in the lightguide, since the cent eri ng is accompl i shed with respect to the out er surface of the fiber, not the core. However, if this t echni que can pr oduce low enough connect or losses, which it appears to do [9.15], it is pot ent i al l y an appr oach which can lead to very cheap connect ors. The connect or s are used in several places in the system. In one appl i cat i on, they are used much in the manner of a connect or on the back of a convent i onal pri nt ed circuit board. The source or det ect or is connect ed by means of a pr ot ect ed opt i cal fiber t o one of the connect or plugs. This plug is fastened to the pri nt ed circuit boar d so t hat it mates with a biconic cont ai ni ng a second plug when the boar d is inserted i nt o the shelf as shown in Fig. 9.5. The bi coni c and An Illustrative Opdcal Communication System 271 Fig. 9.5. Back plane and circuit board single fiber conneclor second pl ug are carri ed on the shelf. Thi s ar r angenl ent t hen al l ows opt i cal and electrical cont act to be made t oget her when the boar d is inserted i nt o the shelf. The obvi ous t ol erance pr obl ems are over come by the use of guidepins, spri ng l oadi ng, and mechani cal l y fl oat i ng shelf har dwar e. The single fiber connect or s are al so used in anot her very i mpor t ant par t of the syst em, the por t i on desi gnat ed as the "l i ght gui de di st ri but i ng panel " (LDP). The ends of the cabl e which a ppe a r at the t ermi nal are made accessible by means of a "f anout ". Thi s is an ar r angement of pr ot ect ed fibers which are collected i nt o an ar r ay connect or at one end, which in t urn mat es with the ar r ay connect or at the end of t he cable. The ot her end of the f anout is compos ed of the plugs of single fiber connect or s, each of which is mol ded ar ound one of the fibers. These plugs are inserted in one side of an ar r ay of biconic pieces which t hemsel ves const i t ut e the LDP, Connect i on is t hen made to the cabl e by means of a single fiber "pat chcor d", which consi st s of a single pr ot ect ed l i ght gui de with a single fiber connect or pl ug on ei t her end. The pat chcor ds can be used to connect the cabl e fiber ei t her to the sources or det ect or s or to ot her cabl e fibers. 9.3.4 Optical Components- Laser, LED, APD The details of t ypi cal devices and associ at ed ci rcui t ry have been discussed in Chaps. 2, 3, 5, so our discussion here will be descri pt i ve only. The Ga AI As laser t r ansmi t t er is of the doubl e- het er ost r uct ur e st ri pe geomet r y, and is capabl e of 272 J. H. Mullins Tabl e 9.2. Repealer characteristics Code Rate Source Power into fiber Receiver Sensitivity AGC range Clock pull-in range Clock jitter BW Unrestricted binary, nonreturn to zero 44.736 Mb/s GaAIAs laser or LED - 3 dB m, average of 1 and 0 (laser) - 20dB m, average of 1 and 0 (LED) Avalanche photodiode - 53dBm average into fiber for 10 ~ bit error rate 70dB electrical (35dB optical) +5Ml l z 60kHz injecting appr oxi mat el y 1 mW of peak power into the fiber. St abi l i zat i on is accompl i shed by feedback, utilizing t he light from the back side of the laser i nt o a p- i - n di ode t o establish stable operat i on. The LED, when it is used instead of the laser, is a doubl e- het er ost r uct ur e Burrus type, with a direct (nonfeedback) driving circuit. The det ect or is an aval anche phot odi ode (APD). At the high bit rat e empl oyed here, t here is 10 to 12 dB of advant age gained by empl oyi ng an APD r at her t han a p- i - n di ode det ect or. This is well wor t h the addi t i onal compl exi t y of the high vol t age circuit. An addi t i onal use is made of the APD, in provi di ng some aut omat i c gain cont r ol (AGC). The mul t i pl i cat i on fact or is al l owed to range from ar ound t wo up to about 100, dependi ng upon the st rengt h of the received signal. The usefulness of this is expl ai ned more fully in the next section. A summar y of the charact eri st i cs of t he repeat er is found in Tabl e 9.2. 9. 3. 5 El ect roni c Co mpo ne nt s - Regenerat or A bl ock di agram of the r egener at or is shown in Fig. 9.6. For purposes of this proj ect the r egener at or (which we define as a one- way repeat er) was i mpl ement - ed upon t hree separat e plug-in modules, cor r espondi ng t o the t hree functions shown in Fig. 9.6. This was done for conveni ence only, however, and is not fundament al to the design. Pract i cal designs will pr obabl y include all of a r egener at or on a single modul e, or perhaps even t wo regenerat ors on a single modul e to make a compl et e repeater. The functions are divided i nt o "receiver", "deci der", and "t ransmi t t er". The receiver includes the APD, its preamplifier, and all the AGC. It is the linear channel of the repeat er and provides a great deal of power gain - over 80 dB, if the mul t i pl i cat i on of the APD is included. Gai n cont r ol is achieved bot h t hr ough t he APD and with a convent i onal AGC amplifier. Toget her, t hey provi de appr oxi mat el y 70dB of (electrical) cont rol , which is set by a peak det ect or on the out put of the receiver to An Illustrative Optical Communication System 273 RECEIVER \ / DECIDER DETECTD & PREAM CONTR( \ / - - TRANSMI TTER ---~ , I Fig. 9.6. Regenerator block diagram st andardi ze the out put pulses to appr oxi mat el y 1 volt peak to peak. The AGC is requi red in or der to make the r egener at or i nt o a practical field component . This r egener at or aut omat i cal l y adjusts itself to a wide range of operat i ng condi t i ons, t hat is, to high loss and low loss spans, wi t hout the need for external opt i cal at t enuat or s or adj ust ment s, or, for t hat mat t er, even for measur ement of the opt i cal loss. The large range is needed because of the "squar e law" nat ur e of t he det ect or. The 70dB electrical range cor r esponds onl y to 35 dB of opt i cal range. Discussion of the details of receiver characteristics which det ermi ne the sensitivity is cont ai ned in Chap. 4. Recei ver characteristics are summari zed in Tabl e 9.2. One especially i mpor t ant poi nt about this design is t hat t here is no equal i zat i on. The filter charact eri st i cs are purel y t hat of a low pass filter to limit noise. Thi s implies t hat the fiber is t aken as possessing onl y flat loss - no significant dispersion. The deci der cont ai ns the convent i onal r egener at or functions of clock ext ract i on and decision. Al t hough it has not hi ng t o do with the opt i cal transmission, as such, the deci der does empl oy a novel f r equency/ phase l ocked l oop [9.17]. This allows the r egener at or to have a very nar r ow j i t t er bandwi dt h and yet be able to handl e a reasonabl e vari at i on (_+ 10 %) in symbol rate. The t ransmi t t er, as has been ment i oned in Sect. 9.3.4, empl oys a laser with opt i cal feedback cont rol . Thi s allows the system to oper at e over a wide range of t emper at ur es (0 to 50 C) and to t ol erat e a qui t e subst ant i al aging change. 9.3.6 Test Equipment No t ransmi ssi on system is compl et e wi t hout a set of accompanyi ng testing procedures and equi pment . Some are common to all repeat ered lines, involving per f or mance moni t ori ng, fault l ocat i ng equi pment , repeat er testers, and so forth. For our purposes here, however, we wish onl y to ment i on t wo instru- ment s which are peculiar t o optical fibers and which were used in the 274 J. H. Mullins i nst al l at i on and testing of the syst em in Chi cago. These are a loss measur i ng i nst r ument and an opt i cal t i me domai n refl ect omet er. The loss measur i ng test set ( LMTS) is a st r ai ght f or war d i nsert i on loss i nst rument . It consi st s of a source, det ect or, and s ome lightguide i nt erfaci ng technique. In the case of the i nst r ument used in t he Chi cago i nst al l at i on, the i nt erface consi st ed o f a f anout si mi l ar to t hat descri bed in Sect. 9.3.3. One end of the f anout mat ed with the cabl e ar r ay splice, and the ot her end t er mi nat ed in single fiber connect ors. The l at t er coul d be pl ugged i nt o ei t her the source or det ect or of the LMTS. In pract i ce, the i nst r ument is cal i br at ed by i nsert i ng a s t andar d fiber pat chcor d bet ween source and det ect or and "zer oi ng" the i nst rument . Then t he unknown replaces the st andar d pat chcor d and the added loss is not ed. Obvi ousl y, when an installed piece of cabl e is being measur ed, the source and det ect or mus t be separ at ed by a consi der abl e di st ance, or else t wo i nst r ument s must be used, which are careful l y cross cal i brat ed. In ei t her case, qui t e consi der abl e st abi l i t y is requi red of t he sour ce and det ect or. To gain addi t i onal sensitivity, the usual t echni ques of such i nst r ument s are awfilable. For exampl e, the source can be modul at ed at a fixed (usually audi o) frequency, and a nar r ow- band filter empl oyed at the receiver, in any case, the i nst r ument requi res access to bot h ends of the cabl e for best results. Some l oop- back t echni ques coul d be appl i ed, but these reduce the sensitive range by a fact or of t wo and compl i cat e the cal i brat i on. The ot her i nst rument , which does not requi re access to bot h ends of the t mknown, is the opt i cal t i me domai n r ef l ect omet er ( OTDR) . Al t hough this i nst r ument does not have the r ange or precision of loss meas ur ement possessed by the LMTS, it does pr ovi de some loss measur ement , plus a weal t h of localized i nf or mat i on such as fiber defects, breaks, posi t i on and condi t i on of splices, and regi ons of high loss. The ability of the i nst r ument to det ect reflective br eaks is si mi l ar to t hat of a convent i onal electrical T DR; t he loss measur e- mcnt ability derives from its ability to det ect and measur e the backwar d Rayl ei gh scat t eri ng [9.18, 193. A bl ock di agr am of the i nst r ument t hat was empl oyed in the Chi cago i nst al l at i on is i l l ust rat ed in Fig. 9.7. The l aser source is a short pulse, l ow dut y cycle, high peak power Ga AI As laser. The det ect or can be ei t her an APD or, for mor e range, a phot omul t i pl i er . In practice, the i nst r ument is usual l y connect ed to the cabl e at one t er mi nat i on, for exampl e, at the LDP (see Sect. 9.3.3) via a single fiber connect or. By use of a si mpl e scope di spl ay of the r et ur ni ng signal, one is abl e to l ocat e splices, breaks, or the cabl e ends out to l ong di st ances - often cor r espondi ng to r ound t ri p fiber losses gr eat er t han 60dB. By the use of different laser pulse wi dt hs it is possible to t rade off sensitivity for selectivity. That is, a shor t pulse localizes breaks accur at el y but t ol erat es less fiber loss, whereas a l ong pulse ret urns a st r onger reflection but gi~ es a less well-localized response. In any case, using the si mpl e scope display, one can general l y localize a splice or br eak to within a few meters. SYNC LASER PULSER An I l l ust r at i ve Opl i cal Co mmu n i c a t i o n Syst em 275 COUPLER FIBER DETECTOR A % . I N , PMT DELAY, LOSS J DISPERSION.,, PROCESSOR SCOPE Fig. 9. 7. Opt i cal t i me d o ma i n r et l ect omet er , bl ock di a gr a m 24 20 16 d> 12 o~ 8 4 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 LENGTH---KILOM ETERS Fig. 9. 8. Loss per uni t l engt h of a 7 0 0 m fiber, as me a s ur e d by t he opt i cal t i me d o ma i n r ef l ect omet er When the OT DR is used to measur e loss, tile backwar d Rayl ei gh scat t er i ng is s moot hed by a boxcar i nt egr at or or si mi l ar t echni que and t he resul t i ng funct i on di fferent i at ed and normal i zed. Thi s gives a measur e of the loss per uni t length. Fi gure 9.8 is the gr aph of the local loss of a 700 m fiber, deri ved f r om meas ur ement s made by the OTDR. Thi s is a par t i cul ar l y i nhomogeneous fiber, but serves to i l l ust rat e the utility of the i nst rument . Losses in splices can al so be measur ed quant i t at i vel y by a meas ur ement of the difference in backscat t er i ng before and aft er the splice. The range of sensitivity of the i nst r ument can obvi ousl y be ext ended by signal aver agi ng techniques. The source is a repetitive pulser, so t hat a much mor e accur at e meas ur ement of ei t her the reflection or the backscat t er can be obt ai ned by use of a gat ed i nt egrat or. Thi s is a circuit which opens a del ayed gat e aft er the injection of each pulse. The reflection appear i ng in this gat e is 276 J. H. Muffins FDP VAULT, FRANKLIN C.O. MH MH : "| -f--CABLE SPUCE ] \ C K C O O O.6mi =0.9km 1 . 