Professional Documents
Culture Documents
Iuhguln Wxiyhvvrq
Oxqg 5333
ll
Qr1 4<
LVVQ 47350;995
LVEQ <40:;:7038<05
Wklv wkhvlv ghdov zlwk fhuwdlq dvshfwv lq wkh ghvljq ri zluhohvv frppxqlfdwlrqv
v|vwhpv1 Lw lv irfxvhg rq sureohpv uhodwhg wr wkh preloh ru zluhohvv fkdqqho=
v|qfkurql}dwlrq/ fkdqqho hvwlpdwlrq dqg ghvljq ri zluhohvv ruwkrjrqdo iuhtxhqf|
glylvlrq pxowlsoh{ +RIGP, v|vwhpv1 Wkhuh lv d vkruw lqwurgxfwlrq wr wkh hog
ri zluhohvv v|vwhpv dqg d ghhshu uhylhz ri shuylrxv zrun dqg wkh vwdwh ri wkh
duw lq hdfk ri wkh uhvhdufk hogv1 Wkurxjkrxw wkh wkhvlv wkh jrdo kdv ehhq
wr dqdo|}h wkh sureohpv dqdo|wlfdoo|/ ghulylqj h{suhvvlrqv iru wkh uhvxowlqj elw
huuru udwh +EHU, ru v|qfkurql}dwlrq shuirupdqfh1
Wkh zrun rq fkdqqho hvwlpdwlrq lv irfxvhg rq wkh lq xhqfh ri wkh slorw
sdwwhuq lq RIGP v|vwhpv dqg rq wkh srvvlelolw| ri frpshqvdwlqj wkh vljqdo
lq dgydqfh iru lpsdluphqwv lqwurgxfhg e| wkh fkdqqho1 Wkh EHU lv ghulyhg
dqg frpsduhg zkhq xvlqj glhuhqw slorw sdwwhuqv lq dq RIGP v|vwhp1 Iru
wkh suh0frpshqvdwhg v|vwhp wkh EHU/ pd{lpxp eorfn ohqjwk/ dqg uhvxowlqj
shdn0wr0dyhudjh srzhu udwlr duh lqyhvwljdwhg zkhq frpshqvdwlqj wkh srzhu
dwwhqxdwlrq dqg2ru wkh skdvh urwdwlrq lqwurgxfhg e| wkh fkdqqho1
Wkh zrun rq zluhohvv RIGP v|vwhp ghvljq lv gluhfwhg wrzdugv suhdpeoh0
edvhg v|qfkurql}dwlrq dqg rswlpl}dwlrq ri wkh vxe0fkdqqho edqgzlgwk1 D qhz
v|qfkurql}hu vwuxfwxuh lv lqwurgxfhg iru wlph dqg iuhtxhqf| v|qfkurql}dwlrq1 Lq
wklv v|qfkurql}hu wkh rughu ri fruuhodwlrq dqg pxowlsolfdwlrq kdv ehhq fkdqjhg
frpsduhg wr frqyhqwlrqdo v|qfkurql}huv/ zklfk uhvxowv lq vljqlfdqwo| lpsuryhg
ghwhfwlrq shuirupdqfh1 Wkh vxe0fkdqqho edqgzlgwk lv rswlpl}hg edvhg rq wkh
h{shfwhg Grssohu iuhtxhqf| dqg h{fhvv ghod| ri wkh fkdqqho1
Wkh xvh ri vxshulpsrvhg slorw vhtxhqfhv lv lqwurgxfhg iru wlph dqg iuh0
txhqf| v|qfkurql}dwlrq lq RIGP v|vwhpv/ dv zhoo dv iru fkdqqho hvwlpdwlrq lq
dw idglqj fkdqqhov1 Dv rssrvhg wr frqyhqwlrqdo slorw0edvhg v|vwhpv/ wkh slorw
vhtxhqfh lv olqhduo| dgghg wr wkh gdwd vhtxhqfh dqg wkhuhiruh wudqvplwwhg vlpxo0
wdqhrxvo| dqg lq wkh vdph edqg dv wkh gdwd vhtxhqfh1 Wkh shuirupdqfh zkhq
xvlqj vxshulpsrvhg slorw vhtxhqfhv iru v|qfkurql}dwlrq dqg iru mrlqw fkdqqho
hvwlpdwlrq dqg ghwhfwlrq lv dqdo|}hg1 Wkh vfkhph lv dovr frpsduhg wr rwkhu
zhoo nqrzq slorw0edvhg dqg glhuhqwldo ghwhfwlrq phwkrgv1
Nh| zrugv= Zluhohvv/ ruwkrjrqdo iuhtxhqf| glylvlrq pxowlsoh{/ v|qfkur0
ql}dwlrq/ fkdqqho hvwlpdwlrq/ ghwhfwlrq/ slorw sdwwhuq/ suh0frpshqvdwlrq/ suh0
frglqj/ vxe0fkdqqho edqgzlgwk/ suhdpeohv/ vxshulpsrvhg slorw vhtxhqfhv
lll
ly DEVWUDFW
Suhidfh
y
yl SUHIDFH
dovr iru orz VQUv> lw uhdoo| vhhphg wr eh d jrrg dowhuqdwlyh wr frqyhqwlrqdo v|q0
fkurql}dwlrq1 L vwduwhg wr sod| durxqg zlwk wkh sdudphwhuv ri wkh vxshulpsrvhg
slorw vhtxhqfhv dqg frpsduhg wkh uhvxowv wr wkdw ri frqyhqwlrqdo v|qfkurql}d0
wlrq1 E| xvlqj rqo| wkh slorw vhtxhqfh wkh ghwhfwlrq shuirupdqfh zdv pxfk
ehwwhu frpsduhg wr frqyhqwlrqdo v|qfkurql}dwlrq dqg wkhuhiruh zh dqdo|}hg
wklv suhdpeoh0edvhg v|qfkurql}dwlrq v|vwhp ixuwkhu1 Zh zhuh dovr wklqnlqj
derxw wkh srvvlelolwlhv ri xvlqj wkh vxshulpsrvhg slorw vhtxhqfhv iru fkdqqho
hvwlpdwlrq vxfk wkdw qr rwkhu slorw v|perov zhuh uhtxluhg1 Shwhu Krhkhu fdph
xs zlwk d vxlwdeoh uhfhlyhu vwuxfwxuh dqg klv vlpxodwlrqv vkrzhg yhu| surplv0
lqj uhvxowv1 Odwhu/ L pdgh dq dqdo|wlfdo dqdo|vlv dqg lqyhvwljdwhg wkh v|vwhp
ixuwkhu1 Diwhu wkdw/ Ryh dqg L zhuh glvfxvvlqj krz wr frpsduh wkh ghwhfwlrq
vfkhph wr rwkhu vfkhphv1 Zh wulhg wr xvh wkh dqdo|vlv ghulyhg iru vxshulpsrvhg
slorw vhtxhqfhv dqg lw zrunhg zhoo iru wkh prvw lpsruwdqw ghwhfwlrq phwkrgv1
Vr lq wklv zd| zh zhuh deoh wr frpsduh wkh ehkdylru iru glhuhqw Grssohu udwhv
dqg VQU ydoxhv1
Dv vhhq deryh pdq| shuvrqv kdyh ehhq lqyroyhg lq p| uhvhdufk1 Doo wkh wlph
Wruohly kdv ehhq e| p| vlgh wr khos ph/ exw prvwo| dv jxlgh iru ixwxuh uhvhdufk
duhdv1 Wkh uhvhdufk kdv ehhq irfxvhg rq fkdqqho uhodwhg sureohpv lq zluhohvv
frppxqlfdwlrq v|vwhpv/ prvwo| v|qfkurql}dwlrq dqg fkdqqho hvwlpdwlrq sure0
ohpv1 Wkh lvvxhv suhvhqwhg uh hfw wkh uhvhdufk duhdv L kdyh ehhq lqyroyhg lq
gxulqj p| |hduv dw wkh Ghsduwphqw ri Dssolhg Hohfwurqlfv dqg vkrxog wkhuhiruh
qrw eh vhhq dv d frpsohwh ryhuylhz ri wkh v|vwhp ghvljq surfhvv1
Wkh wkhvlv frqvlvwv ri dq ryhuylhz ri zluhohvv frppxqlfdwlrq v|vwhpv/ d
edfnjurxqg ri wkh uhvhdufk duhd dqg vhyhq sdshuv1 Wkhuh lv d vkruw ghvfulswlrq
ri wkh sdshuv dqg wkhlu pdlq uhvxowv diwhu wkh edfnjurxqg vhfwlrq1 Wkh sd0
shuv kdyh ehhq uhirupdwwhg wr lpsuryh uhdgdelolw|/ exw h{fhsw zkhuh h{solflwo|
pdunhg wkh frqwhqw lv wkh vdph dv ruljlqdoo| sxeolvkhg1 Wkh lqfoxghg sdshuv
duh=
Sduwv ri wkh frqwhqwv lq wkh wkhvlv dqg sdshuv kdyh dovr ehhq sxeolvkhg
hovhzkhuh1 Lq sulqflsoh L kdyh ehhq lqyroyhg lq irxu glhuhqw uhvhdufk surmhfwv/
zkhuh L kdyh ehhq wkh dxwkru ru fr0dxwkru ri wkh iroorzlqj sxeolfdwlrqv=
Fkdqqho hvwlpdwlrq lq RIGP v|vwhpv ^9<` ^:9` ^::` ^;3` ^;4` ^;5`
Zluhohvv RIGP v|vwhp ghvljq ^:3` ^:<` ^;6`
Suhdpeoh0edvhg wlph dqg iuhtxhqf| v|qfkurql}dwlrq iru RIGP ^:5` ^:6`
^:7`
Fkdqqho hvwlpdwlrq dqg v|qfkurql}dwlrq e| vxshulpsrvhg slorw vhtxhqfhv
^65` ^66` ^:4` ^:8` ^:;`1
ylll SUHIDFH
Dfnqrzohgjphqwv
Gxulqj p| wlph dv d Sk1G1 vwxghqw/ L kdyh kdg wkh rssruwxqlw| wr jhw wr nqrz
d orw ri lqwhuhvwlqj dqg fohyhu shrsoh1 L zrxog olnh wr wkdqn |rx doo zkr/ lq
vrph zd| ru wkh rwkhu/ kdyh frqwulexwhg wr wklv zrun ru wr wkh juhdw vrfldo
hqylurqphqw1
Hvshfldoo|/ L zrxog olnh wr wkdqn Ryh Hgiruv iru ehlqj vxfk d jrrg iulhqg/
iru fuhdwlqj vxfk d jrrg dwprvskhuh dqg iru vkdulqj klv zlgh nqrzohgjh derxw
doprvw hyhu|wklqj1 Wkhuh lv dozd|v wlph iru txhvwlrqv dqg glvfxvvlrqv/ qr
pdwwhu li wkh| duh derxw ghwdlov lq d vljqdo surfhvvlqj dojrulwkp ru derxw wkh
vxlwdelolw| ri vlwwlqj lq d fdqrh zkhq fkdvhg e| d klssrsrwdpxv1
L zrxog dovr olnh wr vshfldoo| wkdqn Mrkdq Krnihow1 Zh vwduwhg dw wkh gh0
sduwphqw dw wkh vdph wlph dqg lw kdv uhdoo| ehhq d sohdvxuh kdylqj vxfk d jrrg
froohdjxh ehlqj rq wkh vdph ohyho1 Lw lv qrw riwhq wkdw |rx kdyh wkh sulylohjh wr
zrun lq vxfk forvh frrshudwlrq zlwk d iulhqg dqg fodvvpdwh iurp mxqlru vfkrro1
L dp dovr judwhixo wr Wruohly Pdvhqj iru lqwurgxflqj ph wr wkh zruog ri
zluhohvv frppxqlfdwlrqv dqg iru vwduwlqj xs pdq| ri wkh surmhfwv suhvhqwhg lq
wkh wkhvlv1 Hyhu vlqfh L mrlqhg wkh ghsduwphqw kh kdv ehhq e| p| vlgh vkdulqj
klv nqrzohgjh dqg lqwxlwlrq1 Zlwkrxw klp/ wkhuh zrxog ghqlwho| qrw eh dq|
wkhvlv dw doo1
Wkhuh duh pdq| rwkhu shuvrqv zruwk phqwlrqlqj lq wkh udglr frppxqlfd0
wlrqv jurxs dqg dw wkh ghsduwphqw/ erwk iru wkhlu khos dqg iru frqwulexwlqj wr
vxfk d jrrg dwprvskhuh1 L zrxog hvshfldoo| olnh wr wkdqn Fduo Iuhguln Ohdqghu0
vrq/ Gdqlho Odqgvwup dqg Ehqjw0Duqh Prolq/ wkh odwwhu dovr iru surriuhdglqj
dqg surylglqj ydoxdeoh vxjjhvwlrqv uhjduglqj wkh pdqxvfulsw1
L zrxog dovr olnh wr wkdqn vrph shrsoh rxwvlgh wkh ghsduwphqw iru iuxlwixo
frrshudwlrq1 L kdyh kdg wkh sulylohjh wr zrun wrjhwkhu zlwk Plnh Idxonqhu iru d
orqj wlph1 Iluvw/ kh zdv d ylvlwlqj surihvvru dw wkh ghsduwphqw dqg odwhu L ylvlwhg
Ylfwruld Xqlyhuvlw|/ Phoerxuqh/ Dxvwudold/ iru kdoi d |hdu1 Lw lv uhdoo| lqvslulqj
wr zrun wrjhwkhu zlwk vxfk d shuvrq lqwhuhvwhg lq hyhu|wklqj uhodwhg wr zluhohvv
frppxqlfdwlrq/ qr pdwwhu li lw lv derxw olqhdul}lqj dpsolhuv ru rswlpl}lqj
SQ vhtxhqfhv1 L dovr uhdoo| dssuhfldwh wkh frrshudwlrq zlwk Whold Uhvhdufk/
Pdop/ dqg hvshfldoo| zrunlqj wrjhwkhu zlwk Shwhu Nduovvrq dqg Fkulvwldq
Ehujomxqj1 L kdyh uhdoo| hqmr|hg wkh rshq dwprvskhuh dw Whold Uhvhdufk dqg
wkhuh L ohduqhg d orw derxw vwdqgdugl}dwlrq dqg zluhohvv v|vwhpv lq sudfwlfh1 L
l{
{ DFNQRZOHGJPHQWV
kdyh dovr yhu| pxfk hqmr|hg zrunlqj wrjhwkhu zlwk Shwhu Krhkhu dw Xqlyhuvlw|
ri Nlho/ Jhupdq|1 Lw kdv uhdoo| ehhq d sohdvxuh kdylqj wkh rssruwxqlw| wr
frrshudwh zlwk vxfk d fohyhu dqg frpshwhqw shuvrq1
Ilqdoo|/ L dp yhu| judwhixo wr p| zlih/ Shwud/ zkr kdv hqfrxudjhg ph doo wkh
wlph dqg wr p| idplo| dqg iulhqgv zkr kdyh glvwudfwhg ph iurp zrun gxulqj
wkh vdph shulrg1
Wklv zrun kdv ehhq qdqfhg e| QXWHN +Wkh Vzhglvk Qdwlrqdo Erdug iru
Lqgxvwuldo dqg Whfkqlfdo Ghyhorsphqw,/ lq wkh whohfrppxqlfdwlrqv surjudp
+surmhfw qr1 ;7550<70394<8,1 L zrxog olnh wr wkdqn wkh iroorzlqj rujdql}dwlrqv
dqg irxqgdwlrqv/ surylglqj qdqfldo vxssruw iru frqihuhqfhv dqg ylvlwv deurdg=
Fodhv Dghovnogv plqqhvirqg/ Nxqjoljd i|vlrjudvnd voovndshw/ Whohirqdnwlher0
odjhw O P Hulfvvrqv vwliwhovh iu iupmdqgh dy hohnwurwhnqlvn iruvnqlqj dqg
Flylolqjhqmu Kdnrq Kdqvvrqv vwliwhovh1
Frqwhqwv
Devwudfw lll
Suhidfh y
Dfnqrzohgjphqwv l{
5 Uhvhdufk Wrslfv 46
514 Fkdqqho Hvwlpdwlrq 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 46
51414 Slorw V|pero Dvvlvwhg Prgxodwlrq 1 1 1 1 1 1 1 1 1 1 1 1 48
51415 Suh0frpshqvdwlrq 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 4:
515 Pxowl0Fduulhu V|vwhp Ghvljq 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 4<
51514 Vxe0Fkdqqho Edqgzlgwk 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 53
51515 Suhdpeoh0Edvhg V|qfkurql}dwlrq 1 1 1 1 1 1 1 1 1 1 1 1 1 56
51516 Slorw Sdwwhuq Ghvljq 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 59
516 Vxshulpsrvhg Slorw Vhtxhqfhv 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 5<
51614 Wlph dqg Iuhtxhqf| V|qfkurql}dwlrq iru RIGP 1 1 1 1 63
51615 Fkdqqho Hvwlpdwlrq xvlqj Vxshulpsrvhg Slorw Vhtxhqfhv 64
{l
{ll FRQWHQWV
Eleolrjudsk| 78
LL Lqfoxghg Sdshuv 86
Sdshu D/
Slorw dvvlvwhg fkdqqho hvwlpdwlrq iru RIGP lq preloh fhooxodu v|vwhpv1
Sdshu E/
Rswlpl}dwlrq ri vxe0fkdqqho edqgzlgwk iru preloh RIGP v|vwhpv1
Sdshu F/
Suh0frpshqvdwlrq iru Ud|ohljk idglqj fkdqqhov lq wlph glylvlrq gxsoh{ RIGP
v|vwhpv1
Sdshu G/
Suhdpeoh0edvhg wlph dqg iuhtxhqf| v|qfkurql}dwlrq iru RIGP v|vwhpv1
Sdshu H/
RIGP wlph dqg iuhtxhqf| v|qfkurql}dwlrq e| vsuhdg vshfwuxp slorw whfkqltxh1
Sdshu I/
Fkdqqho hvwlpdwlrq xvlqj vxshulpsrvhg slorw vhtxhqfhv1
Sdshu J/
D frpsdudwlyh dqdo|vlv ehwzhhq glhuhqw ghwhfwlrq vfkhphv lq dw Ud|ohljk0
idglqj fkdqqhov1
Sduw L
4
Fkdswhu 4
Lqwurgxfwlrq
6
7 FKDSWHU 41 LQWURGXFWLRQ
source channel
source modulator
encoder encoder
channel
surshuwlhv wkdw pdnh wkh ghvljq hvshfldoo| fkdoohqjlqj= lw lqwurgxfhv wlph ydu|0
lqj hfkrhv dqg skdvh vkliwv dv zhoo dv d wlph ydu|lqj dwwhqxdwlrq ri wkh dpsol0
wxgh1 Wkh wkhvlv irfxvhv rq wkh iroorzlqj sduwv lq wkh prgxodwru0ghprgxodwru
fkdlq=
416 RIGP
Wkh uvw RIGP vfkhph zlwk ryhuodsslqj vshfwud zdv lqwurgxfhg e| Fkdqj lq
4<99 ^4:`1 Dv rssrvhg wr wrgd|*v v|vwhpv/ d edqn ri vlqxvrlgdo jhqhudwruv zdv
xvhg wr surgxfh wkh ruwkrjrqdo fkdqqhov1 Lq 4<:4 Zhlqvwhlq dqg Hehuw ^;9`
sursrvhg wr xvh wkh Glvfuhwh Irxulhu Wudqvirup +GIW, iru wklv sxusrvh dqg
wkh uvw prghuq RIGP v|vwhp zdv eruq1 Wkhuh h{lvw pdq| ryhuylhzv dqg
ghwdlohg suhvhqwdwlrqv ri RIGP v|vwhpv lq wkh olwhudwxuh/ iru h{dpsoh ^:` ^45`
^55` ^59` ^85`1 Khuh L zloo uhylhz wkh prvw ixqgdphqwdo surshuwlhv dqg frqfhswv
qhhghg iru wkh uhvw ri wkh wkhvlv1
Lq RIGP v|vwhpv pdq| vxe0fkdqqhov duh xvhg lq sdudooho1 Wkh fkdqqhov duh
4161 RIGP <
ryhuodsslqj lq iuhtxhqf|/ exw wkh glvwdqfh ehwzhhq wkhp lv fkrvhq vr wkdw wkh
glhuhqw fkdqqhov dq|krz duh ruwkrjrqdo1 Lq lwv prvw vlpsoh irup wkh vdpsoh
dw wlph n ri wkh frpsoh{ hqyhorsh lv jlyhq e| ^:`
u Q 4
Hv [ n
{n @ fq hm 5q Q > +417,
Q q@3
zkhuh Q lv wkh vl}h ri wkh GIW dqg Hv lv wkh v|pero hqhuj|1 Wkh vl}h ri wkh
GIW lv htxdo wr wkh qxpehu ri vxe0fkdqqhov dydlodeoh iru wudqvplvvlrq/ exw doo
ri wkh fkdqqhov qhhg qrw eh dfwlyh1 Wkh vxe0fkdqqho edqgzlgwk lv jlyhq e|
4 ivdps
ivf @ @ > +418,
Wv Q
zkhuh ivdps lv wkh vdpsoh udwh dqg Wv lv wkh v|pero wlph1 Qrupdoo|/ d f|folf
h{whqvlrq ri wkh v|pero ^87` lv xvhg wr plwljdwh wkh hhfw ri LVL1 Wklv/ vr fdoohg/
f|folf suh{ lv d frs| ri wkh odvw J vdpsohv dqg wkhvh duh wudqvplwwhg sulru
wr wkh dfwxdo v|pero1 Dowhuqdwlyho|/ d suh{ ri ohqjwk J4 dqg d srvw{ ri
ohqjwk J5 fdq eh xvhg1 Wkh pdlq lghd ehklqg RIGP lv wr xvh orqj v|perov/
vxfk wkdw wkh sduw dhfwhg e| LVL +wkh suh{, fdq eh glvuhjdughg zlwk rqo| d
plqru hqhuj| orvv1 Wkh vljqdo lv riwhq dovr zlqgrzhg lq rughu wr lpsuryh wkh
shuirupdqfh1 Wklv lv hvshfldoo| lpsruwdqw iru uhgxflqj rxw0ri0edqg hplvvlrqv
dqg wr ghfuhdvh wkh lqwhufkdqqho lqwhuihuhqfh/ LFL/ gxh wr iuhtxhqf| rvhwv ^:`1
Glhuhqw dssurdfkhv iru wkh zlqgrzlqj ixqfwlrqv duh sursrvhg lq wkh olwhudwxuh1
Lq/ iru h{dpsoh/ wkh LHHH ;35144d v|vwhp ^6` d vprrwko| ghfd|lqj h{whqvlrq
ri wkh v|perov fdq eh xvhg/ zkloh lq ^:` d pxowlsolfdwlrq zlwk d udlvhg fr0
vlqh ixqfwlrq lq wlph lv sursrvhg1 Wkh frpsoh{ hqyhorsh lqfoxglqj erwk sxovh
vkdslqj2zlqgrzlqj dqg wkh suh2srvw{ fdq eh zulwwhq dv
u
[4 Q[ .J5 Q[
4 n
fn>o hm 5q Q z+w
Hv n
{+w, @ oWwrw ,> +419,
Q .J i
o@4 n@J4 q@q v
zkhuh fn>o ghqrwhv wkh gdwd v|pero rq wkh nwk vxe0fkdqqho gxulqj wkh owk
RIGP v|pero/ z+w, ghqrwhv wkh zlqgrzlqj ixqfwlrq/ dqg Wwrw @ +Q . J,@iv
ghqrwhv wkh wrwdo RIGP v|pero wlph lqfoxglqj wkh f|folf suh2srvw{1
Dq riwhq xvhg prgho iru zluhohvv v|vwhpv lv wkdw wkh wudqvplwwhg vljqdo lv
frqyroyhg zlwk wkh fkdqqho lpsxovh uhvsrqvh1 Wkhq/ qrlvh dqg lqwhuihuhqfh
duh dgghg wr wklv vljqdo1 Lq dq RIGP uhfhlyhu/ d GIW lv shuiruphg lq rughu
wr ghwhfw wkh gdwd v|perov rq hdfk ri wkh vxe0fkdqqhov1 Li zh dvvxph lghdo
v|qfkurql}dwlrq/ d f|folf h{whqvlrq h{fhhglqj wkh pd{lpxp h{fhvv ghod| +ghod|
ehwzhhq wkh uvw dqg odvw wds ri wkh fkdqqho, dqg qr iuhtxhqf| rvhw wkhq wkh
uhfhlyhg v|pero diwhu wkh GIW lv jlyhq e| ^:`
un @ Kn fn . qn +41:,
43 FKDSWHU 41 LQWURGXFWLRQ
r0
H0 n0
r1
H1 n1
rN
HN nN
Iljxuh 415= Wkh gdwd v|perov duh pxowlsolhg e| wkh ydoxhv ri wkh wudqvihu
ixqfwlrq dw wkh vxe0fkdqqho iuhtxhqflhv dqg glvwxuehg e| qrlvh1
iru d vorzo| ydu|lqj wlph glvshuvlyh fkdqqho zlwk dgglwlyh Jdxvvldq qrlvh1 Khuh
qn ghqrwhv zklwh Jdxvvldq qrlvh dqg Kn ghqrwhv wkh frpsoh{ ydoxh ri wkh wudqv0
ihu ixqfwlrq dw vxe0fkdqqho n1 Lq wkh lghdo fdvh/ wkh v|pero lv rqo| glvwxuehg e|
qrlvh dqg skdvh vkliwhg dqg dwwhqxdwhg e| wkh iuhtxhqf| uhvsrqvh/ dffruglqj
wr Iljxuh 415/ diwhu ghprgxodwlrq dqg wudqvplvvlrq ryhu d pxowlsdwk fkdqqho1
Wklv lv lq frqwudvw wr vlqjoh0fduulhu v|vwhpv/ zkhuh wkh uhfhlyhg vljqdo jhqhudoo|
lv fruuxswhg e| LVL1
D eorfn ghvfulswlrq ri d frqyhqwlrqdo RIGP prgxodwru dqg ghprgxodwru
lv suhvhqwhg lq Iljxuh 4161 Iluvw wkh frghg elwv duh ihg wr wkh prgxodwru zkhuh
jurxsv ri orj5 P elwv duh pdsshg wr P 0du| v|perov/ zkhuh P lv wkh vl}h ri
wkh v|pero doskdehw rq hdfk fduulhu1 Wkhq Q ri wkhvh v|perov duh jurxshg
lqwr rqh RIGP v|pero +qrwh/ krzhyhu/ wkdw vrph ri wkh rxwhu fkdqqhov pd|
eh }hur0ydoxhg,1 Hdfk gdwd v|pero lv dvvljqhg wr d vshflf vxe0fkdqqho e|
wkh LGIW/ wkh lqyhuvh glvfuhwh Irxulhu wudqvirup1 Qrupdoo| wklv odwwhu rshud0
wlrq lv shuiruphg dv dq lqyhuvh idvw Irxulhu wudqvirup +LIIW, gxh wr lwv orzhu
frpsoh{lw|1 Wkhq wkh f|folf h{whqvlrq lv dgghg dqg qdoo| sxovh vkdslqj lv
shuiruphg ehiruh wkh vljqdo lv xs0frqyhuwhg dqg wudqvplwwhg ryhu wkh fkdqqho1
Dw wkh uhfhlyhu/ wkh uhfhlyhg vljqdo lv grzq0frqyhuwhg dqg vdpsohg1 Wkhq
wkh vdpsohv duh ihg lq sdudooho wr wkh fkdqqho hvwlpdwru/ wkh IIW dqg wkh
dxwrpdwlf jdlq frqwuro2frduvh v|qfkurql}hu/ uhvshfwlyho|1 Wkh dxwrpdwlf jdlq
frqwuro +DJF, wudfnv wkh lqfrplqj srzhu dqg dgmxvwv wkh ohyho vr wkdw wkh
ixoo g|qdplf udqjh ri wkh dqdorj0wr0gljlwdo frqyhuwhu lv xvhg1 Wkh wdvn ri wkh
v|qfkurql}hu/ ru pruh suhflvh wkh frduvh v|qfkurql}hu/ lv wr qg wkh vwduw ri hdfk
gdwd sdfnhw dqg hdfk RIGP v|pero/ dqg wr qg dq hvwlpdwh ri wkh iuhtxhqf|
rvhw vr wkdw frpshqvdwlrq fdq eh pdgh ehiruh wkh vdpsohv duh ihg wr wkh IIW
4161 RIGP 44
coded
bits c cyclic ext. x
modulation IFFT pulse shaping
channel
channel
estimation
soft
decisions demodu- r
decision FFT
lation
AGC/
synch.