0 mi = 1. 6 k m / / / , . I V WABASH AULT I C.O. Fig. 9.9. Rout e of the Chi cago lightwave system i nst al l at i on accumul at ed over many pulses. By use of such aver agi ng t echni ques it may be possi bl e to pr oduce an i nst r ument t hat will per f or m at least the field loss meas ur ement s which are now per f or med by the loss measur i ng set, as well as t aki ng care of the local fault det ect i on needs. 9. 4 Configuration and Installation of the System 9.4.1 Route Description The rout e of the Chi cago l i ght wave syst em is i ndi cat ed schemat i cal l y in Fig. 9.9. The onl y manhol es shown oil the di agr am are the five ill which splices were made. Ther e are act ual l y 32 manhol es al ong tile rout e, as descri bed below. The syst em connect s t wo cent ral offices, (CO), Wa ba s h and Frankl i n, and one subscri ber l ocat i on, the Brunswi ck building, via under gr ound t el ephone ducts. The t ot al length of the run is appr oxi mat el y 2.5 km, of which appr oxi - mat el y 0.9 km lies bet ween the Brunswi ck bui l di ng and the Frankl i n CO, and 1.6 bet ween the Frankl i n and Wabas h COs. As i ndi cat ed by Fig. 9.9, t here ar e 12 cabl e splices, count i ng tile ones to the fanout s in the equi pment bays, all using the ar r ay connect or s discussed in Sect. 9.3.2. Five of these are in manhol es, t hree in cabl e vaults, and the r emai nder at the equi pment bays. Ther e arc a t ot al of ten cabl e sections, all precut to length and connect or i zed before installation. Al ong the r out e itself t here are 32 manhol es, onl y five of which cont ai n splices. The cabl e was pulled t hr ough the others. The l ayout and rout e condi t i on was per haps t ypi cal of congest ed met r opo- litan areas. All manhol es were in streets, and were often very deep, SOlnetimes An Illustrative Optical Communication System 277 F ' E o ~ - T v / A t l A s , c . o . 1 v i d e o I ~ . I I v , o ~ o I I IIASEIANO': ~ 1 I v , D ~ o I J ' - ' ~ ' ~ I I I I A S E I I A N D I " 1 E N C O D E R I V I V I D E O I. ~ -' BASEIlAND r i o~'~R i ~ 1 I I I I I L _ _ _ _ - L . . . . r l i l T I . EAOOU~RTE. S I I L I I U I I D I N O J ' ~ I I B R U N S W IC K B U IL D IN G I I 1 I i I l ( T R U N K ]PO TS & D A T A ) l _ _ _ _ . J L _ _ _ _ _ I _ _ - - I ._ - - . . . . .~ , PROTECTEO n L , ' ~ _ _ ~ L . . i o o s - - ' ) L D D S . u s O F F iC ~ j Fig. 9.10. Syslcm block diagram as deep as 10m. Because of the pr oxi mi t y of the Chi cago River, among ot her things, (the river was within a bl ock al ong part of the route) the manhol es were usually part l y filled with water. The state of congest i on of manhol es and duct s is a good sample of the t ype of envi r onment t hat will nor mal l y be encount ered. It provi des a real test for the t echnol ogy. 9.4.2 System Configuration Fi gure 9.10 is a di agram of the system fron-I t he poi nt of view of services. The open triangles represent lightwave t ransmi t t ers or receiveJ's. The lightwave line here is denot ed by the desi gnat i on FT3. The F stands for fiber and t he T3 is a remi nder that the line is operat i ng at the t hi rd level, in anal ogy with T1. Ther e are several services carri ed on the system, but perhaps the simplest use is t hat for digital interoffice t runks, as illustrated in the link between the Fr ankl i n and Wabash COs. This empl oys a st andar d Bell System digital multiplex, the M 13, to interface the DSI , or T 1, sources. The spare line shown is aut omat i cal l y switched in when the per f or mance moni t or indicates unsatisfac- t or y er r or performance. The FT3 line can carry 28 DSI lines (672 voice circuits), al t hough onl y 24 are i mpl ement ed in this installation. The link between the Frankl i n CO and t he Brunswick building carries a mix of subscri ber t el ephone traffic (POTS), voi ceband dat a on digital subscri ber 278 J. H. Mullins carri er and digital dat a service (DDS) data. Al t hough a fully-loaded FT3 line coul d car r y 1120 voice circuits using the par t i cul ar subscri ber carri er empl oyed here (Bell System SLC-40), onl y 78 voice circuits, one 2. 4kb/ s voi ceband dat a circuit and t wo 4.8 kb/s DDS lines are act ual l y implemented. Thi s is all placed on t hree of the avai l abl e 28DS1 channel s in the M13 feeding this link. A st andby FT3 line is used here, as well. The ot her FT3 links are used for Pi ct ur ephone Meet i ng Service (PMS). This AT&T service uses a commerci al 4 MHz bandwi dt h monoc hr ome TV signal for conferences. For purposes of t ransmi t t i ng the signal over t he FT3 link, the video signal is digitally encoded i nt o a 4-bit companded differential PCM with a sampling rat e of about 11 MHz. Two t wo-way vi deo links are provi ded. One runs from the subscri ber building (Brunswick) t o the Wabash CO, which itself is adj acent to the television oper at i ng cent er (TOC). At this poi nt connect i on is made to the nat i onwi de net work. The opt i cal fibers on this link are pat ched t hr ough at the Fr ankl i n CO at the lightguide di st ri but i ng panel, using no electronics. The second, short er, PMS link serves a public r oom in the Illinois Bell Tel ephone (IBT) headquart ers, and runs f r om the Fr ankl i n CO to the Wabash CO. Since this is an eval uat i on and not i nt ended to be per manent , all FT3 links are backed up by convent i onal facilities. In part i cul ar, the vi deo links are backed up by aut omat i cal l y switched systems using shielded vi deo pairs. 9. 4. 3 I ns t al l at i on o f t he Cabl e The installation of the cable itself, per haps one of the most crucial tests of the system from a practical viewpoint, was the r esul t , of a col l abor at i on bet ween personnel from Bell Labor at or i es (BTL), West ern Electric (WE), and Illinois Bell (IBT). The cable sections, 10 in number, were precut and connect or i zed by BTL personnel before installation. Two shor t i nt rabui l di ng connect or i zed opt i cal cables were installed, by WE personnel, from the cable vaul t to the equi pment bay in t he Frankl i n CO and a similar i nt rabui l di ng cable was installed in the Wabash CO. The remai ni ng seven connect ori zed cables were installed in under gr ound ducts al ong t he r out e with five out si de pl ant manhol e splices l ocat ed as indicated in Fig. 9.9, Since t here was no cable vault in the Brunswick Building the cable entering the building ran directly to the equi pment bay. Pr i or to the installation of the 12 mm OD optical cable a pol yet hyl ene i nner duct with an out si de di amet er of 24 mm was installed inside the old existing tile duct by I BT personnel. The i nner duct provi ded a cont rol l ed envi r onment for the opt i cal cable as well as a simple met hod of pressurizing ar ound the opt i cal cable. No probl ems were encount er ed in installing and splicing the i nner duct. The opt i cal cables were pulled i nt o the i nner duct and spliced by BTL with the assistance of IBT personnel. Special opt i cal cable installation equi pment was An Illustrative Oplical Communication System 279 designed and built by BTL including sheaves and a special reel and reel handl i ng assembly which allowed cable to be paid out in opposi t e di rect i ons from an i nt ermedi at e point. The t wo-way cable pull reduces the cable pulling t ensi on when cables are placed. All ten cables were pulled in wi t hout any fiber breakage in the installation process. The five cable manhol e splices and seven inside building cable splices were all of the gr ooved chip fiber ar r ay vari et y (see Sect. 9.3.2). The optical splices were enclosed and prot ect ed by a special case inside a modified st andar d t el ephone cable splice case. The splice cases also permi t t ed pressuri zat i on cont i nui t y of the i nner duct. 9.4.4 Installation of Equipment The equi pment was installed by BTL, WE, and 1BT personnel as well, and proceeded most l y in a convent i onal manner. In fact the maj ori t y of the equi pment , from the st andpoi nt of bot h quant i t y and cost, was st andar d t el ephone equi pment and was handl ed in the usual manner. The onl y excep- tions were the lightguide and rel at ed component s, such as the lightguide di st ri but i ng panels (LDP). These were installed by BTL personnel. Even the installation of the regenerat ors, because of the single fiber connect or arrange- ment descri bed in Sect. 9.3.3, was a simple plug-in operat i on. Ther e are, of course, places where special care must still be observed. Lightguides are very small structures, and it takes very little di rt or ot her cont ami nant s to bl ock or degrade a single fiber splice or connect i on. The single fiber connect or s must be clean before t hey are mated, and the mass (array) splice still requires care and a relatively high skill level. Experience and furt her devel opment are expect ed to case these remai ni ng probl ems consi derabl y. Experience with this first installation gives reason for opt i mi sm that the equi pment installation in the fut ure will be a r out i ne and simple thing. 9. 5 Pe r f o r ma nc e and Res ul t s 9.5.1 Cable Characteristics The fibers and cables used in this installation are similar to t hose empl oyed in an earlier system experi ment carri ed out in 1975 at the Bell Labor at or i es- West ern Electric facilities in At l ant a I-9.20 22]. The cable used for t he earlier experi ment cont ai ned 144 fibers (12 ri bbons of 12 fibers), ot herwi se it was very similar to the 24 fiber cable used in Chicago. The per f or mance of the cabled fibers in bot h cases is illustrated in Fig. 9.11. We observed a significant i mpr ovement in fiber loss characteristics, from the At l ant a to the Chi cago cables, br ought about part l y by slightly l ower losses in the fibers themselves and part l y by lower mi cr obendi ng losses in the cabl i ng 280 , I . t t . M u l l i n s 3 0 2 4 6 25 i . u - - 2 0 u . 15 =E z 10 35 B 10 12 .82/u.m CABLE LOSS IdB/km) A T L A N T A EXP. - 1 9 7 5 1 3 8 T R A N S MI T T I N G FI BERS ME A N = 6 . 0 d B / k m CHI CAGO PROJECT - 1976 70 TRANSMI TTI NG FIBERS MEAN -- S. l d B/ k m 14 Fi g. 9. 1 I . P e r f o r ma n c e o f t he ca bl ed f i be r s Z , 6 7 8 WA B A S H- F RA NK L I N LOSS ( d B / k m) AT 0.82,u.m F RANKL I N- BRUNSWI CK LOSS ( d B / k m AT 0. 82, u. m) ~. ASH--FRANKLI N: 1. 62 km = 8. 3 d B / k m FRANKLI N- BRUNSWI CK: 0. 94k m 8. 8 d B / k m 9 1 0 11 Fig. 9. 12. l nst ~l l l cd ~md s pl i c e d c a b l e l os s e s process. In the At l ant a cabl e the added mi cr obendi ng loss was about L.5 dB/ km, whereas in the Chi cago cabl e it was nearer to 0.7 dB/ km. However, cabl e losses al one are not the entire st ory. Splice losses, in an i nst al l at i on such as this, play a significant role ill det er mi ni ng the final span loss. In the Chi cago i nst al l at i on tile aver age cabl e section was about 350m. An Illustrative Optical Communication System 281 This length is det ermi ned, not by the manuf act ur i ng process, which is capabl e of pr oduci ng cables of 1 kin or perhaps much great er in length, but by the nat ur e of the r out e itself. A typical met r opol i t an r out e is made up of many short sections owi ng to manhol e pl acement , r out e congest i on, access re- strictions, and ot her items t hat have very little to do with the t echnol ogy. Consequent l y frequent splicing is unavoi dabl e and splice loss is a key item. Even t hough the splices are r emar kabl y good and losses average less t han 0.5 dB per splice, the effect may be large. In this 2.5 km rout e, for example, t here are 12 splices, count i ng the ones at the equi pment fralnes. The result of this is seen in Fig. 9.12, which gives the effective loss in dB/ km for the installed, spliced cable in Chicago. The average increase over the unspliced cable amount s to mor e t han 3 dB/ km. Clearly the reduct i ons of cable loss proj ect ed for the fut ure must be accompani ed by lower splice losses, longcr cable sections, or both. 9.5.2 Electro-Optical Performance The per f or mance of the system as a whol e has been r emar kabl y good. The er r or per f or mance is out st andi ng, which is not surprising consi deri ng t hat the spans are qui t e short compar ed to the capabilities of the system. The longest span, the Brunswick to Wabash PMS link, is onl y 2. 5kin, whereas the system (as discussed in Sect. 9.5.3) is capabl e of spanni ng 6 to 7 km with excellent er r or performance. The links are short enough in most cases to requi re the use of some optical at t enuat i on even with the 70 dB AGC (for the lasers onl y t here is an LED link which does not requi re at t enuat i on). Consequent l y, the er r or per f or mance is expect ed to be good, and from the st andpoi nt of the devices, the greatest interest is cent ered on reliability. The lasers and APDs used in this system were also used in the early experi ment in Atlanta. Of the 12 lasers, some 60,000 device hours have been accumul at ed in Chi cago and related tests (as of l Oct ober 1977), with onl y one device failure. The failed device was a laser t ransmi t t er, and the defect was not a failure to emit, but a change of dynami c characteristics which made it unstable, so t hat it pr oduced poor pulses. Since these lasers were of an earlier (1975) vintage, their observed reliability lends suppor t to the encour agi ng report s now being received of long-lived lasers, appr oachi ng 10~'h [9.23, 24]. In general, device per f or mance has been excellent. 9.5.3 Current Performance and Future Prospects The system has been carryi ng coml nerci al traffic since May, 1977, and has been responsible for no service outages. The er r or per f or mance is moni t or ed cont i nuousl y by means of a bit-by-bit compar i son on a single DSI (T1 rate) channel l ooped t hr ough the system. Oil 80 % of the days, the moni t or records 282 J. H. Mullins no errors. Tile average over the ent i re peri od has amount ed to one er r or - second per day, t hat is, one second duri ng which t here is an error. Such out st andi ng per f or mance indicates t hat the i nherent er r or per f or mance is excellent, which is to be expected as discussed in Sect. 9.5.2. Mor e i mpor t ant l y, however, it also indicates t hat the system is relatively insensitive to the vari ous spuri ous electrical di st urbances which abound in t el ephone offices and busy met r opo- litan areas. One of the interesting results of the test is an i ndi cat i on of the repeat er spacing t hat can be expect ed from a real system. Designed repeat er spacing is not simply deri ved from the average cable loss, tile average t ransmi t t er power, and the average repeat er sensitivity. Much mor e goes into it. The role of splice losses has al ready been discussed (Sect. 9.5.1). In addi t i on t here are single fiber connect or losses, device aging, and some addi t i onal margin t hat any wise designer will build into the system to t ake care of unknowns. In addi t i on to all of the above it must be remembered t hat all of these i mpai rment s have distributions, and the designer must set his limits not to the average case (for then hal f of his systems woul dn' t work !) but for some poi nt on the tail of the i mpai r ment di st ri but i on in the di rect i on of "wor st case", the t hree sigma poi nt being a common choice. This comes about because of the nat ur e of a digital system. The dependence of er r or rat e on signal-to-noise rat i o is so steep t hat a very slight degr adat i on of a system oper at i ng near its noise limit will pr oduce t ot al l y unaccept abl e performance. In ot her words, every link of a repeat ered line must perform satisfactorily or the end-t o-end system will not. Even so, based oil t he results of the tests in Chi cago it seems clear t hat the designed spacing will be in excess of 6 kin, with relatively rninor i rnprovement s capabl e of bri ngi ng it up to 7 or possibly 8 [9.7]. Earlier tests in At l ant a indicated this also, with some links being ext ended to over 10kin by use of special low loss splices [9.20, 22], Similar results have been report ed from .lapan [9.25]. 9.6 Concluding Remarks This chapt er has discussed some of the general characteristics of fiber opt i c transmission t hat make it suitable for appl i cat i on in the t el ephone plant, The part i cul ar appl i cat i on emphasi zed, t hat of digital carri er for interoffice t runk- ing, appears to provi de one of the earliest economi cal l y at t ract i ve oppor t uni t i es for widespread applicatiol2. The Bell Labor at or i es system under goi ng a test in the Illinois Bell Tel ephone Company in Chi cago was designed specifically for this kind of appl i cat i on, and was confi gured to be as nearly a "pract i cal " system as the state of the t echnol ogy would permit. So far, results have been very encouragi ng. Ot her systems for the same general ki nd of appl i cat i ons are now being tested in tile USA, in the UK, in Europe, and in Japan [9.25, 26-]. Somewhat An Illustrative Optical Communication System 283 di f f erent c ho i c e s h a v e been ma d e in t he s e t est s wi t h res pect t o c abl e des i gn, s pl i ci ng, c o n n e c t o r des i gn, and i n f o r ma t i o n rate. On e t hi ng s e e ms cert ai n at t hi s s t a g e - fi ber o pt i c s wi l l be appl i e d t o i nt e r of f i c e t r unki ng, a nd s o o n . We expect t hat t he n u mb e r o f t he s e a ppl i c a t i o ns wi l l a l s o g r o w rapi dl y, s o t hat it wi l l be o n e o f t he ma j o r e ar l y us es o f fi ber opt i cs . F r o m t he s t a n d p o i n t o f pure v o l u me , h o we v e r , i nt erof f i ce t r unki ng c a n n o t c o mp a r e t o t he s ubs cri ber o r "l oop" pl ant . The a mo u n t o f c a bl e in t he t e l e p h o n e pl ant d e v o t e d t o t r unks is but a s mal l f ract i on o f t hat us e d f or s ubs cri bers . Co n s e q u e n t l y , wh e n l i g ht wa v e t r a n s mi s s i o n begi ns t o be e c o n o mi - c al l y c o mp e t i t i v e wi t h c o ppe r pai rs in t he s ubs c r i be r pl ant , we c an e xpe c t t o s ee a very l arge g r o wt h in t he mar ke t . T o reach thi s poi nt , be c a us e o f t he di s pe r s e nat ur e o f t he s ubs c r i be r pl ant , it wi l l be ne c e s s ar y f or fi bers t o pr o v e c o mp e t i t i v e wi t h a mu c h s ma l l e r i n f o r ma t i o n f l o w per fiber - pe r haps as l i ttl e as o n e v o i c e gr ade ci rcui t . The n, if we al s o ma i nt a i n t he abi l i t y o f t hat s a me fiber t o ha ndl e a l arge i n f o r ma t i o n f l ow, we wi l l h a v e al s o t a ke n a l arge s t ep t o wa r d t he day o f t he "'wired city". R e f e r e n c e s 9.1 S.D.Personick: In Fumt ament al s ~1' OplicaI Fi her CommunicutioJls, ed. by M.K. Barnoski (Academic Press, New York 1976) p. 172 9,2 J.A. Lombardi, R.E. Maurer, W.P. Michaud, J.P. Fitzsimnaons, W.J.Maybach, A,Anufl; J.F.Graczyk, J.J. Ludwick, V..I, Mackey, J. D. Moore, R.S. Burncll, W.A.Janicki, F.E.Weber, D. V. Anderson, B. B. Garg, P, 1:. M olander : Conf. Record Intern. Collf. oil Conllllun. ( 1 EEE, San Francisco 1975) p. 39-1 9.3 P.E. Rubin, H.I. Maunsell, R.B. Robrock, C.A.Van Roesgen, M.A. Rich, .1.M.Trccker, F. D. Waldhauer, A. l).Close, J.M.Stopyra, W.G. Hamme~t : Conl: Record Intern. Conl~ on Commun. (IEEE, San Francisco 1975) p. 48-1 9.4 Bob Johston, Wally Johnston: Telesis 5, 66 (1977) 9.5 R.B.Moore, T.L.Maione, P. Brostrup-Jensen, J.R.Davis, P,T. Haury, J.l).Moore: Conf. Record Intern. Conf . on Communications (IEEE, Seallle 1973) p. 22-1 9.6 T.k.Maione, D.D.SelI: IEEE Trans. COM-25, 517 (1977) 9.7 M.I.Schwarlz, W.A. Reenstra, J.H.Mullins: Post-l)eadlinc Papers of 1977 lntern. Conf. Integrated Optics and Optical Fiber Comm. ([ECE/IEE Japan, Tokyo 1977) p. 55 9.8 M.I.Schwartz, R.A. Kempf, W.B.Gardner: Proc. 2nd Europ. Conf. Optical Fibre Commun., Paris (1976) p. 311 9.9 J.B, MacChesney, P.B.O'Connor, H.M.Presby: Proc. ll1:,E 62, 1278 (1974) 9. [0 Tech. Dig. Intern. Conf. Integrated Optics and Oplical Fiber Commun. (1ECI,/IEE Japan, Tokyo 1977) H. Murata, T.Nakahara, S.Tanaka: p. 281 N.Vchida, M.Koyama, S. hnao, M. Hoshikawa, T. Naruse: p. 285 A.Cook, S.G. Foord, M. M. Ramsay, A. R.Gilberl : p. 297 9.11 C.M.Miller: Bell Syst. Tech. d. 54, 1547 (1975) 9.12 D.L.Brisbce: Appl. Opt. 15, 796 (1976) 9.13 Y. Kohanzadeh: Appl. Opt. 15, 793 (1976) 9.14 A.Tardy, R.Jocteur: Proc. 2nd Europ. Conf. Oplical Fibre ColllnlUll., Paris (1976) p. 261 9.15 P. Runge, L.Curtis, W.C. Young: Tech. Dig. Topical Meeting on Optical Fiber Transmission 11 (IEEE/OSA, Williamsburg, Va. 1977) p. WA4 284 .1. II. Mul l i ns 9.16 N. Suzuki, M. Koyama, N. Kurachi, Y. Koyama, H. Furuta, S.Oguro : Tech. Dig. Intern. Conf. Integrated Optics and Optical Fiber Commun. (IECE/IEE Japan, Tokyo 1977) p. 351 9.17 J.A. Bcllisio: Conf. Record Intern. Conf. oil Commun. (IEEE, Philadelphia 1976) p. 10-17 9.18 M.K.BarnoskLS.M.Jensen: Appl. Opt. 15, 2112 (1976) 9.19 S.D.Personick: Bell Syst. Tech. J. 56, 355 (1977) 9.20 Ira Jacobs: Bell I,ab. Rec. 54, 291 (1976) 9.21 I.Jacobs, S.E.Miller: IEF, E Spectrum 14, 32 41 (1977) 9.22 R.S. Kerdock, D.II.Wolaver: Conf. Record Nat. Telecommun. Conf. (IEEIZ, Dallas 1976) p. 37-2-1 9.23 T. Kobayashi, K. Wakila, T. Kawakami, G. lwane, Y. Horikoshi, Y. Seki, Y. Furukawa : Post- Deadline Papers of the 1977 Intern. Conf. Integrated Optics and Optical Fiber Conlmun. (IECE/IEE Japan, Tokyo 1977) p. 33 9.24 R.L. Hartman, N. E. Schumaker, R. [,. H artman : Appl. Phys. Lett. (1977) to be published 9.25 Tech. Dig. Intern. Conf. Integrated Optics and Optical Fiber Commun. (IECE/IEE Japan, Tokyo 1977) T. Miki, M.Koyama, H.lshio, K. Aoyama: p. 541 1. Kobayashi, M. Koyama, T. Matsumoto, K. Aoyama : p. 545 K.Nakagawa, Y.Okamo, E. Yoneda, H.Hiramalsu, Y.Ohgushi, Y. Mincjima, A.Miyata: p. 549 9.26 H. Mizugucbi, C.P.Sandbank, H.P. Unger, M. Tr e he ux , C. Boisrobert: Invited Papers Tech. Dig. 1977 Conf. Integrated Oplics and O01ical Fiber Commun. (IECE/IEI~ ,lapan, Tokyo 1977) p. 503 10. Li g ht S o u r c e s - An Up d a t e H. Kressel and M. Et t enber g With 3 Figures Since the compl et i on of Chap. 2, cont i nued effort in light sources for opt i cal communi cat i on has been largely devot ed t o four main areas: 1) Pr oduci ng het eroj unct i on lasers of I nGaAs P/ I nP for emission in the 1.3 to 1.55 jam range ; 2) The devel opment of het er oj unct i on l nGaAs P/ I nP LEDs compar abl e in qual i t y t o the earlier AIGaAs devices; 3) Devel opi ng A1GaAs st ruct ures with i mpr oved lateral mode stability at relatively large power emission val ues; 4) St udy of noise propert i es beari ng on the appl i cat i on of lasers to com- muni cat i on systems. 10.1 InGaAsP]lnP Devices The I nGaAs P/ l nP het er oj unct i on st ruct ures have become of increasing interest for the 1.3 to 1.55 pm spectral region useful for opt i cal communi cat i on. Technol ogi cal progress has led to bet t er devices. The mat eri al synthesis t echnol ogy produces reproduci bl e l at t i ce-mat ched st ruct ures and the ready availability of InP subst rat es is helpful for the l ong-t erm commer ci al viability of these devices. The t hreshol d current densities which are achi eved are appr oach- ing t hose obt ai ned with AIGaAs devices (i.e., mi ni mum values in the 1-2 kA/ cm z range). Al t hough the initial work was at 1.3 pro, t he laser emission is being moved t o 1,55 gm because losses as low as 0. 2dB/ km [ I 0. I ] are achieved in optical fibers at 1.55 pm. With removal of a hnos t all t he wat er from the fiber, the loss is basically due to Rayleigh scat t eri ng which decreases as 24. Thus, t he ul t i mat e loss at 1.55 jam is a fact or of 2 l ower t han at 1.3 jam. A basic absor pt i on mechani sm in silica-based fibers restricts the wavelengths to less t han 1.6 p.m [10.1]. Therefore, longer wavelengths will pr obabl y not be requi red unless ent i rel y new fiber mat eri al s are devel oped. The qual i t y of 1.3 and 1.55 gm emission l nGaAs P lasers has i mproved rapidly. The last r emnant of absor pt i on due to OH in the fiber is current l y near 1.4 jam and thus 1.3 and 1.55 jam devices avoi d 1.4 jam or 1.6 jam absorpt i on. Several single mode index-guided st ruct ures have been devel oped at bot h 1.3 and 1.55 jam [-10.2-10]. Such lasers have been used in demonst r at i on l abor at or y systems t o t ransmi t up t o 800 Mb/ s over 60 km [10.13]. These lasers have been 286 H, Kr e s s e l and M. Etlenberl pr epar ed by liquid phase epi t axy and vapor phase epitaxy. Li qui d phase epi t axy is the current favorite. The lowest t hreshol d cur r ent lasers have been obt ai ned with the buri ed het er oj unct i on st ruct ure grown by liquid phase epi t axy [10.11, 14]. The i mpor t ance of a low t hreshol d cur r ent in I nGaAsP lasers is due to the fact t hat these lasers have a much steeper t emper at ur e dependence of t hreshol d com- pared to A1GaAs lasers. Therefore, the heat i ng of the device can severely limit high t emper at ur e operat i on. Thus, the l ower the operat i ng cur r ent density (and the power dissipation), the better. The buri ed het eroj unct i on st ruct ure is very suitable for I nGaAsP/ l nP. The fabri cat i on is easier t han with AIGaAs because t here is no oxi dat i on of the InP aft er the et chi ng step pri or to the InP regrowt h. Fur t her mor e, the facet damage limit for I nGaAsP is hi gher t han for AIGaAs [10.14] so t hat the small emi t t i ng area of this st ruct ure does not severely limit the out put power due to mi r r or damage. In fact, it does not appear necessary to prot ect the facet with a dielectric coat i ng as is the case for the l ong-t erm reliable oper at i on of AIGaAs lasers. Buried het er oj unct i on 1.3 gm lasers have been made with t hreshol d current s as low as 8.5 mA [ i 0. 