eorfn1 Wkh RIGP ghprgxodwru lv yhu| vhqvlwlyh wr iuhtxhqf| rvhwv ^97` dqg
lq rughu wr dyrlg LFL iuhtxhqf| rvhw frpshqvdwlrq kdv wr eh pdgh ehiruh wkh
GIW1
Wkh fkdqqho hvwlpdwru lv xvhg wr hqdeoh frkhuhqw prgxodwlrq vfkhphv1 Wkh
wlph glvshuvlyh fkdqqho uhvxowv lq d skdvh vkliw dqg dq dpsolwxgh dwwhqxdwlrq
dffruglqj wr +41:, dqg wkhuhiruh wkh ydoxhv ri wkh wudqvihu ixqfwlrq dw wkh gli0
ihuhqw vxe0fkdqqho iuhtxhqflhv kdyh wr eh hvwlpdwhg lq rughu wr xvh frkhuhqw
ghwhfwlrq1 Lq dgglwlrq/ li wkhuh lv d vpdoo wlplqj rvhw/ wklv zloo uhvxow lq olq0
hduo| lqfuhdvlqj skdvh ehwzhhq wkh vxe0fkdqqhov1 Wkh fkdqqho hvwlpdwru fdq
hvwlpdwh dqg frpshqvdwh iru wkhvh skdvh vkliwv vlqfh wkh| fdq eh uhjdughg dv
d sduw ri wkh fkdqqho wudqvihu ixqfwlrq ^59`1
Wkh IIW eorfn shuirupv d GIW rq wkh uhfhlyhg RIGP v|pero lq rughu wr
uhwulhyh wkh gdwd v|perov1 Wkhvh v|perov duh wkhq sdvvhg wr wkh ghprgxodwlrq
vwdjh zkhuh wkh lq xhqfh ri wkh fkdqqho lv uhpryhg vr wkdw kdug ru vriw ghflvlrqv
fdq eh pdgh1 Wkhvh ghflvlrqv duh wkhq sdvvhg wr wkh fkdqqho ghfrghu dv lq
Iljxuh 4141
wr shuirup dq IIW lq rughu wr ghprgxodwh wkh gdwd/ exw wkh htxdol}hu fdq rq
wkh rwkhu kdqg eh uhdoo| vlpsoh vlqfh d wlph glvshuvlyh fkdqqho rqo| uhvxowv lq
d pxowlsolfdwlrq e| wkh ydoxh ri wkh fkdqqho wudqvihu ixqfwlrq diwhu wkh IIW1
Lq d vlqjoh0fduulhu uhfhlyhu wkhuh lv qr IIW eorfn/ exw wkh htxdol}hu pxvw eh
ghvljqhg wr plwljdwh wkh LVL lqvwhdg1 Wkh frpsoh{lw| ri wkh htxdol}hu fdq eh
txlwh kljk li wkh LVL vsdqv ryhu vhyhudo v|perov/ exw li lw vsdqv ryhu rqo| d ihz
v|perov wkh vroxwlrq lv vwudljkwiruzdug1 Dq hvwlpdwh ri wkh frpsoh{lw| ri wkh
htxdol}hu lv irxqg e| frpsdulqj wkh pd{lpxp h{fhvv ghod| ri wkh fkdqqho wr
wkh v|pero wlph> li wkh LVL vsdqv ryhu vhyhudo v|perov/ wkhq RIGP lv wr suhihu
wr wkh xvh ri orqj +dqg wkhuhe| frpsoh{, htxdol}huv ^:`1
Dqrwkhu dgydqwdjh ri pxowl0fduulhu prgxodwlrq lv wkdw lw lv srvvleoh wr xvh
dgdswlyh prgxodwlrq rq d vxe0fkdqqho edvlv1 Wklv phdqv wkdw glhuhqw prg0
xodwlrq ohyhov fdq eh xvhg rq glhuhqw vxe0fkdqqhov1 Rq d edg vxe0fkdqqho
zkhuh wkh fkdqqho dwwhqxdwlrq lv kljk/ d orz0ohyho prgxodwlrq vfkhph/ vxfk dv
elqdu| SVN/ fdq eh xvhg dqg rq d jrrg vxe0fkdqqho d kljkhu rughu prgxodwlrq/
vxfk dv 490TDP/ fdq eh xvhg1 Wklv phdqv wkdw wkh fkdqqho lv pruh h!flhqwo|
xvhg dqg wkdw wkh elw huuru udwh +EHU,/ fdq eh ghfuhdvhg vljqlfdqwo| frp0
sduhg wr {hg prgxodwlrq vfkhphv ^59`1 Qrwh krzhyhu/ wkdw dgdswlyh vfkhphv
uhtxluh erwk h{wud kdugzduh dqg frqwuro vljqdolqj1
Wkh pdlq glvdgydqwdjhv ri xvlqj pxowl0fduulhu v|vwhpv/ ru RIGP v|vwhpv
lq sduwlfxodu/ duh wkh vhqvlwlylw| wr iuhtxhqf| rvhwv dqg wkh odujh g|qdplf
udqjh ri wkh vljqdo wr eh wudqvplwwhg ^:`1 Iuhtxhqf| rvhwv fdxvh erwk LFL dqg
dwwhqxdwlrq ri wkh ghvluhg vljqdo1 Wkh vhqvlwlylw| wr iuhtxhqf| rvhwv fdq eh
h{sodlqhg e| wkh odujh iuhtxhqf| vlgh orehv ri wkh vljqdo dqg wkh vhqvlwlylw|
fdq eh ghfuhdvhg e| xvlqj d zlqgrz ixqfwlrq zlwk d pruh dwwudfwlyh iuhtxhqf|
fkdudfwhulvwlfv frpsduhg wr wkh rqh uhvxowlqj iurp wkh uhfwdqjxodu wlph gr0
pdlq zlqgrz ^:`1 Wkh odujh g|qdplf udqjh fdq eh xqghuvwrrg e| orrnlqj dw
wkh RIGP vljqdo dv d vxp ri Q vlqxvrlgv1 Vrphwlphv wkh zdyhirupv dgg
frqvwuxfwlyho|/ zklfk phdqv wkdw d shdn lv surgxfhg1 Iru prvw ri wkh wlph/
krzhyhu/ wkh zdyhirupv kdyh glhuhqw vljqv vxfk wkdw wkh vxp dfklhyhv orz
dpsolwxgh1
Fkdswhu 5
Uhvhdufk Wrslfv
46
47 FKDSWHU 51 UHVHDUFK WRSLFV
fhlyhg zlwkrxw lq xhqfh iurp hdfk rwkhu/ wkh surfhvvhv ri fkdqqho hvwlpdwlrq
dqg ghwhfwlrq fdq eh vhsdudwhg1 Dovr/ wudlqlqj v|perov duh vxlwdeoh iru sdfnhw0
edvhg frppxqlfdwlrq vlqfh wkhvh uhvxow lq urexvw ehkdylru zlwk qr vljqlfdqw
wudqvlhqw hhfwv ^7;`1
Lq wkh jhqhudo fdvh wkh rswlpdo uhfhlyhu/ lq wkh vhqvh ri plqlpxp vh0
txhqfh huuru suredelolw| zkhq qr sulru fkdqqho hvwlpdwhv duh dydlodeoh/ lv wkh
pd{lpxp0olnholkrrg +PO, uhfhlyhu pdnlqj mrlqw ghwhfwlrq dqg fkdqqho hvwlpd0
wlrq ^7;`1 Wkh wdvn lv wr qg d fkdqqho hvwlpdwh dqg d gdwd vhtxhqfh wkdw pdnh
wkh dfwxdo uhfhlyhg vhtxhqfh dv suredeoh dv srvvleoh/ l1h1 wr qg
+a a
f> k,PO @ duj pd{ i+umf> k,
f>k
+514,
zkhuh u lv wkh uhfhlyhg vhtxhqfh/ f lv wkh gdwd vhtxhqfh/ k lv wkh vhtxhqfh ri wkh
fkdqqho uhvsrqvhv/ i+, ghqrwhv wkh suredelolw| ghqvlw| ixqfwlrq dqg duj pd{
ghqrwhv wkh dujxphqw pd{lpl}lqj wkh ixqfwlrq1 Wklv dssurdfk ohdgv/ krzhyhu/
wr frpsoh{ uhfhlyhu vwuxfwxuhv dqg wkhuhiruh rwkhu ohvv frpsoh{ dowhuqdwlyhv duh
ri lqwhuhvw1 Lq fdvh fkdqqho hvwlpdwlrq dqg gdwd ghwhfwlrq fdq eh shuiruphg
vhsdudwho|/ wkh vwuxfwxuh fdq eh vlpsolhg e| ghulylqj wkh fkdqqho hvwlpdwhv
ehiruh ghwhfwlrq wdnhv sodfh1 Jlyhq d fhuwdlq fkdqqho hvwlpdwh k/ a wkh rswlpdo
ghwhfwru/ lq wkh vhqvh ri plqlpdo vhtxhqfh huuru suredelolw|/ lv wkh Ylwhuel
ghwhfwru ^89`1 Wklv htxdol}hu jlyhv wkh pd{lpxp0olnholkrrg vhtxhqfh/ zklfk
phdqv wkdw lw pdnhv d ghflvlrq lq idyru ri wkh vhtxhqfh wkdw pd{lpl}hv wkh
suredelolw| ri uhfhlylqj wkh dfwxdo uhfhlyhg vhtxhqfh/ l1h1
l , @ uSl >
K+S +517,
sSl
zkhuh S @ ^S3 > S4 > = = = > SO4 ` lv wkh vhw ri slorw v|pero srvlwlrqv xvhg iru wkh
hvwlpdwh/ sSl lv wkh slorw v|pero dw srvlwlrq Sl dqg uSl lv wkh fruuhvsrqglqj
uhfhlyhg ydoxh diwhu wkh GIW1 Wkh fkdqqho hvwlpdwh/ K a n>o / ri vxe0fkdqqho n dw
wlph o fdq wkhq eh frpsxwhg e| olqhdu owhulqj
O
[ 4
a n>o @
K l ,>
zn>o +Sl , K+S +518,
l@3
zkhuh O lv wkh qxpehu ri slorw v|perov xvhg iru wkh hvwlpdwh dqg zn>o +Sl , lv wkh
fkdqqho hvwlpdwlrq frh!flhqw xvlqj wkh slorw v|pero dw srvlwlrq Sl iru wkh gdwd
v|pero dw srvlwlrq +n> o,1 Li wkh fkdqqho lv zlgh0vhqvh vwdwlrqdu|/ wkh rswlpdo
O 4 yhfwru ri owhu frh!flhqwv lv wkh vroxwlrq wr wkh Zlhqhu0Krsi htxdwlrq
^64`
zkhuh n>o ghqrwhv wkh O 4 furvv0fruuhodwlrq yhfwru zlwk hohphqwv n>l +l, @
+Sl ,` dqg ghqrwhv wkh O O dxwrfruuhodwlrq pdwul{ zlwk hohphqwv
H^Kn>o K
l, K
l>m @ H^K+S +Sm ,`1 Wkh phdq vtxduh huuru +PVH, fdq eh xvhg dv d shu0
irupdqfh phdvxuh ri wkh hvwlpdwlrq owhu/ exw fdq dovr eh xvhg wr fdofxodwh wkh
uhvxowlqj EHU1 Wkh PVH xqghu lghdo frqglwlrqv +nqrzq fruuhodwlrq pdwulfhv,
lv jlyhq e| ^64`
zkhuh 5K lv wkh yduldqfh ri wkh frpsoh{ frpsrqhqwv ri wkh fkdqqho1 Ilqdoo|/
diwhu whglrxv ghulydwlrqv/ wkh EHU iru T0SVN fdq eh fdofxodwhg dv ^89`
4 n>o
Sn>o +huuru, @ t > +51;,
5 5 5 5
n>o
zkhuh
5K
n>o @ Hv s 5 +51<,
+ K . 5q ,+ 5K . Mn>o ,
pre-compensate
f estimate
uplink downlink
51415 Suh0frpshqvdwlrq
Suh0frpshqvdwlrq/ vrphwlphv fdoohg suhfrglqj/ lv d whfkqltxh zkhuh wkh vljqdo
lv frpshqvdwhg ehiruh lw lv wudqvplwwhg lq rughu wr lpsuryh wkh shuirupdqfh lq
vrph vhqvh1 Wkh whfkqltxh fdq dovr eh xvhg wr pryh frpsoh{ ixqfwlrqv iurp
wkh uhfhlyhu wr wkh wudqvplwwhu/ wkhuhe| hqdeolqj ohvv frpsoh{ whuplqdo vwuxf0
wxuhv zlwk shuirupdqfh vlplodu wr pruh frpsoh{ rqhv1 Frpshqvdwlrq fdq/ iru
h{dpsoh/ eh xvhg wr plwljdwh LVL lq vlqjoh0fduulhu v|vwhpv wr dyrlg wkh qhhg
iru htxdol}dwlrq dw wkh uhfhlyhu ^78` ^;<` ^<3` ru iru dqwhqqd glyhuvlw| xvlqj rqo|
rqh uhfhlyh dqwhqqd ^44` ^5;` ^84` ^9;`1 Wkh whfkqltxh fdq dovr eh xvhg wr frp0
shqvdwh wkh idglqj fkdqqho lq rughu wr lpsuryh wkh shuirupdqfh ^F` ^7:` ru
wr pdlqwdlq ruwkrjrqdolw| ehwzhhq wkh glhuhqw xvhu vljqdov lq pxowl0fduulhu
vsuhdg0vshfwuxp +PF0VV, v|vwhpv ^6;` ^;8` ^;;`1 Gxh wr wkh vlpsolflw| dqg ur0
exvwqhvv d yduldqw iru dqwhqqd glyhuvlw| lv xvhg lq wkh GHFW v|vwhp ^5` zkhuh
wkh dqwhqqd uhfhlylqj wkh prvw srzhu gxulqj xsolqn lv xvhg iru grzqolqn wudqv0
plvvlrq1 Dovr lq wkh XPWV2Z0FGPD v|vwhp d vlpsoh skdvh frpshqvdwlrq
v|vwhp lv xvhg lq wlph glylvlrq gxsoh{ +WGG, prgh wr doljq wkh skdvhv ri
glhuhqw sdwkv ^68`1
Wkh whfkqltxh uholhv rq wkdw wkh fkdqqho fkdudfwhulvwlfv duh uhflsurfdo dqg
frqvwdqw gxulqj rqh uhfhlyh0wudqvplw0f|foh1 Wkhuhiruh suh0frpshqvdwlrq lv vxlw0
deoh zkhq WGG lv xvhg dqg wkh wlph ehwzhhq uhfhlylqj prgh dqg wudqvplvvlrq
prgh lv vkruw frpsduhg wr wkh frkhuhqfh wlph ri wkh fkdqqho1 Wkh fkdqqho
fkdudfwhulvwlfv duh lghqwlhg lq rqh gluhfwlrq dqg wkhq frpshqvdwlrq lv pdgh
sulru wr wudqvplvvlrq lq wkh rwkhu gluhfwlrq/ dv ghvfulehg lq Iljxuh 5141 Vlqfh
frpshqvdwlrq lv pdgh lq dgydqfh/ wkh pd{lpxp eorfn ohqjwkv duh ghwhuplqhg
e| wkh wlph0fruuhodwlrq ixqfwlrq ri wkh fkdqqho1 Wkh vkdsh ri wklv fruuhodwlrq
ixqfwlrq lv/ dprqj rwkhu wklqjv/ ghwhuplqhg e| wkh Grssohu iuhtxhqf|1 Kljk
4; FKDSWHU 51 UHVHDUFK WRSLFV
Grssohu iuhtxhqflhv qdwxudoo| fdoo iru vkruw eorfnv dqg wkhuhiruh wkh whfkqltxh
lv prvw vxlwdeoh iru v|vwhpv rshudwlqj dw shghvwuldq vshhgv1
Zkhq ixoo| frpshqvdwlqj wkh skdvh dqg dpsolwxgh ri d dw0idglqj fkdqqho
wkh frpsoh{ hqyhorsh ri wkh wudqvplwwhg vljqdo lv jlyhq e|
zkhuh {+w, lv wkh frpsoh{ hqyhorsh ri wkh xqfrpshqvdwhg vljqdo dqg K+w, a lv
wkh fkdqqho hvwlpdwh1 Li wkh fkdqqho lv Ud|ohljk idglqj wkh vljqdo xfwxdwlrqv
ehfrph vhyhuh dqg wkhuhiruh wkh dpsolwxgh kdv wr eh olplwhg/ hlwkhu e| olplwlqj
wkh rxwsxw srzhu ru e| lqklelwlqj wkh wudqvplwwhu zkhq wkh fkdqqho dwwhqxdwlrq
lv odujh ^F` ^7:`1 Lq d Ud|ohljk idglqj fkdqqho wkh idglqj glsv fdxvh prvw ri
wkh elw huuruv1 Li lw lv srvvleoh wr frpshqvdwh iru wkhvh wkhuh zloo/ iru xqfrghg
v|vwhpv/ eh d odujh lpsuryhphqw lq shuirupdqfh1 Lq ^7:` lw zdv vkrzq wkdw
lw lv srvvleoh wr ghfuhdvh wkh phdq wudqvplwwhu srzhu e| : gE +iru d {hg
EHU ri 4306 , zkhq lqwurgxflqj srzhu frpshqvdwlrq lq d qrq0frkhuhqw IVN
v|vwhp1 Lq ^F` lw lv vkrzq wkdw wkh EHU fdq eh ghfuhdvhg 53 wlphv frpsduhg
wr dq xqfrpshqvdwhg v|vwhp iru frkhuhqw TSVN dw He @Q3 @ 43 gE1 Wklv/
krzhyhu/ uhvxowv lq dq lqfuhdvh ri wkh shdn0wr0dyhudjh srzhu udwlr e| 43 gE vr
wklv vwudwhj| pljkw eh ri olplwhg lqwhuhvw iru zluhohvv dssolfdwlrqv1
Srzhu frpshqvdwlrq kdv dovr ehhq sursrvhg wr pdlqwdlq ruwkrjrqdolw| eh0
wzhhq xvhuv lq PF0VV v|vwhpv ^;8`1 Lq dq PF0VV v|vwhp/ ru pruh vshflfdoo|
lq dq PF0FGPD v|vwhp/ wkh gdwd v|pero iurp rqh xvhu lv vsuhdg wr pdq|
vxe0fkdqqhov dqg wkhq gdwd v|perov iurp vhyhudo xvhuv duh wudqvplwwhg rq wkh
vdph vxe0fkdqqhov vlpxowdqhrxvo| ^6<`1 Vlqfh hdfk gdwd v|pero lv wudqvplw0
whg ryhu vhyhudo vxe0fkdqqhov/ d iuhtxhqf| glyhuvlw| jdlq fdq eh dfklhyhg zkhq
dgglqj wkh frpsrqhqwv iurp wkh glhuhqw vxe0fkdqqhov1 Li frkhuhqw ghwhfwlrq
lv xvhg wkhq htxdol}dwlrq lv uhtxluhg lq wkh vdph zd| dv lq frqyhqwlrqdo RIGP
v|vwhpv1 Krzhyhu/ wkh ruwkrjrqdolw| ehwzhhq xvhuv fdq qrw eh suhvhuyhg lq
d pxowlxvhu v|vwhp zkhq htxdol}lqj wkh xsolqn vljqdo lq wkh uhfhlyhu1 Wkhuh0
iruh pxowlxvhu ghwhfwlrq ru suh0frpshqvdwlrq lv uhtxluhg lq rughu wr dfklhyh
jrrg shuirupdqfh ^;8`1 Wkh txhvwlrq iru suh0frpshqvdwlrq lq wklv dssolfdwlrq
lv zkhwkhu rqh fdq drug wkh uhvxowlqj lqfuhdvh lq shdn0wr0dyhudjh srzhu udwlr
uhsruwhg lq ^F`1
Suh0frpshqvdwlrq fdq dovr eh xvhg iru dqwhqqd glyhuvlw|1 Wkh whfkqltxh
kdv ehhq xvhg wr fkrrvh wudqvplw dqwhqqd lq wkh GHFW v|vwhp ^5`/ wr dfklhyh
frqvwuxfwlyh dgglwlrq iurp wzr dqwhqqdv ^44` ^5;` ru iurp dq duud| ri dqwhq0
qdv ^84`1 Lq ^44` d glyhuvlw| jdlq ri 9043 gE zdv uhsruwhg +dw EHU 4306 , iru d
v|vwhp xvlqj wkh dqwhqqd glyhuvlw| duudqjhphqw dqg dq htxdol}hu +wr plwljdwh
LVL,/ frpsduhg wr d v|vwhp xvlqj rqo| wkh htxdol}hu1 Wklv jdlq lv ri wkh vdph
rughu dv wkh 43 gE jdlq wkdw fdq eh dfklhyhg +iru TSVN dw EHU 4306 , xvlqj wzr
uhfhlyh dqwhqqdv lq d Ud|ohljk idglqj fkdqqho ^89`1 Dqwhqqd suh0frpshqvdwlrq
lv wkhuhiruh dq h!flhqw whfkqltxh wr jhw forvh wr rswlpdo dqwhqqd glyhuvlw| xvlqj
5151 PXOWL0FDUULHU V\VWHP GHVLJQ 4<
rqo| rqh uhfhlyh dqwhqqd1 Lq ^9;` wkh vzlwfkhg dqwhqqd glyhuvlw| v|vwhp zdv
jhqhudol}hg wr RIGP v|vwhpv1 Lq wklv v|vwhp/ vrph ri wkh vxe0fkdqqhov zhuh
wudqvplwwhg iurp rqh dqwhqqd dqg wkh uhpdlqlqj vxe0fkdqqhov zhuh wudqvplw0
whg iurp wkh rwkhu dqwhqqd1 Wklv v|vwhp dovr uhvxowhg lq d forvh wr rswlpdo
glyhuvlw| jdlq/ dvvxplqj lqghshqghqw vxe0fkdqqhov dqg lghdo v|qfkurql}dwlrq
dqg fkdqqho hvwlpdwlrq1 Wkh txhvwlrq lv/ krzhyhu/ krz wr ghulyh wkh qhfhvvdu|
fkdqqho hvwlpdwhv lq dq h!flhqw zd|1 Lw lv srvvleoh wr xvh vrph nlqg ri slorw0
edvhg fkdqqho hvwlpdwlrq rq d shu vxe0fkdqqho edvlv/ exw lw lv kdug wr xvh wkh
fruuhodwlrq ehwzhhq wkh vxe0fkdqqhov iru hvwlpdwlrq lq wkh uhfhlyhu vlqfh glhu0
hqw dqwhqqdv duh xvhg1 Dowhuqdwlyho|/ skdvh0rqo| suh0frpshqvdwlrq ^F` fdq eh
xvhg lq rughu wr doljq wkh skdvhv dqg dyrlg fkdqqho hvwlpdwlrq dw wkh uhfhlyhu1
D jhqhudo sureohp zkhq xvlqj suh0frpshqvdwlrq lq dq| irup lv wkdw wkh
fkdqqho qhhgv wr eh frqvwdqw/ ru doprvw frqvwdqw/ gxulqj hdfk uhfhlyh0wudqvplw0
f|foh1 Lq ^F` wkh pd{lpxp eorfn ohqjwk lv fdofxodwhg zkhuh wkh lqvwdqwdqhrxv
EHU jhwv zruvh wkdq wkdw ri glhuhqwldo ghwhfwlrq1 Lw lv vkrzq wkdw iru skdvh0
rqo| suh0frpshqvdwlrq d eorfn ohqjwk ri iG Weorfn 3=59 fdq eh xvhg ehiruh wkh
shuirupdqfh ghwhulrudwhv1 Wr frpsduh wklv qxpehu wr dq h{lvwlqj WGG v|vwhp
wklv fruuhvsrqgv wr d eorfn ohqjwk 73 wlphv orqjhu wkdq wkh rqh lq wkh GHFW
v|vwhp dw 4<33 PK} dqg d yhorflw| ri 43 np2k1 Wkh frqfoxvlrq dovr krogv zkhq
d zruvw0fdvh ghvljq lv pdgh iru wkh fkdqqho hvwlpdwruv ^F`1 Vlplodu eorfn ohqjwkv
fdq suredeo| eh h{shfwhg iru wkh rwkhu suh0frpshqvdwhg v|vwhpv/ exw wklv lv ri
frxuvh ghshqghqw rq wkh vhqvlwlylw| ri wkh uhfhlyhu wr glhuhqw nlqgv ri huuruv1
Wkh EHU lv ghshqghqw rq wkh delolw| ri pdnlqj fkdqqho suhglfwlrqv1 Li rqo| wkh
odvw hvwlpdwhv lq wkh xsolqn eorfn duh xvhg iru frpshqvdwlrq ri wkh grzqolqn
eorfn/ wkh shuirupdqfh lv ghjudghg frpsduhg wr wkh xvh ri rswlpdo suhglfwlrq1
Iru wkh skdvh0rqo| frpshqvdwlrq v|vwhp lq ^F` wkh pd{lpxp eorfn ohqjwk lv
dssur{lpdwho| iG Weorfn 3=38 zkhq wkh odvw hvwlpdwh ri wkh xsolqn eorfn lv
xvhg iru skdvh frpshqvdwlrq1 Lq ^6;`/ zkhuh wkh srzhu lv suh0frpshqvdwhg lq
dq PF0FGPD v|vwhp/ wkh pd{lpxp eorfn ohqjwk iru d wzr0irog ghjudgdwlrq lq
wkh EHU lv iG Weorfn 3=41 Erwk wkhvh qxpehuv fdq/ iru h{dpsoh/ eh frpsduhg
wr wkh dyhudjh gxudwlrq ri d 43 gE idglqj gls zklfk lv iG Wgls 3=46 iru d
Ud|ohljk idglqj fkdqqho zlwk lvrwurslf vfdwwhulqj ^;:`1
Lq wklv vhfwlrq wkh edfnjurxqg ri wkh uhvhdufk duhdv uhodwhg wr RIGP v|vwhp
ghvljq lv suhvhqwhg1 Wkh wrslfv suhvhqwhg fryhu rswlpl}dwlrq ri wkh vxe0fkdqqho
edqgzlgwk/ slorw sdwwhuq ghvljq dqg suhdpeoh0edvhg v|qfkurql}dwlrq1 V|vwhp
ghvljq ri wkh sk|vlfdo od|hu lqfoxghv pruh wrslfv wkdq suhvhqwhg khuh/ exw wkhvh
duh rxw ri vfrsh iru wkh wkhvlv1 Rwkhu wrslfv lqfoxgh/ iru h{dpsoh/ fkrlfh ri wkh
prgxodwlrq irupdw/ ghvljq ri wkh frglqj vfkhph dqg pxowlsoh dffhvv vfkhph1
53 FKDSWHU 51 UHVHDUFK WRSLFV
wkh LFL lqfuhdvhv wkh xvhixo srzhu dxwrpdwlfdoo| ghfuhdvhv e| wkh vdph dprxqw1
Lq wkh iroorzlqj/ zh wuhdw LFL lq pruh ghwdlo/ vlqfh wklv lv d yhu| lpsruwdqw lvvxh
lq RIGP v|vwhp ghvljq1
Dq lpsruwdqw hhfw wr frqvlghu lq wkh v|vwhp ghvljq lv lqwhuihuhqfh gxh wr
fkdqqho fkdqjhv gxulqj dq RIGP v|pero1 Wkh duhd kdv/ iru h{dpsoh/ ehhq
vwxglhg lq ^76` ^94`1 Wklv lqwhuihuhqfh lv fdxvhg e| pryhphqwv ri hlwkhu wkh
wudqvplwwhu/ wkh uhfhlyhu ru uh hfwlrq srlqwv dqg lw lv kdug/ li qrw lpsrvvleoh/
wr frpshqvdwh iru ^94`1 Wudqvplwwhu ru uhfhlyhu pryhphqwv uhvxow lq wkh vdph
nlqg ri lqwhuihuhqfh dv rvfloodwru rvhwv ^97`/ exw lq d zluhohvv hqylurqphqw wkh|
fdxvh d Grssohu vsuhdg udwkhu wkdq d iuhtxhqf| vkliw dqg wkhuhiruh wkh| duh
pxfk kdughu wr frpshqvdwh1 Wkh srzhu ri wkh LFL gxh wr wkh wlph0vhohfwlyh
idglqj lv sursruwlrqdo wr wkh vtxduhg qrupdol}hg Grssohu iuhtxhqf| +iG Wv ,5 /
dqg wkhuhiruh lw lv lqyhuvho| sursruwlrqdo wr wkh vtxduh ri wkh vxe0fkdqqho edqg0
zlgwk ivf5 1 Wkh yduldqfh ri wkh htxlydohqw qrlvh fdxvhg e| wkh wlph0vhohfwlyh
fkdqqho fdq eh fdofxodwhg dv ^97`
5
5wv @ +iG Wv ,5 > +5145,
9
zkhq wkh uhfhlyhg vljqdo srzhu lv qrupdol}hg wr 41
Lq ^E`/ wkh vxe0fkdqqho edqgzlgwk lv rswlpl}hg zlwk uhvshfw wr wkh sk|vlfdo
skhqrphqd ghvfulehg deryh/ wkh hqhuj| orvv dqg wkh LFL gxh wr wkh wlph0
vhohfwlyh fkdqqho1 Wkh rswlpdo vxe0fkdqqho edqgzlgwk lv ghshqghqw rq wkh
vljqdo0wr0qrlvh udwlr vlqfh wkhuh lv d edodqfh ehwzhhq wkh LFL dqg wkh qrlvh>
wkhuhiruh lw lv kdug wr jlyh dq h{dfw h{suhvvlrq iru d vxlwdeoh vxe0fkdqqho edqg0
zlgwk1 Krzhyhu/ iru preloh dssolfdwlrqv zkhuh He @Q3 @ 43 58 gE d qru0
pdol}hg Grssohu iuhtxhqf| ri iG Wv 3=34 3=35 vhhpv vxlwdeoh ^E` ^99`1 Wklv
fruuhvsrqgv wr d ghjudgdwlrq lq VQU ri ohvv wkdq 4 gE ^97` dqg uhvxowv lq d
vljqdo0wr0lqwhuihuhqfh udwlr h{fhhglqj 63 gE ^76`1
LFL fdq dovr eh fdxvhg e| iuhtxhqf| rvhwv dqg iru frqvxphu surgxfwv/
zkhuh orz0frvw lpsohphqwdwlrqv duh lpsruwdqw/ wklv pljkw eh d vljqlfdqw
vrxufh ri lqwhuihuhqfh1 Wkh lqwhuihulqj qrlvh ohyho fdq eh dssur{lpdwhg e|
^97`
5
5ir +i Wv ,5 > +5146,
6
zkhuh i ghqrwhv wkh iuhtxhqf| rvhw1 Dv vhhq lq +5146,/ wkh yduldqfh lv djdlq
sursruwlrqdo wr wkh vtxduh ri wkh qrupdol}hg iuhtxhqf| rvhw/ %5f @ +i Wv ,5 /
dqg wkhuhiruh lqyhuvho| sursruwlrqdo wr wkh vtxduh ri wkh vxe0fkdqqho edqg0
5 1 Qrupdoo|/ iuhtxhqf| rvhw frpshqvdwlrq lv pdgh sulru wr wkh GIW
zlgwk ivf
dqg wkh lqwhuhvwlqj sdudphwhu lv wkhuhiruh wkh uhpdlqlqj iuhtxhqf| rvhw diwhu
frpshqvdwlrq1 Wkh h{suhvvlrq iru wkh yduldqfh ri wkh htxlydohqw qrlvh fdxvhg
e| iuhtxhqf| rvhw lv vlplodu wr wkh qrlvh fdxvhg e| wkh wlph0vhohfwlyh fkdqqho1
Wkhuhiruh/ iru wkh vdph ohyho ri lqwhuihuhqfh wkh pd{lpxp iuhtxhqf| rvhw lv
55 FKDSWHU 51 UHVHDUFK WRSLFV
ri wkh vdph rughu dv wkh Grssohu iuhtxhqf|/ w|slfdoo| %f 3=34 3=35/ exw wklv
vkrxog eh vhhq rqo| dv d urxjk jxlgholqh1
Rwkhu hhfwv wkdw qhhg frqvlghudwlrq duh wkh lq xhqfh ri skdvh qrlvh ri wkh
rvfloodwru dqg wkh hhfw ri vdpsolqj huuruv1 Skdvh qrlvh kdv ehhq vwxglhg lq/ iru
h{dpsoh/ ^9` ^88` ^93` dqg lw ohdgv wr wzr nlqgv ri glvwruwlrq1 Iluvw/ wkhuh lv dq
LFL whup/ zkrvh yduldqfh lqfuhdvhv zlwk wkh qxpehu ri vxe0fduulhuv1 Lw lv kdug
wr jhw qxphulfdo h{suhvvlrqv iru wkh yduldqfh vlqfh wklv lv ghshqghqw rq wkh
prgho ri wkh skdvh qrlvh lqfoxglqj srvvleoh ihhgedfn lq wkh flufxlw1 Iru Zlhqhu
skdvh qrlvh/ wkh fruuhvsrqglqj VQU orvv gxh wr wkh LFL lv sursruwlrqdo wr wkh
rvfloodwru olqhzlgwk dqg lqyhuvho| sursruwlrqdo wr wkh vxe0fkdqqho edqgzlgwk
^88`1 Wkh skdvh qrlvh dovr ohdgv wr d skdvh huuru frpprq wr doo vxe0fkdqqhov
zlwklq dq RIGP v|pero1 Wkh yduldqfh ri wklv glvwruwlrq ghfuhdvhv dv wkh vxe0
fkdqqho edqgzlgwk ghfuhdvhv ^93`1 Wkh fruuhodwlrq ri wkh frpprq skdvh huuru
lv lq jhqhudo orz ehwzhhq vxffhvvlyh RIGP v|perov/ vr frpshqvdwlrq rq d
v|pero0e|0v|pero edvlv lv uhtxluhg li wkh yduldwlrqv duh vljqlfdqw1
Vdpsolqj rvhwv ohdg wr d skdvh urwdwlrq ri wkh v|perov1 Ixuwkhu/ li vrph
sdwkv ri wkh fkdqqho idoo rxwvlgh wkh jxdug lqwhuydo/ LFL dqg LVL rffxu1 Wkh
uhvxowlqj lqwhuihuhqfh fdq eh dssur{lpdwhg dv Jdxvvldq qrlvh zlwk yduldqfh ^97`
[ 5 %l q %l r5
5wr mdl +w,m 5 > +5147,
l
Q Q
wkurxjk wkh wlph rvhw w dqg wkh vdpsoh wlph Wvdps @ 4@ivdps 1 Dv ehiruh/ Q
ghqrwhv wkh qxpehu ri vxe0fkdqqhov/ J ghqrwhv wkh ohqjwk ri wkh f|folf suh{/
l ghqrwhv wkh ghod| ri sdwk l/ dqg dl +w, ghqrwhv wkh fkdqqho wds ydoxh ri sdwk
l1 Wkh yduldqfh ri wkh LFL dqg LVL gxh wr vdpsolqj rvhwv lv kljko| ghshqghqw
rq wkh srzhu ghod| suroh ri wkh fkdqqho ^97` dqg wkh glvwxuedqfh lv lq sulqflsoh
sursruwlrqdo wr wkh vxe0fkdqqho edqgzlgwk ^99`1 Li pdq| vxe0fkdqqhov duh xvhg/
wkh uhodwlyh lpsruwdqfh ri vdpsohv fruuhvsrqglqj wr sdwkv idoolqj rxwvlgh wkh
f|folf suh{ ehfrphv vpdoohu dqg wkhuhiruh wkh ryhudoo glvwruwlrq ghfuhdvhv1
Ilqdoo| vrph uhpdunv derxw wkh uhodwlrq ehwzhhq wkh vxe0fkdqqho edqg0
zlgwk dqg wkh ohqjwk ri sdfnhwv lq zluhohvv v|vwhpv1 Wkh qxpehu ri vxe0
fkdqqhov xvhg ghwhuplqhv wkh plqlpxp qxpehu ri v|perov lq d sdfnhw dqg
wkhuhiruh dovr wkh judqxodulw| zkhq fkrrvlqj wkh sdfnhw ohqjwkv1 Iru odujh gdwd
sdfnhwv wklv pljkw qrw eh d sureohp/ exw li wkhuh duh pdq| vkruw sdfnhwv/ ru
wkhuh duh orz0udwh frqqhfwlrqv zlwk frqvwudlqwv rq wkh ghod|/ wklv hhfw kdv wr
eh frqvlghuhg zkhq ghwhuplqlqj wkh vxe0fkdqqho edqgzlgwk1
5151 PXOWL0FDUULHU V\VWHP GHVLJQ 56
uhshdwhg erwk ehwzhhq wkh v|perov dqg zlwklq wkh eorfn/ zklfk phdqv wkdw
frduvh v|qfkurql}dwlrq fdq eh shuiruphg zlwklq wkh eorfn dqg qh iuhtxhqf|
v|qfkurql}dwlrq fdq eh shuiruphg e| xvlqj dgmdfhqw eorfnv1 Ilqdoo|/ lq ^G`
d uhshdwhg p0vhtxhqfh +pd{lpdo0ohqjwk vhtxhqfh, lv xvhg1 Wklv dssurdfk lv
prwlydwhg e| wkh dwwudfwlyh fruuhodwlrq surshuwlhv iru wkhvh vhtxhqfhv1
Wkhuh duh dovr phwkrgv edvhg rq sulqflsohv rwkhu wkdq uhshdwhg v|perov1
Lq ^75` wkh slorw v|perov lqwhqghg iru fkdqqho hvwlpdwlrq duh xvhg wrjhwkhu
zlwk wkh f|folf suh{ wr ghulyh dq lpsuryhg wlph v|qfkurql}dwlrq vljqdo1 Lq
^46` wkh suhdpeoh lv jhqhudwhg xvlqj rqo| wkh kljkhvw ru orzhvw kdoi ri wkh vxe0
fkdqqhov1 Lq wkh v|vwhp/ wlph v|qfkurql}dwlrq lv dfklhyhg e| frpsdulqj wkh
vshfwudo frqwhqw lq wkh wzr kdoyhv ri wkh qrupdo wudqvplvvlrq edqgzlgwk1 Lq
^98` d vshfldo orz0udwh vljqdolqj fkdqqho xvlqj iuhtxhqf| vkliw nh|lqj +IVN, lv
xvhg iru v|qfkurql}dwlrq sxusrvhv1 Wklv dssurdfk zdv prwlydwhg e| frpsoh{lw|
dujxphqwv vlqfh wkhq d vlpsoh IVN uhfhlyhu fdq eh xvhg gxulqj vohhs prgh1 Lq
^8<` frqwlqxrxv slorw fduulhuv zhuh xvhg wr ghulyh d iuhtxhqf| rvhw hvwlpdwh1
Dqrwkhu dssurdfk lv edvhg rq vkruw srzhu yduldwlrqv lq wkh wudqvplwwhg vljqdo
^95`1 E| xvlqj vr fdoohg qxoo v|perov wkh uhfhlyhu rqo| kdv wr orrn iru fhuwdlq
yduldwlrqv lq wkh uhfhlyhg srzhu wr qg d frduvh wlph hvwlpdwh/ zklfk fdq eh
shuiruphg xvlqj d orz0frpsoh{lw| uhfhlyhu zlwkrxw shuiruplqj dq IIW1
Wkh pdlq glhuhqfh ehwzhhq wkh dssurdfk lq ^G` dqg wkh rwkhu fruuhodwlrq0
edvhg phwkrgv lv wkdw lq ^G` fruuhodwlrq zlwk wkh nqrzq vhtxhqfh lv pdgh sulru
wr pxowlsolfdwlrq zlwk wkh vhfrqg sduw ri wkh uhshdwhg vhtxhqfh zkhq fdofx0
odwlqj wkh wlplqj v|qfkurql}dwlrq1 Wklv v|qfkurql}hu lv fdoohg d suh0fruuhodwlqj
v|qfkurql}hu1 Wkh v|qfkurql}dwlrq vljqdo fdq iru erwk v|qfkurql}huv eh fdofx0
odwhg dv
OS4 NS4
^n> d` @ s ^n q oU d`u^n q oU`
o@3 q@3
N 4 & +5149,
S
s ^n q +o . T,U d`u^n q +o . T,U` >
q@3
zkhuh d lv wkh odj ehwzhhq wkh nqrzq v|qfkurql}dwlrq vhtxhqfh dqg wkh uh0
fhlyhg vhtxhqfh/ O lv wkh qxpehu ri surgxfwv xvhg iru wkh v|qfkurql}dwlrq
vljqdo/ s^n` lv wkh nqrzq vhtxhqfh/ T lv dq h{wud ghod| zklfk fdq eh xvhg wr
lpsuryh wkh shuirupdqfh ri wkh iuhtxhqf| rvhw hvwlpdwh/ N lv wkh fruuhodwru
ohqjwk dqg U lv wkh uhshwlwlrq ohqjwk ri wkh vhtxhqfh1 Wkh rqo| glhuhqfh
ehwzhhq wkh frqyhqwlrqdo fruuhodwlrq0edvhg v|qfkurql}hu dqg wkh v|qfkurql}hu
lq ^G` lv krz wkhvh sdudphwhuv duh fkrvhq1 Wkh frqyhqwlrqdo v|qfkurql}hu lv
vkrzq lq Iljxuh 515 dqg wklv lv rewdlqhg e| fkrrvlqj wkh fruuhodwlrq ohqjwk wr
N @ 4> wkh qxpehu ri surgxfwv xvhg iru wkh hvwlpdwh wr O @ U/ wkh nqrzq
vhtxhqfh wr s^n` @ 4 dqg T @ 41 Wkh suh0fruuhodwlqj v|qfkurql}hu lv vkrzq lq
Iljxuh 5161 Wkh v|qfkurql}dwlrq vljqdo lv rewdlqhg e| fkrrvlqj wkh qxpehu ri
surgxfwv wr O @ 4> wkh fruuhodwlrq ohqjwk wr wkh uhshwlwlrq ohqjwk N @ U dqg
T @ 4= Wkh suh0fruuhodwlrq zlwk wkh nqrzq vhtxhqfh phdqv wkdw wkh qrlvh lv
5151 PXOWL0FDUULHU V\VWHP GHVLJQ 58
received samples, r
timing
( )*
freq.