11] and t hey have oper at ed at t emper at ur es as high as 95 C [10.14]. The devices have also been life-tested wi t hout significant degr adat i on for peri ods exceedi ng 5000 h at 50 and 70'~C at 5r oW out put [.10.14]. Buried het er oj unct i on lasers have also been made emi t t i ng at 1.551am [10.8, 10, 11]. At this time single mode lasers at 1.3 to 1.55 lam are still in the devel opment stage, whereas mul t i mode devices are commerci al l y available. Reliable single mode 1.3 and 1.55 lam devices shoul d be fully devel oped in the next few years. The main out st andi ng pr obl em with lasers in the 1.3 to 1.55 gm region is t hei r st rong t emper at ur e dependence of t hreshol d current . Thi s dependence is charact eri zed by a T o value, as discussed in the Chap. 2, i.e. [ 1 , , o c e x p ( A T / T o ) ] . AIGaAs lasers exhibit T o values between 100 and 200<'C, whereas I nGaAs P lasers, bot h at 1.3 and 1.55 l~m, have T o values rangi ng from 30 to 70 C, with T o = 6 0 ' C bei ng typical. With To=60 C, an increase in the active region t emper at ur e from 10 to 70<~C increases the t hreshol d cur r ent by a fact or of about 2.7. This active region t emper at ur e increase can be due to ei t her an ambi ent t emper at ur e change or j unct i on heat i ng duri ng cw operat i on. We illustrate in Fig. 10.1 the large effect t hat a T o = 60 C value has on cw oper at i on by compar i ng the pulsed and cw t hreshol ds for I nGaAs P lasers and Al GaAs lasers ( To= 170C). It is evi dent t hat the highest cw oper at i ng t emper at ur e of the long wavelength laser is limited compar ed to the AIGaAs laser. To maximize the I nGaAsP laser oper at i ng t emperat ure, very low t hreshol d current devices or devices with very low t hermal resistance and electrical resistance have to be made. In general, however, it is necessary to limit the heat sink t emper at ur e of the I nGaAsP lasers by t hermoel ect ri c or ot her heat sink t emper at ur e cont rol . The steep t emper at ur e dependence ,~f t hreshol d is t hought to be due to an Auger nonr adi at i ve r ecombi nat i on process [-10.15] which is pr obabl y present in ot her similar low bandgap materials. For example, AIGaAsSb lasers also exhibit the same relatively low T o values [10.16]. Light Sources - an Update 287 7OO 6OO 5 0 0 4OO 5 0 0 2 0 0 I 0 0 0 0 L a s e r I ~ ~ C W ; P u l s e d 1.5/zm 0 . 8 F . m II 3 / J . m I I n G a A s P T = I T O C T = 6 0 C I CW; Pul sed R = 4 D I 0 . 8 F m Zf h ( 2 2 " C ) = 100mA T h e r ma l R e s . = 5 O o C / W I A I Ga A s /cw / / / / / / / / " / P u I s e d J ~ ~ C W .....-. / .~. ~ P u l s e d I I I I I I I I I I I 0 2 0 5 0 4 0 5 0 6 0 7 0 8 0 9 0 100 A mb i e n t T e mp e r a t u r e ( C) F i g . 10.1. Calculated pulsed and cw threshold currents for comparable A1GaAs 0.8/.un and l uGaAsP 1.3 p.m lasers. Values chosen are somewhat worse than the best achievable to empbasize the comparison and to lake into account nonuniformities in real devices - n n I E taJ o Q. I 0 t . . . 8 4 p . m I 0 5 I 0 1 5 2 0 2 5 D I S T A N C E ( k i n ) 50 Fig. 10.2. l)istance possible at various data rates between 2 and 100 Mb/s for a 0.84 p.m laser with a Si APD receiver compared lo a 1.3 gill InGaAsP LED/ I nGaAs PIN receiver system. Parameters used are those for conl- mercially available devices and fibers. Pessimislic wtlues have been chosen for the 1.3 pm components 288 H. Kressel and M. Ettenber 9 Ther e has also been consi derabl e activity in the devel opment of l nGaAs P LEDs because not onl y are the losses in fibers much l ower at 1.3 1.55 gin t han at ~0. 8 lam, but the spectral or mat eri al dispersion in fibers at 1.3 lam is close to zero [10.17]. Since the limitation in t ransmi ssi on rat e and distance in optical fibers is due to a combi nat i on of loss and dispersion, a lowering of bot h allows 1.3 IJm LED- dr i ven systems to compet e with some systems empl oyi ng 0.8 lain laser diodes, as illustrated in Fig. 10.2 [10.18]. I nGaAs P LEDs have been made in the surface-emi t t i ng [10. 19-23] and edge-emi t t i ng confi gurat i on [10.24]. The compar i son of these st ruct ures at 1.3 lam is similar t o t hat at 0.8 lain, as discussed in detail in Chap. 2. Ther e are differences, however, between 0.8 lain and 1.3 lam LEDs. One difference is due to the Auger or ot her nonr adi at i ve r ecombi nat i on in the long wavelength materials. As a result, t here is a sat ur at i on of the light out put [10.25] and bot h the highest radi ance out put and efficiencies of LEDs at 1.3 lain and 1.55 gm is a hal f to a t hi rd of t hose at 0.8 lam. For example, typical 0.8lain LEDs at r easonabl e drive condi t i ons (100-200 mA) can coupl e 100-150 ~tW i nt o 50 lain core, 0.2 NA graded index fibers. By cont rast , less t han 50 laW of coupl ed power is typical for 1.3 lam and 1.55 lam LEDs under similar condi t i ons. The modul at i on rates for 1.3 lain and 1.55 [am LEDs are similar to t hose of 0.8 lain LEDs. However, the spectral linewidths differ. Even for spectral widths as small as 200/~ at 0.8 lam, the di st ance-bandwi dt h pr oduct of a graded index fiber is limited t o less t han 0 . 2 GHz . k m [10.26]. Good graded-i ndex mul- t i mode fibers have a pr oduct of ~ 1 GHz . k m which is limited by modal dispersion. In or der not to significantly i mpact the di st ance-bandwi dt h pr oduct at 1.3 lam, t he source must have a spectral wi dt h of less t han 600/~ [10.26]. Such a nar r ow spectral width can be achi eved in edge-emi t t i ng LEDs [10.24] due to the increased selective absor pt i on in these structures, but it has not been achieved in surface-emitting LEDs where the spectral width tends t o exceed 900/~ [10. 19-23]. LED- dr i ven fiber opt i c systems at 1.3 lam shoul d be compet i t i ve with 0.8 gin laser systems at least up to rates of about 100 Mb/ s (limited by t he modul at i on rat e of t he LED). Fur t her mor e, the LED system' s great advant ages present l y are in reliability and lower cost. In addi t i on to havi ng a si mpl er and t hus mor e reliable driving circuitry, the 1. 3gin LEDs coul d t o be significantly mor e reliable t han typical 0.8 lam lasers. The best 0.8 lain laser reliability dat a predict a medi an time t o failure ( MTTF) at r oom t emper at ur e of 105-106 h and 25,000 h at 70 C [10.17]. By cont rast , dat a for 1.3 gm LEDs predi ct MTTF of > 109h at r oom t emper at ur e and > 105 h at 70C [10.28, 29]. In fact, such devices have been oper at ed for over 10,000 h at 170 C [10.28, 29] maki ng t hem the most reliable light-emitting devices devel oped to date. Since the i nt ernal degr adat i on process of lasers and LEDs is similar, l nGaAs P lasers and LEDs shoul d event ual l y have compar abl e reliability, al t hough this remai ns to be demonst r at ed. At this time, obt ai ni ng long-lived I nGaAsP lasers appears to be subst ant i al l y mor e difficult t han for LEDs. Facet damage is not believed to be a fact or in this difference. Rat her, it is the steep Light Sources - an Update 289 t emper at ur e dependence of the t hreshol d current which makes the laser cw out put very dependent on mi nor changes in the t hermal and electrical resistances. An increase in these paramet ers is reflected in a decrease in the power out put at const ant current owi ng to the increase in j unct i on t empera- ture. Thus, compar ed to AIGaAs lasers, I nGaAs P lasers will show a larger power decrease for equal changes in electrical and t hermal resistance. Not e t hat as the last r emnant s of OH are r emoved from the fiber, the whol e range of wavelengths from 1.2 to 1. 55gm can be used maki ng wavelength mul t i pl exi ng practical with several wavelengths. One should be able to pr oduce LEDs and lasers of I nGaAs P/ I nP of uni f or m charact eri st i cs and qual i t y across the 1.3 to 1.55/am spectral region. For wavelengths between 1.2 and 0.9 gin, anot her mat eri al s)istem will have t o be devel oped such as l nGa P/ I nGa As [10.30] or AIGaSb/ AIGaAsSb. 10. 2 AI GaAs Devi ces The work on AIGaAs lasers has been ai med at the refi nement of si ngl e-mode st ruct ures and increasing their power out put and reliability. In addi t i on, some work has been on movi ng the emission to shor t er wavelengths ( < 0.8 gm). Such lasers are not being designed for fiber optics, because the loss in fibers increases with shor t er wavelengths, but for dat a recordi ng and scanning where the shor t er wavelength allows a reduced spot size. Some of the single-mode st ruct ures which cont i nue to be devel oped for optical fiber systems are the buri ed het eroj unct i on (BH), [10.31] the transverse j unct i on stripe (TJS) laser [10.30], the pi ano-convex wavegui de (PCW) [10. 33-35], t he channel subst rat e pl anar (CSP) [10.36], and the const ri ct ed doubl e het er oj unct i on laser ( CDH) [10.37]. These are illustrated in Fig. 10.3 [10.38]. These devices are capabl e of out put s of a few milliwatts in a single spatial mode. It has been consi st ent l y found t hat lasers which oper at e in a single spatial mode due t o t he built-in dielectric wavegui de operat e in a single l ongi t udi nal mode as well, and are thus true single mode lasers [10.38]. Such lasers are commonl y called "i ndex-gui ded lasers". On the ot her hand, laser st ruct ures such as oxi de stripe, pr ot on- bombar ded stripe, nar r ow stripe [10.39], pl anar Zn diffused [10.40], or V-groove [10.41], are called "gain- guided lasers" because the gain profile pri mari l y det ermi nes the effective lateral waveguide. In general, these devices, while capabl e of pr oduci ng moder at e out put powers (5-10 roW) in single spatial modes, pr oduce out put s with multi- l ongi t udi nal (spectral) modes. Thei r spatial stability with aging [10.42], power out put and t emper at ur e is significantly worse t han for the index-guided devices and this affects the coupl i ng stability to optical fibers. Bot h index and gain- guided lasers have exhibited oper at i ng lives of 5000 to 10,000 h at 70 C which ext rapol at es, assumi ng a 0.6 to 0.9 eV "act i vat i on ener gy" for degr adat i on, to 105 to l 0 b h at r oom t emper at ur e [10.43]. The reliable si ngl e-mode out put of i ndex-gui ded lasers has been increased in some structures. High out put s at 0.83 gm are useful in high dat a- r at e bus 290 H. Kressel a nd M. Ettenherg Zn DIFFUSED p-AIGaAs ~ O X I D E n-GoAs ~ l n-AJ-Go As n-A] GoAs / ,"~.. -~-- 1' n- GaAs I~ ' P " ' ~ SUBSTRATE (o) .METAL (~) ~ n - G a A s ~ ' ~ . ~ - - n- A. I GaA s p* ~ n - G a A s Y / / / / / / / , ~ / J I ~ - n - A , t G o A s i-GoAs 2/zm SUBSTRATE (b) METAL ZnDIFFUSION. .,,~,... /OXIDE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . / n - G a A s r ~ / . / _ / Z / _ / _ / _ / ~ I - - / AIGaAs " , ~ - - ; ~ n - A J G a A s n- GaAs --~ ~,~ SUBSTRATE 5Frn METAL OXIDE ( ~ ~ j ~ - -AJ Ga As P ! _ 7 " - / p - Ga As ~ n-AJ~GaAs n-AJ~ GaAs n- GaAs SUBSTRATE (c) ( d) J 5 F r o ( e) /METAL . . . . . ~ T /OXIDE , p-GoAs p- A. t GaAs A I G a A s n - A~ Go As n - GaAs MI S ORI E NT E D SUBST RAT E Fi g. 10. 3a- e. Some o f t he i ndex- gui de d si ngl e mo d e s t r uc t ur e s : (a) bur i e d b e t e r o j u n e t i o n ( BH) ; (b) t r a ns ve r s e j u n c t i o n s t r i pe ( TJ S) ; (c) p i a n o - c o n v e x wa ve gui de ( PCW) ; (d) c h a n n e l s ubs t r a t e pl a- na r ( CSP) ; a n d (e) c ons t r i c t e d d o u b l e h e t e r o j u n c t i o n ( CDH) distribution systems where distribution losses might outweigh the loss in the fiber. Three structures have produced stable single spatial mode cw outputs in excess of 15 mW: the buried heterojunction large optical cavity laser (BH-LOC) [10.46], the CSP [10.47], and the constricted double heterojunction large optical cavity (CDH-LOC) laser [10.48]. The power of these devices was increased by making the mode larger in the direction perpendicular to the junction either by adding another layer as in the LOC concept or by producing a very thin active region (< 1000A) as was done in the high-power CSP. The CDH-LOC ~levice has produced the highest power in a single-mode" up to 40 mW cw in both single spatial and longitudinal modes and up to 100 mW at 50% duty cycle in the fundamental spatial mode [10.49]. Light Sources - an Update 291 10. 3 Appl i cati ons Are t here advant ages in usi ng onl y i ndex-gui ded lasers in opt i cal com- muni cat i ons ? The answer depends on t he specific appl i cat i on and the pr oper - ties of the avai l abl e gai n-gui ded devices. Li near out put s can be obt ai ned f r om ma ny t ypes of devices up t o a few milliwatts. However , the i ndex-gui ded lasers general l y exhi bi t a shar per t hr eshol d knee wi t h less s pont aneous emi ssi on bel ow t hreshol d. Fur t her mor e, if a power out put in excess of 10 mW is needed in a single spat i al mode, t hen selected i ndex-gui ded lasers are essential. In addi t i on, the i ndex-gui ded lasers exhi bi t a mor e st abl e far-field pat t er n with time, t hus ensuri ng t hat the coupl i ng efficiency to l ow numer i cal aper t ur e fibers r emai ns relatively const ant over the life of the system. Not e, however, t hat the single l ongi t udi nal - mode emi ssi on charact eri st i c of well const r uct ed i ndex-gui ded lasers can act ual l y be det r i ment al in some appl i cat i ons. Modal noise has been descri bed in mul t i mode fibers resul t i ng f r om the det r i ment al sel f-i nt eract i on of a coher ent wave pr opagat i ng t hr ough the fiber [10.44]. Thus, l aunchi ng a si ngl e-mode wave i nt o such fibers is act ual l y det r i ment al t o syst em per f or mance. Anot her pot ent i al pr obl em area with single l ongi t udi nal - mode emi ssi on is t hat caused by opt i cal feedback when t he laser is close t o a reflecting surface (the tip of a fiber, for exampl e). The opt i cal feedback i nt o t he laser can cause i nst abi l i t y in the i nput [10.45]. The effect of the feedback depends on the di st ance bet ween the laser and t he refl ect or relative to the coher ence l engt h of the laser emission. Thus, a single- mode l aser with a l ong coher ence length is mor e difficult to use t han a mul t i - mode device. Finally, not e t hat in Fabr y- Per ot lasers t here is no st r ong i nt ernal gai n- dependent mechani sm for fixing a single l ongi t udi nal mode. Thus, a t emper at ur e change (or any change in the gai n versus wavel engt h profile) will shift the preferred l ongi t udi nal mode. The utility of si ngl e- mode fiber syst ems is likely to gr ow as the bandwi dt h limits are pushed [10.50]. By usi ng opt i cal i sol at ors with si ngl e- mode fibers, the det r i ment al feedback effects can be l argel y avoi ded. Pul se code modul at i on at 8 Gb/ s has been demons t r at ed wi t h A1GaAs lasers [10.51]. Wi t h I nGa As P lasers, modul at i on rat es to 2 Gb/ s have been r epor t ed [10.52]. References 10.1 T.Miya, T.Terunuma, T.Hosaka, T.Miyashita: Electron. Lett. 15, 106 (1969) 10.2 A.Doi, N.Chinone, K.Aiki, R.lto: Appl. Phys. Lett. 34, 393 (1979) 10.3 H.Kano, K.Sugiyama: J. Appl. Phys. 50, 7934 (1979) 10.4 l.P.Kaminow, R.E.Nahory, M.A.Pollack, L.W.Stulz, J.D. DeWintcr: Electron. Len. 15, 763 (1979) 10.5 K.Kishino, Y.Suematsu, Y.ltaya: Electron. Len. 15, 134 (1979) 10.6 R.J.Nelson, P.D.Wright, P.A.Barnes, R.L.Brown, T.Cella, R.G.Sobers: Appl. Phys. Letl. 36, 358 (1980) 10.7 E.Oomura, T. Murotani, H.Higuchi, H.Namizaki, W.Susaki: 1EEE J. QE-17, 646 (1981) 10.8 H.Nagai, Y.Noguchi, K.Takahei, Y.Toyoshima, G.lwane: Japan ,1. Appl. Phys. 19, L218 (1980) 292 II. Kressel and M. Ettenbert 10.9 K. Moriki, K. Wakao, M. Kitamura, K. Iga, Y.Suematsu : Japan J. Appl. Phys. 19, 2191 (1980) I 0.10 K. Utaka, K. Kobayashi, K. Kishino, Y.Suematsu : Electron. Lett. 16, 455 (1980) 10. i1 LMito, K. Kaede, M. Kitamura, K. Kobayashi, S. Mat sushi t a: Tech. Digest IOOC '81, Paper Tu A3, San Francisco, CA (1981) [0.12 K.Sakai, F. Tanaka, Y.Noda, Y. Matsushima, S.Akiba, T. Yamamot o: IEEE J. QE-17, 1245 (1981) 10.13 E. l wa ha s hi : l EEEJ . QE- l ?, 891 (1981) 10.[4 M. Hirao, S.Tsuji, K.Mizuishi, A.Doi, M. Nakamur a: J. Optical Commun. I, 10 (1980) 10.15 G. H. B. Thompson, G. D. Henshal l : Electron. Lett. 2.6, 42 (1980) I 0.16 L.M. Dolginov, A. F. Drakin, C.V. Druzhinina, P.G. Eliseev, M.G. Milvidsky, V. A. Skripkin, B. N. Suerolov: 1EEE J. QE-17, 593 (1981) 10.17 D.N.Payne, W. A. Gambl i ng: Electron. Lett. 111, 176 11975) 10.18 l ). J. Channi n, M. Et t enberg: Proc. SPI E Technical Symposium on l~fi'ared Fibers, Los Angeles, CA 11981) p. 40 10.19 A.G. Dentai, T.P. Lee, C.A. Burrus: Electron. Lett. 13, 484 11977) 10.20 K.Oe, S. Ando, K. Sugiyama: Japan J. Appl. Phys. 16, 1693 11977) 10.21 l. Umebu, O.Hasegawa, K. Aki t a: Electron. kett. 14, 499 11978) 11/.22 O. Wada, S. Yamakoshi, M. Abe, Y. Nishitani, T. Sakurai: IEEE J. QE-17, 174 (1981) 10.23 R.C.Goodfe[low, A.C.Carter, [.Griffith, R.R. Bradley: IEEE Trans. ED-2.6, 161 (1979) 10.24 G.H.O[sen, F.Z. Hawrylo, D. J. Channin, D. Botez, M. Ettenberg: IEEE J. QE-17, 2130 (1981) 10.25 R.C.Goodfellow, A.C.Carter, C.J. Rees, R. Davi s: Trans. E1)-28, 365 (1981) 10.26 D. J. Channi n: Private communi cat i on 10.27 M. Ettenberg, H.Kressel: IEEE.I. QE- | 6, 186 (1980) 10.28 S. Yamakoshi, O. Wada, S. Komiya, T. Sakurai : IEEE J. QE-17, 167 (198l) 10.29 R. Yeats, Y.G.Chai, T. l ). Gi bbs, G. A. Antypas : I EEE Electron Dev. Lett. EDL-2, 234 (1981) 10.30 C.J.Nuese, G.H.Olsen, M. Ettenberg, J. J. Gannon, T. J. Zamerowski : Appl. Phys. Lett. 29, 8O7 (1976) 10.31 T. Tsukada: J. Appl. Phys. 11, 4899 (1974) 10.32 S. Ni t a : I EEEJ . QE-15, 1208 11979) 10.33 T. Furuse et al.: 5th Eur. Conf. Opt. Commtm., Amsterdam, The Netherlands (1979) Paper 22 10.34 R. D. Burnham, D.R.Scifres: Appl. Phys. Lett. 27, 510 (1975) 10.35 P.A. Kirkby, G. H. B. Thompson: J. Appl. Phys. 47, 4578 11976) 10.36 K.Aiki, N. Nakamura, T. Kuroda, J. Umeda: Appl. Phys. Lett. 48, 649 (1977) 10.37 D. Botez: Appl. Phys. Lett. 33, 872 (1978); and IEEE ,I. QE-17, 2290 11981) 10.38 D. Botez: J. Optical Commun. 1, 42 (1980) 10.39 P.M. Asbeck, D.A. Cammack, J.J. Daniele : Appl. Phys. Lett. 3& 504 (1978) 10.40 H.Yonezu, l. Sukama, K. Kobayashi , T. Kamejima, M. Ueno, Y. Nanni chi : Japan 1. Appl. Phys. 12, 1584 (1973) 10.41 P. Marschall, F.Schlosser, C. Wock : Electron. Left. 15~ 38 (1979) 10.42 M. Ettenberg: J. Appl. Phys. 52, 3845 11981) 10.43 For a review see M. Ettenberg, H. Kressel: IEEE J. QE-16, 186 (1980) 10.44 R. E. Epwort h: Tech. Digest Topical Meeting on Optical Fiber Commun. , Paper ThD1, Washington, DC (1979). See also R.E. Epwort h: Laser Focus 109 (September 1981) 10.45 R. Lang, K. Kobayashi : IEEE J. QE-16, 347 (1980) 10.46 N. Chinone, K.Saito, R.lto, K. Aiki, N.Shige: Appl. Phys. Lelt. 35, 513 11979) 10.47 S. Yonezawa, M.Ojima, T.Sugiyama, Y. Fukui, K. Aiki: CLEO '81, Washington, DC (1981) 10.48 D. Botez: Appl. Phys. Lett. 36, 190 (1980) 10.49 D. Botez: Tech. Digest IEDM '81, Washington, DC 11981) 10.50 An i nt roduct ory treatment is given by A. B. Sharma, S.J. Halme, I~4. M. Butusov : Optical Fibre Systems and Their Components, Springer Ser. Opt. Sci., Vol. 24 (Springer Berlin, Iteidelberg, New York 1981) 10.51 W.Susaki, K. lkeda, K. Shi rahat a: Proc. IOOC, San Francisco, CA (1981) paper TuB4 10.52 J. Yamada, S. Machida, T. Mukai, H. Kano, K. Sugiyama : Electron. Left. 15, 596 (1979) 11. Phot ode t e c t or s and Recei vers - An Updat e R. G. Smi t h With 2 Figures At the time of writing of Chaps. 3 and 4 the principal t hrust of fiber communi cat i on was in devel opi ng systems for use in the 0.8 t o 0.9 lam region where GaAIAs sources oper at ed and where silicon det ect ors were essentially unchallenged. At t hat time, however, t he pot ent i al for i mpr oved system per f or mance at wavelengths bet ween 1.3 and 1.6lam had been realized and wor k was under way t o devel op sources and det ect ors for this spectral region. In the i nt erveni ng time a greal deal of progress has been made. Thi s new chapt er will review the advances made in t he devel opment of det ect ors and receivers for this spectral region. 11.1 Mat eri al f or Phot odet ect ors Several material systems have been investigated for use in fabri cat i ng phot ode- t ect ors for the 1.3 t o 1.6 laln spectral region. These include Ge and several alloys i ncl udi ng In~Gat _xAs~P 1 _:, and GaxA| ~ _xSb~,As 1 _~,. Of t he above t he maj or effort has gone into the devel opment of the I nGaAs P system which can be grown lattice mat ched to the bi nary hap with an energy gap cont i nousl y vari abl e between 1.35 eV (haP) and 0.75 eV (In0.53Ga0.47As) cor r espondi ng to wavelengths between 0.92 and 1.65 lain [11.1]. 11.1.1 Germanium Both p - i - n ' s and aval anche phot odi odes have been made f r om Ge for some time. Several recent papers have discussed new, i mpr oved results obt ai ned with this system [11. 2-8]. Al t hough some i mpr ovement s have been made, the use of these devices cont i nues to be limited by excessive dark currents. Typi cal dar k current s in Ge aval anche phot odet ect or s which under go mul t i pl i cat i on are 50 to I00 nA at r oom t emper at ur e and up to a fact or of 10 hi gher at the maxi mum oper at i ng t emper at ur es of fiber systems. These dar k current s severely limit the useful aval anche gain, in t urn reduci ng much of t he advant age of APDs. The use of Ge APDs appears limited to ext remel y high bit rat e systems where larger dar k current s are permissible or t o appl i cat i ons where the devices can be cooled. The noise propert i es of these devices are also limited by an unfavorabl e rat i o of i oni zat i on coefficients, c~/ / ~0. 5, which t ends to make the excess 294 R. G. Smith noise/factor large (Fig. 3.7). It is the dark current however, which severely restricts the utility of these devices. In the limit M = 1 (i.e., a p-i-n), the dark current is still ~ 100 nA, considerably greater t han t hat of state-of-the-art III-V p i n detectors. As a result, most of the current effort is on developing I I I - V phot odi odes. 11.1.2 l nGaAsP The InxGa l_xAs~,Pl_y material system has been intensively studied recently. Various composi t i ons of this material can be grown lattice mat ched to InP [11.9, 10], ranging in the bandgap from t hat of the binary, Eg= 1.35 eV, to the ternary, In0.53Gao.47As, with Eg=0. 75eV. These materials t hus span the spectral range of current interest, 2 = 1.3 to 1.6 lam. Whereas the maj ori t y of the effort in material growth has been done using liquid phase epitaxy, there are reports of impressive results obt ai ned with vapor phase epitaxy [11.11 17]. Some recent work has also been report ed on the molecular beam epitaxial growt h of lno.saGao.4vAs [11.18]. 11.1.3 GaA1AsSb This material system may be grown lattice mat ched to GaSb substrates with a range of bandgaps covering the 1.3 to 1.61am spectral range [11.19-22]. Al t hough detectors with high quant um efficiencies have been demonst r at ed [11.22], leakage currents, believed to be surface related, are excessive. This material system does have the interesting feature t hat the hole ionization ratio exhibits a resonant characteristic as a function of the alloy composi t i on [11.21]. This effect, at t ri but ed to impact ionization initiated by holes from the split-off valence band, has produced ratios /3/c~20. This high ratio is capable of produci ng low excess noise factors and is potentially useful provided the dark current can be kept low. 11.2 p- i - n Photodiodes Because of early problems encount ered in fabricating avalanche phot odi odes with low dark current (described below), a maj or effort has been put on developing p- i - n photodiodes. The material system which has undergone the most development is the InxGa 1 _xAsyPl y [11.11, 13-16] family, especially the t ernary limit, Ino.53Gao.47As [11.12-14, 16, 27-35]. Whe~'eas for an emitter the bandgap must be precisely determined to produce the desired emission wavelength, in a detector the essential requirement is t hat the bandgap energy be smaller t han the phot on energy. The t ernary lno.53Gao.47As with a bandgap Photodetectors and Receivers an Update 295 of 0.75 eV is t hus capabl e of det ect i ng signals at bot h the 1.3 and 1.55 gm fiber wi ndows. Because of this fact, mos t cur r ent det ect or wor k uses this com- posi t i on. Da r k current s, on t he ot her hand, t end t o be s omewhat l arger in the nar r ower bandgap mat eri al . By using layers of different composi t i ons, dual wavel engt h det ect or s capabl e of abs or bi ng t wo different wavel engt hs in separ at e regi ons have been demons t r at ed [11. 