R
offset
arg L
K
received samples, r
timing
( )*
freq. R
offset
arg
uhjduglqj wkh ghvljq ri wklv sdwwhuq1 Ndlvhu jlyhv d jrrg ryhuylhz ri fkdqqho
hvwlpdwlrq dqg uhodwhg wrslfv iru pxowl0fduulhu v|vwhpv lq ^6<`/ exw lw lv kdug
wr qg dqdo|wlfdo wuhdwphqwv1 Lq ^D` lw zdv vkrzq wkdw wkh slorw sdwwhuq kdv
d odujh lq xhqfh rq wkh uhvxowlqj shuirupdqfh1 E| ghvljqlqj wkh sdwwhuq lq
dq h!flhqw zd| lw lv srvvleoh wr lqfuhdvh wkh Grssohu udqjh ryhu zklfk wkh
v|vwhp fdq rshudwh e| xs wr whq wlphv ^D` frpsduhg wr d v|vwhp xvlqj d slorw
sdwwhuq vlplodu wr d vlqjoh0fduulhu fdvh1 Lq wkh sdshu/ d Ndopdq owhu zdv xvhg
wr ghvljq wkh slorw sdwwhuq e| vhdufklqj iru wkh ehvw sodfh wr sxw wkh qh{w
slorw v|pero1 Pxfk zrun fdq eh sxw lqwr wkh ghvljq ri wkh slorw sdwwhuqv/ dqg
rswlpdo fkdqqho hvwlpdwlrq owhuv fdq eh xvhg iru wkh dqdo|vhv1 Lq uhdo0wlph
rshudwlrq ohvv frpsoh{ owhuv fdq qdwxudoo| eh xvhg1 Lq ^86` d Zlhqhu owhu zdv
xvhg iru fkdqqho hvwlpdwlrq dqg wkh wkhruhwlfdo v|pero huuru udwh zdv ghulyhg
edvhg rq wkh PVH ri wkh hvwlpdwlrq owhu1 Wkh lq xhqfh ri wkh slorw vsdflqj lq
wlph dqg iuhtxhqf| zdv wkhq lqyhvwljdwhg xvlqj d uhfwdqjxodu sdwwhuq1 Xqghu
rswlpdo frqglwlrqv/ l1h1 zkhq wkh fruuhodwlrq ixqfwlrqv duh nqrzq/ wkh dxwkruv
lq ^86` frqfoxghg wkdw d slorw vsdflqj forvh wr wkh vdpsoh wkhruhp zdv vx!flhqw1
Krzhyhu/ iru zruvw0fdvh ghvljqv/ ru zkhq ohvv frpsoh{ qrq0rswlpdo fkdqqho
hvwlpdwlrq owhuv duh xvhg/ wkh glvwdqfh ehwzhhq wkh slorw v|perov kdv wr eh
ghfuhdvhg e| dw ohdvw d idfwru wzr1 Wkh pd{lpxp slorw glvwdqfh lq iuhtxhqf|
+phdvxuhg lq qxpehu ri vxe0fkdqqhov, gxh wr wkh vdpsolqj wkhruhp lv ^4<`
4
QI > +514:,
pd{ ivf
zkhuh pd{ ghqrwhv wkh pd{lpxp h{fhvv ghod| ri wkh fkdqqho1 Wkh pd{lpxp
slorw glvwdqfh lq wlph lv vlploduo| jlyhq e| +516,1 Wkh pd{lpxp slorw glvwdqfh
lq wlph lv wkhuhiruh sursruwlrqdo wr wkh frkhuhqfh edqgzlgwk/ qrupdol}hg e|
wkh vxe0fkdqqho edqgzlgwk1 Lq ^64` wkh whup edodqfhg ghvljq zdv lqwurgxfhg/
phdqlqj wkdw wkh hvwlpdwlrq xqfhuwdlqw| lq wkh wlph gluhfwlrq dqg lq wkh iuh0
txhqf| gluhfwlrq vkrxog eh wkh vdph1 Dv d uxoh ri wkxpe wkh dxwkruv sursrvhg
wzr wlphv ryhuvdpsolqj/ uhvxowlqj lq wkh iroorzlqj jxlgholqh4 =
zkhuh QI lv wkh slorw vsdflqj lq wkh iuhtxhqf| gluhfwlrq1 Wklv jxlgholqh frlq0
flghv zlwk wkh uhvxowv lq ^86`1
Wkh sxeolfdwlrqv phqwlrqhg deryh ghdo pdlqo| zlwk uhfwdqjxodu slorw sdw0
whuqv dffruglqj wr Iljxuh 5171 Dqrwkhu dowhuqdwlyh lv wr xvh d kh{djrqdo sdwwhuq
dqg wklv dssurdfk zdv lqyhvwljdwhg lq ^57`1 Wkh dxwkruv frpsduhg wkh shuiru0
pdqfh ri glhuhqw sdwwhuqv dqg frqfoxghg wkdw d kh{djrqdo julg jlyhv ehwwhu
shuirupdqfh vlqfh lw ohdgv wr d pruh h!flhqw vdpsolqj ri wzr0glphqvlrqdo vlj0
qdov1 Krzhyhu/ wr pdnh d pruh jhqhudo frqfoxvlrq/ lqyhvwljdwlrqv iru pruh
fkdqqhov wkdq wkh KI fkdqqho dgguhvvhg lq ^57` duh uhtxluhg1
4 Lq wkh ruljlqdo sxeolfdwlrq ^64` edodqfhg ghvljq zdv ghqhg dv s( Ar u A E 4@ srS u
8 E *ec exw wklv zdv fkdqjhg wr +514;, lq ^6<`1
5; FKDSWHU 51 UHVHDUFK WRSLFV
f
NF
NT
data symbol pilot symbol
Wkh lpsruwdqw frqfoxvlrq lq doo wkh sdshuv lv wr vsuhdg wkh slorw v|perov
lq erwk wlph dqg iuhtxhqf|/ rwkhuzlvh wkhuh zloo eh d shuirupdqfh ghjudgdwlrq
gxh wr ryhukhdg ru lqdffxudwh fkdqqho hvwlpdwhv1 Lq wkh KlshuODQ25 v|vwhp
^7` wkh slorw v|perov duh frqwlqxrxvo| wudqvplwwhg dw wkh vdph vxe0fkdqqhov
gxulqj wkh zkroh eorfn dqg wklv surshuw| lv wkhuhiruh qrw ixooohg1 Hdfk gdwd
eorfn lv suhfhghg e| d suhdpeoh/ exw iru d orqj eorfn lw lv kdug wr pdnh ixoo
xvh ri wkh wlph fruuhodwlrq ri wkh fkdqqho zkhq ghulylqj wkh fkdqqho hvwlpdwhv1
Lq ^;3` lw zdv vkrzq wkdw lw zdv srvvleoh wr jdlq 40418 gE mxvw e| uhduudqjlqj
wkh slorw v|perov vr wkh| duh orfdwhg dw glhuhqw vxe0fkdqqhov1 Wklv vwxg|
zdv shuiruphg iru dq xqfrghg v|vwhp zlwkrxw xvlqj wkh suhdpeoh iru fkdqqho
hvwlpdwlrq/ exw vlpxodwlrqv vkrzhg ^53` wkdw wkh vdph jdlq frxog eh dfklhyhg
lq d frghg v|vwhp lqfoxglqj wkh suhdpeoh iru fkdqqho hvwlpdwlrq1
Wkh fkdqqho prgho iru wkh wlph fruuhodwlrq xvhg lq ^D` lv dq dxwr0uhjuhvvlyh
prgho/ zklfk dovr kdv ehhq xvhg e| rwkhu uhvhdufkhuv ^69` ^;7`1 Lw lv kdug wr
frpsduh wkh uhvxowlqj slorw sdwwhuqv lq ^D` gluhfwo| wr wkh jxlgholqhv deryh
vlqfh wkh vkdsh ri wkh fruuhodwlrq ixqfwlrqv lv glhuhqw1 Lq dgglwlrq/ wkhuh
lv d glhuhqfh lq wkdw rqo| rqh slorw v|pero dw d wlph lv xvhg lq ^D`/ zkloh
pdq| slorw v|perov duh wudqvplwwhg vlpxowdqhrxvo| lq wkh rwkhu lqyhvwljdwhg
v|vwhpv1 Krzhyhu/ wkh jhqhudo uhvxow vwloo krogv wkdw wkh slorw v|perov vkdoo
eh vsuhdg erwk lq wkh wlph dqg lq wkh iuhtxhqf| gluhfwlrq vr wkdw wkh wlph
fruuhodwlrq dqg iuhtxhqf| fruuhodwlrq ri wkh fkdqqhov fdq eh xvhg wr lpsuryh
wkh fkdqqho hvwlpdwhv1 Lq ^4<` wkh lq xhqfh ri wkh hgjh hhfwv zdv phqwlrqhg
5161 VXSHULPSRVHG SLORW VHTXHQFHV 5<
data xk
pilot {1,-1} pk
Iljxuh 518= Jhqhudwlrq ri wkh vljqdo zkhq xvlqj vxshulpsrvhg slorw v|perov1
dqg lw zdv vxjjhvwhg wr kdyh pruh slorw v|perov dw wkh rxwhu v|perov lq d
sdfnhw wr plwljdwh wklv hhfw1 Wklv fruuhvsrqgv zhoo wr wkh uhvxow ri wkh slorw
sdwwhuq vhdufk lq ^D`/ zkhuh wkh rxwhu fkdqqhov duh xvhg iru slorw v|perov pruh
iuhtxhqwo|1
{n @ xn . sn > +514<,
zkhuh sn lv wkh nqrzq slorw vhtxhqfh zlwk hqhuj| Hv @ H^msn m5 `Wv dqg xn lv
wkh xqnqrzq gdwd sduw zlwk hqhuj| +4 , Hv @ H^mxn m5 `Wv 1 Wkh srzhu ri wkh
slorw vhtxhqfh lv rqo| d vpdoo iudfwlrq ri wkh wudqvplwwhg srzhu/ exw wkh slorw
vhtxhqfh fdq dq|krz eh xvhg iru fkdqqho hvwlpdwlrq ru wlph dqg iuhtxhqf|
v|qfkurql}dwlrq1 Wkh slorw vhtxhqfh fdq eh wuhdwhg hlwkhu dv h{wud qrlvh ru dv
d nqrzq glvsodfhphqw ri wkh frqvwhoodwlrq1 Lq wkh odwwhu fdvh wkh glvwxuedqfh
rq wkh gdwd vhtxhqfh fdq eh pdgh qhjoljleoh/ exw rwkhuzlvh wkh glvwruwlrq rq
wkh gdwd sduw lv dq|krz vpdoo1
Wkh lghd ri vxshulpsrvlqj d slorw vhtxhqfh kdv ehhq sursrvhg e| ydulrxv
dxwkruv iru glhuhqw dssolfdwlrqv1 Kroghq dqg Ihkhu sursrvhg lq ^67` d v|v0
whp iru fduulhu uhfryhu| edvhg rq vxshulpsrvhg slorw vhtxhqfhv1 Wkh| fdoohg
wkh whfkqltxh vsuhdg0vshfwuxp slorw whfkqltxh +VVSW, vlqfh wkh slorw vhtxhqfh
xvhg zdv d SQ0vhtxhqfh/ zklfk zdv olqhduo| dgghg wr wkh lqskdvh gdwd fkdq0
qho1 E| pxowlso|lqj wkh lqfrplqj vljqdo zlwk wkh nqrzq SQ0vhtxhqfh lw lv
srvvleoh wr uhfryhu wkh fduulhu vljqdo lq dq hdv| zd|1 Diwhu gh0vsuhdglqj/ wkh
fduulhu vljqdo lv qduurzedqg dqg wkhuhiruh pxfk ri wkh glvwruwlrq iurp wkh gdwd
63 FKDSWHU 51 UHVHDUFK WRSLFV
vljqdo fdq eh owhuhg rxw1 Frpsduh wklv wr wkh surfhvvlqj jdlq dfklhyhg lq
FGPD v|vwhpv1 Wkh phwkrg vkrzhg wr eh urexvw iru orz VQUv/ wkh v|vwhp
zdv lpsohphqwhg lq kdugzduh dqg wkh fduulhu uhfryhu| xqlw zrunhg zhoo grzq
wr He @Q3 @ 3 gE1 Ehmmdql dqg Ehoruh sursrvhg d v|vwhp xvlqj d yduldqw ri vx0
shulpsrvhg slorw vhtxhqfhv ^43` iru xqghuzdwhu frppxqlfdwlrq1 Wkh jrdo wkhuh
zdv wr hqdeoh wudfnlqj ri d udslgo| fkdqjlqj fkdqqho e| lqwurgxflqj d frqwlq0
xrxv slorw vhtxhqfh1 Lq wkh v|vwhp/ wzr glhuhqw ruwkrjrqdo sxovh vkdshv zhuh
xvhg iru wkh gdwd dqg slorw vhtxhqfh1 Wkhuhiruh/ wkhuh zdv d orvv lq vshfwudo
h!flhqf|/ exw wkh v|vwhp vkrzhg wr eh urexvw iru kljk Grssohu iuhtxhqflhv1
Wklv whfkqltxh lv vlplodu wr wkh XPWV2Z0FGPD v|vwhp lq xsolqn/ zkhuh wkh
gdwd dqg slorw vhtxhqfhv duh wudqvplwwhg vlpxowdqhrxvo| ^68`/ exw xvlqj wkh
ruwkrjrqdo lqskdvh dqg txdgudwxuh fkdqqhov/ uhvshfwlyho|1 Erwk wkhvh phwk0
rgv phdq wkdw wkh vshfwudo h!flhqf| lv ghfuhdvhg vlqfh qr gdwd lv wudqvplwwhg
rq wkh slorw fkdqqho1 Vxshulpsrvhg slorw vhtxhqfhv kdyh dovr ehhq xvhg iru
iudph v|qfkurql}dwlrq ^9:`1 Wkh dxwkruv xvhg d qrq0ruwkrjrqdo slorw vhtxhqfh
wr ghwhuplqh wkh vwduw ri iudphv lq d zluhohvv v|vwhp1 Wkh slorw vhtxhqfh zdv
qrw xvhg iru elw v|qfkurql}dwlrq dqg wkhuhiruh wkh lq xhqfh ri wkh gdwd frxog
eh uhpryhg ehiruh fruuhodwlqj wkh uhfhlyhg slorw vhtxhqfh zlwk wkh nqrzq slorw
vhtxhqfh1 Wkhq/ zkhq iudph v|qfkurql}dwlrq zdv dfklhyhg/ wkh lq xhqfh rq
wkh gdwd v|perov zdv uhpryhg e| vxewudfwlqj wkh slorw vhtxhqfh1 Wklv whfk0
qltxh dovr vkrzhg jrrg v|qfkurql}dwlrq shuirupdqfh frpsduhg wr frqyhqwlrqdo
v|vwhpv/ hvshfldoo| lq idvw Ud|ohljk idglqj fkdqqhov zkhuh wkh orqj frqwlqxrxv
slorw vhtxhqfh phdqv wkdw wlph glyhuvlw| lv dfklhyhg1
Lq sdshu ^H` vxshulpsrvhg slorw vhtxhqfhv duh xvhg iru wlph dqg iuhtxhqf|
v|qfkurql}dwlrq lq RIGP v|vwhpv dqg lq sdshu ^I` wkh whfkqltxh lv xvhg iru
mrlqw fkdqqho hvwlpdwlrq dqg ghwhfwlrq lq dw Ud|ohljk idglqj fkdqqhov1 Wkh
odwwhu sdshu lv qrw uhvwulfwhg wr pxowl0fduulhu v|vwhpv/ exw fdq eh xvhg iru dq|
v|vwhp zkhuh wkh wlph glvshuvlrq lv qhjoljleoh1 Wkh v|qfkurql}hu vwuxfwxuh lq
^H` kdv orz frpsoh{lw| dqg lv vxlwdeoh iru lpsohphqwdwlrq exw wkh frpsoh{lw|
ri wkh fkdqqho hvwlpdwru lq ^I` lv txlwh kljk dqg wkhuhiruh orz0frpsoh{lw| ds0
sur{lpdwlrqv duh ri lqwhuhvw1 Wkh vfrsh ri ^I` kdv ehhq wr vkrz wkh srwhqwldo
ri xvlqj vxshulpsrvhg slorw vhtxhqfhv iru fkdqqho hvwlpdwlrq dqg wr ghulyh dq
rswlpdo uhfhlyhu vwuxfwxuh1 Wkh sdshu krzhyhu dovr dgguhvvhv frpsoh{lw| lvvxhv
dqg srvvleoh vlpsolhg uhfhlyhu vwuxfwxuhv1
{1 Wkh whfkqltxh lv vxlwdeoh iru erwk wudfnlqj dqg dftxlvlwlrq dqg lw lv yhu|
h{leoh lq wkdw wkh srzhu ri wkh slorw vhtxhqfh fdq eh dgmxvwhg wr wkh vshflf
qhhgv1 Lq dftxlvlwlrq prgh doo wkh wudqvplwwhg srzhu fdq eh ghyrwhg wr wkh
slorw vhtxhqfh dqg wkhq wkh wudqvplwwhg vljqdo lv htxlydohqw wr wkh suhdpeoh
ghvfulehg lq ^G`1 E| xvlqj doo wkh srzhu iru wkh slorw vhtxhqfh gxulqj dftxlvl0
wlrq/ wkh dftxlvlwlrq wlph fdq eh pdgh vpdoo1 Gxulqj wudfnlqj/ wkh slorw srzhu
fdq eh orz zlwkrxw orrvlqj shuirupdqfh vlqfh wkh sdudphwhuv gr qrw fkdqjh
wkdw idvw dqg dyhudjlqj ryhu vhyhudo v|perov fdq eh xvhg1
Wkh v|qfkurql}dwlrq vljqdo lv fdofxodwhg lq wkh vdph zd| dv iru wkh suh0
fruuhodwlqj v|qfkurql}hu dffruglqj wr +5149, dqg Iljxuh 5161 Wkh uhfhlyhg vljqdo
lv uvw fruuhodwhg zlwk wkh nqrzq vhtxhqfh ri ohqjwk U= Wkhq wkh fruuhodwhg
ydoxhv duh pxowlsolhg lq sdluv dqg qdoo| O ri wkhvh surgxfwv duh vxpphg lq
rughu wr jhw wkh v|qfkurql}dwlrq vljqdo1 Wkh rqo| glhuhqfh frpsduhg wr wkh
suhdpeoh v|qfkurql}dwlrq vfkhph lv wkdw wkh gdwd vhtxhqfh lv wudqvplwwhg dw
wkh vdph wlph dv wkh v|qfkurql}dwlrq vhtxhqfh dqg wkhuhiruh dfwv olnh dq dggl0
wlrqdo glvwxuedqfh1 Wkhuhiruh/ wkh fruuhodwru ohqjwk ru wkh qxpehu ri surgxfwv
xvhg iru wkh v|qfkurql}dwlrq vljqdo kdv wr eh lqfuhdvhg frpsduhg wr wkh fdvh
zlwk wkh v|qfkurql}dwlrq vljqdo rqo|/ l1h1 wkh suhdpeoh1 Wkh dpsolwxgh ri wkh
v|qfkurql}dwlrq vljqdo lv xvhg iru wlph v|qfkurql}dwlrq dqg zkhq wkh fruuhfw
wlplqj lv irxqg wkh iuhtxhqf|0rvhw hvwlpdwh lv jlyhq e| wkh skdvh dqjoh ri wkh
v|qfkurql}dwlrq vljqdo1
Lq wkh olwhudwxuh qr vlplodu zrun fdq eh irxqg iru RIGP v|vwhpv/ dw ohvw dv
idu dv L nqrz1 Wkh phwkrg kdv vrph vlplodulwlhv wr v|qfkurql}dwlrq lq FGPD
v|vwhpv1 Riwhq lq FGPD v|vwhpv/ iru h{dpsoh lq wkh XPWV2Z0FGPD v|v0
whp/ wkhuh lv d vshfldo v|qfkurql}dwlrq fkdqqho wudqvplwwhg rq wrs ri wkh gdwd
fkdqqhov1 Vlqfh wkh v|qfkurql}dwlrq fkdqqho fdq eh vkduhg ehwzhhq wkh xvhuv/
wkh ryhukhdg lv vpdoo dq|zd|1 Krzhyhu/ dv rssrvhg wr wkh vxshulpsrvhg slorw
vhtxhqfhv/ wkh v|qfkurql}dwlrq fkdqqho lq wkh FGPD v|vwhp lv ruwkrjrqdo wr
wkh gdwd fkdqqhov e| phdqv ri glhuhqw frghv1 Wklv phdqv vrphzkdw ehwwhu
ghwhfwlrq surshuwlhv/ vlqfh wkh v|qfkurql}dwlrq vfkhph edvhg rq vxshulpsrvhg
slorw vhtxhqfhv uholhv rq vwdwlvwlfdo dyhudjlqj iru ghwhfwlrq ri wkh v|qfkurql}d0
wlrq vljqdo1
Lq wkh vsuhdg0vshfwuxp slorw v|vwhp lq ^67` vxshusrvlwlrq ri wkh slorw vh0
txhqfh dqg gdwd vhtxhqfh zdv pdgh lq wkh dqdorj grpdlq1 Wklv phdqv wkdw
iru wklv v|vwhp elw v|qfkurqrxv wudqvplvvlrq frxog qrw eh jxdudqwhhg1 Wklv lv
lq frqwudvw wr wkh vxshulpsrvhg slorw vhtxhqfhv lq ^H`/ zkhuh wklv surshuw| lv
hvvhqwldo iru wlph v|qfkurql}dwlrq dqg odwhu iru fkdqqho hvwlpdwlrq lq ^I`1
qrw ixooohg/ dqg iru frqyhqwlrqdo v|vwhpv wklv ehfrphv dq lpsruwdqw olplwdwlrq
dw kljk Grssohu iuhtxhqflhv1 Glhuhqw dssurdfkhv fdq eh xvhg iru ghwhfwlrq
edvhg rq vxshulpsrvhg slorw vhtxhqfhv1 Wkh slorw vhtxhqfh fdq eh vhhq dv d
slorw wrqh wkdw lv vsuhdg ryhu wkh zkroh wudqvplvvlrq edqgzlgwk ru dv d vpdoo
vkliw ri wkh vljqdo frqvwhoodwlrq1 Lq ^I`/ d v|vwhp edvhg rq mrlqw pd{lpxp
olnholkrrg ghwhfwlrq ri wkh gdwd dqg wkh fkdqqho lv ghvfulehg1 Wklv phdqv wkdw
wkh slorw vhtxhqfh dqg wkh gdwd vhtxhqfh erwk duh xvhg iru fkdqqho hvwlpdwlrq
dqg ghwhfwlrq1 Rwkhu srvvleoh dssurdfkhv zrxog eh wr uvw hvwlpdwh wkh fkdqqho
e| khos ri wkh slorw vhtxhqfh dqg wkhq ghwhfw wkh gdwd edvhg rq wklv hvwlpdwh1
Lq ^67` wkh v|qfkurql}dwlrq vljqdo zdv h{wudfwhg dqg diwhu wkdw wkh glvwxuedqfh
ri wkh gdwd vhtxhqfh zdv ghfuhdvhg e| fkdqjlqj wkh ghwhfwlrq wkuhvkrog zlwk
uhjdugv wr wkh slorw vhtxhqfh1 Lq ^9:` wkh gdwd vljqdo zdv uvw ghwhfwhg dqg lwv
lq xhqfh rq wkh v|qfkurql}dwlrq vljqdo zdv uhpryhg ehiruh h{wudfwlrq ri wkh
v|qfkurql}dwlrq vljqdo1 Krzhyhu/ wkh odwwhu lv qrw srvvleoh iru gdwd ghwhfwlrq
dqg fkdqqho hvwlpdwlrq vlqfh wkh lq xhqfh ri wkh gdwd fdq qrw eh uhpryhg
ehiruh h{wudfwlrq ri wkh v|qfkurql}dwlrq sdudphwhuv dqg ylfh yhuvd1
Mrlqw fkdqqho hvwlpdwlrq dqg ghwhfwlrq lv zhoo ghvfulehg lq ^7;`1 Wkh dlp lv
wr qg wkh gdwd vhtxhqfh dqg fkdqqho hvwlpdwh wkdw pd{lpl}hv wkh suredelolw|
ghqvlw| ixqfwlrq ri wkh uhfhlyhg vhtxhqfh dffruglqj wr +514,1 Dvvxph wkdw wkh
uhfhlyhg vljqdo vdpsoh dw wlph n fdq eh ghvfulehg e|
un @ kn fn . qn > +5153,
zkhuh wkh qrlvh qn dqg pxowlsolfdwlyh fkdqqho fn duh mrlqwo| Jdxvvldq yduldeohv
dqg wkdw wkh gdwd v|perov duh lqghshqghqw udqgrp yduldeohv zlwk htxdo sure0
delolwlhv1 Lw fdq eh vkrzq wkdw ^7;` wkh rswlpdo mrlqw hvwlpdwlrq0ghwhfwlrq uxoh
fdq eh irupxodwhg dv
fPO > k,
+a a @ duj pd{ i +u> kmf, @ duj pd{ i +kmu> f,i+umf,= +5154,
f>k f>k
D vxlwdeoh vwudwhj| lv wkhuhiruh wr pd{lpl}h i +kmu> f, wr qg wkh PDS +pd{l0
pxp d srvwhulrul, fkdqqho hvwlpdwh iru hdfk srvvleoh gdwd vhtxhqfh/ f/ dqg xvh
wklv dv d wuxh sdudphwhu zkhq vhohfwlqj wkh vhtxhqfh zlwk wkh odujhvw olnholkrrg1
Vlqfh wkh fkdqqho dqg qrlvh yduldeohv duh dvvxphg wr eh Jdxvvldq/ wkh PDS
hvwlpdwh lv htxdo wr wkh plqlpxp phdq vtxduh huuru hvwlpdwh/ zklfk lq wxuq
fdq eh fdofxodwhg xvlqj d Zlhqhu owhu ^73`1 Wklv lv wkh dssurdfk lq ^I`1 Iru
hdfk k|srwkhvlv ri wkh gdwd vhtxhqfh wkh fkdqqho hvwlpdwh lv fdofxodwhg xvlqj d
Zlhqhu owhu1 Wklv fkdqqho hvwlpdwh lv wkhq xvhg wr fdofxodwh wkh phwulf vxfk
wkdw wkh prvw suredeoh vhtxhqfh fdq eh irxqg1 Lq ^I` erwk dq rswlpdo dqg dq
dssur{lpdwh phwulf lv ghulyhg1 Wkh xvh ri wkh rswlpdo phwulf uhtxluhv wkdw doo
srvvleoh gdwd vhtxhqfhv duh wdnhq lqwr dffrxqw zkhq orrnlqj iru wkh vhtxhqfh
uhvxowlqj lq wkh orzhvw phwulf/ zkhuhdv wkh dssur{lpdwh phwulf lv fdq eh xvhg
iru d ohvv frpsoh{ ghwhfwru e| xvlqj shu0vxuylyru surfhvvlqj ^8:`1
Lq ^J` wkhuh lv d frpsdulvrq ehwzhhq fkdqqho hvwlpdwlrq dqg ghwhfwlrq xvlqj
vxshulpsrvhg slorw vhtxhqfhv dqg rwkhu ghwhfwlrq dqg hvwlpdwlrq phwkrgv1 Wkh
5161 VXSHULPSRVHG SLORW VHTXHQFHV 66
Vxppdu| dqg
Frqwulexwlrqv
Lq wklv fkdswhu/ vrph jhqhudo uhvxowv duh suhvhqwhg dqg wkh frqwhqwv dqg pdlq
uhvxowv lq hdfk sdshu duh ghvfulehg1 Wkhuh lv dovr d suhvhqwdwlrq ri wkh frq0
wulexwlrqv lq hdfk sdshu wr wkh vshflf uhvhdufk hog1 Wkhuhdiwhu/ wkhuh lv d
glvfxvvlrq ri srvvleoh gluhfwlrqv iru ixwxuh zluhohvv frppxqlfdwlrq v|vwhpv1
Ilqdoo|/ vrph jhqhudo frqfoxvlrqv duh suhvhqwhg dqg srvvleoh ixuwkhu zrun lq
wkh uhvhdufk hogv lv glvfxvvhg1
68
69 FKDSWHU 61 VXPPDU\ DQG FRQWULEXWLRQV
Wkh irfxv ri wkh wkhvlv lv sureohpv uhodwhg wr wkh zluhohvv preloh fkdqqho1
Wklv fkdqqho kdv vrph lpsruwdqw surshuwlhv wkdw fdoo iru vshfldo dwwhqwlrq> wkh
fkdqqho fkdqjhv udslgo| dqg lw kdv odujh dpsolwxgh xfwxdwlrqv1 Wkhuhiruh/
wkh v|vwhp kdv wr eh rswlpl}hg dffruglqj wr wkhvh vshflf surshuwlhv1 Wkh