36-38]. Bot h mesa and pl anar p - i - n phot odi odes have been fabri cat ed wi t h the maj or i t y of the results r epor t ed on the si mpl er mesa st ruct ure. Int erest in pl anar devices [11.12, 14] is par t l y due to t he possibility of obt ai ni ng passi vat ed devices. Earl y wor k using SiO 2 as a diffusion mask and al so as a passi vat i ng l ayer pr oduced devices with excessively large dar k current s [11.14]. Removal of the SiO 2 reduced the dar k cur r ent by several orders of magni t ude, but r ecoat i ng with SiO 2 r et ur ned the dar k cur r ent to pr evi ousl y obser ved values. Pol yi mi de films have recent l y been r epor t ed to pr oduce surface passi vat i on of l n Ga As P devices [11.39]. The ar ea of surface passi vat i on is not well under- st ood and wor k is under way in t he devel opment of effectively passi vat ed devices. 11.2.1 Dark Current The dar k cur r ent of phot odet ect or s is ext remel y i mpor t ant and has been ext ensi vel y studied. As ment i oned above, Ge devices have excessively large dar k current s which limit t hei r utility. The si t uat i on is bet t er in the case of some I I I - V devices. The dar k cur r ent behavi or of I nGa As P devices can be di vi ded i nt o t wo regi ons of interest. At low vol t ages, cor r es pondi ng to oper at i on of the devices as p - i - n det ect ors, the dar k cur r ent in the best devices shows a sat ur at ed char act er i st i c with the dar k cur r ent var yi ng as the depl et ed vol ume. For vl / 2 vol t age dependence. Da r k uni f or ml y doped devices this cor r es ponds t o a --b~,~ cur r ent densities cor r espondi ng to 10- s A/ cm 2 or 10-1 A/ cm 3 ar e t ypi cal of the best mesa devices f abr i cat ed in Ino. s3Gao. 4vAs. In I nP the best val ues are appr oxi mat el y three orders of magni t ude smaller. The t emper at ur e dependence of these current s varies as e x p ( - Eg/nkT), where Eg is the band gap, k T is the Bol t zmann energy and n is a const ant , with a value n ~ 1 for diffusion current s, and n ~ 2 for gener at i on- r ecombi nat i on current s. Obser ved val ues of n r ange bet ween 1 and 2 with t ypi cal val ues bet ween 1.5 and 1.8 near r oom t emper a- ture. The cor r es pondi ng t emper at ur e rise for an i ncrease of 10X in dar k cur r ent is 36 C and 43 C for n = 1.5 and 1.8, respectively, and Eg=0. 75 eV. Since mos t of the dat a are t aken on mesa devices, par t of the obser ved current s may be surface related. Some evi dence exists t hat in the best devices the dar k cur r ent scales wi t h ar ea i ndi cat i ng a bul k origin. Ther e is at least one r epor t t hat excess dar k current s are associ at ed with cryst al l i ne defects [11.40]. Fur t her wor k is needed to det er mi ne the origin of dar k current s, especially for aval anche phot odet ect or s. 296 R. G. Smith The second region of interest for dar k cur r ent is associated with the appr oach t o breakdown. 111 essentially all devices with bandgaps less t han appr oxi mat el y 1 eV, br eakdown exhibits a soft behavi or r at her t han the har d br eakdown associated with the aval anche process. Several expl anat i ons have been given but it is now generally accept ed t hat the soft charact eri st i c is due to t unnel i ng [11. 41-44] which is mor e pr onounced in materials with small bandgaps and small effective carri er masses. In at least one case the t emper a- t ure and voltage dependence of the dar k cur r ent of an lno.53Gao.47As homoj unct i on device has been fit over a wide t emper at ur e range using onl y gener at i on- r ecombi nat i on and t unnel i ng current s [11.323. Because of the exponent i al dependence of the cur r ent on vol t age in t he t unnel i ng regime, much of the l i t erat ure cited quot es the dark cur r ent at 0.5 V~. In or der t o keep the dar k current s low in homoj unct i on devices, (with the / ~n j unct i on and hence the high field region in the nar r ow bandgap material), and to obt ai n a wide depl et i on region for good absor pt i on efficiency and low capaci t ance, the dopi ng of the absor bi ng mat eri al must be kept low, typically in t he mid 10 a 5 cm 3 range. An analysis of the basic requi rement s for the design of p- i - n det ect ors may be found in [ 11.44]. 11.2.2 Quantum Efficiency The quant um efficiency of a phot odet ect or is affected by surface reflection, r adi at i on passing t hr ough absor bi ng regions before it reaches the depl et i on region, or by r ecombi nat i on of the phot ogener at ed el ect ron hole pairs, ei t her at surfaces or within the bulk. Recombi nat i on is most likely to occur in undepl et ed regions. Fi gure 1 shows an In0.53Ga0.~TAs p- i - n st ruct ure typical of mesa devices descri bed in the l i t erat ure consisting of an n +-InP substrate, an n-l nP epitaxial buffer layer, and an uni nt ent i onal l y doped Ino.53Gao.47As layer 5-10 [am thick. The p-n j unct i on is formed in the t er nar y layer by diffusion and the lateral device di mensi ons are limited by formi ng a mesa. Since the absor pt i on coefficient in the absor bi ng t er nar y is ~ 104cm -1, the absor pt i on length is 1 [am. Such a device may be i l l umi nat ed t hr ough the InP subst rat e which is t r anspar ent t o radi at i on in the 1.3 to 1.6 [am region or from the top. In the l at t er case some of the radi at i on is absor bed in the heavily doped p-region where it is subject to surface and bulk recombi nat i on. For i l l umi nat i on from the subst rat e ~back or rear illumination), free-carrier absor pt i on can occur if the subst rat e is heavily doped. For devices of this t ype quant um efficiencies of 65-70 % have been obt ai ned for rear i l l umi nat i on and 40-50 % or often l ower values obt ai ned for front illumination. Cor r ect i ng for surface reflection, which can be reduced using antireflection coatings, [11.45] the rear illuminated devices have nearly 100% internal quant um efficiency. Hi gher efficiencies can be obt ai ned with front illuminated devices if a t r anspar ent cap l ayer is grown over the absor bi ng l ayer and the p-n j unct i on placed in the cap l ayer [11.25]. Photodetectors and Receivers an Update 297 h lt 1 ~ . ~ A u - Z n TOP CONTACT p-InO'53GaO'47As / n - - I n o . 5 3 Gao,47As "x~ n + - I n P SUBSTRATE I N\ ' t hv Fig. l l . l . Photodetector structure with ternary active region Au- Sn BACK CONTACT 11.2.3 Speed of Response In most III V mat eri al s t he carriers reach velocities appr oachi ng 10 v cm/s at fields ~104V/ cm. The transit time of carriers is thus ~10ps / l am in t he depl et i on region. Since depl et i on widths of at most a few mi crons are required, rise times of several tens of pi coseconds are possible. Recent l y a p- i - n det ect or with a 30 ps rise time has been r epor t ed [11.45a]. 11. 3 Aval anc he Phot ode t e c t or s In or der for an aval anche phot odet ect or to pr oduce significant sensitivity i mprovement s, t he dar k cur r ent which undergoes mul t i pl i cat i on must be kept small. As described above, t unnel i ng current s are excessively large as break- down is appr oached in a homoj unct i on device maki ng such homoj unct i on APDs of little use in receiver design. It has been demonst r at ed, however, t hat a het er ost r uct ur e device with separat e absor bi ng and mul t i pl yi ng regions [11.39, 46 52] (sometimes referred to as a SAM st ruct ure) can have low dar k current s at moder at e to high gains. In such a device the aval anche gain occurs in the wide bandgap InP mat eri al while the absor pt i on occurs in the smaller bandgap materials. An exampl e of such a st ruct ure is shown in Fig. 11.2. Successive layers of p-InP, n-InP, and n-Ino.53Gao.4vAs are gr own on a p+- I nP substrate. 298 R. G. Smith b i l e AU-Sn TOP CONTACT n - - I n 0 . 5 ~ GO . . . . I - p+- ~P SUBSTRATE L~\~I h ~ Au- zn BACK CONTACT Fig. 11.2. Phot ode t e c t or s t r uct ur e wi t h I n P / l n Ga As act i ve regi on for r educed dar k c ur r e nt With pr oper dopi ng of the n-InP layer [11.52], the fields can be sufficiently large to pr oduce aval anche gain in the I nP and still be low enough in t he t er nar y material t o avoi d tunneling. If the dopi ng and thickness of t he n-InP layer are t oo low, then t unnel i ng occurs before aval anche br eakdown and a noisy device results. On the ot her hand, if the dopi ng is t oo high, the depl et i on region may not reach the t er nar y with the result t hat the quant um efficiency is small. This is due to the fact t hat the phot ogener at ed holes cannot sur mount the val ence-band di scont i nui t y in the absence of an appl i ed electric field. Devices with the pr oper dopi ngs have been fabri cat ed and gains up to several hundr ed obt ai ned with reasonabl e dark current s. In addi t i on to devices with the aval anche region in the InP layer, there has been a lot of work report ed oll I nGaAsP aval anche phot odi odes with vari ous st ruct ures and mat eri al composi t i ons. Papers [11.39, 46 69] include much of the relevant material on this subject. In or der to obt ai n low noise aval anche gain, it is necessary to initiate the aval anche with the mor e ionizing carri er species. In InP, holes are the mor e ionizing carri er and the st ruct ure of Fig. 11.2 is designed to feed phot o- generat ed holes into the aval anche region. The second requi rement for low noise is t hat the rat i o of i oni zat i on coefficients be as large as possible. Ther e are several report s of measurement s of t he i oni zat i on coefficients in t he l i t erat ure [-11.49, 7(~75]. Al t hough the results of different aut hor s are onl y in fair agreement , t he general consensus is t hat in InP, [~/c~2-3, [11.70, 71] a val ue similar to t hat of Ge. Ther e are several possibilities of i mprovi ng this situation. For example, in Ga~ _xAlxSb, t here is a nar r ow range of composi t i ons where the energy of the split-off valence band nearl y equals the bandgap energy. For Photodetectors and Receivers an Update 299 this case a r esonant enhancement of [1 gives / ~/ c~20, roughl y an or der of magni t ude i mpr ovement over InP [11.21]. Anot her met hod for enhanci ng the rat i o c~//~ which uses a mul t i -quant um-wel l st ruct ure of al t ernat i ng wide and nar r ow gap mat eri al s has been pr oposed [11.76]. In this st ruct ure the electrons, upon leaving the high bandgap l ayer and ent eri ng the nar r ow gap material, effectively have a great er energy and are mor e capabl e of initiating ionizing collisions. These and ot her means of synthesizing mat eri al s and st ruct ures to i mpr ove the per f or mance of aval anche devices are an area of cur r ent research. The ul t i mat e utility of the I nGaAs P/ I nP or the GaA1SbAs devices depends upon t he level to which the dar k current s can be reduced. In general, t he present best dar k cur r ent levels must be reduced by roughl y an or der of magni t ude in or der for the devices to be useful at tile highest t emper at ur es (70-90 C) wi t hout cooling. 11. 4 Ot her De t e c t or s I21 addi t i on to work on p i -n and aval anche devices, t here are report s in tile l i t erat ure on the devel opment of phot ot r ansi st or s [11. 77-84], phot oconduct or s [11.85 87], pnpn det ect ors [11.88], and the use of reverse-biased LEDs as det ect ors [11.89]. Monol i t hi c phot oconduct or - ampl i f i er [11.87] and p-i-n FET combi nat i ons [11.90] have also been report ed. A new t ype of maj ori t y carri er det ect or, fabri cat ed by mol ecul ar beam epitaxy, in which the base region is fully depl et ed has recent l y been r epor t ed [11.91]. Much of this work is in the earl y expl or at or y phase and none of the devices has demonst r at ed a perfor- mance level compar abl e to t hat of an APD or p-i-n coupl ed with a low-noise amplifier. It can be shown, for example, t hat a phot ot r ansi st or is equi val ent in per f or mance to a p-i-n/bipolar amplifier [11.92]. As is discussed below, a p-i-n/GaAs FET gives superi or per f or mance t o a bi pol ar for most bit rates and hence is expect ed to be superi or to a phot ot r ansi st or . 11. 5 Recei vers In t he 0.8--0.9 gm spectral region, Si aval anche phot odi odes are empl oyed in receivers where the ul t i mat e in sensitivity is needed. | n addi t i on, when using an APD, amplifiers with great er dynami c range and somewhat hi gher noise levels can be used. Because of the earl y difficulties encount er ed with dar k current s in Ge and I I I - V aval anche phot odi odes, a different appr oach was t aken to receiver design for l ong-wavel engt h systems. The appr oach t aken at this time is to design the ultimate in low noise amplifier and use it with a p- i - n det ect or. In designing an ul t ra-l ow-noi se receiver, it is necessary to minimize all noise sources which in t urn involves reduci ng all st ray capaci t ances and increasing t he i nput det ect or l oad resistor (see Chap. 4 for details). The use of ext remel y 300 R. G. Smith low input capacitance GaAs FET tran.sistors ( C<0. 5pF, ,qm~30mS), low capacitance detectors, ( C<0. 5 pF), al ong with large input resistors, R > 1 Mf~, has allowed the design of amplifiers giving performance levels within 5-8 dB of those obt ai ned with APDs at 0.8 tam [11.93-98]. The performance level of receivers at about 1.3 lain is not yet equal to t hat at 0.8 gm but is 7-10 dB better t han obt ai ned at 0.8 lain with tr-i-n detectors. One of the principal drawbacks of these receivers is t hat t hey must be equalized to a large degree in order to restore the pulse response (Chap. 4). The result is a decrease in dynami c range. Typical dynami c ranges are report ed to be ~ 2 0 d B compared to ~ 4 0 d B for 0.8~tm APD receivers [11.99]. One possible solution is to shunt the input current away from the amplifier at high signal levels. Such an approach may require adapt i ve equalization if the receiver poles are altered by the shunt i ng scheme. Anot her approach is to include an active at t enuat or in the optical pat h to provide at t enuat i on of large optical signals. There are several reports of results obt ai ned using p-i-n/FET receivers in 1.3 I.tm systems at bit rates up to 274 Mb/s. The reader is referred to the references for details [11.100--104]. In addi t i on to the devel opment of high sensitivity receivers, there has been considerable work in the development of low cost units, especially for use in dat a links. In these receivers compromises have been made between sensitivity, dynami c range and cost. Several of these receivers have been integrated, and include functions, moni t ors, etc. A review of this subject may be found in [lJ.105]. 11. 6 Heterodyni ng Detectors for optical fiber communi cat i on systems are basically phot on counters, convert i ng each absorbed phot on into an electron-hole pair (in an APD, <M> - 1 addi t i onal carriers are produced by the avalanche gain process). Such a process ignores the phase of the electromagnetic field. Receivers using phot on detectors are generally limited in sensitivity by the amplifier noise except at very high S/N levels. Digital receivers, even those with the best report ed sensitivities, are 10-20dB away from the quant um limit, which corresponds to 21 phot ons received per mark. On the other hand, com- muni cat i on systems using CO 2 lasers [11.106, 107] and forms of i nst rumen- t at i on using He- Ne lasers [11.108] empl oy het erodyne or homodyne detection. In such systems the i ncomi ng signal is mixed with an optical local oscillator to produce a difference frequency which is then processed using st andard techniques. In such a receiver, the noise is domi nat ed by the shot noise of the local oscillator while the signal power is also proport i onal to the local oscillator power. In the limit of large local oscillator powers the S/N ratio is i ndependent of the amplifier noise and is determined by the characteristics of the source and the local oscillator. Photodetectors and Receivers - an Update 301 For a het er odyne system to be pract i cal bot h the source and the local osci l l at or must be spectrally pure and frequency stabilized with respect to each other. Wi t h gas lasers high spectral pur i t y and frequency st abi l i t y have been demonst r at ed. On the ot her hand, LEDs have ext remel y large l i newi dt hs and most laser diodes do not pr oduce a pure frequency, especially when modul at ed at high frequencies. Up to the present t i me little work has been done in the area of frequency st abi l i zat i on of laser diodes. Recently, however, there have been several report s of experi ment s in whi ch laser di odes have been stabilized by cont r ol of the t emper at ur e and i nj ect i on current . Short t erm (20 mi n) stabilities of 1 MHz have been report ed. These results poi nt to the i nt erest i ng possi bi l i t y of usi ng hct crodyne det ect i on for fiber communi cat i ons. The payoff woul d be sensitivities appr oachi ng the qua nt um l i mi t and the possibility of achi evi ng a high level of opt i cal mul t i pl exi ng. Di scussi ons of vari ous aspects of het erodyne det ect i on can be f ound in [1 1.109-111]. In addi t i on to the pr obl ems of laser st abi l i zat i on and t racki ng, the depol ari zi ng propert i es of fibers need to be addressed. Invest i gat i on of this area of fiber optics is relatively new and more i nt erest i ng results can be ant i ci pat ed in the fut ure [11.112]. Re f e r e nc e s I 1.1 For a summary of III-V devices for 1.3-1.6 ~tm fiber systems see: IEEE J. QE-17, (1981) 11.2 H. Ando, H. Kanbe, T. Kimura, T. Yamaoka, T. Kaneda : IEEE J. QE-14, 804 (1978) 11.3 M.C.Brain: Electron. Lett. 15, 821 (1979) 11.4 T. Kaneda, S. Kagawa, T. Mikawa, Y.Toyama, H. Ando : Appl. Phys. Len. 36, 572 (1980) 11.5 T.Mikawa, S. Kagawa, T.Kaneda: Fujitsu Sci. Tech. J. 16, 95 (1980) 11.6 T. Mikawa, S. Kagawa, T.Kaneda, T.Sakurai, H. Ando, O.Mikami: IEEE J. QE-17, 210 (1981) 11.7 H.Kaube, G.Grosskopf, O.Mikami, S.Machida: 1EEE J. QE-17, 1534 (1981) 11.8 H.Kanbe, G.Grosskopf: Jpn. J. Appl. Phys. 17, L767 (1980) 11.9 J.J.Hsieh : IEEE J. QE-17, 118 (1981) 11.10 B. de Cremoux : 1EEE J. QE-17, 123 (1981) 11.11 G.H.Olsen, t-I. Kressel: Electron. Lett. 15, 141 (1979) 11.12 N.Susa, Y.Yamauchi, H.Kanbe: Electron. Lett. 15, 238 (1979) 11.13 S.B. Hyder, R.R.Saxena, S.H.Chiao, R. Yeats: AppL Phys. Lett. 35, 787 (1979) 11.14 N.Susa, Y.Yamauchi, H. Ando, H.Kanbe: IEEE Electron. Dev. Len. EDL-I, 55 (1980) 11.15 G. ll.Olsen, T.J.Zamerowski: IEEE J. QE-17, 128 (1981) 11.16 N.Susa, Y. Yamauchi:J. Cryst. Growth 51, 518 (1981) 11.17 N.Susa, Y. Yamauchi, H.Kanbe: IEEE J. QE-I6, 542 (1980) 11.18 T.P. Lee, C.A. Burrus, A.Y.Cho, K.Y.Cheng, D.D. Manchon, Jr. : Appl. Phys. Lett. 37, 730 (1980) 11.19 F.Capasso, M.B.Panish, S.Sumski: 1EEE J. QE-17, 273 (1981) 11.20 H.D.Law, R.Chin, K.Nakano, R.A.Milano: IEEE J. QE-17, 275 (1981) 11.21 O.Hidebrand, W.Kuebart, K.W.Benz, M.H.Pilkuhn: 1EEE J. QE-17, 284 (1981) 11.22 F.Capasso, M.B.Panish, S.Sumski, P.W.Foy: Appl. Phys. Lett. 36, 165 (1980) 11.23 M.A.Washington, R.E.Nahory, M.A.Pollack, E.D.Beebe: Appl. Phys. Lett. 33, 854 (1978) 302 R. G. Smith 11.24 T.F'.Lee, J.C.Campbell, K.Ogawa, A.R. McCormick, A. G. Dent ai , C.A. Burrus: Electron Lett. 15, 388 (1979) 11.25 F.Capasso, R.A.Logan, A. Hutchinson, D.D. Manchon: Electron. Lett. 16, 893 (1980) 11.26 K. Takahashi, T. Murot ani , M. Ishii, W.Susaki, S. Takami ya: IEEE J. QE-17, 239 (1981) 11.27 K. Bachman, J. LSha y: Appl. Plays. Lett. 32, 446 (1978) 11.28 K. Ahmad, A. W. Mabbi t t : Tech. Dig., Inter. Elect. Dev. Meeting (Washington, D. C. 1978) p. 646 11.29 T.P.Pcarsall, M. l ' apuchon: Appl. Phys. Lett. 33, 640 (1978) 11.30 R.E. Leheny, R.E. Nahory, M. A. Pollack: Electron. Left. 15, 713 (1979) 11.31 T.P.Lee, C.A.Burrus, A.G. Dentai, K. Ogawa: Electron. Lett. 16, 155 (1980) 11.32 S.R. Forrest, R.F. Leheny, R. E. Nahory, M. A. Pol l ack: Appl. Phys. Lett. 37, 322 (1980) 11.33 T. P. Pear sal I : I EEE J. QE-16, 709 (1980) 11.34 R.F.Leheny, R. E. Nahory, M.A.Pollack, E.l).Beebe, J.C. DeWi nt er: IEEE J. QE-17, 227 (1981) 11.35 T.P.Lee, C. A. Burrus, A. G. Dent ai : I EEE J. QE-17, 232 (1981) 11.36 T.P.l,ee, J. C. Campbell, K.Ogawa, A. R. McCormick, A.G. Dentai, C. A. Burrus: Electron. Left. 15, 388 (1979) 11.37 J. C. Campbell, T.P.Lee, A.G. Dentai, C. A. Burrus: Appl. Phys. Lett. 34, 401 (1979) 11.38 J. C. Campbell, A. G. Dent ai , T.P.Lee, C. A. Burrus: IEEE J. QE-16, 601 (1980) 11.39 V. Diadiuk, S.H.Groves, C.E. Hurwitz, G. W. Isel er: IEEE J. QE-17, 260 (1981) 11.40 T.P. Lee, C.A. Burrus: Appl. Phys. Lett. 36, 587 (1980) 11.41 S.R. Eorrest, M. D. Domeni co, Jr., R. G. Smith, H. J. Stocker: Appl. Phys. Lett. 36, 580 (1980) 11.42 H. Ando, I-I. Kanbe, M.Ito, T. Kaneda : Jpn. J. Appl. Phys. 19, L277 (1980) 11.43 T. P. Pearsall: Electron. Lett. 16, 771 (1980) 11.44 S.R. For r e s t : I EEEJ . QE-17, 217 (1981) 11.45 S.Sakai, M. IJmeno, T. Aoki, M.Tobe, Y. Amemi ya: IEEE J. QE-15, 1077 (1979) 11.45a T.P. Lec, C.A.Burrus, K.Ogawa, A.G. Dent ai : Electron. Lett. 17, 431 (1981) 11.46 K.Nishida, K.Taguchi, Y. Mat sumot o: Appl. Plays. Left. 35, 251 (1979) 11.47 N.Susa, H. Nakagome, O. Mikami, H. Ando, H. Ka n b e : I EEEJ . QE-16, 864 (1980) l l . 48 V.Diadiuk, S.H.Groves, C.E. Hurwitz: Appl. Phys. Lett. 37, 807 (1980) 11.49 H. Kanbe, N.Susa, H. Nakagome, H. Ando: Electron. Lett. 16, 163 (1980) 11.50 N.gusa, H. Nakagome, H. Ando, H. Ka n b e : l EEEJ . QE-17, 243 (1981) 11.51 Y. Matsushima, K.Sakai, Y, Noda: IEEE Electron. De'~. Lett. EDL-.?., 179 (1981) 11.52 O. K. Ki m, S.R. Forrest, W.A. Bonner, R. G. Smi t h: Appl. Phys. Lett. 39, 402 (1981) 11.53 H.D.Law, L. R. Tomasetta, K. Nakano, J.S. Harri s: Appl. Phys. Lett. 33, 416 (1978) 11.54 C.E. Hurwitz, J.J. Hsieh: Appl. Phys. Lett. 32, 487 (1978) 11.55 Y.Takanashi, Y. Hori koshi : Jpn. J. AppL Phys. 17, 2065 (1978) 11.56 l-l.D.Law, kR. Tomas et t a, K. Nakano: Appl. Phys. Lett. 33, 920 (1978) 11.57 Y. Matsushima, K.Sakai, S. Akiba, T. Yamamot o : Appl. Phys. Lett. 35, 466 (1979) 11.58 R. Yeats, S. El.Chiao: Appl. Phys. Lett. 34, 581 (1979) 11.59 M.Feng, J. D. Oberstar, T. H. Wi ndhorn, L. W. Cook, G. E. St i l l man: Appl. Phys. Lelt. 34, 591 (1979) 11.60 S. Akiba, S.Sakai, Y. Matsushima, T. Yamamot o: Nat' l. Cony. Rec. of 1ECE Japan (April 1979) p. 859 11.61 K.Taguchi, Y. Mat sumot o, K. Ni shi da: Electron. Lett. 15, 453 (1979) 11.62 T.P. Lee, C.A.Burrus, A. G. Dent ai : IEEE J. QE-15, 30 (1979) 11.63 H.D. Law, K. Nakano, L. R. Tomas et t a: l EEE J. QE-15, 549 (1979) 11.64 T.P. Lee, C.A.Burrus, A. G. Dentai, A.A.Ballman, W.A. Bonner: Appl. Phys. Lett. 35, 511 (1979) 11.65 Y.Takanashi, Y. Hori koshi : Jpn. J. Appl. Phys. 19, L163 (1970) I 1. 66 F.Osaka, K. Nakazima, T. Kaneda, T.Sakurai, N. Susa: Electron. Lett. 16, 716 (1980) 11.67 N. Susa: Jpn. J. Appl. Phys. 19, L675 (1980) 11.68 Y. Matsushima, K.Sakai, S.Akiba, T. Yamamot o: Jpn. J. Appl. Phys. 19, 573 (1980) 11.69 Y. Matsushima, K.Sakai, S. Akiba, T. Yamamot o: Sixth Eur. Conf. on Opl. Comm., York (1980) pp. 226 230 Phot odet ect ors and Receivers - an Updat e 303 11.70 l. Umebu, A. N. M. M. Choudhur y, P. N. Robson: Appl. Phys. Lett. 36, 302 (1980) 11.71 C. A. Armi ent o, S.H.Groves, C. E. Hurwi t z: Appl. Plays. Lett. 35, 333 (1979) 11.72 T. l' . Pearsall, R. E. Nahory, M. A. Pol l ack: Appl. Phys. Lett. 27, 330 (19751 I 1. 73 M. lto, T. Kaneda, K. Nakaj i ma, Y.Toyoma, T. Yamaoka, T. Kot ani : Electron. Lett. 14, 428 (1978) 11.74 T.R.Pearsall, R. E. Nahory, M.A.Pollack : Appl. Plays. Lett. 28, 403 (1976) 11.75 H.D. Law, K. Nakano, L. R. Tomasetta, J. S. Harri s: Appl. Phys. Lett. 33, 948 (1978) 11.76 R.Chin; N. Hol onyak, Jr., G.E.Slillman, J. Y. Tang, K. Hess: Electron. Lett. 16, 467 (1980) 11.77 H. Beneking, P. Muschel, G. Schul : Electron. Lett. 12, 395 (1976) 11.78 M. Konagi , K.Katsukawa, K. Takahashi : J. Appl. Plays. 48, 4389 (1977) 11.79 R. A. Milano, T. H. Wi ndhorn, E. R. Anderson, G.E.Stillman, R.A.Dupuis, P. D. Dapkus: Appl. Plays. Lett. 34, 562 (19791 11.80 K.T. Alavi, R. J. Markunus, G.C. Fonst ad: Tech. Dig., hater. Elect. Dev. Meeting (Washington, D. C. 19791 p. 643 1.81 M.Tobe, Y. Amemiya, S.Sakai, M. Umeno: Appl. Phys. Lett. 37, 73 (1980) 1.82 H.Beneking, N. Grote, W. Roth, M. N. Svi l ans: Electron. Left. 16, 602 (1980} 1.83 P. D. Wright, R.J.Nelson, T.Cella: Appl. Phys. Lett. 37, 192 (1980) 1.84 J. C. Campbell, A. G. Dentai, C. A. Burrus, J. F. Fer guson: 1EEE J. QE-17, 264 (1981) 1.85 J. C. Gammel , G. M. Met ze, J. M. Bal l ant yne: IEEE T. ED-28, 841 (1981) 1.86 J. C. Gammel , H. Ohno, J . M. Ba l l a nt yne : I EEEJ . QE-17, 269 (1981) 1.87 J. Barnard, H. Ohno, C. E. C. Wood, L. F. East man: IEEE Elect. Dev. Lett. EI)L-2, 7 (19811 11.88 J. A. Copel and, A. G. Dent ai , T.P. Lee: IEEE J. QE-14, 810 (1978) 11.89 T.Ozeki, Y.Oe,natsu, T. l t o, M. Yamamot o, Y. Unno: Optics. Lell. 2, 103 (19781 11.90 R.F.Leheny, R. E. Nahory, M.A.Pollack, A.A. Balhnan, E.D. Beebe, J. C. DeWinter, R. J. Mar t i n: Electron. Lett. 16, 353 (19801 11.91 C. Y. Chen, A. Y. Cho, P. Garbinski, C.G. Bethea, B.F. Levine : Post Deadline paper, IOOC Conf. San Francisco (19811 11.92 K.Ogawa, J. C. Campbel l : Tech. Dig., IOOC, San Francisco (1981) 11.93 L. R. Tomasetta, H. D. Law, R.C.Eden, 1.Deyhimy, K. Nakano: 1EEE J. QE-14, 800 (19781 11.94 D. R. Smith, R. C. Hooper, 1. Garrett: Opt. and Quan. Electr. 