wdvn ri wkh v|qfkurql}hu dqg fkdqqho hvwlpdwru lv wr hqdeoh frppxqlfdwlrq
dovr xqghu vhyhuh frqglwlrqv1 Lq wkh wkhvlv/ zh kdyh wkhuhiruh vwxglhg urexvw
v|qfkurql}dwlrq dqg fkdqqho hvwlpdwlrq whfkqltxhv/ vxlwdeoh dovr iru orz VQUv
dqg idvw idglqj fkdqqhov1 Wkh zrun kdv ehhq irfxvhg rq ixwxuh v|vwhpv dqg
v|vwhpv xqghu vwdqgdugl}dwlrq1 Krzhyhu/ wkh sulqflsohv duh wkh vdph iru h{lvw0
lqj v|vwhpv dqg wkh dqdo|vhv fdq eh dssolhg wr wkhvh dv zhoo1 Lq wkh wkhvlv wkh
lq xhqfh ri wkh slorw sdwwhuq lq RIGP v|vwhpv lv vwxglhg dqg jxlgholqhv iru
wkh ghvljq duh jlyhq1 V|qfkurql}dwlrq lq RIGP v|vwhpv edvhg rq suhdpeohv lv
dovr dqdo|}hg dqg d qhz v|qfkurql}hu vwuxfwxuh lv sursrvhg1 Ixuwkhu/ wkh vxe0
fkdqqho edqgzlgwk lq zluhohvv RIGP v|vwhpv lv rswlpl}hg zlwk uhjdugv wr wkh
fkdudfwhulvwlfv ri wkh fkdqqho1 Wkh xvh ri suh0frpshqvdwlrq lv dovr lqyhvwljdwhg
e| ghulylqj h{suhvvlrqv iru wkh EHU dqg ghwhuplqlqj ihdvleoh eorfn ohqjwkv1
Ilqdoo|/ wkh xvh ri vxshulpsrvhg slorw vhtxhqfhv lv dqdo|}hg dqg sursrvhg iru
v|qfkurql}dwlrq dqg fkdqqho hvwlpdwlrq lvvxhv1
Wkh uhtxluhphqwv rq wkh v|qfkurql}hu dqg fkdqqho hvwlpdwru duh qrw vwdwlf
ryhu d orqjhu wlph1 Wkh| fkdqjh zkhq qhz whfkqrorjlhv dqg whfkqltxhv duh
lqwurgxfhg1 Lpsuryhg vhplfrqgxfwru whfkqrorj| kdv phdqw wkdw pruh frpsoh{
dojrulwkpv wkdq ehiruh fdq eh xvhg1 Wkh qhz dojrulwkpv/ xvhg dv dq h{dpsoh
iru ghwhfwlrq ru huuru fruuhfwlrq/ duh riwhq pruh dffxudwh wkdq wkh suhylrxvo|
xvhg rqhv/ exw dovr pruh vhqvlwlyh wr wkh dffxudf| ri wkh v|qfkurql}dwlrq sd0
udphwhuv1 Huuruv wkdw suhylrxvo| zhuh klgghq gxh wr d olplwhg uhvroxwlrq dqg
glg qrw dhfw wkh uhvxow fdq sod| dq lpsruwdqw uroh lq qhz v|vwhpv zlwk ehwwhu
uhvroxwlrq1 Wkh jrdo lv wkdw wkh vxssruwlqj ixqfwlrqv vkrxog qrw eh d olplwd0
wlrq ri wkh v|vwhp shuirupdqfh1 Frpsoh{lw| lv krzhyhu vwloo dq lpsruwdqw lvvxh
iru wkh dojrulwkpv/ vlqfh wkh frpsoh{lw| vkrxog eh vshqw rq ixqfwlrqv surylglqj
wkh odujhvw jdlq1
slorw sdwwhuqv irxqg lq wkh olwhudwxuh1 Lw lv vkrzq wkdw wkh slorw sdwwhuq kdv
d odujh lq xhqfh rq wkh uhvxowlqj elw huuru udwh1 D zhoo0ghvljqhg slorw sdwwhuq
phdqv wkdw wkh fkdqqho fkdqjhv fdq eh wudfnhg ehwwhu/ zlwk d fruuhvsrqglqj
shuirupdqfh lpsuryhphqw1 Lq d qxphulfdo h{dpsoh lw lv vkrzq wkdw wkh EHU
fdq eh ghfuhdvhg e| d idfwru ri 8 ru wkdw wkh v|vwhp fdq frsh zlwk 43 wlphv
kljkhu Grssohu iuhtxhqf| zkhq xvlqj d vxlwdeoh slorw sdwwhuq lqvwhdg ri rqh
vlplodu wr d vlqjoh0fduulhu fdvh1
Wkh pdlq frqwulexwlrq ri wkh sdshu lv wr vkrz wkh lpsruwdqfh ri d zhoo0
ghvljqhg slorw sdwwhuq dqg wkh odujh lpsuryhphqwv wkdw fdq eh dfklhyhg1 Wkh
sdshu lv rqh ri wkh uvw sdshuv xvlqj dq dqdo|wlfdo dssurdfk wr rswlpl}h wkh
slorw sdwwhuq1 Wkh dqdo|vlv lv jhqhudo dqg fdq eh xvhg wr ghulyh wkh dprxqw ri
slorw v|perov uhtxluhg dqg wr qg d vxlwdeoh slorw sdwwhuq iru dq| RIGP0edvhg
v|vwhp1
41 skdvh0rqo| frpshqvdwlrq
Wkh EHUv dqg lqfuhdvh lq shdn0wr0dyhudjh srzhu udwlrv duh fdofxodwhg iru wkh
glhuhqw phwkrgv dvvxplqj shuihfw fkdqqho nqrzohgjh1 Wkh pd{lpxp eorfn
ohqjwkv duh fdofxodwhg iru wkh skdvh frpshqvdwhg v|vwhp/ vlqfh wklv lv dvvxphg
prvw dssursuldwh iru zluhohvv dssolfdwlrqv1 Lw lv frqfoxghg wkdw eorfn ohqjwkv
pxfk orqjhu wkdq wkh rqh lq wkh GHFW v|vwhp fdq eh xvhg ehiruh wkh lqvwdq0
wdqhrxv shuirupdqfh jhwv zruvh wkdq wkdw ri d glhuhqwldo v|vwhp1
Wkh pdlq frqwulexwlrqv duh wkh ghulydwlrq ri wkh uhvxowlqj xqfrghg EHU
iru wkh skdvh0rqo| frpshqvdwlrq phwkrg dqg wkh dqdo|vlv ri wkh lq xhqfh ri
erwk wkh xsolqn eorfn ohqjwk dqg wkh grzqolqn eorfn ohqjwk/ uhvshfwlyho|1 Wkh
fdofxodwlrqv duh shuiruphg erwk iru shuihfw nqrzohgjh ri wkh fkdqqho fruuhodwlrq
ixqfwlrqv dqg xqghu fruuhodwlrq plvpdwfk1 Wkh pd{lpxp eorfn ohqjwk lv d yhu|
lpsruwdqw ghvljq sdudphwhu lq wkh ghvljq ri suh0frpshqvdwhg wlph glylvlrq
gxsoh{ v|vwhpv dqg wr p| nqrzohgjh qr dqdo|wlfdo dqdo|vlv ri lwv lq xhqfh rq
wkh shuirupdqfh h{lvwv lq wkh olwhudwxuh1 Lq rwkhu sdshuv wkh fkdqqho lv dvvxphg
wr eh frqvwdqw ru vlpxodwlrqv duh xvhg wr lqyhvwljdwh wkh lq xhqfh ri eorfn
ohqjwk dqg Grssohu iuhtxhqf|1 Rwkhu frqwulexwlrqv lq wklv sdshu lqfoxgh wkh
ghulydwlrq ri wkh xqfrghg EHU iru wkh wkuhh phwkrgv zkhq dvvxplqj shuihfw
fkdqqho nqrzohgjh dqg wkh fruuhvsrqglqj lqfuhdvh lq shdn0wr0dyhudjh srzhu
udwlr1
skdvh lv xvhg iru wkh iuhtxhqf|0rvhw hvwlpdwh1 Lq wkh sdshu/ wkh glvwulexwlrq
ri wkh v|qfkurql}dwlrq vljqdo lv ghulyhg iru DZJQ fkdqqhov/ dqg yhulhg e|
vlpxodwlrqv1 Wkh suhvhqwhg v|qfkurql}hu vwuxfwxuh lv ri orz frpsoh{lw| dqg lw
lv vkrzq wkdw wkh suhvhqwhg phwkrg zrunv zhoo dovr dw orz VQUv1 Wkh srzhu
ri wkh slorw vhtxhqfh fdq eh dgmxvwhg wr wkh vshflf uhtxluhphqwv/ exw lw fdq
lq jhqhudo eh orz gxh wr vxssuhvvlrq ri qrlvh dqg gdwd zkhq shuiruplqj wkh
fruuhodwlrq1
Wkh pdlq frqwulexwlrqv lqfoxgh wkh lqwurgxfwlrq ri vxshulpsrvhg slorw vh0
txhqfhv iru v|qfkurql}dwlrq sxusrvhv lq RIGP v|vwhpv1 Wkh sursrvhg phwkrg
lv lqghshqghqw ri wkh f|folf suh{ dqg lv yhu| h{leoh1 Wr p| nqrzohgjh/ wkh
ghulydwlrq ri wkh glvwulexwlrq ixqfwlrqv ri wkh wlph v|qfkurql}dwlrq vljqdo lv
dovr qhz/ hyhq wkrxjk vlplodu v|qfkurql}dwlrq vwuxfwxuhv fdq eh xvhg lq FGPD
v|vwhpv1 Dqrwkhu frqwulexwlrq lq wkh sdshu lv wkh jhqhudol}dwlrq ri rwkhu v|q0
fkurql}dwlrq phwkrgv vxfk wkdw wkh dqdo|vlv fdq eh dssolhg wr wkhvh dv zhoo1
udwhv dqg sdfnhw edvhg frppxqlfdwlrq1 Wklv phdqv wkdw gdwd udwhv ri xs wr
6;7 nelw2v fdq eh rhuhg ryhu zlgh duhdv1 Lq d ihz |hduv wkh qhz Z0ODQ
v|vwhpv zloo dovr eh rq wkh pdunhw1 Wkhvh v|vwhpv zloo zrun lq xqolfhqvhg
iuhtxhqf| edqgv dqg rhu gdwd udwhv ri durxqg 53 Pelw2v/ exw zlwk d olplwhg
fryhudjh1 Ilqdoo|/ wkh uvw frpphufldo Eoxhwrrwk ghylfhv zloo eh odxqfkhg vrrq1
Wklv phdqv wkdw vkruw glvwdqfh zluhohvv frppxqlfdwlrq zlwk gdwd udwhv ri xs
wr durxqg :33 nelw2v lv vrrq dydlodeoh iru d zlgh udqjh ri hohfwurqlf ghylfhv1
L wklqn wkdw wkh lpsruwdqfh ri wkh odwwhu wzr v|vwhpv fdq qrw eh ryhuhv0
wlpdwhg1 Wkhvh v|vwhpv zloo rhu idvw zluhohvv frppxqlfdwlrq iru iuhh dqg
fdq eh xvhg dv fkhds olqnv ehwzhhq dq| hohfwurqlf ghylfh dqg wkh dffhvv srlqw1
Wklv phdqv wkdw lw lv srvvleoh iru orz0frvw ghylfhv wr kdyh dffhvv wr idvw dqg
dgydqfhg frppxqlfdwlrq olqnv zkhuh wkh preloh whuplqdo dfwv pruh olnh d uhod|
vwdwlrq1 Suredeo| wkhuh zloo eh d zlgh udqjh ri orz0frvw wklug sduw| surgxfwv
rhulqj qhz vhuylfhv xvlqj lqgluhfw dffhvv wr wkh h{lvwlqj qhwzrun1
L dovr wklqn wkdw wkh Z0ODQ v|vwhpv zloo kdyh d yhu| lpsruwdqw uroh lq
wkh qhdu ixwxuh1 H{fhsw iru fryhudjh dqg pd|eh prelolw|/ wkhvh v|vwhpv duh
dfwxdoo| ixooolqj doo wkh uhtxluhphqwv vwdwhg e| vrph ri wkh ohdglqj dfwruv lq
wkh duhd iru d irxuwk jhqhudwlrq v|vwhp1 Wklv phdqv wkdw |rx fdq dfklhyh
gdwd udwhv kljk hqrxjk erwk iru ylghr dqg kljk0txdolw| dxglr dssolfdwlrqv1 Lw
lv dovr srvvleoh wr eh |rxu rzq rshudwru dqg rhu eurdgedqg vhuylfhv lq dq
r!fh ru d uhvlghqwldo duhd iuhh ri fkdujh1 Ixuwkhu/ |rx fdq vhw xs |rxu rzq
shuvrqdo duhd qhwzrun zkhuh hohfwurqlf ghylfhv lq |rxu krph fdq frppxqlfdwh
zlwk hdfk rwkhu1 Krshixoo| wkhvh v|vwhpv zloo eh xvhu0iulhqgo| hqrxjk qrw wr
suhyhqw wkhlu xvdjh1 Wkh xvh ri wkh xqolfhqvhg edqgv lv yhu| lpsruwdqw vlqfh
wklv phdqv wkdw lq |rxu rzq v|vwhp wkhuh zloo eh qr uhoxfwdqfh xvlqj lw gxh
wr kljk wudqvplvvlrq ihhv1
Lq wklv vfhqdulr wkhuh lv qr fhoo sodqqlqj dqg wkh lqwhuihuhqfh ohyho zloo vrph0
wlphv eh kljk1 Wkh kljk wudqvplvvlrq udwhv phdqv wkdw/ lq rughu wr olplw wkh
wudqvplwwhg srzhu wr dssursuldwh ohyhov/ wkh hqhuj| shu elw zloo eh ghfuhdvhg
frpsduhg wr wrgd|*v v|vwhpv1 Wkhuhiruh/ wkh v|vwhpv qhhg wr eh deoh wr dgdsw
wkh wudqvplvvlrq wr wkh suhvhqw vlwxdwlrq dqg rshudwh lq kljko| glvwxuehg hq0
ylurqphqwv1 Idvw dqg uholdeoh v|qfkurql}dwlrq dqg fkdqqho hvwlpdwlrq duh lp0
sruwdqw lq rughu wr suhyhqw wkh v|vwhpv iurp ehlqj olplwhg e| wkhvh ixqfwlrqv/
udwkhu wkdq e| qrlvh ru lqwhuihuhqfh1
Pd|eh wkh prvw lpsruwdqw wklqj wkdw L kdyh ohduqhg gxulqj wkhvh |hduv lv
wkdw zluhohvv frppxqlfdwlrq v|vwhpv duh frpsoh{ v|vwhpv dqg wkh wuxh fkdo0
ohqjh lv wr rswlpl}h wkh v|vwhp dqg jhw hyhu|wklqj wr zrun wrjhwkhu1 Lw lv qr
xvh wr rqo| lpsuryh fhuwdlq sduwv dv orqj dv wkh uhvw ri wkh v|vwhp fdq qrw
pdnh xvh ri wkhvh lpsuryhphqwv1 Wr ph lw vhhpv wkdw lq wkh kljko| rswlpl}hg
v|vwhpv wkh glhuhqw ixqfwlrqdo eorfnv ehfrph pruh ghshqghqw rq hdfk rwkhu/
wkh lqirupdwlrq h{fkdqjh ehwzhhq wkh eorfnv kdv ehfrph pruh lpsruwdqw1 Wkh
fkdqqho hvwlpdwru/ iru h{dpsoh/ kdv qrw rqo| wr gholyhu fkdqqho hvwlpdwhv wr
wkh htxdol}hu/ exw dovr wr wkh ghfrghu vxfk wkdw wkh ghfrglqj fdq eh rswlpl}hg>
dqg pd|eh wr wkh wudqvplwwhu vxfk wkdw wkh wudqvplwwhg vljqdo fdq eh dgmxvwhg
6161 FRQFOXVLRQV DQG GLVFXVVLRQ 76
wr wkh vshflf fkdqqho1 Lq rughu wr fkdoohqjh wrgd|*v olplwv wkh uhfhlyhu kdv wr
lqfrusrudwh pruh dqg pruh ri wkh dydlodeoh lqirupdwlrq/ riwhq wr wkh sulfh ri
lqfuhdvhg frpsoh{lw|1 Wkhuh lv dovr d txhvwlrq zkhwkhu wklv lqfuhdvh lq frp0
soh{lw| fdq eh drughg dqg krz odujh wkh uhvxowlqj jdlq lv1 Dv dozd|v/ lw lv d
txhvwlrq ri frvw yv1 shuirupdqfh1
Lq rughu wr ixooo wkh uhtxluhphqwv rq v|qfkurql}dwlrq dqg fkdqqho hvwlpd0
wlrq L wklqn wkdw lw lv zhoo zruwk vshqglqj d ihz shufhqw ri wkh wudqvplwwhg
srzhu dqg edqgzlgwk rq nqrzq slorw v|perov1 Wkh slorw v|perov phdq wkdw
idvwhu dqg ohvv frpsoh{ dojrulwkpv fdq eh xvhg/ riwhq uhvxowlqj lq pruh urexvw
ghwhfwlrq dqg ehwwhu shuirupdqfh1 Wkh orvv fdxvhg e| wkh slorw v|perov lv riwhq
vpdoo frpsduhg wr wkh jdlq wkdw fdq eh dfklhyhg/ vr slorw v|perov duh jhqhudoo|
d jrrg lqyhvwphqw
Lq RIGP v|vwhpv lw lv uhdoo| lpsruwdqw wr olplw wkh lq xhqfh ri wkh iuh0
txhqf| rvhw1 Wklv hhfw riwhq uhvxowv lq xqghvludeoh frqvwudlqwv lq wkh v|vwhp
ghvljq/ hvshfldoo| zkhq xvlqj orz frvw whuplqdov zkhuh/ lq jhqhudo/ wkh dffxudf|
ri wkh rvfloodwruv lv orz1 Wkhuhiruh/ L wklqn wkdw lw lv lpsruwdqw wr orrn iru idvw
dqg urexvw hvwlpdwlrq whfkqltxhv dv zhoo dv fkhds dqg dffxudwh rvfloodwruv1 Lq
rughu wr jhw h!flhqw v|vwhpv lw lv lpsruwdqw wkdw wkh shuirupdqfh lv qrw olp0
lwhg e| kdugzduh lpsdluphqwv/ exw udwkhu e| wkh fkdudfwhulvwlfv ri wkh sk|vlfdo
fkdqqho ru e| lqwhuihuhqfh iurp rwkhu xvhuv1
Uhjduglqj ixwxuh zrun rq wkh pdwhuldo suhvhqwhg lq wklv wkhvlv/ L wklqn wkdw
wkh sdshu rq suh0frpshqvdwlrq udwkhu hdvlo| fdq eh h{whqghg wr dffrxqw iru dq0
whqqd glyhuvlw| dv zhoo1 Wklv pdnhv wkh sdshu pruh frpsohwh e| dovr fryhulqj
v|vwhpv vlplodu wr wkh dqwhqqd suh0frpshqvdwlrq v|vwhp lq wkh XPWV2Z0
FGPD v|vwhp1 Wkh suhdpeoh0v|qfkurql}dwlrq sdshu fdq eh h{whqghg wr lq0
foxgh suhdpeoh0edvhg fkdqqho hvwlpdwlrq1 Lq lwv suhvhqw irup/ wkh irfxv lv rq
dftxlvlwlrq ri wlph dqg iuhtxhqf| v|qfkurql}dwlrq sdudphwhuv rqo|1 Lq sdfnhw
edvhg v|vwhpv wkhuh duh/ krzhyhu/ vrph lqlwldo wudqvlhqwv zkhq d sdfnhw duulyhv
dqg lw lv qrw phdqlqjixo wr pdnh d qdo fkdqqho hvwlpdwh ehiruh uhdfklqj vrph
nlqg ri vwhdg| vwdwh1 Dq|zd|/ wkh uhtxluhg lqirupdwlrq lv wkhuh/ diwhu fruuh0
odwlrq wkh fkdqqho lpsxovh uhvsrqvh lv dydlodeoh dv d frduvh fkdqqho hvwlpdwh1
Wkh xvh ri vxshulpsrvhg slorw vhtxhqfhv lv dovr zruwk ixuwkhu lqyhvwljdwlrqv1
Iluvw ri doo/ lw zrxog eh lqwhuhvwlqj wr orrn dw ohvv frpsoh{ uhfhlyhu vwuxfwxuhv1
Wkh whfkqltxh vhhpv surplvlqj/ hvshfldoo| iru idvw idglqj fkdqqhov/ exw wkh
frpsoh{lw| lv vwloo wrr kljk1 Lw zrxog eh yhu| lqwhuhvwlqj wr frpelqh v|qfkur0
ql}dwlrq dqg fkdqqho hvwlpdwlrq wr dfwxdoo| jhw d zrunlqj v|vwhp vxlwdeoh iru
lpsohphqwdwlrq1
77 FKDSWHU 61 VXPPDU\ DQG FRQWULEXWLRQV
Eleolrjudsk|
78
79 ELEOLRJUDSK\
^5:` V1 Kd|nlq1 Dgdswlyh Ilowhu Wkhru|1 Suhqwlfh Kdoo/ Xsshu Vdggoh Ulyhu/
QM/ XVD/ 6ug hglwlrq/ 4<<91
Lqfoxghg Sdshuv
86
Paper A
Paper A
This part has been published as:
Fredrik Tufvesson and Torleiv Maseng, "Pilot Assisted Channel Estimation for
OFDM in Mobile Cellular Systems", Proceedings of IEEE Vehicular Technology
Conference, Phoenix, USA, pp. 1639-1643, May 1997.
IEEE 1997, reprinted with permission.
Pilot Assisted Channel Estimation for OFDM
in Mobile Cellular Systems
Fredrik Tufvesson and Torleiv Maseng
Department of Applied Electronics, Lund University,
Box 118, S-221 00 Lund, Sweden
E-mail:Fredrik.Tufvesson@tde.lth.se
Abstract
The use of pilot symbols for channel estimation introduces overhead and it is
thus desirable to keep the number of pilot symbols to a minimum. The number
of needed pilot symbols for a desired bit error rate and Doppler frequency is
highly dependent on the pilot pattern used in orthogonal frequency division mul-
tiplexed, OFDM, systems. Five different pilot patterns are analysed by means of
resulting bit error rate, which is derived from channel statistics. Rearrangement
of the pilot pattern enables a reduction in the number of needed pilot symbols up
to a factor 10, still retaining the same performance. The analysis is general and
can be used for performance analysis and design of pilot patterns for any OFDM
system.
1 Introduction
1
2 PILOT PATTERN DESIGN
data
transmitter decision
channel
pilot estimator
pattern
receiver
Figure 1: The pilot pattern is known both by the receiver and the transmitter. The pur-
pose is to minimize the number of transmitted pilot symbols without increasing the
bit error rate.
thereby minimizing the overhead introduced by the pilot symbols. The problem is to
decide where and how often to insert pilot symbols. The spacing between the pilot
symbols shall be chosen small enough to enable reliable channel estimates but large
enough not to increase the overhead too much. This paper includes among all an
algorithm of how to design a suitable pilot estimation pattern.
2 Estimation Strategies
1. Measure all channels at the same time, compare to a broadband single-carrier sys-
tem.
2. Measure the channels in increasing order, one at a time.
3. Measure the channels one at a time in a smart, but predetermined, way. The mea-
surement order is derived upon channel statistics and is optimal in the sense that
the total bit error rate is kept at a minimum each symbol time.
4. A pilot pattern presented by T. Mueller et al. [6], where the pilots are located with
equidistant spacings in time and frequency.
PILOT PATTERN DESIGN 3
64
1) 2)
......
...
Pilot
symbol
Channel
Data
symbol
2
1 ... ...
1 2 Time 65
3) 4) 5)
...
...
..
5. A pilot pattern by P. Hoeher [7], used in [4]. The pilot symbol locations are
shifted one step in frequency each pilot interval.
For comparison the same amount of pilots is used, one of 64 sent symbols. This
means that only 1.6% extra overhead is introduced by the pilots, but this is not suffi-
cient for large Doppler frequencies in some of the cases.
3 System
At different signal to noise ratios the resulting bit error rate from each pilot pattern is
evaluated using the algorithms given in Section 8. A matched filter receiver and
coherent BPSK or QPSK are used. Additive white Gaussian noise is assumed for
every sub-channel. The channel has delays and Rayleigh distributed amplitudes
according to the COST 207: Typical Urban profile. The complex parts of the trans-
fer function are assumed to change according to a first order auto-regressive process
as described in Section 6. The reason for using this model is to get a linear system
which rather easy can be hand-led algebraically. A Kalman filter is used to estimate
the frequency response of the channel. From the filter, a time dependent covariance
matrix is given as described in Section 7. This is used to calculate the expected bit
error rate for each channel. See figure 3 for a description of the system.
4 Results
The resulting bit error rate curves of the pilot patterns are presented for different
Doppler frequencies in figure 4.
4 PILOT PATTERN DESIGN
I Typical urban X
PSK channel, H +
Real BER I decision H
matched filter
BER for QPSK when different pilot patterns are used, Eb/No=10
Pattern 1
Pattern 2
Pattern 3
Pattern 4
Pattern 5
Coherent
-1
10
BER
-2
10 -4 -3 -2 -1
10 10 10 10
fd*Ts
Figure 4: Resulting bit error rate as a function of Doppler value when the different
pilot patterns are used. The same number of pilot symbols are used for all curves.
For a given Doppler frequency the pilot pattern used sets the limit for the lowest pilot
density to be used, alternatively for a given pilot density it limits the maximum Dop-
pler frequency allowed. The calculations are made at 1800 MHz using 10 kHz chan-
nel separation between 64 OFDM channels carrying in total 640 ksymbols/s. No
intersymbol interference, perfect synchronization and known Doppler frequency, fd,
is assumed. 1.6% of the sent symbols are pilot symbols and the average bit error rate
of the 64 channels is presented.
The bit error rate is degraded both by imperfect channel estimates and noise dis-
turbances. The pilot pattern used determines the conversion between noise limited
PILOT PATTERN DESIGN 5
BER for QPSK when different pilot patterns are used, fdTs=0.001
0
10
Pattern 1
Pattern 2
Pattern 3
Pattern 4
Pattern 5
-1 Coherent
10
BER
-2
10
-3
10
-4
10
0 5 10 15 20 25 30
Eb/No
Figure 5: Bit error rate at different signal to noise ratios. Note the difference between
the error floors.
It is interesting to study the resulting pattern when estimating the channel that
gives the lowest bit error rate (strategy 3). A steady state pilot pattern is often
achieved where only few of the sub-channels are measured, see figure 6. Channel
estimates of the other sub-channels are achieved by filtering.
To see the effect of mismatch between the pilot pattern design parameters and the
actual parameters, the optimal pilot patterns (pattern 3) for three Doppler values were
60 60
50 50
40 40
Channel
Channel
30 30
20 20
10 10
20 40 60 80 100 20 40 60 80 100
Time Time
Figure 6: Resulting pilot pattern when sending a pilot at the channel which gives the
lowest total bit error rate. Note that only few channels will be used for pilot symbols.
6 PILOT PATTERN DESIGN
used when moving at another speed. The actual Doppler frequency was as before
assumed to be known by the estimator, just to see the effect of the pilot pattern with-
out influences of the estimation algorithm. The designed pilot patterns for the typi-
cal urban environment were also used in hilly terrain and rural area specified in
[2] to see the influence of the propagation environment on the bit error rate, see
figure 7. As seen in the figure, the pilot patterns are robust to mismatches in the
design parameters.
5 Discussion
To minimize the bit error rate it is desirable to spread the pilot symbols both in time
and in frequency, as seen in figure 4 and figure 5. Normally a worst case design is
made for the channel estimation system and then we suggest to tailor the pilot pattern
to each base station site. A suitable pilot pattern can be calculated once the propaga-
tion environment and maximum expected speed in the particular cell is known. In
such a system, the pilot pattern used in the cell is among the information transferred
to the mobile when it logs into a new cell. When designing the pilot pattern one also
has to take the estimation algorithm into account. Here the estimation algorithm is
used only for evaluation and the complexity of the used algorithm is not a problem.
In some cases the pilot pattern has to aid the estimation algorithm to enable a less
complex one. The estimation algorithm used here, the Kalman filter based on an AR-
process, has no delay and the received signal can be detected immediately, i.e. no
future pilot symbols is taken into account when making the channel estimate. If the
BER when using pattern designed for another Doppler freq. or channel, Eb/No=10
-1
10
Rural
Hilly
Urban
fd*Ts=0.02
fd*Ts=0.002
fd*Ts=0.0002
-2
10 -4 -3 -2 -1
10 10 10 10
fd*Ts
Figure 7: Changes in the bit error rate due to Doppler mismatch and power delay pro-
file mismatch. The pilot patterns designed for fdTs=0.02, 0.002 and 0.0002 were used
at different Doppler values and the designed patterns for typical urban environment
were used in hilly terrain and rural area.
PILOT PATTERN DESIGN 7
received signal is stored in a memory, the pilot symbols can be used in both time
directions and the time spacings between pilot symbols can be increased, retaining
the same performance.
The bit error rates within the sub-channels differ depending on where the pilots
are located. When minimizing the total bit error rate (pattern 3) the channels of the
sides get higher error rate, see example in figure 8.
6 Channel Model
0.12
0.11
0.1
BER
0.09
0.08
0.07
0.06
80
60
40
70
60
20 50
40
30
20
0 10
Time 0
Channel
Figure 8: The bit error rate becomes higher for the side channels when the total bit
error rate is kept at a minimum
8 PILOT PATTERN DESIGN
N
h ( , t ) = an ( t ) ( t n ) (1)
n=1
The tap coefficients, an(t), and the tap delays, n , are chosen according to the COST
207 Typical Urban model in the GSM specification [2]. The transfer functions,
H(f,t), are obtained by the Fourier transform and these are the functions we want to
estimate for the different carriers. These channel transfer functions are regarded as
flat fading and constant during a symbol time.
A first order AR-process is used to model how the different taps may change from
one time instant to another. If we look at all the transfer functions at the same time, it
is possible to set up a state-space model of the form:
H( k + 1 ) = H(k) + v( k)
(2)
y(k) = C(k)H(k) + e(k)
The matrix is a diagonal N*N-matrix (here treated as a scalar) with elements
k AR 2 f d T s
e (3)
that define the AR-process. Ts is the symbol time including any cyclic prefix or guard
space. The white noise v(k) has covariance matrix R 1 = F R GSM F , where RGSM cor-
responds to the multipath intensity profile described in [2]. The vector y(k) is the
measured transfer functions. C(k) is an observation vector with ones only at the posi-
tions (channels) measured at time KTs and e(k) is measurement noise with a diagonal
covariance matrix R2.
The parameter kAR in the auto-regressive process for the channel changes is cho-
sen to adjust the memory so that it corresponds to the memory of Jakes model.
Channels corresponding to the U-shaped spectrum given in [8] were simulated and
then estimators based on an AR-model with different kAR were used, see Figure 9:
Simulations were performed with one sub-channel (Eb/No= 10, fdTs=0.002) with
every tenth symbol as a pilot symbol. In the figure the bit error rates of the nine data
symbols are shown. figure 9 shows that the best fit, in this case, is reached for
kAR0.15.
The adjustment of the AR-process can also be seen as an adjustment of the band-
width. If we set the 90% power bandwidth of the AR process equal to the bandwidth
of the Doppler spread, see figure 10, a value of kAR0.158 can be calculated. This is
the value used for all bit error rate calculations throughout the paper.
PILOT PATTERN DESIGN 9
Simulated BER for QPSK, estimates based on AR-models with different fdTs
0.15
0.13
BER
0.12
0.11
0.1
0.09 -1 0 1
10 10 10
fdTs used in estimator (fdTs of estimator)/(fdTs of simulated channel)
Figure 9: Bit error rate for a simulated channel by Jakes when estimates are based on
an AR-process. The minimum value is reached for kAR=0.15.
Figure 10: Power density spectrum of the channel by Jakes and a first order AR-pro-
cess. The latter is adjusted to have the 90% power bandwidth equal to the Doppler
spread.
7 Channel Estimator
For the analysis and pilot pattern design a Kalman filter is used to estimate the trans-
fer functions. The Kalman filter is causal and uses measurements up to time k to esti-
mate the transfer functions, H(k). The Kalman filter is given [3] by (4)-(8), where X
denotes conjugate transpose of X:
H ( k k ) = H ( k k 1 ) + P ( k k 1 ) C ( k ) (4)
1
{ C ( k ) P ( k k 1 ) C ( k ) + R 2 } ( y ( k ) C ( k ) H ( k k 1 ) )
10 PILOT PATTERN DESIGN
H( k + 1 k) = H( k k 1) + (5)
K(k )[y( k ) C( k )H(k k 1 )] = H(k k )
K ( k ) = P ( k k 1 ) C ( k ) (6)
1
[ C ( k ) P ( k k 1 ) C ( k ) + R 2 ]
P ( k + 1 k ) = P ( k k 1 ) + R 1 (7)
K ( k ) [ C ( k ) P ( k k 1 ) C ( k ) + R 2 ] K ( k )
P ( k k ) = P ( k k 1 ) P ( k k 1 ) C ( k ) (8)
1
[ C ( k ) P ( k k 1 ) C ( k ) + R 2 ] C ( k ) P ( k k 1 )
The reconstruction error H (k k ) = H (k ) H (k k ) is given by:
H(k k ) = [ K( k)C( k) ]H( k 1 k 1 ) + v(k 1 ) (9)
1
P ( k k 1 ) C ( k ) [ C ( k ) P ( k k 1 ) C ( k ) + R 2 ]
{ e( k ) + C( k) v( k 1 ) }
The Kalman filter is optimal in the sense that the variance of the reconstruction error
is minimized. The matrix P(k|k) is the variance matrix and this is used together with
K to make an estimate of the bit error rate, which in turn is used to decide the order in
which the channels are going to be measured. For pattern 3, the channel is chosen
that minimizes the total bit error rate of the channels after the measurement. The
matrices P(k|k) and K are independent of the measured values and therefore it is pos-
sible to precompute the order in which the channels are going to be measured.
The bit error rate is calculated for BPSK and QPSK. A matched filter receiver is
assumed. The sampled output of this filter is given by
2
j ------ ( n 1 )
M
Xm = 2 Hm e + em (10)
where is the signal energy, Hm is the current transfer function at channel m, n is the
signal alternative among M sent and em is white gaussian noise. The bit error rate at
channel m is calculated as [1]
PILOT PATTERN DESIGN 11
1
P bmBPSK = --- ( 1 m )
2
1 m (11)
P bmQPSK = --- 1 --------------------
2
2 m
2
where
E [ X m Hm ]
m = ------------------------------------------------ (12)
2
E [ X ] E [ H ]
2
m m
Hm are the outputs of the Kalman filter. These are not known in advance and there-
fore (13) -(15) are used. If matrix notation is used and signal alternative 1 is used for
the pilot symbol (does affect the analysis here, but in practise different pilot symbol
alternatives should be used), the expectations for all the channels can be calculated
as:
E[ X(k )H( k) ] = E[{ 2H( k) + N(k )} { H( k ) H( k) } ] (13)
= 2 E [ H ( k ) H ( k ) ] 2 E [ H ( k ) H ( k ) ]
2
E [ X ( k ) ] = E [ { 2 H ( k ) + N ( k ) } { 2 H ( k ) + N ( k ) } ] (14)
2
= 4 E [ H ( k ) H ( k ) ] + E [ N ( k ) N ( k ) ]
2
E[ H(k ) ] = E[{ H( k ) H( k )} { H( k ) H(k ) } ] (15)
= E [ H ( k ) H ( k ) ] 2E [ H ( k ) H ( k ) ] + E [ H ( k ) H ( k k ) ]
The expected values E [ X ( k ) H ( k ) ] , E [ X ( k ) 2 ] and E [ H ( k ) ] are independent of the
2
measured values but dependent upon which channels that are measured. Therefore
they are time dependent. The expectations can be calculated as:
E [ H ( k ) H ( k ) ] (16)
= E [ { H ( k 1 ) + v ( k 1 ) } { H ( k 1 ) + v ( k 1 ) } ]
2
= E [ H ( k 1 ) H ( k 1 ) ] + E [ v ( k 1 ) v ( k 1 ) ]
2
= R1 ( 1 )
E [ N ( k ) N ( k ) ] = R 2 (17)
12 PILOT PATTERN DESIGN
E [ H ( k ) H ( k k ) ] = E [ H ( k ) H ( k ) H ( k ) H ( k k ) ] = (18)
E [ H ( k 1 ) H ( k 1 ) + R 1 R 1 C ( k ) a
H ( k 1 ) H ( k 1 ) C ( k ) a ]
= ( E [ H ( k 1 ) H ( k 1 ) ] + R 1 ) ( I C ( k ) a )
where
E[H(k k)H(k k ) ] = P(k k) (19)
1
a = P ( k k 1 ) C ( k ) { C ( k ) P ( k k 1 ) C ( k ) + R 2 } (20)
The bit error rate calculation is compared and verified by simulations. Proakis [1]
gives expressions for the bit error rate when estimating a constant Rayleigh channel
using different numbers of pilot symbols, see figure 11.
The bit error rate when using only one pilot symbol corresponds to estimating a
rapidly changing channel or the first estimate of an unknown channel. Then, only the
latest measurement is useful. In a similar way, the bit error rate when estimating a
constant channel with use of (infinitely) many pilot symbols corresponds to that of
coherent detection. Normally in a practical system the effect of the estimation is
somewhere between these two cases.
-1
10
BER
-2
10
-3
10
-4
10
0 2 4 6 8 10 12 14 16 18 20
Number of pilot symbols used for the estimate
Figure 11: Bit error rate for 2PSK when estimating a constant Rayleigh channel with
several pilot symbols at different signal to noise ratios.
PILOT PATTERN DESIGN 13
9 Conclusions
The bit error rate for pilot assisted QPSK modulation is calculated when using differ-
ent pilot patterns. The ability to estimate the channel reliably when it changes due to
e.g. vehicle movements is highly dependent on the pilot pattern used. By rearranging
the pilot pattern it is, in some cases, possible to handle 10 times higher Doppler fre-
quency alternatively possible to reduce the number of needed pilot symbols the same
amount, still retaining the same bit error rate. Alternatively the new pilot pattern
could be used to reduce the bit error rate up to a factor 5, even more in a low noise
environment. The pilot symbols in the proposed pilot pattern are spread out both in
time and frequency. For a given propagation environment, e.g. a base station site, it is
possible to pre-calculate a suitable pilot pattern.
References
Paper B
This part has been published as:
Abstract
The bit error rate in orthogonal frequency division multiplex (OFDM) sys-
tems is affected by the number of sub-channels used, changes in the channel
characteristics and, to some extent, the excess delay of the channel. This paper
presents an analytical expression for the bit error rate on Rayleigh fading chan-
nels when interchannel interference (ICI) caused by channel changes during a
symbol and energy loss due to the cyclic prefix are regarded. This expression is
used to optimize the number of sub-channels, and thereby the sub-channel band-
width in the system. It is argued that the system can be optimized neglecting the
effect of imperfect channel estimation and on a worst case assumption for the
Doppler frequency and signal to noise ratio. The analysis is general and can be
used for performance analysis and optimization of any mobile OFDM system.
1 Introduction
OFDM, orthogonal frequency division multiplexing, is used and proposed for several
broadcast systems [1] [2] and there is a growing interest in using the technique for the
next generation land mobile communication system. In OFDM systems the informa-
tion signal can be seen as divided among and transmitted by several narrowband sub-
carriers. The bandwidths of the sub-carriers depend on the bit rate sent on each of
them and consequently, for a given total bitrate the sub-carrier bandwidth is depen-
dent on the number of sub-channels used. The number of sub-channels used is set in
the system design and the problem is to find a good trade off between bandwidth,
limitations by the hardware and the physical channel. As the sub-carrier bandwidth is
reduced, the symbol time on each carrier gets longer, the channel changes during a
symbol gets larger and channel compensation gets more difficult. When the changes
in the channel characteristics during a symbol time become evident, the orthogonality
between the subchannels is lost and interchannel interference (ICI) arises. On the
other hand as the carrier bandwidth and the sub-channel symbol rate increases,
intersymbol interference (ISI) becomes a problem. This problem can be avoided by
1
2 SUB-CHANNEL BANDWIDTH
introducing a cyclic prefix [3], but this results in an energy loss. In between these two
extremes there exists an optimal bandwidth where the bit error rate is minimized, this
bandwidth is found in this paper.
2 System description
When designing an OFDM system one often start with a requirement on the total bit
rate. Assume that a total bitrate of Rtot bits per second is required. If these are equally
divided between the channels then the bitrate
R tot
R 1 = --------- (1)
N
is transferred on each of the N sub-channels. The symbol time for M-ary modulation
on each sub-channel becomes
log 2M N log 2M
T sub = --------------- = -------------------- . (2)
R1 R tot
In the following the influence of some of the most important design parameters are
discussed.
copy
f
N*f
f
t
Tcp Ts
Tsub
Figure 1: Eight symbols sent on four sub-channels. The striped area indicates inser-
tion of the cyclic prefix.
1 1 R tot
f = ----- = ------------------------- = ------------------------------------------ (3)
Ts T sub T cp N log 2M T cp R tot
Insertion of a cyclic prefix means that the sub-channel bandwidth has to be increased
in order to keep the bit rate R1 constant. The bit rate determines the symbol time Tsub
and if a part of this is used for the cyclic prefix, then the time Ts has to be shortened
which in turn leads to increased sub-channel bandwidth.
The cyclic prefix also leads to a power loss, cp . The receiver uses the energy
received during the time Ts and discards the energy corresponding to the duration of
the cyclic prefix. The remaining signal energy can be calculated as cp E s where
Ts T cp R tot
cp = ---------- = 1 -------------------- . (4)
T sub N log 2M
Even though the channel is perfectly estimated there are some variations in the trans-
fer function during each symbol interval. These variations becomes evident when
many sub-channels are used due to the long symbol time, they are hard to track and
result in interchannel interference, ICI. When the number of sub-channels is suffi-
ciently large the resulting ICI can be modeled as additive white Gaussian noise,
which is added to the channel noise [6] with spectral density N0. The variance of the
ICI-noise, 2ICI = E [ n ICI n ICI ] , is calculated as[6]:
4 SUB-CHANNEL BANDWIDTH
N1
log 2M T cp
= 2E s 1 ---- ------ ( N n ) J 0 2 f D n --------------- --------
2 1 2
ICI (5)
N N 2 R tot N
n=1
If we neglect the effect of channel estimation errors, the equivalent signal to noise
ratio after the cyclic prefix is removed and the ICI-noise is added becomes
E E s cp
-----s- = -----------------------
-. (6)
N 0 eq 2
N 0 + ICI
The BER for QPSK/4QAM in a Rayleigh fading channel with coherent detection can
then be calculated as
Es
p ( y ) Q y ------ dy = ---
1
pb = N 0 eq 2
(7)
y=0
1 NE s ( 2N T cp R tot )
--- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-
2 N1
2 log 2M T cp
N ( 4N 0 + 10E s ) NE s ( 8 + T cp R tot ) 16E s ( N n ) J 0 2 f D n --------------- --------
R tot N
n=1
3 dB
-1
10
13 dB
-2
10
BER
23 dB
-3
10
33 dB
loss incl.
no loss
-4
10
3 4 5
10 10 10
Figure 2: The BER curves for Rtot=320, 640 and 1280 kbit/s overlap each other
totally in this case. Coherent QPSK is used in a Rayleigh fading channel with Tcp=10
s and fD=50 Hz when the number of sub-channels and signal to noise ratio are var-
ied. The dotted lines represent coherent detection without losses.
3 dB
-1
10
13 dB
-2
10
23 dB
BER
-3
10 33 dB
-4
10 10 Hz
50 Hz
200 Hz
-5
10
3 4 5
10 10 10
Sub-channel bandwidth/Hz
Figure 3: Bit error rate for coherent detection when the Doppler frequency is varied,
fD={10, 50, 200} Hz, Rtot=320 kbit/s and Es/N0= {3, 13, 23, 33} dB. The dotted lines
represent coherent detection without losses.
6 SUB-CHANNEL BANDWIDTH
The radio channel corrupts the transmitted signal and in order to make coherent
detection possible we have to know the impact of the channel. Both the amplitude
and phase are corrupted by the fading channel, whose characteristics vary because of
movements of the mobile terminal. In order to keep track of the channel characteris-
tics pilot symbol assisted modulation, PSAM, can be used. This means that known
training symbols are multiplexed into the data stream at certain sub-channels and cer-
tain times. The receiver interpolates the channel information derived from the pilot
symbols and makes channel estimates for the data symbols. Since the pilot symbols
carry no data the pilot density is to be kept at a minimum not to increase the overhead
too much. In order to make it possible for the receiver to achieve nearly maximal
channel information from each of the pilots the pilot pattern can be made balanced
[7]. This means that the spacings between the pilots are approximately the same in
both frequency and time when normalized by the minimum expected coherence
bandwidth and the minimum expected coherence time respectively. Few sub-chan-
nels means that we have to insert several data symbols between the pilot symbols in
the time domain. Many sub-channels gives us the opportunity to use several of the
channels for pilots simultaneously without increasing the pilot density, see figure 4.
Let NT and NF denote the pilot time distance respectively pilot frequency distance
in number of symbols, fDmax denote the maximum expected Doppler frequency (Hz)
and max denote the maximum expected excess delay (s) of the channel, then the con-
dition for balanced design becomes [7]
f f f f
t t t t
N=1 N=2 N=4 N=8
Pilot symbols Data Symbols
NT
f
Data
Pilot
NF
Figure 5: An example of a balanced pilot pattern with pilot time distance NT=5 and
pilot frequency distance NF=4.
1
f Dmax T sub N T = max f N F = --- (9)
4
An example of a pilot pattern with NF=4 and NT=5 is given in figure 5.
With a fixed time-frequency block, edge problems can arise when designing the
pilot pattern. In order to get rid of this problem in the following and in order to sup-
press the bit error rate at the edges, pilots are put on the outer sub-channels respec-
tively time instants and then sufficiently many pilot symbols are evenly distributed
between these edge pilots in frequency respectively time in order not to exceed the
pilot distances given by (9).
To see the effect of the channel estimation only, the bit error rate was calculated
without ICI-noise and energy losses due to the cyclic prefix. For the estimation a
two-dimensional Wiener filter [7] was used, which is optimal in the sense that the
variance of the estimation error is minimized. Furthermore, known Doppler fre-
quency, perfect synchronization, a COST 207 Typical Urban channel [8] and a
matched filter receiver was assumed, see Appendix A for more details. The resulting
bit error rate under these assumptions with a bit rate of 320 kbit/s, 200 Hz Doppler
frequency is presented infigure 6.
The reason why the different pilot patterns virtually do not affect the bit error rate
is explained by the fact that the pilot symbols are placed at approximately the same
frequencies and time instances independent of the sub-channel bandwidth and sym-
bol time used. There are some changes in the bit error rates due to the fact that the
pilot distances can not be below one, due to numerical reasons in the pilot pattern
design and due to the slight increase in the total bandwidth to estimate when few
channels are used, but these changes are rather small.
8 SUB-CHANNEL BANDWIDTH
3 dB
-1
10
13 dB
-2
10
BER
23 dB
-3
10 33 dB
-4
10
3 4 5
10 10 10
Sub-channel bandwidth/HZ
Figure 6: Changes in the bit error rate caused by the pilot pattern, Es/N0={3, 13, 23,
33} dB. As seen, the pilot pattern has virtually no effect on the bit error rate when
balanced design is used.
3 Complete System
In a real system the channel estimation is of course not perfect. The bit error rate
when both channel estimation and the losses due to the cyclic prefix and ICI-noise
are included is lower bounded by the bit error rate for coherent detection in (7) and
the deviation gets smaller as the pilot density increases. The bit error rate for the
same parameters as in last section (Rtot=320 kbit/s fD=200 Hz, Typical Urban chan-
nel, Tcp=10 s) but with the losses included is presented in figure 7.
4 Discussion
More narrowband sub-channels can be used when terminal movements are slow
SUB-CHANNEL BANDWIDTH 9
3 dB
-1
10 13 dB
-2
10 23 dB
BER
-3
33 dB
10
est. incl.
coherent
-4
10
3 4 5
10 10 10
Sub-channel bandwidth/Hz
Figure 7: Bit error rate for coherent detection and for the total system when channel
estimation errors are included, Es/N0={3, 13, 23, 33} dB.
since the channel is not expected to change during the longer symbol time. This
means that the relative duration of the cyclic prefix becomes shorter and that the
bandwidth expansion is kept small. However, the sub-channel bandwidth has to be
optimized on a worst case assumption for the Doppler frequency, otherwise the
increase in bit error rate can become severe for high Doppler values.
The effect of frequency offsets is not included in the analysis and this effect
becomes more significant when many narrowband sub-channels are used. The carrier
to interference ratio due to frequency offsets is proportional to f 2 [9], thus moving
the optimal sub-channel bandwidth towards broader channels when frequency offsets
are present. Finally, the calculations are based on the assumption that the sub-chan-
nels are flat fading, which may not be the case when broadband sub-channels are
used in some channels, e.g. hilly terrain.
10 SUB-CHANNEL BANDWIDTH
5 Conclusions
The impact of the number of sub-channels, and thereby the sub-channel bandwidth,
used in a mobile OFDM-system was analysed. An analytical expression for the bit
error rate in Rayleigh fading channels was presented where the impact of the cyclic
prefix and the channel changes during a symbol time are encountered. This BER
showed an optimum for the sub-channel bandwidth to be used depending on Doppler
frequency, length of cyclic prefix and signal to noise ratio. The effect of channel esti-
mation on the optimum was also examined and this showed that the optimization can
be performed on the coherent system, without taking the channel estimation into
account. The excess delay of the channel, high noise levels and efficient usage of the
spectrum call for many narrowband sub-channels while a rapidly moving terminal
call for the use of fewer more broadband sub-channels. Finally it was argued that the
sub-channel bandwidth has to be designed on a worst case assumption for the Dop-
pler frequency and the signal to noise ratio.
In the following section the channel model, estimation algorithm and bit error rate
calculations are presented in detail.
N
h ( , t ) = an ( t ) ( t n ) (A.1)
n=1
The tap coefficients, an(t), and the tap delays, n , are chosen according to the COST
207 Typical Urban model in the GSM specification [8]. The transfer functions,
H(f,t), are obtained by the Fourier transform of (A.1) and these are the functions we
want to estimate for the different carriers. These channel transfer functions are
regarded as flat fading.
For the estimation and analysis a two-dimensional Wiener filter [7] is used, which
is optimal in the sense that the variance of the reconstruction error is minimized. The
sampled signal from the matched filter receiver of a pilot symbol at channel m in time
k is given by
SUB-CHANNEL BANDWIDTH 11
2
j ------ ( p 1 )
k k M k
Xm = Es Hm e + nm (A.2)
where Es is the signal energy, p is the signal alternative sent among the M possible
ones and n is white Gaussian noise. The noise terms have a variance given by (7), but
the covariance matrix R ke = E [ n k ( n k ) ] is not diagonal due to the correlation between
the channels introduced by the ICI. The filter coefficients are given by
k k T 1 T
m = ( ( m ) ) (A.3)
where = E [ XX ] is the covariance matrix for the received pilot symbols and
k k k
m = E [ H m ( X m ) ] is a cross-covariance vector between the transfer function to be esti-
mated and the received pilot symbols. The estimates of the channel transfer functions
can then be calculated as
k k T
Hm = ( m ) X (A.4)
where X is a vector of the sampled values from all of the pilot symbols.
k 2 k T 1 k
J m = H ( m ) ( m ) (A.5)
and this matrix is in turn used to calculate the QPSK bit error rate for each of the
symbols. The BER is given by [10] [11]
1
P b = --- ---------------------- (A.6)
2 2
2 2
where
2
H
= E s -------------------------------------------------- . (A.7)
2 2 2
( H + e ) ( H + J )
12 SUB-CHANNEL BANDWIDTH
References
[1] Digital broadcasting systems for television, sound and data services. Euro-
pean telecommunications standards institute, prETS 300 744, Valbonne,
France, 1996
[2] Radio broadcasting systems; Digital Audio Broadcasting (DAB) to mobile,
portable and fixed receivers. European telecommunications standards institute,
ETS 300 401, Valbonne, France, 1995
[3] A. Peled and A. Ruiz, Frequency domain data transmission using reduced
computational complexity algorithms, Proc. IEEE Int. Conference Acoustics,
speech, signal processing, Denver, USA, pp 964-967, 1980
[4] S. Hara, M. Mouri, M. Okada and N. Morinaga, Transmission performance
analysis of multi-carrier modulation in frequency selective fast Rayleigh fading
channel, Wireless Personal Communications, nr 2, Kluwer Academic Publish-
ers, Dordrecht, Netherlands, pp 335-356, 1996
[5] M. Wahlqvist, C. stberg, J.-J. van de Beek, O. Edfors and P. O. Brjesson, A
conceptual study of OFDM-based multiple access schemes: Part 1 - Air inter-
face requirements. Technical report Tdoc 116/96 ETSI STC SMG2 meeting no
18, Helsinki, Finland, May 1996
[6] M. Russell and G. L. Stber, Terrestrial digital video broadcasting for mobile
reception using OFDM, Wireless Personal Communications, vol 2, nr 1-2,
Kluwer Academic Publishers, Dordrecht, The Netherlands, pp 45-66, 1995
[7] P. Hoeher, S. Kaiser and P. Robertson, Pilot-symbol-aided channel estimation
in time and frequency, In K. Fazel and G. P. Fettweis, editors, Multi-Carrier
Spread-Spectrum, chapter IV, pp 169-178. Kluwer Academic Publishers, Dor-
drecht, The Netherlands, 1997.
[8] European Telecommunications Standard Institute, GSM 05.05, version 4.6.0,
July 1993.
[9] Y. Zhao and S.-G. Hggman, Sensitivity to Doppler shift and carrier frequency
errors in OFDM systems -- The consequences and solutions, IEEE Vehicular
Technology Conference, Atlanta, USA, April 29 - May 2, pp. 1564-1568, 1996.
[10] S. K. Wilson, Digital audio broadcasting in a fading and dispersive channel,
Ph.D. thesis, Stanford University, USA, 1994.
[11] J. G. Proakis, Digital communications, third edition, McGraw-Hill, New
York, USA, 1995.
Paper C
This part will be published as:
Fredrik Tufvesson, Mike Faulkner and Torleiv Maseng, "Pre-Compensation for Ray-
leigh Fading Channels in Time Division Duplex OFDM systems", Wireless Personal
Communication, 2000.
Kluwer Academic Publishers 2000, reprinted with permission.
Pre-Compensation for Rayleigh Fading
Channels in Time Division Duplex OFDM
systems
Fredrik Tufvesson1, Mike Faulkner2 and Torleiv Maseng1
Abstract
Coherent demodulation results in good detection performance but requires
channel estimation. Fading pre-compensation (precoding) at the transmitter can
lead to low-complex receiver structures with good performance capabilities,
without the need for channel estimation. Time division duplex systems based on
orthogonal frequency division multiplex (OFDM) are particularly suited to this
because intersymbol interference effects can be neglected, simplifying transmit-
ter adjustments. Methods that involve amplitude and/or phase pre-compensation
are compared in terms of resulting bit error rate and increase in peak-to-average
power ratio. Dynamic channels degrade the performance as the block lengths get
longer. For a certain block length the performance degrades below that of tradi-
tional differential decoding. A block length of up to 40 times that used in the
Digital European Cordless Telephone system, DECT, is possible using channel
estimation and ideal Wiener prediction.
1 Introduction
In time division duplex (TDD) systems the same frequency is used both by the base
station and the mobile terminal, but at different time instants. In these systems the
possibility of making predictions of the channel impact is rather straightforward. The
used radio channel is often assumed to be reciprocal, see e.g. [1] [2], since we are
interested in compensating the multipath channel and not the interference. If the
block length is short compared to the speed of change of the channel, then predictions
of the channel influence are likely to be accurate for adjacent blocks and we can
adjust the signal to be transmitted to achieve better performance.