10, 293 (19781 11.95 K.Ogawa, E. L. Chi nnock: Electron. Lett. 15, 650 (19791 11.96 D. R. gmi t h, R.C. l l ooper, K. Ahmad, D.Jenkins, A.W. Mabbi t t , R. Nicklin: Electron. Lett. 16, 69 (19801 11.97 D. R. Smith, A.K.Chatterjee, M.A.Z. Rejman, D.Wake, B. R. Whi t e: Electron. Left. 16, 750 (1980) 11.98 R. C. Hooper, M.A.Z. Rejman, S.T.D. Ritchie, D.R.Smith, R. B. White: Sixth Eur. Conf. Opt. Comm. York (1980) 11.99 R.G.Smith, C.A.Brackett, H. W. Rei nbol d: Bell Syst. Tech. J. 57, 1809 (1978) 1 1.100 D. Gloge, A. Albanese, C. A. Burrus, E. k. Chi nnock, J. A. Copel and, A.G. Dentai, T.P. Lee, Ti Li, K. Ogawa: Bell Syst. Tech. J. 59, 1365 (1980) 11.101 M. J. O' Mahony: Electron. Lett. 16, 752 (1980) 11.102 J. E. Mi dwi nt er: IEEE J. QE-17,911 (1981) 11.103 K.Ogawa, E. L. Chinnock, D.Gloge, P. Kaiser, S.R.Nagel, S. J. Jang: Electron. Lctt. 17, 71 (1981) 11.104 T. Ki mura: Proc. IEEE 68, 1263 (1980) 11.105 S.D.Personick, N. L. Rhodes, D. C. Hanson, K. H. Chan: Proc. IEEE 68, 1254 (1980) 11.106 M. C. Tcich: Proc. IEEE 56, 37 (1968) 11.107 M. C. Tei ch: Proc. IEEE 57, 786 (1969) 11.108 L.E. Drai n: The Laser Doppler Technique (Wiley, New York 1980) 11.109 M. Favre, J. Jeunhomme, l . Joi ndot , M. Monerie, J. C. Simon : I EEE J. QE-17, 897 (1981) 11.110 Y. Yamamoto, T. Ki mu r a : l EEEJ . QE-17, 919 (1981) I 1.111 S. Saito, Y. Yamamot o, T. Ki mura : IEEE J. Q E-I 7, 935 (1981) 11.112 See IEEE J. QE-17 (June 1981) special issue for papers on single-mode technology Subject Index AIGaAs 10,17 19,37 Analog receiver 151 APD see Avalanche photodiode APD detectors 266 Array com~ectors 276 Attenuation, optical 281 Attenuators, optical 273 Automatic bias control 186 Automatic gain control 272 Available transmission loss 262 Avalanche photodiode 66, 92, 95, 272 effective k-value 72 excess noise fuctor 72, 141 gain 70, 139, 141 noise 72 temperature dependence 70 Bandwidth, effective 128 Bell System SLC-40 278 Bias circuit noise 97 Biconic guide 270 Boxcar integrator 275 Broadband amdog systems 262 Broadening of modes 232, 234 Cable 268 Atlanta 280 Chicago 280 comlectorized 278 instalhttion 278 pulling tension 279 sections 276, 278 splicing 274, 279 vaults 276 Cabled fibcrs, perlbrmancc 279 Central offices 276 Channel, linear 91,93 Channel systems 263 repeater spacing 264 Characteristic frequency 250 Coding see Transmitters Coherent light injection 235, 236 Collector current, optimum 119 Commercial traffic 281 Companded differential PCM 278 Condition of laser instability 232 Connectors, precision 270 array 269 losses, single fiber 282 Cooling see Thermoelectric cooling Costs, comparative 265 Coupling efficiency 162, 206 CUlTents dark 65, 75 76, 95, 140 gateleakage 112 optimum colletor 119 shunt noise 97, 110 Cutoff fi'equency 257 Dark current 65, 75 76, 95, 140 Decider 272 273 Degradation 289 Delaylimc 217, 219 Detector 273 Device failure 281 Device hours 281 Diffused-stripe laser 224 Diffusion process 225 Digital carrier 266 data service (DDS) 278 hierarchy 263,266 multiplexing 264, 277 regenerators 264 subscriber carrier 277 terminals 264 Digital systems 262,264 encoded voice channels 263 Diode heating 244 Dislocations 12, 13, 54, 55 Dispersion 93, 147, 155, 261 Dislortion 165, 189, 197, 249 Distributed feedback injection lasers Distributions, impairments 282 Double section injection laser 231 Drivers analog 189 digital 170, 180 emitter-coupled 173, 180, 183 237 306 Subject Index Drivers emitter-follower 175 laser 182 LED 171 series 171,180 shunt 175,180,182 DSI (TI rate) channel 281 Ducts, congestion 277 Dynamic behavior 214, 229 Dynamic range 126 F.conomic considerations 260 261,266 Electric arc fusing 269 Emission properties 226 Emission spectrum 226, 230, 232 modulation influenced 229 spatially resolved 229, 230 l-poxy, silica filled 270 Equalization 132. 146.147.155 FET and bipolar amplifiers 125 Equalizer 93. 110. 125. 138 Equip,nent standard telephone 279 installation 279 Error performancc 281. 282 litching techniques, preferential 269 Exchange area paired cable 265 Exponential pulse shape 105.110 Extinction ratio 143. 183 Fauout 271. 274 Feedback bias control 182. 183. 185. 189 drive control 185 linearization 189 stabilizalion 272 Feed forward 191 FET 112.156 Fiber alignlnent 269 array, grooved chip 279 breaks 274 comlectors 207 single 269,279 demountable 269,270 defects 274 dispersion 273 modes 43, 44 patchord 274 splices 274 Fibers 265 see also Lightguide bandwidth 268 chemical wlpor deposition 268 cost 265 crosstalk 266 coupling laser 34 35 LED 40 45 dimensional tolerances 269 economics 26 I graded index 2 information rate 266 loss 261,262, 268, 274 muhimode 2 numerical aperture 268 ribbons 268 sheaths 268 signal-to-noise ratio 262 single inode I specific costs 265 Figure of merit bipolar 121 FET 114 Filaments 227 Fillers 93~ 109, 273 Floating shelf hardware 271 Frequency spectrum of the self-pulsations Front end design 124 high impedance 125 inlegraling 125 transimpedance 126, 1311 Gaussian approximation 133 noise 134 pulse shape 105 Harmonic amplitudes 250 Harmonic distortion 238,253 total 249, 253 Heat capacity 246 Heterojunction s t r uc t ur e 285 Heterodyning 300 Iloieburning 214, 237 hnpact ionization 67 Impulse response 95 Index of refraction, effective 215 [nGaAs 10,37 lnGaAsP 10,14,22,37 Injection, coherent light signal 234 hmer duet 278 InP/GalnAsP injection lasers 238 Integrals (noise analysis) 101 Integrator, gated 2"S lntcrsymbol interference 93, 105, 146 Junction heating 243 k-wtlue(APD) 141 Kinks 227,234 233 Subj ect I ndex 307 Lar ge- s i gnal r e s pons e to a st ep ctJrrenl pul s e 237 Las er char act er i s t i cs 161 excess noi se 168 far field pa t t e r ns 3(1, 31 ki nks 36, 227, 234 l i neari t y 165, 193 mo d e s 3(1, 34, 227, 232 moda l i nst abi l i t y 165+ 168, 185, 193, 196 near field pa t t e r ns 30, 31 power 244 pul s at i ons 59, 168 speed 164 t e mpe r a t ur e effects 20. 55, 56 t hr es hol d 16, 17, 24, 25 Las er out put , t i me resol ved 229, 231 Las er t her mal pr oper t i es 27, 28 Laser s di rect mo d u l a t i o n 217 gi an- gui ded 289 i ndex- gui ded 289-291 l ong-l i ved 28 I opt i cal feedback 273 pa c ka gi ng 195 rel i abi l i t y 166 s i mul a t or 196 t ypes bur i ed heterojunction 24, 289, 290 doubl e het er oj unct i on 289 c ha nne l e d s ubs t r at e, pl anar 34, 35, 224 doubl e- dovet ai l s t r uct ur e 34 h o mo j u n c t k m 10 l ow mes a st ri pe 224 oxi de st r i pe 224 pr ot on- i mpl a nt e d 224 st r i pe cont act 15, 23 25 t r ans ver s e- j unct i on- s t r i pe 224, 237, 289 V-groove 224, 238 Lat t i ce ma t c h 13, 14 LED see Li ght - emi t t i ng di ode Li ght - emi t t i ng di ode Bur r us 38 50 char ct er i st i cs 161 edge emi l t er 38- 50 l i ueari t y 165, 189 modul a t i on 45- 50 pa c ka gi ng 195 radi,' mce 40- 43 rel i abi l i t y 166 spect r al wi dt h 163 speed 164 Li g h t g u i d e s y s t e ms 263 a dva nt a ge s 263, 265 cr osst al k 262 el ect ri cal i nt erference 262 Li ght gui des a c c ur a c y 270 cabl es 264 di s t r i but i ng panel 271, 279 i nf or mat i on capaci t y 265 FT3 277 T3 277 Li ne r epeat er s 260 Loa di ng coils 264 Locki ng r ange 236 ['or s ync hr oni z a t i on 235 Lopp- back t ecf i ni ques 274 Los s effective 281 t r a ns mi s s i on 265 me a s ur i ng i ns t r ume nt 274 Low- f r equency r es pons e 246 Mat er i al di s per s i on 163 Met al l i c medi a awt i l abl e l oss 262 cost 261 equal i zat i on 263 s i gnal - t o- noi s e rat i o 262 Mi c r obondi ng l osses 279, 280 Mi cr ol ens es 206 Mode - gui di ng me c h a n i s m 224 Mode l ocki ng 232 Mode st abi l i t y 285 Mode t r a ns mi s s i on 260 Mo d e s 201 f unda me nt a l 229 f i r st - or der t r ans ver s e 229 Modul a t i on 167, 193,243 de pt h 223 l i mi t at i ons 243 rat es 288 s i nus oi dal of l asers 220 smal l si gnal 222, 237 Mo n o c h r o ma t i c near-fi el d 229 Mo n o c h r o me TV si gnal 278 Mul t i mode fibers 291 Ne a r field 226, 230. 235 Noi s e anal ys i s 93 ff. ampl i f i er 9"7 base r esi st ance 119, 121 bi as ci rcui t 97, 112 bi pol ar fi' onl end I 17 bit rat e de pe nde nc e 1 I0, 116 ci rcui t 97 cor ner fi ' equency I I I dar k cur r ent 95 equi val ent i nput cur r ent 103 eqt, i val ent o u t p u t vol t age 97, 102 308 Subject Index Noise FET fi'ont end 112 FET vs bipolar 123 input 103 measurement 128 normalization 98 output 97,102 photodetector 95 series 110 shot 97, II5, 117 shunt 110 signal dependent 95 thermal 115 Nonlinear characteristics 249 Normal failure distribution 56 Optical cable, pressurizing 278 communication systems 238 contact 271 power 99 Optically coupled lasers 234 Paired cable 266 Patchcord, single fiber 271 Pattern effccts 219, 220, 235 Peak detector 272 Personick integrals 103 Phase distortion maximum 257 minimum 254 Phase-shift modulation 193 Photodetectors 91,293 awdanche 66ff., 95, 297 channel stop 79 dark current 295 fabrication 78 germanium 81 gcttering 79 guard ring 68, 79 linearity 66, 77 long wavelength 80ft. noise 95 ft. passivation 68, 79 p-i-n 63,95 quantum efficieny 296 reach-through 68 silicon 63 structures 67 III-V materials 82 Photodiodes, avalanche 294 Photomultiplier 274 Phototransistor 299 Photon lifctimc 215 Physical environment 264 Picturephone Meeting Service (PMS} 278 p-i-n diode 272 Postamplifier 92 Power dissipation 244 Power gain 272 Preamplifier 92 Precision grooved chips 269 Probability density function 133 Pulsations, microwave spectrum of 232 Pulse code modulation (PCM) 219, 220, 223, 229, 235, 263 Quantum effects 262 Quantum efficiency 63 effect of surface recombination 81 Raised cosine 104 Rate equations 214 monomode 216 small-signal analysis 220 Rayleigh scattering, backward 274. 275 Receivers 272, 299 equalization 273 Receiver sensitivity s e e Sensitivity Recombination interface I1 13 nonradiative 10, 54 radiative 10, 47,48 Regenerator 272 Relaxation oscillations 217,220, 225,229 Reliability 288 Repeaters characteristics 272 rcpeatered line 282 spacing 282 Responsivity 65, 92 Reverse c u r r e n t s , s ed Dark currents Routing manholes 276 metropolitan 281 splices 276 congestion 281 Saturation 77 Self-pulsations 231-234 Sensitivity analog 151 avalanche detector 139, 153 bit rate depcndence t38 circuit noise limit 152 digital 129, 132 effect of dark cur ent 145 effect of source extinction ratio 143 effect of transmitter and channel parameters 146 p-i-ndetector 136, 152 shot m~ise limit 152 Subject Index 309 Separation A2 227 Sheaths 264 Shielded video pairs 278 Shot noise 65, 262 Signal averaging techniques 275 fault detection 276 field loss measurements 276 Signal-to-noise ratio 152 dependence of error rate 282 requirements 261 Single longitudinal mode enaission 237 Single-mode fibers 291 Span loss 280 Spatial hole burning 225 Spectral broadening 228 Spectral halfwidtb 228 Spectrum accelerated lifetest 55 58 degradation 52 59 gradual laser degradation 54, 55 laser 29 32 dark lines 54 facet coatings 29, 53 facet damage 53 and LED phonon kick 54 and LED reliability 52 59 mirror damage 53 I . ED 48, 53 t i me- r esol ved 229 Speed of response 65, 70 Spl i ce losses 269, 275, 280, 281 Spont aneous emi ssi on 220 rates 215 Spur i ous el ect r i cal di st ur bances 282 Squar e l aw 273 St a t i o n a r y sel f - pul sat i ons 231 Subscr i ber o r ' qoop' " phmt 283 Subscr i ber t el ephone t r af f i c 227 Systems block diagram 268 characteristics 268 digital 260 error performance 281 interbuilding 260 mtercity 260 interoffice trunks 260 king 260, 266 Taper launcher 204 Taps, optical fiber 185 Telephone duct, costs 265 Teleptlone plant 264 Terminal 268 costs 264 functions 267 Television operating center (TOC) 278 Thermal resistance 244 Thermal time constant 246 Thermally-induced nonlinearities 243 Thermoelectric cooling 166, 183, 195 Threshold current 243 density 4. 217 Time domain reflectometer, optical 274 Transfer function 94, I 11,252 Transfer molding process 270 Transient response 244 laser 218 Transimpedance. front end 126, 130 Transmission system equipment 273 maintenance and repair 267 testing procedures 273 Transmitters 161,272 see al so Drivers encoding 168, 177 laser protection 196 subsystem considerations 194 Trunking digital system 263 DS1 263, 277 DS3 266 interoffice 263, 277, 282 long voice pair 264 metropolitan 263 required system characteristics 266 system length 263 TI 263,277 urban-suburban area 263 Underground syslems ducts 264 electronics 264 Video signal 278 Voiceband data 277 Waveguide 201