Pre-compensation can be used for different purposes, for example to control the
1
2 PRE-COMPENSATION
output power [3], to reduce interference from other users [4], or to modify the trans-
mitted signal in order to reduce intersymbol interference (ISI) [1]. In this paper we
examine the possibilities and limits when using pre-compensation of the channel
impact on the downlink symbols in orthogonal frequency division multiplex (OFDM)
systems. Figure 1 shows the principle of the technique. In OFDM systems it is
information
to be sent mobile
pilot terminal
pilot symbols transmitter
multiplexer
received antenna
information receiver
data received
base symbols information
receiver equalizer
station
pilot
symbols channel estimate
antenna channel estimator
channel
estimate information
to be sent
transmitter pre-compen-
sation
straightforward to compensate for a time dispersive channel since its impact after the
FFT only is an amplitude attenuation and a phase shift of the data symbols. Channel
estimation is normally a complex operation to perform and every mobile terminal has
to make channel estimates in order to use coherent detection. By pre-compensation it
is possible to move the task of channel estimation to the base station, but still achieve
the performance benefits of coherent detection at the mobile terminal [5]. This means
that cheap, less complex, mobile receivers can be used. We address here the problem
of channel pre-compensation in a simple uncoded OFDM system, but the results can
be applied for other systems where the channel can be modeled as a single complex
tap or together with coding and diversity arrangements.
Compensating channel attenuation means that we transmit more power where the
channel attenuation is high and less power where the channel attenuation is low. This
contradicts the water pouring theorem, which states that more power and higher order
modulation should be used where the channel attenuation is low in order to maximize
PRE-COMPENSATION 3
the transmission capacity. However, here the aim is to use fixed modulation and
enable a low-complex receiver.
The paper is organized as follows: In section 2, the system and the model for the
OFDM system are described. In section 3 we then present the three analyzed com-
pensation methods, and in section 4 we derive their performance in case where the
channel is known by the base station. In section 5 the resulting increase in peak-to-
average power ratio is derived for the three methods, respectively. We then proceed
in section 6 with the method where only the phase shift of the channel is compen-
sated and we derive the maximum block length before the performance gets worse
than that of differential detection. Finally, the conclusions are presented in section 7.
2 System
In the paper we use an OFDM TDD-FDMA system with M sub-channels, see Figure
2 for a description of the transmitter and receiver. The OFDM signal is transmitted
through a time dispersive Rayleigh fading channel. It is assumed that the fading is
flat for each of the sub-channels and constant during a symbol interval. We assume
further a cyclic prefix exceeding the maximum excess delay, and that the demodu-
lated signals do not suffer from interchannel interference nor ISI.
Under these assumptions we can use a model for the OFDM system where the data
s1[k] r1[k]
data
S/P IFFT P/S D/A h(t,) A/D S/P FFT P/S
sM[k] rM[k]
n(t)
signal is transmitted on separate, but correlated, Rayleigh fading channels with addi-
tive Gaussian noise. The transmitted OFDM signal is convolved with the channel
impulse response, h ( t, ) , but after demodulation the convolution only results in a
phase shift, m [ k ] , and a power gain, ym[k], of the data symbols. The sampled com-
plex representation of the received signal on channel m at time k can therefore be
written as [6]
j m [ k ]
rm [ k ] = ym[k ]e sm [ k ] + nm [ k ] , (1)
where sm[k] is the complex representation of the transmitted data signal and nm[k] is
independent samples of white Gaussian noise.
3 Compensation Methods
The base station makes estimates of the channel amplitude and channel phase e.g.
by using known pilot symbols from the uplink. Then the signal to be transmitted is
compensated. If we let u q[ k ] denote the complex representation of the uncompen-
sated signal alternative q, the complex representation of the compensated signals,
u q [ k ] , becomes
j
( uq [ k ] ) 1 = e u q[ k ] (2)
0 y < a2
( uq [ k ] )2 = 2 j y a2 (3)
------ e u q[ k ]
y
PRE-COMPENSATION 5
3 j
------ e u q[ k ]
a3 y < a3
( uq [ k ] ) 3 = , (4)
3 j y a3
------ e u q[ k ]
y
for the three methods respectively. Here 2 and 3 are power normalization coeffi-
cients. The probability density function of the power gain, y, in a Rayleigh channel
y
with unity mean is p Y ( y ) = e . Hence, if the channel is perfectly estimated and
E q denotes the symbol energy before compensation, then the respective mean trans-
mitted symbol energies are
( E q ) 1 = E q (5)
y
e
( E q ) 2 = 2 E q ------- dy (6)
y
a2
a3
3 y 3 y
( Eq )3 = ------ e E qdy +
a3 -----
y
-e E qdy (7)
0 a3
y
a 3
= 3 E q ----- ( 1 e ) + ------- dy .
1 e
a3 y
a3
In order to maintain the same transmitted mean energy for the three methods, the
power normalization coefficients are consequently chosen as
y
2 = 1 ------- dy
e
(8)
y
a2
y
a3
3 = 1 ----- ( 1 e ) + ------- dy .
1 e
(9)
a3 y
a3
6 PRE-COMPENSATION
The performance limit for each of the given methods is achieved when all downlink
transfer functions are known by the base station and the channel is compensated per-
fectly. This is of course not the situation in a practical case, but it is anyhow interest-
ing to get a lower limit for the resulting bit error rate (BER).
The BER, Pb, when the Rayleigh fading channel is perfectly known is for the first
method the same as for ordinary coherent detection of QPSK signals, i.e. [6]
2E b y
( Pb )1 = Q y ---------- e dy
N0
(10)
0
1 E b N 0
= --- 1 --------------------------
- .
2 1 + E b N 0
Here N0 denotes the density of the noise and E b denotes the bit energy before com-
pensation, E b = E q 2 for QPSK. The BER, when the base station knows the chan-
nel impact, can for the two other methods be calculated as
a2
1 y 2E b y
( Pb )2 = --- e dy +
2 Q 2 ---------- e dy
N0
(11)
0 a2
1 a2 a2 2E b
= --- ( 1 e ) + e Q 2 ----------
2 N0
a3
3 2E b y 2E b y
( Pb )3 = Q y -----
- ---------- e dy +
a3 N0 Q 3 ---------- e dy
N0
(12)
0 a3
1 3 ( E b N 0 ) 2E b 1
- Q 2a 3 + 3 ---------- --- .
= --- + ----------------------------------------
2 a 3 + 3 ( E b N 0 ) N0 2
The resulting BER is shown in Figure 3 together with the curves for differential
QPSK [6].
At low signal to noise ratios phase-only compensation appears to be best for the
majority of power compensating limits. At medium and high signal to noise ratios
PRE-COMPENSATION 7
0
10
1
10
2
10
BER
3
10
4
10 diff.
ph. only
5
10 Tx off
Tx sat.
6
10 5 4 3 2 1 0 1 5 4 3 2 1 0 1
10 10 10 10 10 10 10 10 10 10 10 10 10 10
compensation limit compensation limit
Figure 3: BER for the compensation methods when the channel is known by the base
station, Eb/N0= 20 dB (left) and 10 dB (right). The compensation limit (i.e. a1, a2 or
a3) is the maximum power amplification (attenuation) of the channel to be entirely
compensated.
there is some value in adding amplitude compensation with peak power saturation,
but the question is whether one can afford the increased peak-to-average power ratio.
For the amplitude compensation methods a low compensation limit means that too
much energy is spent on compensating very bad channels, while a high limit means
that the transmitter is saturated or off even when communication normally is advis-
able.
The compensation limits, a2 and a3, can be seen as measures of the dynamic range
for the power compensation. In practical measurements the peak-to-average ratio of
the transmitted power is often used, which for the pre-compensation methods is the
product of the peak-to-average ratio of the modulation method, e.g. OFDM, and the
increase due to the compensation. For a single sub-channel the increase in peak-to-
average power ratio can be calculated as 1, 2 a 2 and 3 a 3 for the three methods
respectively, but for an OFDM signal the increase is dependent on the correlation
between the sub-channels and which particular channels that are compensated. If we
compensate the phase only, then the transmitted signal is a sum of sinusoids with dif-
ferent phases and there is no increase in the peak-to-average power ratio. If we com-
pensate the amplitude as well, then the transmitted signal is a sum of weighted
sinusoids and we get a larger peak-to-average power ratio. In Figure 4 we present
8 PRE-COMPENSATION
10
ph. only
8
Tx off
peaktoaverage ratio (dB)
Tx sat.
6
high corr. full corr.
4
low corr.
2
3 2 1 0 1
10 10 10 10 10
compensation limit
simulated values of the increase in peak-to-average power ratio when applying the
different methods in channels with various delay spreads. We use here the OFDM
based wireless local area network (LAN) specified in ETSI BRAN [7]1 as an exam-
ple. The system uses 52 sub-channels and has a sub-channel bandwidth of 312.5 kHz.
For the simulations we use three different channel models:
An indoor channel with short delay spread (channel A, rms =50ns), resulting
in large sub-channel correlation.
A channel with long delay spread modeling large open buildings (channel E,
rms =250ns). This gives low correlation between the sub-channels.
The simulated values of the increase in peak-to-average power ratio are achieved as
the difference between the 99.5% peak-to-average power ratio levels for the uncom-
pensated and pre-compensated OFDM signal.
In Figure 4 we see that we get a larger increase in peak-to-average power ratio if the
sub-channels are correlated, i.e. the delay spread is short. A large peak in the trans-
mitted signal is achieved when the sub-channels contributing to the peak are ampli-
fied due to the amplitude compensation. When the correlation is small, the increase
becomes smaller. This is because the probability that all sub-channels contributing to
the peak are amplified becomes lower. The simulations are performed for a specific
system, but similar results can be expected for other OFDM systems if we take into
account the relation between the coherence bandwidth of the channel and the OFDM
bandwidth.
In practice the transfer functions are not known, but they have to be estimated e.g. by
transmitting known so-called pilot symbols. The impact of the channel on the pilot
symbols is observed, and the channel coefficients for the data symbols are achieved
by filtering and prediction. Different alternatives can be used to form these channel
estimates. Here we use a two-dimensional Wiener filter, which uses both the correla-
tion in time and in frequency to derive the channel estimates. The Wiener filter is
optimal in the sense that the mean squared error is minimized. For further details
about channel estimation in OFDM systems see e.g. [8].
The channel predictions for the downlink block become obsolete after a certain
time. This means that the length of the downlink block is limited before the BER
reaches an unacceptable level. This particular level is dependent on the application.
Here we compare the BER to that of differential detection and use the latter as a
benchmark since the complexity of the mobile terminals is almost the same. The
maximum block length is a function of the Doppler frequency, fD, in the system since
it determines the speed of change in the channel. In the following we assume a classi-
cal U-shaped Doppler spectra such that the time correlation can be described by a
zeroth-order Bessel function. We also concentrate on the performance of the phase-
only compensated system, since there is no increase in peak-to-average power ratio.
In order to investigate the influence of the observation time we study the performance
of a single sub-channel in the OFDM system. We use almost noise free pilot symbols
10 PRE-COMPENSATION
uplink downlink
for channel estimation where the observation time t1, the time between the first and
last pilot symbol, is varied, see Figure 5. The low-noise pilot symbols are used in the
analysis since we want to separate the impact of the time varying channel. In the
analysis we have not compensated the symbol energy of the data symbols due to the
boosted pilots. For long and medium block lengths the performance is limited by the
time varying channel and not by the noise. Besides, in a well designed OFDM system
the noise affecting the pilot symbols are averaged over several symbols when calcu-
lating the channel estimates. Therefore, the variance of the channel estimation error is
much lower than the variance of the noise affecting data symbols.
In Figure 6 the BER as a function of prediction time t2, the time since the last
pilot symbol, is presented for different observation times. All the pilot symbols, are
0
10
fDt1=0
fDt1=0.01
fDt1=0.05
1
10 fDt1=0.30
BER
2
10
3
10
0 0.05 0.1 0.15 0.2
prediction time (fD*t2)
Figure 6: BER as a function of prediction time, fDt2, for different observation times,
fDt1. Eb/N0=20 dB for the dashed lines and Eb/N0=10 dB for the solid ones. The dash-
dotted lines represent differential detection.
PRE-COMPENSATION 11
equally distributed within the observation time t1. In this example pilot symbols at
P=4 different time instants are used for prediction. The signal to noise ratio, Eb/N0, is
set to 20 dB and 10 dB respectively, except for the pilot symbols where we use a 30
dB higher signal to noise ratio. We assume that the Doppler frequency is known and
use a Wiener filter for estimation and prediction of the channel. The BER is derived
2
from the mean squared error, e , of the estimate. For QPSK it is given by the corre-
lation coefficient, , between the backrotated received samples and the channel esti-
mate as [6, appendix C]
1
P b = --- ---------------------- . (13)
2 2
2 2
The correlation coefficient when using pilot symbols can be calculated as
Es y
est = --------------------------------------------------------- , (14)
2
( Es y + N0 ) ( Es y + e )
where y = E [ y ] is the mean power of the channel. The BER for DQPSK is also
given by (13), but with
Es
diff = ------------------ . (15)
Es + N0
The BER of the downlink block is determined by the ability of making correct
channel predictions. This ability is dependent on the location of the pilot symbols in
the uplink block, especially their spreading in time, the observation time. Longer
observation time means in general better predictions. An observation time of zero
corresponds to keeping the last channel estimate in the uplink block as a channel pre-
diction for the whole downlink block. By just adding some information about the
trend of the channel values, a considerable decrease in the BER could be achieved.
But the enhanced predictions are naturally dependent on the validity of the channel
model and the ability to suppress the noise affecting the pilot symbols.
At a certain prediction time, the differential system and the pre-compensated sys-
tem have the same bit error performance, see Figure 6. This point, the break even
point, is evaluated in Figure 7 as a function of the observation time and the number of
used pilot symbols using the same assumptions as before. For block lengths smaller
than the ones in Figure 7 the (instantaneous) BER is lower for a pre-compensated
system than for a differential system. Beyond the break even point, differential detec-
tion without pre-compensation gives better results for similar receiver complexity. As
an example, look at Figure 6, Eb/N0=10 dB and the curve for observation time
fDt1=0.05. After a certain time fDt2=0.17 the BER is the same as for a differential sys-
12 PRE-COMPENSATION
0.35 P=1
P=2
0.3
P=3
P=4
0.25
P=5
break even (fD*t2)
0.2
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4
observation time (f *t )
D 1
Figure 7: Break even time where the BER for a pre-compensated system starts to
exceed the one for a differential system, Eb/N0= 10 dB. P is the number of time
instants in the uplink block used for pilot symbols.
tem. This point can be found in Figure 7. For P=4 and an observation time of 0.05 the
break even point is fDt2=0.17. In Figure 7 we clearly see the effect of the observation
time. By using all the pilot symbols for prediction and not only the last estimate there
is an increase in the break even time of at least a factor 3. We also see the importance
of using sufficiently many pilot symbols to ensure that there always are pilot symbols
with high correlation to the point to be predicted. Pilot symbols at 3 or 4 time instants
in the uplink block seems enough for this.
To compare the above block lengths to an existing TDD system we can use the
Digital European Cordless Telephone (DECT) system, which has a block length of
368 s [9]. For a terminal at pedestrian speed (10 km/h, 1900 MHz carrier frequency)
this results in a normalized block length of fDt=0.0065. With pre-compensation,
Eb/N0=10 dB, P=4 and a known Doppler frequency the break even time becomes
fDt2=0.26 (when the observation time is equal to the break even time). This means
that it is possible to use approximately 40 times longer block lengths than the one in
the DECT system before the BER becomes worse than that of differential detection.
In a real system the statistics of the time variations may not be known and one
often avoids the requirement of measuring Doppler frequency. The estimation and
prediction filters are therefore often adjusted to the worst case scenario for the chan-
nel changes. This of course decreases the feasible block lengths in the system. In Fig-
ure 8 the break even points are given when the Wiener filter is designed for k times
PRE-COMPENSATION 13
0.15
k=1
k=2
k=4
k=7
break even (fD*t2)
0.1 k=10
0.05
0
0 0.02 0.04 0.06 0.08 0.1
observation time (f *t )
D 1
Figure 8: Break even time when the Wiener filter is adjusted to k times the actual
Doppler frequency and pilot symbols at P=4 time instants are used for prediction,
Eb/N0= 10.
higher Doppler frequency than the actual one. If a pedestrian for example reduces the
speed to 2.5 km/h when the Wiener filter is designed for 10 km/h, the mismatch fac-
tor k is 4 and the break even time is reduced from fDt2=0.18 to 0.07. But this means
on the other hand that absolute block length measured in number of symbols or the
time t2 is longer in the slow but mismatched case since the Doppler frequency is
smaller there, ( t 2 ) 2.5 km/h 0.016 s instead of ( t 2 ) 10 km/h 0.010 s. The feasible
block length in number of symbols is in the general case still determined by the worst
case Doppler frequency and therefore mismatch is not regarded as a problem.
7 Conclusions
accurate. For unknown Doppler frequencies where the filters are adjusted to the
worst case scenario, the feasible block lengths are determined by the maximum Dop-
pler frequency, though the filters are mismatched for lower Doppler frequencies.
References
Fredrik Tufvesson was born in Lund, Sweden in 1970. He received the M.S. degree
in Electrical Engineering from Lund University in 1994, the Licentiate Degree in
1998 and he is currently a Doctoral student at the Department of Applied Electronics,
Lund University. His research interests include channel estimation and synchroniza-
PRE-COMPENSATION 15
Mike Faulkner received the B.S.E.E. from Queen Mary College, London Univer-
sity, U.K., the M.E. degree from university of New South Wales, Australia and the
Ph.D. degree from the University of Technology, Sydney in 1970, 1978 and 1993
respectively. Between 1988 and 1996 he spent three periods at Lund University, Swe-
den, the last time as a Visiting Professor working in radio hardware design. Since
1977 he has been a Lecturer and then Professor at Victoria University of Technology,
Melbourne, Australia. His current interests are signal processing, radio circuit design
and spread spectrum.
Fredrik Tufvesson and Ove Edfors, "Preamble-based Time and Frequency Synchro-
nization for OFDM Systems", submitted to IEEE Journal on Selected Areas in Com-
munications, January 2000.
Preamble-based Time and Frequency
Synchronization for OFDM Systems 1
Fredrik Tufvesson, student member, IEEE, and
Ove Edfors, member, IEEE
Department of Applied Electronics, Lund University, Sweden
Abstract
Fast and reliable time and frequency synchronization is crucial for packet-
based orthogonal frequency division multiplex (OFDM) systems. Various syn-
chronization methods are analyzed to determine their performance when using
preambles based on repeated short pseudo noise (PN) sequences or conventional
preambles based on repeated OFDM data symbols. We make analytical evalua-
tions for AWGN channels and simulate the performance for one- and two-tap
Rayleigh fading channels. Conventionally, the synchronization signal is calcu-
lated as a sum of products between received samples. However, it is shown that
switching the order of summation and multiplication improves the detection per-
formance in terms of lower probability of false detection and of missing the syn-
chronization signal. The false detection probability is decreased by several
orders of magnitude for reasonable parameter settings. The performance in terms
of frequency-offset estimation is similar for both preambles. The PN-based pre-
ambles have low peak-to-average power ratio and they make it possible to use
analog-to-digital converters with only one-bit quantization in stand-by mode.
Therefore, the PN-based preambles allow for a great reduction in stand-by mode
power consumption.
1 Introduction
There is great interest in using orthogonal frequency division multiplex (OFDM) for
high-speed wireless local area network applications. Standardization of systems
where OFDM is used is occurring both in the United States [1], Japan [2] and Europe
[3]. The systems are based on packet data transmission and therefore fast and reliable
time and frequency synchronization with low overhead is crucial to the systems.
1 Parts of this work have been presented at the IEEE Vehicular Technology Confer-
ence, Amsterdam, The Netherlands, September 1999.
1
2 PREAMBLE-BASED SYNCHRONIZATION
ing samples one symbol or half a symbol apart. The timing signal is then the ampli-
tude of the correlator output. Time synchronization is achieved when the amplitude
exceeds a certain threshold. The frequency-offset estimate is then obtained from the
phase of the same signal. When using a cyclic prefix the timing signal has a flat
region, which causes an uncertainty about where the actual start of the packet is [6].
On the other hand, if no cyclic prefix is used, the frequency-offset estimate is cor-
rupted by intersymbol interference. This has to be suppressed by weighting the
received samples [7].
There are low power variations within the preamble, which means relaxed
requirements on the automatic gain control (AGC).
The amplitude of the synchronization signal is small when the timing is wrong
compared to the amplitude when the timing is correct. This means low probability
of false detection and of missing the correct synchronization signal.
To our knowledge, this is a novel approach for OFDM system synchronization and
no analysis of the statistics of the synchronization signal has been made before. The
idea of using one-bit quantization has been presented before for conventional OFDM
synchronization [10] but there the quantization loss was large. By using the technique
presented here this loss becomes low enough to make a one-bit solution attractive.
2 System Description
The preambles are designed for an OFDM system with M sub-channels. The aim is to
estimate the start of the packet and achieve a reliable frequency-offset estimate using
a single-symbol preamble, but other preamble lengths are of course also possible. We
compare two preambles, a PN-based and a conventional one. The comparison is per-
formed using two alternative synchronizer structures, the conventional OFDM syn-
chronizer and a pre-correlating synchronizer.
The PN-based preamble was derived from a R/2 chip long m-sequence, where R
denotes the repetition length of the preamble. The sequence was oversampled twice
and repeated until the total length became 2R+G samples, as shown in Figure 1,
where G denotes the length of the guard interval. In order to decrease the power vari-
R R G
AGC m-seq m-seq guard
G R R
AGC guard OFDM OFDM
ations within the preamble we used an offset QAM constellation for transmission
where the quadrature sequence was delayed one sample compared to the inphase
sequence. The preamble can be generated using the FFT of the PN-sequence as input
to the OFDM modulator so no extra device is required for preamble generation.
The main parts of the pre-correlating synchronizer are presented in Figure 2. First,
pre-correlation of length K with the known sequence is performed. When the
received sequence coincides with the known sequence, a large peak occurs with a
phase angle determined by the channel. When there is no synchronization signal, the
output is noise only. The correlator output is multiplied by the conjugate of the output
delayed PR times. The positive integer P denotes an extra delay, which can be used to
decrease the variance of the frequency-offset estimate. Most often no extra delay is
used and P=1. Correlation peaks spaced PR samples apart produce a large synchroni-
zation signal with a (small) phase shift proportional to the frequency-offset. The
maximum frequency offset is inversely proportional to the pre-correlator length. If
the pre-correlator length is limited, e.g. due to a large frequency offset, and the varia-
tions of the synchronization signal is too large it is possible to sum consecutive peaks
to decrease the variance. However, normally a long correlator is preferred and no
summation is necessary.
samples
K
| |2
cK-1 cK-2 c3 c2 c1 c0
x
timing
( )*
PR
freq. est.
arg
[ k, a ]
Figure 2: Structure of the pre-correlating synchronizer when using only one pre-corr-
elator product for synchronization (L=1).
PREAMBLE-BASED SYNCHRONIZATION 5
For both methods the absolute value of the synchronization signal is compared to
the short time mean of the received power. If a threshold is exceeded, time synchroni-
zation is achieved and the frequency estimate is available from the phase of the syn-
chronization signal.
For the analytical performance evaluation we consider an AWGN channel. The base
band representation of the received signal is
2 c t
r 0 ( t ) = s 0 ( t ) exp j -------------- + + n 0 ( t ) , (1)
T
s
where n0(t) is white Gaussian noise with one-sided spectral density N0, and s0(t) is
the transmitted preamble. Further, is the carrier phase and c = f T s is the fre-
quency-offset to be estimated when normalized by the sub-channel spacing 1/Ts.
The synchronization signals from the two synchronizer structures are represented
in a common framework. Let r[k] denote the sampled received signal and c[k] the
sequence used for pre-correlation. If the pre-correlating synchronizer is used then the
pre-correlator length is equal to the repetition length, i.e. K=R. If no pre-correlator is
samples
PR
| |2 ( )*
x
timing
used, as for conventional OFDM synchronization, then K=1 and all c[k]=1. Further,
let a denote the slip between the received sequence and the locally generated pre-cor-
relation sequence. In the following analysis this slip is assumed to be an integer. The
received sampled signal is multiplied by the conjugate of the known pre-correlation
sequence to get a remodulated signal, which is summed over length K to get the
pre-correlator output. The remodulated signal can be calculated as
c [ k a ] r [ k ] (2)
2 c k
= s c [ k a ] s [ k ] exp j --------------- + c + n [ k ] ,
M
2
where * denotes complex conjugation, s is the power of the transmitted signal, M is
the number of sub-channels in the OFDM system, n[k] is white Gaussian noise with
2
variance n and s[k] is the transmitted preamble.
The synchronization signal for both the conventional and the pre-correlating syn-
chronizer is given by
K 1
L1
[ k, a ] = c [ k n lR a ] r [ k n lR ]
(3)
l = 0 n = 0
K 1
c [ k n ( l + P ) R a ] r [ k n ( l + P ) R ] ,
n = 0
where, again, L is the number of products used for the synchronization signal, P is the
extra delay when calculating the products and K is the pre-correlator length. When
using the pre-correlator we often have only one product for synchronization and no
extra delay so L=1 and P=1, without the pre-correlator these parameters are set to
L=R and P=R.
3 Performance Evaluation
In this section we analytically analyze the performance of time and frequency syn-
chronization for the two preambles and the two synchronizer structures in AWGN
channels. To compare different parameter choices, we normalize the synchronization
2 2
signal as [ k, a ] = [ k, a ] ( LK s ) . As we will see below, when the timing is
correct the mean of the synchronization signal is almost unity, and when the timing is
wrong the mean is close to zero.
In subsection 3.1, we analyze the distribution of the timing signal when the timing
is correct, i.e. when a=0 and there is no slip between the received and known
sequence. In subsection 3.2 we derive the distribution of the timing signal when the
timing is wrong. By wrong timing we mean that there is no synchronization signal
present and the synchronizer is fed by white Gaussian noise only. We concentrate on
the cases where the synchronization signal consists of one, two or a large number of
products. Normally, only one product is used together with the pre-correlating syn-
chronizer since this results in the best performance. Further, in subsection 3.3, we
analyze the performance of time synchronization when using the low-resolution
mode. In subsection 3.4 we summarize our results on time synchronization and
finally, in subsection 3.5, we analyze the performance of the frequency-offset esti-
mate. For the figures and numerical examples in this section we use an AWGN chan-
nel with 0 dB SNR and a preamble length of 60 samples (R=30), except for the
evaluation of the frequency-offset estimate where we have other SNR values and the
preamble length is 64 for the OFDM-based preamble. The extension is used in that
case in order to get a full OFDM symbol as a preamble.
c K 2 c
[ k, 0 ] = ------- sinc --------- exp j ------------ k + --- + n [ k, 0 ] ,
1 1
(4)
L M M 2
2 2 2
where n [ k, 0 ] is white Gaussian noise with variance = n ( KL s ) and the
8 PREAMBLE-BASED SYNCHRONIZATION
2 2
output has been normalized by LK s . Define, in the same way as in [5], inphase
as the phase of the sum,
2 c PR
arg [ k, 0 ] [ k PR, 0 ] -------------------- . (5)
M
L
For reasonable SNR values the noise-times-noise products are negligible and the
inphase part of the synchronization signal [ k, 0 ] is much larger than its quadrature
part. The amplitude, i.e. the timing signal, can therefore be approximated by the
inphase components as
2 c K
--- sinc --------
1
[ k, 0 ] [ k, 0 ] = L M
- + (6)
L
c K 2 c ( k PR + 1 2 )
InPhase n [ k, 0 ] ------- sinc --------- exp j -------------------------------------------------
1
L M M
c K 2 c ( k + 1 2 )
+ ------- sinc --------- exp j ------------------------------------ n [ k PR, 0 ] .
1
L M M
The approximation, [ k, 0 ] , is Gaussian with mean
2 c K
E [ [ k, 0 ] ] = sinc --------- , (7)
M
and variance
2 c K
var ( [ k, 0 ] ) = n sinc --------- ( LK s ) .
2 2
(8)
M
The timing signal is always positive and, since the expected value is large compared
to the variance, the approximation as a Gaussian gives a negligible probability of
negative values. When there is a frequency offset, the mean decreases slightly since
the samples do not add totally in phase. However, this loss is small in the interesting
frequency-offset range. The cumulative density function (CDF) at the correct timing
point can be expressed as
x E [ [ k, 0 ] ]
F [ k, 0 ] ( x ) F [ k, 0 ] ( x ) = 1 Q ------------------------------------ , (9)
var ( [ k, 0 ] )
When the timing is correct, the distribution of the timing signal with the conven-
PREAMBLE-BASED SYNCHRONIZATION 9
E [ conv [ k, 0 ] ] = 1 , (10)
and variance
2
n
var ( conv [ k, 0 ] ) = -------------- . (11)
2
LK s
Monte Carlo simulations, not presented here, show a good agreement between the
true values and the approximations.
When using the pre-correlating synchronizer, the normalized complex outputs of the
2 2 2
sub-correlators have zero mean and variance = n ( KL s ) . If the synchroniza-
tion signal consists of one product only, i.e. L=1, we can rewrite the timing signal as
[ k, a ] = [ k lR, a ] [ k ( l + P ) R, a ] . (12)
Since the terms are the amplitude of complex Gaussians, they are Rayleigh distrib-
uted with probability density function (PDF)
2x x2
f [ k, a ] ( x ) = ------ exp ------ . (13)
2
2
The two terms in (12) are independent. By using the transformation theorem on their
joint distribution, the PDF of the timing signal can be calculated as
2 2 x 1 2 x 2
f [ k, a ] ( x ) = ------ --- exp ------ b + ----- db (14)
2 b 2 2
b
0
2 2 x
= ------ xk 0 2 ------ ,
2 2
where k0(x) denotes the zeroth order modified Bessel function. The CDF of the tim-
ing signal is, consequently, given by
10 PREAMBLE-BASED SYNCHRONIZATION
F [ k, a ] ( ) = f [ k, a ] ( x ) dx = 1 2 ------ k 1 2 ------
2 2
(15)
0
2 2
K s K s
= 1 2 -------------- k 1 2 -------------- .
2
n 2n
For L=2 we are not able to derive exact expressions for the distributions of the
timing metric, but we determine bounds for the CDFs instead. By an upper bound we
mean that if the CDF of A is upper bounded by the CDF of B then F A ( x ) F B ( x ) .
The opposite is true for a lower bound. In appendix A we show that the CDF of
[ k, a ] is lower bounded by the CDF of an Erlang-2 variable X ,
2 2
F X ( x ) = 1 1 + ------- x exp ------- x . (16)
2
2
We also show that the CDF of the timing signal [ k, a ] when L=2 is upper
bounded by the CDF of the maximum, Y , of two independent exponential variables,
2 2
F Y ( y ) = 1 exp ------- y . (17)
2
When the number of terms, L, in the sum is large, the distribution of the timing
signal can be calculated in the same way for the pre-correlating synchronizer and the
conventional synchronizer. In such cases, the distribution of the real and the imagi-
nary part of the timing signal can be approximated as Gaussian due to the central
limit theorem. The real and imaginary part of the normalized timing signal,
Re { conv } and Im { conv } , both have zero mean and a variance
2 2 2
Re { = L( ) 2 . (18)
conv }
2 2
x
2 2 2 s
F [ k, a ] ( x ) = 1 exp ---------------------------- = 1 exp x K L ------ . (19)
2
conv
2 Re { } 2n
conv
The timing signal from the pre-correlating synchronizer is generally smaller com-
pared to the same signal from the conventional synchronizer when the timing is
wrong. This in turn means that the false detection probability of the pre-correlating
PREAMBLE-BASED SYNCHRONIZATION 11
There is a high correlation between the sign of the received signal and the sign of the
pre-correlating sequence when the timing is correct. The correlation is zero when the
timing is wrong and therefore it is possible to have a low-resolution operation mode
where only the sign of the received signal is used to find a coarse timing signal. The
probability that the sign of the received sample is equal to the sign of the pre-correlat-
ing sequence can, for the I-channel and the Q-channel respectively, be calculated as
2 c k
p I, equal [ k ] = 1 Q 2 ( s n ) cos --------------- + c
2 2
M
(20)
2 c k
p Q, equal [ k ] = 1 Q 2 ( s n ) sin --------------- + c ,
2 2
M
when the timing is correct. Note that if the phase of the received signal is between
and 2 then the probability of equal signs in e.g. the quadrature channel is lower than
0.5. However, the same is true for the second part of the preamble so that the product
will in general be positive anyway. The correlator adds all the equal and unequal
samples and therefore the two complex parts of the correlator output, low res ,
have a binomial distribution. The probability that the real part of pre-correlator has nI
samples of equal signs is given by
K nI K nI
P Re ( equal ( n I ) = ( p I, equal ) ( 1 p I, equal ) (21)
low res ), n I
with a corresponding expression for the imaginary part. For small frequency offsets
and correct timing the real part of the synchronization signal is much larger than the
imaginary part, so the absolute value can be approximated by the real part only. The
12 PREAMBLE-BASED SYNCHRONIZATION
0.8
wrong
timing
0.6 correct
CDF
timing
0.4 p.c. L=1
p.c. L=2 u.b.
p.c. L=2 l.b.
0.2
conventional
correct
0
0 0.5 1 1.5
normalized timing signal,
Figure 4: CDFs of the timing signals when using the conventional synchronizer or the
pre-correlating synchronizer (p.c.) with L=1 or L=2. In case of L=2 and wrong tim-
ing we show an upper (u.b.) and lower bound (l.b.) for the distribution.
real part of the synchronization signal is the sum of the products between the two real
parts of the correlator outputs and the two imaginary parts of the correlator outputs.
The correlator output is given by low res = 2n I K + i ( 2n Q K ) . The probabil-
ity function of the products is calculated by adding all the probabilities for values
resulting in a certain product and the probability function, P Re ( ( n ) , of the
low res )
sum is then achieved by convolution. For larger frequency offsets we have a similar
distribution, but the signal is not dominated by the real part.
When the timing is wrong, the input can be seen as random and the probability
that the sign of the received sequence is equal to the sign of the pre-correlating
sequence is PI,equal=PQ,equal=50% for the I and Q-channel, respectively. In this case
none of the parts dominate and we derive bounds for the distribution. The absolute
value of the synchronization signal is larger than the maximum absolute value of the
real and imaginary part of the synchronization signal and smaller than the sum of the
amplitudes of the real and imaginary part respectively. The CDF of the synchroniza-
tion signal is therefore upper bounded by the CDF of the maximum absolute value of
the real and imaginary part and lower bounded by the CDF of the sum of the absolute
values of the real and imaginary parts, i.e.
0.8 wrong
timing
0.6
CDF
correct
0.4 timing
0.2
wrong u.b.
wrong l.b.
correct
0
0 200 400 600 800 1000 1200
lowresolution timing signal
Figure 5: CDFs of the low-resolution timing signal. For wrong timing we have a
lower and an upper bound for the distribution.
the low-resolution timing signal using the PN-based preamble and the same parame-
ters as in Figure 4. For wrong timing we show an upper and a lower bound for the
distribution. In the figure it can be seen that the probability is very small that the tim-
ing signal is larger when the timing is wrong than when it is correct.
For the timing signals there is a relationship between the probability of missing the
synchronization signal and the probability of false detection, which is dependent on
the chosen detection threshold. We plot this relationship, the so-called receiver oper-
ation characteristics, in Figure 6 for the conventional and pre-correlated timing signal
together with an upper and lower bound for the low-resolution timing signal. Note
that we plot the probability of missed synchronization signal and not, as most often,
the probability of detection vs. the probability of false detection. We see in the figure
that the timing signal from the pre-correlating synchronizer results in much better
performance, compared to the conventional synchronizer. For a probability of missed
synchronization signal of 10-4 we get a false detection probability of 3*10-2 for the
conventional synchronizer while we get 2*10-8 for the pre-correlating synchronizer.
The lower variance for the pre-correlated timing signal when the timing is wrong also
means that it is possible to decrease the threshold for detection without having too
many false alarms. The low-resolution synchronization signal also results in good
detection performance, in AWGN channels it competes well with the conventional
full-resolution synchronizer. When the probability of missed synchronization signal
is 10-4 the false detection probability from the low-resolution synchronization signal
14 PREAMBLE-BASED SYNCHRONIZATION
0
10
2
10
missed sync probability
4
10
6
precorr
10 conventional
lowres l.b.
lowres u.b.
8
10 10 8 6 4 2 0
10 10 10 10 10 10
false detection probability
is in the order of 10-3. Compare this to the false detection probability of the conven-
tional, but full resolution, synchronization signal that for these parameters is 3*10-2.
In [10] a synchronizer based on the cyclic prefix using an A/D-converter with one-bit
quantization is analyzed. This system is similar to conventional preamble synchroni-
zation but the repetition due to the cyclic prefix is used instead of the repetition in the
preamble. There the low-resolution synchronizer gave a significant loss compared to
conventional full-resolution synchronization. Here the performance of the low-reso-
lution synchronizer is actually better than that of the conventional full resolution syn-
chronizer.
M
c = ------------------- arg ( [ k, 0 ] ) . (23)
( 2 PR )
This phase increment method gives a (nearly) unbiased frequency estimate of a com-
plex exponential in white noise [11] and can cope with frequency-offsets which are
smaller than c < M ( 2PR ) .
var ( c ) (24)
2 2 2
M
2 1 n 1 n
= -------------------------------- --- ------------------------------------------ + ----------- - .
-----------------------------------------
2 2 2 2
( 4 KLPR ) L s sinc ( c K M ) 2KP 2s sinc 2 ( c K M )
The performance of the frequency-offset estimate is similar for the conventional and
the pre-correlating synchronizer. For low SNRs the variance from the pre-correlating
synchronizer is slightly lower. In this case the effective SNR after the pre-correlator
is often high enough that the second term in (24) can be neglected. In Figure 7 we
plot the theoretical standard deviation of the frequency-offset estimation error
together with simulated results for the conventional synchronizer using the OFDM-
based preamble and the pre-correlating synchronizer using the PN-based preamble.
The preamble length was 60 (K=30, L=1) for the PN-based preamble and 64 (K=1,
L=32) for the OFDM-based one in an AWGN channel with SNR 0 10 and 20 dB,
respectively. The simulated values agree well with the theoretical values. For the pre-
correlating synchronizer at large frequency offsets it is possible to observe the loss
due to the sinc-term in (24). However, as seen in the figure this increase in the stan-
dard deviation is small.
0.4
precorr K=30
0.2 conv L=32
sim conv
0.1
sim precorr
0 dB
standard deviation
0.04
10 dB
0.02
0.01
20 dB
0.004 3 2 1 0
10 10 10 10
frequency offset,
Figure 7: Theoretical and simulated values of the standard deviation of the fre-
quency-offset estimation error. AWGN channel with SNR 0, 10 and 20 dB. Note that
the OFDM-based preamble is slightly longer than the PN-based one.
16 PREAMBLE-BASED SYNCHRONIZATION
In multipath channels the timing signal from the pre-correlating synchronizer has a
peak for each of the resolvable channel taps. The correlation peaks have the same
phase, except for a small distortion due to the noise, and their amplitudes correspond
to the tap-power for each delay. For time synchronization the synchronizer can either
use the largest peak, a sum of the largest peaks or a sum of all peaks within a time
window corresponding to the excess delay of the channel to find the correct timing.
In OFDM systems we want a timing signal that has a peak when the largest part
of the channel impulse response is within the window set by the guard interval. This
is straightforward to achieve with a rectangular window with length as the guard
interval. To get sharp synchronization peaks and to minimize the influence between
peaks it is important to use a sequence with good autocorrelation properties. Mini-
mizing the influence is especially important for the frequency-offset estimate, other-
wise there will be a small increase in the standard deviation of the estimation error.
The PN-based preamble used together with the pre-correlating synchronizer makes it
possible to resolve taps spaced one chip (i.e. two samples) apart.
As before, we compare the timing signals when the timing is correct and when the
timing is wrong so that the input to the synchronizer in the latter case is noise only.
The pre-correlating synchronizer in this and the next figure used the largest peak and
not a sum to find the timing. In Figure 9 we present simulated receiver operation
characteristic for the two-tap Rayleigh fading channel. The delay between the taps
was 4 samples such that we got two fully resolvable Rayleigh fading taps when using
the PN-based preamble with the pre-correlating synchronizer. Due to the delay the
preambles were extended by 4 samples. For a probability of missed synchronization
signal of 10-3 the probability of false detection is here 2*10-1 for the conventional
synchronizer, 10-2 for the low-resolution synchronizer and 7*10-4 for the pre-corre-
lating synchronizer, respectively.
PREAMBLE-BASED SYNCHRONIZATION 17
1
10
missed sync probability
2
10
3
10
4
10 conventional
lowresolution
precorr, PN
5
10 precorr, OFDM
5 4 3 2 1 0
10 10 10 10 10 10
false detection probability
1
10
missed sync probability
2
10
3
10
4
10 conventional
lowresolution
precorr, PN
5
10 precorr, OFDM
5 4 3 2 1 0
10 10 10 10 10 10
false detection probability
The pre-correlating synchronizer gives better detection performance both for the
one-tap and the tow-tap channel. The reason is again that the mean amplitude is
lower for that timing signal when the timing is wrong. The low-resolution synchro-
nizer also performs well compared to the conventional one. It also gives lower proba-
bility of false detection and lower probability of missing the synchronization signal.
When using the pre-correlating synchronizer the difference between the OFDM-
based and the PN-based preambles is small. Both preambles have good autocorrela-
tion properties and therefore their performance is similar. The PN-based preamble
has a slightly better performance when the threshold is low and the false detection
probability thereby is high. This is because the received samples are weighted
equally for the PN-based preamble while some samples have higher weight for the
OFDM-based preamble. For short preambles this implies that the variance of the tim-
ing signal is slightly smaller when the timing is wrong.
5 Conclusions
0.04 10 dB
standard deviation
0.02
20 dB
0.01
conventional
precorr, PN
0.004 precorr, OFDM
3 2 1 0
10 10 10 10
frequency offset
L1
[ k, a ] = [ k lR, a ] [ k ( l + P ) R, a ] (25)
l=0
L21
= d 2k + 1 c 2k + 1 + b 2k + 1 e 2k + 1 .
k=0
All the terms (b, c, d and e) in (25) are independent zero mean Gaussian variables; d
2 2 2
and b are real valued and have variance b = d = E [ bb ] = 2 , whereas c and
e are complex valued with variance of the both complex parts
2 2 2
Re { c } = Im { c } = . Defining new variables as g k = d 2k + 1 Re { c 2k + 1 } and
h k = d 2k + 1 , the joint probability function for g and h becomes
g 2 1 2 1
f G, H ( g, h ) = ------------------------------ exp ----------------- --- ------ h --- .
1 1
(26)
2 Re { c } d 2 h 2
h
Re { c } d
Integrating [13] (3.478.4) over h gives the PDF for the products, i.e.
= --------------------------- k 0 -----------------------
1 g
fG ( g ) = f G, H ( g , h ) d h Re { c } d Re { c } d
(27)
2 2 g
= ---------- k 0 ------------- .
2 2
Naturally this PDF also applies for b 2k + 1 e 2k + 1 and for the corresponding imagi-
nary parts. The characteristic function of the product, g, can be calculated as [13]
(6.671.14)
ig 2 2 g ig
g ( ) = fG ( g )e dg = ------------ k 0 ------------- e dg
2 2
(28)
a
2 1
= ------- ------------------------------------------ .
2
( 2 2 )2 + 2
PREAMBLE-BASED SYNCHRONIZATION 21
The PDF of the sum of products, i.e. the real part of , is given by the inverse trans-
form of the product between the L characteristic functions,
L 2
1 2 2 2 2
L
i x
f Re { } ( x ) = ------ -------
2 2 ------2- +
e d . (29)
For L=2, we get a Laplacian random variable with PDF [13] (3.389.5)
1 2x
f Re { } ( x ) = -------------- exp ------------- . (30)
2
2 2
2 2
F X ( x ) = 1 1 + ------- x exp ------- x . (31)
2
2
The CDF of [ k, a ] when L=2 is upper bounded by the CDF of the maximum, Y ,
of two independent exponential variables,
2 2
F Y ( y ) = 1 exp ------- y . (32)
2
6 Acknowledgment
The authors would like to thank Prof. Mike Faulkner, Victoria University of Technol-
ogy, Australia, for helpful discussions and suggestions.
References
Fredrik Tufvesson, Mike Faulkner, Peter Hoeher and Ove Edfors, "OFDM Time and
Frequency Synchronization by Spread Spectrum Pilot Technique", 8th IEEE Com-
munication Theory Mini Conference in conjunction to ICC99, Vancouver, Canada,
June 1999, pp. 115-119.
IEEE 1999, reprinted with permission.
OFDM Time and Frequency Synchronization
by Spread Spectrum Pilot Technique
Fredrik Tufvesson1, Mike Faulkner2, Peter Hoeher3 and Ove Edfors1
1
Department of Applied Electronics, Lund University, Box 118, SE-
221 00 Lund, Sweden, e-mail: Fredrik Tufvesson@tde.lth.se
2School of Communications and Informatics, Victoria University of
Technology, Melbourne, Australia
3Information and Coding Theory Lab, University of Kiel, Kiel,
Germany
Abstract
Correct time and frequency synchronization is important for the perfor-
mance of orthogonal frequency division multiplex (OFDM) systems. We evalu-
ate a system where a pseudo noise sequence (PN-sequence) is used for
estimation of these synchronization parameters. The PN-sequence is superim-
posed on the OFDM signal. The system is evaluated by means of the variance of
the frequency estimation error and the probability for correct timing synchroni-
zation. Both theoretical and simulated results are presented. The proposed tech-
nique is very flexible and works also at low signal to noise ratios. The frequency
offset range has been significantly increased compared to conventional OFDM
synchronizer schemes.
1 Introduction
Time and frequency synchronization are two important issues for the performance of
orthogonal frequency division multiplex (OFDM) systems. Correct frequency syn-
chronization is crucial in order to preserve orthogonality of the sub-channels. Timing
synchronization is necessary in order to locate the cyclic prefix and identify the start
of new packets or frames. Often the synchronization process is divided into two parts,
acquisition and tracking. Burst transmission acquisition includes a continuous search
for new packets, finding the beginning of the symbols as well as the packets and
making a coarse frequency estimate so that demodulation can be performed. For con-
tinuous transmission, acquisition means finding the start of symbols and frames and
making a coarse frequency estimate. In tracking mode the synchronization parame-
ters are continuously updated to keep the performance of the demodulator as good as
possible under changing conditions.
1
2 SYNCHRONIZATION USING SUPERIMPOSED PILOTS
This paper presents a new method for acquisition and tracking of OFDM synchro-
nization parameters, which is based on a continuous transmission of a pseudo noise
(PN) sequence. Similar techniques has previously been proposed for synchronization
of single carrier systems [6] and for frame synchronization in OFDM systems [7]. In
this paper we propose and investigate a system for frequency and timing synchroni-
zation, including both symbol and frame synchronization. The system can cope with
very large offsets, up to half the OFDM bandwidth, and gives the frequency estimate
in one step. The synchronization signal has a sharp peak and is independent of the
OFDM parameters used. It can be used for any length of the cyclic prefix and does
not increase the transmitted bandwidth.
1 c [k ]
power
PN
t
data
acq. tracking
3 System Description
r [ k ] = s [ k ] + n[ k ] (1)
2 c k
j --------------- + c
M
= ( s c [ k ] + 1 s d [ k ] ) e + n[ k ] .
where c[k] is the transmitted PN-sequence, d[k] is the OFDM data sequence and n[k]
is white Gaussian noise with spectral density N0. Further, c is the carrier phase and
c = f T s denotes the frequency offset to be estimated when normalized by the
sub-channel spacing. Both the PN-sequence and the data sequence has unity power
and the amount of power used for the PN-sequence is controlled by the code power
2 2
ratio . The received power is given by s E s T s = E [ s [ k ] ] , whereas the
2 2
noise power is given by n ( N 0 T s ) = E [ n [ k ] ] . The received signal is
despread by the local PN-sequence to get
j ( 2 c k M + c )
c [ k a ] r [ k ] = s c [ k a ] c [ k ] e (2)
j ( 2 c k M + c )
+ 1 s d [ k ] c [ k a ] e + n[k ].
For time synchronization we want to find the integer slip, a, between the two codes.
For the purpose of time and frequency estimation, the two latter terms in (2) are act-
ing as disturbances. For data demodulation, no code multiplication is performed in
the receiver and the first and the last terms cause disturbances. The resulting SNR
becomes 2s ( ( 1 ) 2s + 2n ) for the code sequence and ( 1 )2s ( 2s + 2n ) for the
data sequence. In acquisition mode, only the code sequence is sent and the code
power ratio equals one. In tracking mode a low code power ratio is then used in
order to minimize the impact on data transmission.
4 SYNCHRONIZATION USING SUPERIMPOSED PILOTS
The despread signals are summed in groups of K samples to get a so called sub-
correlation. The synchronization signal, [ k, a ] , for a given slip, a, is then achieved
as the sum of L products between sub-correlations spaced KP samples apart (P is the
delay in number of correlator lengths),
L1 K 1
[ k, a ] = c [ k n lK a ] r [ k n lK ]
(3)
l = 0 n = 0
K 1
c [ k n ( l + P ) K a ] r [ k n ( l + P ) K ] .
n = 0
Time synchronization is given by the absolute value of this signal whereas the fre-
quency offset estimate is given by the phase. The synchronizer structure is given in
Figure 3. This scheme can be seen as a modification of the schemes in [4] and [5].
Here, the length of the sub-correlators is K instead of 1 as in [5], and the differential
detectors work on the sub-correlator outputs instead of the correlator outputs as in
[4].
In this section we analyze the distribution of the timing metric in an AWGN chan-
nel. Firstly we analyze the case when the timing is correct and later when it is wrong.
The calculations are quite tedious and therefore we just summarize the results and
verify them by means of simulations. Often it is desirable to have a signal which is
2 2
independent of received power. By normalizing as [ k, a ] = [ k, a ] ( LK s )
we get a timing metric with a mean close to one when the timing is correct and just
above zero when it is wrong.
K1
c [ k n a ] r [ k n ]
r[k]
n=0
freq. est.
( )* arg
L1
timing
z-KP l=0 abs
When the timing is correct, a=0, the distribution of the timing metric can be ana-
lyzed in a similar way as in [3]. The frequency offset entails that the chips are not
added totally in phase in the sub-correlator, which leads to an energy reduction. We
assume that no ISI is present and c M . Normalizing the sub-correlator signal by
1
LK s gives the normalized output, [ k, 0 ] , where [4]
2 2
2 c
[ k, 0 ] = ------- sinc ( c K M ) exp j ------------ k + --- + n [ k, 0 ] .
1 1
(4)
L M 2
All differential products then add coherently since they have nearly the same phase.
For medium and high SNR:s, the contribution from the noise vector perpendicular to
the signal vector can be neglected. If we consider the data signal as Gaussian, or if K
is large, [ k, 0 ] can be approximated as a Gaussian variable with mean
2
E [ [ k, 0 ] ] = sinc ( c K M ) (5)
and variance
2 2 2
( ( 1 ) s + n ) sinc ( c K M )
var [ [ k, 0 ] ] = ----------------------------------------------------------------------------- . (6)
2
LK s
As seen in (6) the variance is inversely proportional to LK. The particular choice of
correlator length has small impact on this product if the observation interval is kept
constant. The probability of missing the sync signal is therefore mainly determined
by the observation interval and the equivalent SNR for the PN-sequence.
When the timing is wrong, a 0 , the sub-correlators do not peak. The sub-corre-
lator outputs, [ k, a ] , can be regarded as Gaussian variables with zero mean. The
timing metric is therefore the absolute value of a sum of products between zero mean
Gaussian variables,
L1
[ k, a ] a0
= a [ k lK, a ] a [ k ( l + P ) K, a ] . (7)
l=0
1 A typo in the original publication has been corrected, the square root was missing.
6 SYNCHRONIZATION USING SUPERIMPOSED PILOTS
0.8
0.6
cdf
0.4
K=1
K=16
0.2
K=128
K=256
0
0 0.5 1 1.5
normalized timing metric
Figure 4: CDF for the normalized timing metric when the timing is wrong (solid) and
correct (dashed), SNR=10 dB, =0.1, observation interval 512 samples and
( K, L ) = { ( 1, 256 ), ( 16, 16 ), ( 128, 2 ), ( 256, 1 ) } . Marker values are obtained by
simulations, whereas the lines represent theoretical values.
In appendix A the distribution of the timing metric is analyzed for different choices
of the number of differential signals. In Figure 4 the cumulative distribution function
(CDF) of the timing metric is presented for various correlator lengths, K, and number
of differential signals, L, when the observation interval is 512 samples, SNR=10dB
and =0.1.
Depending on how the detection threshold is set we will get a specific probability
of missing the sync signal and a probability of false alarm. In Figure 5 we plot the
receiver operation characteristics (ROC) for different parameter choices. It is of
course desirable to be in the lower left corner where both probabilities are low. Here,
again, we see the importance of using long sub-correlator lengths. By correlating
samples before differential detection we are not so sensitive to single noise samples.
In conventional OFDM synchronization no correlation is performed before differen-
tial detection The proposed technique will therefore give better performance, also
when only used as a preamble in acquisition mode.
The frequency estimator structure is the same both in tracking and acquisition
SYNCHRONIZATION USING SUPERIMPOSED PILOTS 7
0
10
missed synch probability
2
10
4
10
K=1
K=16
6
10 K=128
K=256
8 6 4 2 0
10 10 10 10 10
false detection probability
mode, but a two step procedure can be used in order to enhance the performance. In
acquisition mode the sub-correlator length, K, and the number of delay elements, P,
are chosen according to the maximum expected frequency offset. In tracking mode,
initial compensation has already been made and therefore larger K and P can be used
to decrease the variance of the frequency estimation error.
The phase increment method used gives a (nearly) unbiased frequency estimate of a
complex exponential in white noise [8] and can cope with frequency offsets which
are upper bounded by1 c < M ( 2PK ) . The normalized frequency estimate when
the timing is known is given by2
M arg ( [ k, 0 ] )
c = -------- ------------------------------ . (8)
KP 2
The frequency offset estimators in [1], [2], [4] and [5] can, from a performance point
of view, be seen as a special case of the proposed estimator in acquisition mode. See
Table 1 for a comparison of parameters, where G denotes the length of the cyclic pre-
fix.
1 A typo in the original paper has been corrected, the factor 2 was missing.
2 A typo in the original paper has been corrected, 2 was missing.
8 SYNCHRONIZATION USING SUPERIMPOSED PILOTS
reference K P L
Nishinaga [5] 1 1 M -1
Fawer [4] M/2 1 1
PN-based M
K < ---------------
M
P < ---------------
M
----- P
2P c 2K c K
2
M
var ( c ) = ----------------------- (9)
2 3
4 K LP
2 2 2 2 2
1 ( 1 ) s + n 1 ( 1 ) s + n
2 2
- + ----------- ---------------------------------------------
--- --------------------------------------------- - .
L
s sinc ( c K M ) 2KP 2s sinc2 ( c K M )
Strictly speaking this is only valid for zero frequency offset, but simulations indicate
[8] that it is a good approximation for other frequency offsets as well. An interesting
property to note is that the same minimum variance can be achieved for different sub-
correlator lengths as long as the number of delay elements are chosen properly.
des=1
1
10 des=4
=10
estimation error variance
des
2
10
3
10
4
10
5
10 3 2 1 0 1
10 10 10 10 10
frequency offset
6 Conclusions
In this section we analyze the distribution of the timing metric when the timing is
not correct. We analyze the case with one or two differential signals, L, or when the
number of differential signals is large. The normalized complex outputs of the sub-
2 2 2 2
correlators have zero mean and variance = ( ( 1 ) s + n ) ( KL s ) if an
a
10 SYNCHRONIZATION USING SUPERIMPOSED PILOTS
[ k, a ] = a [ k lK, a ] a [ k ( l + P ) K, a ] , (10)
where the two terms are Rayleigh distributed. By the transformation theorem we can
derive the joint PDF for [ k, a ] and a [ k lK, a ] and then integrate over
a [ k lK, a ] to get the probability density function (PDF) for the timing metric.
The PDF for incorrect timing becomes
2 2 d
f [ k, a ] ( d ) = -------- dk 0 2 -------- , (11)
2 2
a a
where k0(x) denotes the zeroth-order modified Bessel function. The CDF of the tim-
ing metric is consequently given by
P { [ k, a ] } = F D ( ) = 1 2 -------- k 1 2 -------- , (12)
2 2
a a
For L=2 we derive bounds for the timing metric. If L is even the metric can be
rewritten as
L1
[ k, a ] = a [ k lK, a ] a [ k ( l + P ) K, a ] (13)
l=0
L21
= d 2k + 1 c 2k + 1 + b 2k + 1 e 2k + 1 .
k=0
All the terms (b, c, d and e) in (13) are independent zero mean Gaussian variables, d
2 2 2
and b are real valued and have variance b = d = E [ bb ] = 2 , whereas c
a
and e are complex valued with variance of the both complex parts
2 2 2
Re { c } = Im { c } = . Defining new variables as g k = d 2k + 1 Re { c 2k + 1 } and
a
h k = d 2k + 1 , the joint probability function for g and h becomes
g 2 1 2 1
f G, H ( g, h ) = ------------------------------ exp ----------------- --- ------ h --- .
1 1
(14)
2 Re { c } d 2 h 2
Re { c } h
d
SYNCHRONIZATION USING SUPERIMPOSED PILOTS 11
Integrating [10, integral 3.478.4] over h gives the PDF for the products, i.e.
= --------------------------- k 0 -----------------------
1 g
fG ( g ) = fG, H ( g, h ) dh Re { c } d Re { c } d
(15)
2 2 g
= ------------ k 0 ------------- .
2 2
a a
Naturally this PDF also applies for b 2k + 1 e 2k + 1 and for the corresponding imagi-
nary parts. The characteristic function of the product, g, can be calculated as [10,
integral 6.671.14]
ig 2 2 g ig
g ( ) = fG ( g ) e dg = ------------ k 0 ------------- e dg
2 2
(16)
a a
2 1
= -------- -------------------------------------------- .
2
( 2 2 )2 + 2
a a
The PDF of the sum of products, i.e. the real part of , is given by the inverse trans-
form of the product between the L characteristic functions,
L 2
1 2 2 2 2
L
i x
f Re { } ( x ) = ------ --------
2 2 2
-------
- +
e d . (17)
a
a
For L=2, we get [10, integral 3.389.5] a Laplacian random variable with PDF
1 2 x
f Re { } ( x ) = ---------------- exp ------------- . (18)
2
2 2
a a
2 2
F X ( x ) = 1 1 + -------- x exp -------- x . (19)
2
2
a a
2 2
F Y ( y ) = 1 exp -------- y . (20)
2
a
2 2
Re { } = L ( ) 2 .
2
(21)
a
2
P { [ k, a ] } = F ( ) 1 exp -------------------- . (22)
2
2 Re { }
References
[1] J.J. van de Beek, M. Sandell and P. O. Brjesson, ML estimation of time and
frequency offset in OFDM systems, IEEE Trans. on Signal Processing, vol.
45, no. 7, July 1997, pp. 1800-1805.
[2] P. Moose, A technique for orthogonal frequency division multiplexing fre-
quency offset correction, IEEE Trans. on Communications, vol.42, no. 10,
Oct. 1994, pp. 2908-14.
[3] T.M. Schmidl and D. C. Cox, Robust frequency and timing synchronization
for OFDM, IEEE Trans. on Communications, vol. 45, no. 12, Dec. 1997, pp.
1616-1621.
[4] U. Fawer, A coherent spread-spectrum diversity receiver with AFC for multi-
path fading channels, IEEE Trans. on Communications, vol. 42, no. 2/3/4,
Feb.-April 1994, pp. 1300-1311.
[5] N. Nishinaga, M. Nakagami and Y. Iwadare, A simple synchronization acqui-
sition method for DS/SS system under carrier frequency offset, IEICE Trans.
on Fundamentals, vol. E80-A, no. 11, Nov. 1997, pp. 2162-2171.
[6] T.P. Holden and K. Feher, A spread spectrum based system technique for syn-
chronization of digital mobile communication systems, IEEE Trans. on Broad-
casting, vol. 36, no. 3, Sept. 1990, pp. 185-194.
[7] A. Steingass, A.J. van Wijngaarden and W.G. Teich, Frame synchronization
using superimposed sequences, in Proc. IEEE International Symposium on
SYNCHRONIZATION USING SUPERIMPOSED PILOTS 13
Fredrik Tufvesson and Peter Hoeher, "Channel Estimation using Superimposed Pilot
Sequences", submitted to IEEE Transactions on Communications, March 2000.
Fkdqqho Hvwlpdwlrq xvlqj Vxshulpsrvhg Slorw
Vhtxhqfhv
Iuhguln Wxiyhvvrqh dqg Shwhu Krhkhun
h
Ghsduwphqw ri Dssolhg Hohfwurqlfv/ Oxqg Xqlyhuvlw|
Er{ 44;/ VH0554 33 Oxqg/ Vzhghq
Id{= .79079 45 << 7;/ H0pdlo= Iuhguln1WxiyhvvrqCwgh1owk1vh
n Lqirupdwlrq dqg Frglqj Wkhru| Ode/ Xqlyhuvlw| ri Nlho
Ndlvhuvwu1 5/ G057476 Nlho/ Jhupdq|/
Id{= .7<07640::8:50736/ H0pdlo= skCwhfkidn1xql0nlho1gh1
Devwudfw
Iru wkh sxusrvh ri ydulrxv v|qfkurql}dwlrq wdvnv +lqfoxglqj fduulhu
skdvh/ wlph/ iuhtxhqf|/ dqg iudph v|qfkurql}dwlrq,/ rqh pd| dgg d nqrzq
slorw vhtxhqfh/ w|slfdoo| d svhxgr0qrlvh vhtxhqfh/ wr wkh xqnqrzq gdwd
vhtxhqfh1 Wklv dssurdfk lv nqrzq dv d vsuhdg0vshfwuxp slorw whfkqltxh
ru dv d vxshulpsrvhg slorw vhtxhqfh whfkqltxh1 Lq wklv sdshu/ zh dsso|
wkh vxshulpsrvhg slorw vhtxhqfh whfkqltxh iru wkh sxusrvh ri fkdqqho hv0
wlpdwlrq +FH,1 Zh sursrvh d wuxo| frkhuhqw uhfhlyhu edvhg rq wkh Ylwhuel
dojrulwkp/ dqg ghulyh lwv sdluzlvh huuru suredelolw|1 Wklv dqdo|wlfdo uh0
vxow jlyhv ixuwkhu lqvljkw dqg ohdgv wr dq dssur{lpdwlrq ri wkh elw huuru
udwh/ zklfk lv frquphg e| Prqwh Fduor vlpxodwlrqv1 Ixuwkhupruh/ zh
suhvhqw d jhqhulf orz0frvw uhfhlyhu vwuxfwxuh edvhg rq uhgxfhg0vwdwh vh0
txhqfh hvwlpdwlrq1 Dprqj wkh glvwlqfw dgydqwdjhv frpsduhg wr frqyhq0
wlrqdo slorw0v|pero0dvvlvwhg FH duh +l, qr edqgzlgwk h{sdqvlrq dqg +ll,
d vljqlfdqwo| lpsuryhg shuirupdqfh lq idvw idglqj hqylurqphqwv1 Wkh
sursrvhg Ylwhuel uhfhlyhu pd| dovr eh xvhg dv dq dowhuqdwlyh uhfhlyhu iru
slorw0v|pero0dvvlvwhg FH/ pdnlqj wklv vfkhph pruh urexvw lq idvw0idglqj
fkdqqhov1
W Sduwv ri wklv zrun zhuh suhvhqwhg dw wkh Lqw1 Zrunvkrs rq Pxowl0Fduulhu Vsuhdg0
Vshfwuxp/ Rehusidhqkrihq/ Jhupdq|/ Vhsw1 4<<< dqg LHHH JOREHFRP *<</ Ulr gh
Mdqhlur/ Eud}lo/ Ghf1 4<<<1
4
5 FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV
4 Lqwurgxfwlrq
Frqvlghu gljlwdo gdwd wudqvplvvlrq ryhu wlph0vhohfwlyh dw0idglqj fkdqqhov1 Dv0
vxplqj frkhuhqw ghprgxodwlrq/ rqh ri wkh pdlq sureohpv lv fduulhu skdvh v|q0
fkurql}dwlrq/ erwk lq whupv ri dftxlvlwlrq dqg wudfnlqj/ sduwlfxoduo| zkhq wkh
fkdqqho lv idvw dqg zkhq d olqh0ri0vljkw frpsrqhqw lv devhqw ^4`^5`1
D srsxodu whfkqltxh wr pdlqwdlq frkhuhqw ghprgxodwlrq iru d zlgh fodvv
ri gljlwdo prgxodwlrq vfkhphv kdv ehhq sursrvhg e| Prkhu dqg Orgjh ^6` ^7`/
dqg lv nqrzq dv slorw0v|pero0dvvlvwhg FH 4 1 Wkh pdlq lghd ri slorw0v|pero0
dvvlvwhg FH lv wr pxowlsoh{ nqrzq slorw v|perov +dovr fdoohg wudlqlqj v|perov,
lqwr dq xqnqrzq gdwd vwuhdp1 Frqyhqwlrqdoo|/ wkh uhfhlyhu uvwo| rewdlqv whq0
wdwlyh fkdqqho hvwlpdwhv dw wkh srvlwlrqv ri wkh slorw v|perov e| phdqv ri
uh0prgxodwlrq/ dqg wkhq frpsxwhv qdo hvwlpdwhv e| phdqv ri lqwhusrodwlrq1
Djkdprkdppdgl hw do1 dqg Fdyhuv zhuh dprqj wkh uvw dqdo|}lqj dqg rswl0
pl}lqj slorw0v|pero0dvvlvwhg FH jlyhq glhuhqw lqwhusrodwlrq owhuv ^8` ^9`1 Gxh
wr wkh slorw v|perov wkh edqgzlgwk voljkwo| lqfuhdvhv1
Lq wklv sdshu/ zh h{soruh d uhodwhg whfkqltxh sursrvhg d gr}hq |hduv djr e|
Pdnudnlv/ Ihkhu dqg Kroghq ^:`^;`1 Wkh whfkqltxh zdv ruljlqdoo| fdoohg vsuhdg0
vshfwuxp slorw whfkqltxh dqg wkh lghd lv wr olqhduo| dgg d nqrzq slorw vhtxhqfh
wr wkh xqnqrzq gdwd vhtxhqfh/ vhh Vhfwlrq 51 Vlplodu phwkrgv duh dssolhg lq
wkh hog ri dxwrpdwlf frqwuro/ zkhuh lqwhqwlrqdo shuwxedwlrq vljqdov duh xvhg/
h1j1/ iru gxdo frqwuro/ forvhg orrs lghqwlfdwlrq dqg h{wuhpxp frqwuro ri g|0
qdplfdo v|vwhpv ^<`1 Iru frppxqlfdwlrq v|vwhpv Pdnudnlv/ Ihkhu dqg Kroghq
dssolhg wkh whfkqltxh wr skdvh v|qfkurql}dwlrq ^:`^;`1 Odwhu/ wkh whfkqltxh kdv
ehhq dssolhg iru wkh sxusrvhv ri iudph v|qfkurql}dwlrq ^43` dqg mrlqw wlph dqg
iuhtxhqf| v|qfkurql}dwlrq ri RIGP vljqdov ^44`/ uhvshfwlyho|1 D nh| ihdwxuh ri
wkh vfkhph lv wkdw wkh slorw lqirupdwlrq dqg wkh gdwd duh wudqvplwwhg lq wkh
vdph glphqvlrq rq wkh vdph fkdqqho1 Dowhuqdwlyh whfkqltxhv pd| eh edvhg rq
wudqvplwwlqj erwk vhtxhqfhv rq vhsdudwh fkdqqhov1 Rqh fdq/ h1j1/ xvh wzr ru0
wkrjrqdo owhuv dv vxjjhvwhg e| Ehmmdql dqg Ehoruh ^45`/ ru xvh dq ruwkrjrqdo
vljqdo glphqvlrq +h1j1 wkh txdgudwxuh frpsrqhqw,/ dv grqh lq wkh XPWV2Z0
FGPD xsolqn ^46`1 Lq erwk fdvhv/ krzhyhu/ wkh edqgzlgwk h!flhqf| zrxog eh
uhgxfhg1
Zlwklq wklv sdshu/ zh dvvxph v|qfkurqrxv v|pero0e|0v|pero vxshusrvlwlrq
ri wkh slorw v|perov/ zkhuh wkh srzhu ri wkh slorw v|perov lv w|slfdoo| pxfk
ohvv wkdq wkh srzhu ri wkh gdwd v|perov1 E| frqvwuxfwlrq/ wkhuh lv qr lqfuhdvh
lq edqgzlgwk zkhq xvlqj d svhxgr0qrlvh slorw vhtxhqfh1 Wkhuhiruh/ zh xvh
wkh qrwlrq ri d vxshulpsrvhg slorw vhtxhqfh whfkqltxh dv vxjjhvwhg lq ^43`/
ehfdxvh vsuhdg0vshfwuxp whfkqltxhv pd| vxjjhvw d zlghqlqj ri wkh vshfwuxp1
Dv rssrvhg wr slorw0v|pero0dvvlvwhg FH/ qr lqwhusrodwlrq lv qhfhvvdu|1 Wkh
vxshulpsrvhg slorw vhtxhqfh whfkqltxh lv wkhuhiruh pruh edqgzlgwk dqg srzhu
4 Fkdqqho hvwlpdwlrq pd| eh vhhq dv d jhqhudol}dwlrq ri fduulhu skdvh v|qfkurql}dwlrq/
vlqfh hvwlpdwhv ri wkh txdgudwxuh frpsrqhqwv fryhu wkh fduulhu skdvh dv zhoo dv uholdelolw|
lqirupdwlrq/ riwhq fdoohg fkdqqho vwdwh lqirupdwlrq1
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 6
uk yk ^
u
Viterbi k-D
detector
fk nk
Khqfh/ wkh dyhudjh hqhuj| ri wkh frpsrvlwh v|perov lv Hv / li wkh vhtxhqfhv duh
pxwxdoo| lqghshqghqw= H^xn sn ` @ 31 Wkh dyhudjh hqhuj| shu lqirupdwlrq elw
lv He @ Hv @ orj5 P 1
Wkh qrupdol}hg srzhu ri wkh slorw v|perov/ / fruuhvsrqgv wr wkh dprxqw
ri nqrzq lqirupdwlrq wudqvplwwhg/ dqg vkrxog eh rswlpl}hg1 Lq slorw0v|pero0
dvvlvwhg FH v|vwhpv wkh uhodwhg sdudphwhu lv wkh vsdflqj ri wkh slorw v|perov1
Lq wkh uhpdlqlqj/ lv dvvxphg wr eh frqvwdqw iru doo n/ vlqfh rxu irfxv lv rq
wudfnlqj1 Lq rughu wr lpsuryh wkh dftxlvlwlrq ehkdylru/ pd| eh odujhu iru wkh
uvw v|perov zlwklq d gdwd iudph ^44`1
Qrwh wkdw lq slorw0v|pero0dvvlvwhg FH v|vwhpv wkh gdwd dqg slorw v|perov
duh ruwkrjrqdo e| phdqv ri wlph0glylvlrq pxowlsoh{lqj/ zkhuhdv iru vxshulp0
srvhg slorw vhtxhqfhv wkh furvv0fruuhodwlrq ehwzhhq wkh gdwd dqg slorw v|perov
w|slfdoo| lv qrq0}hur1 Krzhyhu/ wkh vxshulpsrvhg slorw vhtxhqfh lv nqrzq dqg
wkhuhiruh wkh glvwxuedqfh fdq eh pdgh vpdoo1 Wkh orvv gxh wr wkh slorw vh0
txhqfh lv ryhu0frpshqvdwhg e| wkh odfn ri lqwhusrodwlrq/ dv zh zloo vkrz lq
Vhfwlrq 81 Wkh wudqvplvvlrq v|vwhp pd| eh lqwhusuhwhg dv d wzr0xvhu v|v0
whp/ zkhuh rqh xvhu wudqvplwv lqirupdwlrq xqnqrzq wr wkh uhfhlyhu/ dqg wkh
rwkhu xvhu wudqvplwv nqrzq lqirupdwlrq1 Fruuhvsrqglqjo|/ wkh wudqvplvvlrq
whfkqltxh lv uhodwhg wr pxowl0xvhu v|vwhpv/ dqg wkh v|qfkurql}dwlrq2ghwhfwlrq
sureohp lv uhodwhg wr pxowl0xvhu ghwhfwlrq1
Jlyhq rxu dvvxpswlrqv/ wkh pdwfkhg owhu rxwsxw vdpsohv/ |n 5 F/ fdq eh
zulwwhq dv
jlyhq d SQ slorw vhtxhqfh/ zh pd| h{whqg rxu uhfhlyhu wr frsh zlwk iuhtxhqf|0
vhohfwlyh idglqj dv zhoo1 Ilqdoo|/ lw lv lqwhuhvwlqj wr qrwh wkdw wkh frqfhsw ri
vxshulpsrvhg slorw vhtxhqfhv lv dssolfdeoh wr prvw h{lvwlqj dqdorj dqg gljlwdo
wudqvplvvlrq vfkhphv dv dq dgg0rq ihdwxuh
6 Uhfhlyhu Vwuxfwxuh
Wkh frqyhqwlrqdo uhfhlyhu iru wkh vxshulpsrvhg slorw vhtxhqfh whfkqltxh vsolwv
wkh uhfhlyhg vljqdo/ dqg ihhgv lw lqwr d fduulhu uhfryhu| xqlw dqg d ghwhfwlrq
xqlw ^:`^;`/ uhvshfwlyho|1 Wkh rxwsxw vljqdo ri wkh fduulhu uhfryhu| xqlw lv ihg
lqwr wkh ghwhfwlrq xqlw1 Lq rughu wr sduwo| frpshqvdwh wkh lqwhuihuhqfh gxh wr
wkh slorw vhtxhqfh/ d vlpsoh fdqfhoodwlrq vfkhph kdv ehhq sursrvhg1
Wkh frqyhqwlrqdo uhfhlyhu iru wkh slorw0v|pero0dvvlvwhg FH whfkqltxh rs0
hudwhv vlploduo|= wkh uhfhlyhg vljqdo lv ghpxowlsoh{hg/ dqg ihg lqwr d fkdqqho
hvwlpdwlrq xqlw +l1h1/ dq lqwhusrodwru, dqg d ghwhfwlrq xqlw ^6`^7`^8`^9`/ uhvshf0
wlyho|1 Lqwhuihuhqfh fdqfhoodwlrq lv qrw qhfhvvdu| e| ghvljq1
Dv rssrvhg wr wkhvh uhfhlyhuv/ zh zloo qrz lqwurgxfh d uhfxuvlyh uhfhlyhu
edvhg rq wkh Ylwhuel dojrulwkp/ zklfk gluhfwo| rxwsxwv hvwlpdwhv ri wkh gdwd
v|perov1 Wklv uhfhlyhu lv sulpdulo| ghvljqhg iru wkh vxshulpsrvhg slorw vh0
txhqfh whfkqltxh/ exw lv dovr dssolfdeoh iru wkh frqyhqwlrqdo slorw0v|pero0
dvvlvwhg FH whfkqltxh1 Glhuhqw fkdqqho hvwlpdwhv duh frpsxwhg iru glhu0
hqw k|srwkhvlv e| phdqv ri wkh sulqflsohv ri shu0vxuylyru surfhvvlqj ^48`^49`1
Lqwhuihuhqfh fdqfhoodwlrq lv grqh lqkhuhqwo|1 Wkh sulqflsoh ri wkh uhfhlyhu lv
wkh iroorzlqj= Wkh uhfhlyhu vwduwv zlwk d k|srwkhvlv ri wkh wudqvplwwhg gdwd
vhtxhqfh1 Dvvxplqj wkh k|srwkhvlv lv fruuhfw/ wkh uhfhlyhg v|perov duh wkhq
xvhg wrjhwkhu zlwk wkh k|srwkhvlv wr fdofxodwh fkdqqho hvwlpdwhv iru hdfk ri
wkh v|perov lq wkh eorfn1 Wkhq wkh glvwdqfh +phwulf, ehwzhhq wkh uhfhlyhg
vhtxhqfh dqg wkh k|srwkhvlv dhfwhg e| wkh fruuhvsrqglqj fkdqqho hvwlpdwhv lv
fdofxodwhg1 Wkh vdph surfhgxuh lv wkhq xvhg iru wkh qh{w k|srwkhvlv1 Ilqdoo|/
wkh uhfhlyhu pdnhv d ghflvlrq lq idyru ri wkh vhtxhqfh zlwk wkh orzhvw glvwdqfh1
Iru wkh fdofxodwlrqv lw lv frqyhqlhqw wr xvh pdwul{ qrwdwlrq1 Zh ghqrwh wkh
uhfhlyhg vdpsoh yhfwru e| | @ +|4 > |5 > ===> |Q ,W @ Fi . q/ zkhuh i dqg q duh
yhfwruv ri wkh idglqj surfhvv dqg wkh qrlvh/ uhvshfwlyho|/ dqg F lv d gldjrqdo
pdwul{ zlwk wkh wudqvplwwhg v|perov rq lwv gldjrqdo= fn>n @ +xn . sn ,1
Wkh fkdqqho hvwlpdwhv duh fdofxodwhg xvlqj wkh O odwhvw uhfhlyhg v|perov1
Dffruglqj wr +4,/ d whqwdwlyh fkdqqho hvwlpdwh/ iano / fdq eh rewdlqhg e| uh0
prgxodwlrq=
|n
ian @ > +5,
x
n . sn
zkhuh o lv dq duelwudu| lqwhjhu qxpehu1 Qrwh wkdw wkh uvw sduw ri wkh ghqrplqd0
wru lv edvhg rq wkh k|srwkhvlv/ zkhuhdv wkh vhfrqg sduw lv nqrzq= wkh slorw v|p0
erov khosv wr uhvroyh wkh skdvh dpeljxlw|1 Jlyhq k|srwkhvhv x nO.4 /
n > = = = > x
9 FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV
zkhuh o @ H^in iano ` @ uo duh wkh hohphqwv ri wkh furvv0fruuhodwlrq yhfwru dv0
vxplqj d fruuhfw k|srwkhvlv1 Ixuwkhu/ o>m @ H^iano ianm ` @ 5q @ m
xno . sno m5
o>m . umo duh wkh hohphqwv ri wkh dxwrfruuhodwlrq pdwul{ / zkhuh o> m @
4> = = = > O/ uo @ uo lv wkh owk dxwrfruuhodwlrq frh!flhqw ri wkh fkdqqho/ dqg
5q @ Q3 @Hv lv wkh yduldqfh ri wkh qrlvh1 Wkh fruuhvsrqglqj yhfwru ri fkdqqho
hvwlpdwhv iru d fhuwdlq k|srwkhvlv fdq frqvhtxhqwo| eh zulwwhq dv
4 |=
i @ DF +8,
Iru h{dpsoh/ iru d eorfn ohqjwk ri 8 uhfhlyhg v|perov dqg O @ 6> wkh fkdqqho
hvwlpdwlrq pdwul{ kdv wkh irup
3 4
d4>3 3 3 3 3
E d5>4 d5>3 3 3 3 F
E F
D@E E d6>5 d6>4 d6>3 3 3 F>
F +9,
C 3 d7>5 d7>4 d7>3 3 D
3 3 d8>5 d8>4 d8>3
zkhuh dl>m duh wkh owhu frh!flhqwv1 Qrwh wkdw wkh rswlpdo owhu frh!flhqwv duh
ghshqghqw rq wkh k|srwkhvlv1 Iru orz ydoxhv wkh glhuhqfh ehfrphv vpdoo/
krzhyhu/ dqg d vwdwlf owhu jlyhv d jrrg dssur{lpdwlrq1
Wkh jrdo lv wr uhfxuvlyho| frpsxwh wkh pd{lpxp0olnholkrrg hvwlpdwh ri wkh
wudqvplwwhg vhtxhqfh1 Wkh uhfhlyhu wkhq whvwv doo srvvleoh k|srwkhvhv dqg gh0
flghv iru wkh vhtxhqfh uhvxowlqj lq wkh orzhvw phwulf/ l1h1 wkh kljkhvw suredelolw|1
Lw lv/ krzhyhu/ kdug wr frpsxwh wkh wuxh pd{lpxp0olnholkrrg hvwlpdwh uhfxu0
vlyho|/ dqg wkhuhiruh zh kdyh wr pdnh dssur{lpdwlrqv wr dfklhyh dq dgglwlyh
phwulf vxlwdeoh iru uhfxuvlyh uhfhlyhuv1 Zh ghulyh wkh phwulf lq Dsshqgl{ D1
Wkh rswlpdo phwulf lv jlyhq e|
K
. |K +FU
rsw @ oq+ghw+!,,
K 5 4
i F . q L, |> +:,
@H^+| |
zkhuh ! ghqrwhv wkh fryduldqfh pdwul{ ri wkh uhfhlyhg vdpsohv/ ! ,
K
, `1 Dvvxplqj wkh k|srwkhvlv lv fruuhfw wklv fryduldqfh pdwul{ fdq eh
+| |
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV :
Iljxuh 5= Wkh fkdqqho hvwlpdwhv iru wkh glhuhqw k|srwkhvhv ri wkh wzr uvw
v|perov1
fdofxodwhg dv
4 4
F
! @ H^+L FD ,||K +L FD
F ,K `
F
@ +L FD 4 ,+FU
K K 5 4 K
i F . q L,+L FDF , +;,
zkhuh Ui lv wkh fryduldqfh pdwul{ ri wkh fkdqqho frqwdlqlqj wkh dxwrfruuhodwlrq
frh!flhqwv1
Zlwk wkh dgglwlyh phwulf lqfuhphqwv lw lv srvvleoh wr xvh wkh Ylwhuel dojr0
ulwkp iru ghwhfwlrq1 Lq Dsshqgl{ D lw lv vkrzq wkdw d vxlwdeoh dssur{lpdwlrq
wkdw |lhogv dgglwlyh phwulf lv jlyhq e|
@ +| FD
F 4 |,K +| FD
F 4 |,= +<,
Wkh fruuhvsrqglqj dgglwlyh phwulf lqfuhphqw xvhg lq wkh Ylwhuel uhfhlyhu lv
wkhq jlyhq e|
fn > fn4 > = = = > fnO , @ m|n fn in m5 >
n + +43,
zkhuh in lv wkh fkdqqho hvwlpdwh edvhg rq wkh fxuuhqw k|srwkhvlv1 Lq Ilj1 5 wkh
fkdqqho hvwlpdwhv iru wkh uvw wzr v|perov lq d irxu0vwdwh wuhoolv lv suhvhqwhg1
Wkh fkdqqho hvwlpdwhv duh edvhg rq wkh k|srwkhvhv dqg rq wkh fruuhodwlrq
ehwzhhq d vshflf fkdqqho hvwlpdwh dqg wkh dfwxdo fkdqqho ydoxh lv kljk rqo|
zkhq wkh k|srwkhvlv lv fruuhfw1 Wkh fruuhodwlrq fdq eh fdofxodwhg dv
% O4 & O4
[ [ xno . sno
H + do iano ,in @ do H^ino in `= +44,
no . sno
x
o@3 o@3
; FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV
Wkh iudfwlrq rq wkh uljkw kdqg vlgh htxdov rqh rqo| iru wkh fruuhfw k|srwkhvlv
wkhuhe| surgxflqj d fruuhfw fkdqqho hvwlpdwh1 Wkh uhfhlyhu fdq eh lqwhusuhwhg
dv dq hvwlpdwru0fruuhodwlqj ghwhfwru/ zkhuh sduwv ri wkh uhfhlyhg vljqdo duh
nqrzq dqg sduwv ri wkh vljqdo duh xqnqrzq1 Lq d uvw vwhs/ wkh xqnqrzq
fkdqqho uhvsrqvh +dqg wkhuhe| wkh uhfhlyhg vljqdo, lv hvwlpdwhg e| wkh uhfhlyhg
v|perov dqg wkh k|srwkhvlv1 Lq wkh qh{w vwhs wklv hvwlpdwh lv fruuhodwhg zlwk
wkh uhfhlyhg v|perov1 Li wkh fruuhodwlrq ehwzhhq wkh fkdqqho hvwlpdwh dqg wkh
dfwxdo fkdqqho ydoxh lv kljk/ wkhq wkh phwulf zloo eh orz dqg wkh k|srwkhvlv
wkdw fruuhvsrqgv wr wkh wudqvplwwhg gdwd zloo eh fkrvhq1
Wkh k|srwkhvhv xn4 > = = = > xnO.4 frqvwlwxwh wkh P O4 vwdwhv ri d wuhoolv/
dqg xn ehorqjv wr wkh dfwxdo eudqfk1 Wkh wuhoolv kdv P eudqfkhv2vwdwh1 Iru wkh
dgglwlyh phwulf wkh idploldu dgg0frpsduh0vhohfw rshudwlrq fdq eh dssolhg dqg
wkh prvw suredeoh vhtxhqfh fdq eh irxqg e| edfn0wudflqj ru uhodwhg rshudwlrqv1
D uxoh ri wkxpe iru wkh ghflvlrq ghod| lv G O = = = 7O1 Gxh wr wkh vhtxhqfh
hvwlpdwlrq/ wkh uhfhlyhu fdq frsh zlwk d orz0srzhu slorw vhtxhqfh1
Iru odujh P dqg O/ wkh frpsoh{lw| fdq eh vljqlfdqwo| uhgxfhg e| dsso|lqj
wkh sulqflsohv ri uhgxfhg0vwdwh vhtxhqfh hvwlpdwlrq ^4:` ^4;`=
N
[ 4 O
[4
in @ do |no @+
xno . sno , . do |no @+a
xno . sno ,> +45,
o@3 o@N
Wr pdnh d ghflvlrq wkh uhfhlyhu fdofxodwhv wkh fxpxodwlyh phwulfv +<, iru
hdfk ri wkh k|srwkhvhv1 Wkh uhfhlyhu pdnhv d fruuhfw ghflvlrq li wkh phwulf
iru wkh fruuhfw sdwk lv vpdoohu wkdq doo ri wkh rwkhu sdwk phwulfv1 Ghqh wkh
a @
dgglwlyh ghflvlrq yduldeoh dv 1 Li a A 3 iru doo k|srwkhvhv wkhq wkh
fruuhfw sdwk zloo eh fkrvhq1 Wkh ghflvlrq yduldeoh fdq eh h{suhvvhg dv
a @ |K ^+L FD
F 4 ,K +L FD
F 4 , +L FDF4 ,K +L FDF4 ,`|=
+46,
Uhzulwh wkh uhfhlyhg vdpsohv dv | @ Gj zlwk
5 6
x4 . s4 3 3 4 3 3
9 3 x 5 . s5 3 3 4 3 :
9 :
G@9 11 1 11 11 11 : +47,
7 1 1 1 1 1 1 8
3 3 xQ . sQ 3 3 4
dqg
i
j@ = +48,
q
Wkh ghflvlrq yduldeoh fdq wkhq eh uhzulwwhq dv
a @ jK GK ^+L FD
F 4 ,K +L FD
F 4 , +L FDF4 ,K +L FDF4 ,`Gj
@ jK Nj=
+49,
a rsw
@ jK GK ^+FU K K 5 4 K K 5 4
i F . q L, +FUi F . q L, `Gj
@ jK N
rsw j> +4;,
lqvwhdg
exw wr pdnh fruuhfw ghflvlrqv wklv kdv wr h{fhhg oq+ghw+!,,oq+ghw+!,,
ri }hur1
glvwulexwlrq ri wkh ghflvlrq yduldeoh dqg fdofxodwh wkh sdluzlvh huuru suredelolw|1
Ghqh qhz Jdxvvldq yduldeohv dv j3 @ Ej> zkhuh E lv d udqn u pdwul{ ri vl}h
5Q u ixooolqj Uj @ EEK 1 Lq ^54` lw lv vkrzq wkdw wkh ghflvlrq yduldeoh
fdq eh uhjdughg dv d vxp ri u lqghshqghqw zhljkwhg "5 0yduldeohv/ zkhuh wkh
zhljkwv duh wkh hljhqydoxhv ri EK NE>
Qv
[
a @ jK Nj
@ q "55>q = +4<,
q@4
Wkh yhfwru ri Jdxvvldq vdpsohv/ j> lv frpsoh{ dqg wkh "5 0yduldeohv wkhuhe|
kdyh wzr ghjuhhv ri iuhhgrp1 Wkh phdq ri wkh ghflvlrq yduldeoh jlyhv d jrrg
lqglfdwlrq ri wkh huuru suredelolw|1 Lw fdq eh fdofxodwhg dv wkh wudfh ^54`
a @ wu+NU
H^` j ,= +53,
p
[ Qv
[
a@
q "55>q q "55>q= +54,
q@4 q@p.4
Dvvxplqj wkdw doo wkh hljhqydoxhv duh glvwlqfw/ zklfk wkh| iru doo sudfwlfdo fdvhv
a fdq eh fdofxodwhg dv ^54`
duh/ wkh suredelolw| ghqvlw| ixqfwlrq ri
; p
A S +q ,Qv 5 h|@+5q ,
A
A %
Q
& |A3
A Tv
? q@4 5
A +q n ,
n@4>n9@q
ia +|, @ Qv +55,
A S
A
A
+%q ,Qv 5 h|@+5q&, | 3=
A
A Qv
= q@p.4 5 T +n q ,
n@4>n9@q
Wkh sdluzlvh huuru suredelolw| iru wkh dgglwlyh ghflvlrq yduldeoh/ l1h1 wkh sure0
delolw| wkdw wkh ghflvlrq yduldeoh +49, ehfrphv qhjdwlyh vr wkdw zh zloo pdnh d
ghflvlrq lq idyru ri wkh huurqhrxv k|srwkhvlv lqvwhdg ri wkh fruuhfw rqh/ lv wkhq
] 3
+q ,Qv 4
Qv
[
Sdgg @ ia +|,g| @ Q = +56,
4 q@p.4
Tv
+n q ,
n@4>n9@q
Wkh fruuhvsrqglqj huuru suredelolw| iru wkh rswlpdo ghflvlrq yduldeoh ghshqgv
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 44
Qrwh wkdw wkh hljhqydoxhv q duh ghshqghqw rq wkh wudqvplwwhg gdwd zklfk
phdqv wkdw wkh huuru suredelolw| ydulhv ghshqglqj rq wkh wudqvplwwhg gdwd
vhtxhqfh1
0
10
PEP, optimal
-2
10
-4
10
-6
10
0
10
20 50
150 100
Eb/N0 30 250 200
hypothesis
Iljxuh 6= Sdluzlvh huuru suredelolwlhv +SHS, iru doo 588 huurqhrxv k|srwkhvhv lq
dq ;0v|pero eorfn xvlqj wkh rswlpdo phwulf dqg wkh rswlpdo owhu frh!flhqwv1
srzhu1 Wkh huuruv zlwk d 4;3 ghjuhh skdvh vkliw lq wkh plggoh ri wkh vhtxhqfh
duh dovr dyrlghg e| lqfuhdvlqj wkh eorfn ohqjwk1 Wkh phwulf iru wkh huurqhrxv
sdwkv lq wkh wuhoolv ehfrphv kljk diwhu d ihz v|perov vr iru orqj eorfnv wkhvh
k|srwkhvhv zloo qrw eh fkrvhq1 Krzhyhu/ dovr iru orqjhu eorfnv wkhuh duh vwloo
frpprq huuruv zlwk d 4;3 ghjuhh skdvh vkliw lq wkh ehjlqqlqj ru qhdu wkh hqg
ri wkh eorfn1 Wklv lv ehfdxvh wkhuh lv qr lqlwldo fkdqqho hvwlpdwh iru wkh uvw
v|perov dqg qr nqrzq hqg0vwdwh iru wkh odvw rqhv1 Li wkh uvw ru odvw v|perov
duh khdylo| glvwxuehg e| qrlvh wkh fkdqqho hvwlpdwhv fdq eh dgmxvwhg wr wkhvh
huurqhrxv v|perov dqg lw wdnhv vrph wlph ehiruh wkh fkdqqho hvwlpdwh ehfrphv
fruuhfw djdlq1 Ilqdoo|/ wkh prvw frpprq jurxs ri huuruv lv zkhuh zh kdyh d
vlqjoh skdvh vkliwhg v|pero lq wkh plggoh ri wkh vhtxhqfh1 Wkhvh huuruv fdq ri
frxuvh eh fruuhfwhg e| xvlqj dq| nlqg ri fkdqqho frglqj diwhu wkh ghwhfwru ru
e| xvlqj mrlqw ghwhfwlrq dqg ghfrglqj1 Wkh huuruv zlwk d vlqjoh skdvh vkliwhg
v|pero phdqv wkdw wkh fkdqqho hvwlpdwh lq wkh qhljkerukrrg ri wkh huuru zloo
eh edvhg rq dq lqfruuhfw k|srwkhvlv1 Exw vlqfh rqo| rqh ri wkh O v|perov xvhg
iru wkh hvwlpdwh lv zurqj/ wkh fkdqqho hvwlpdwlrq huuru dq|krz ehfrphv txlwh
vpdoo1
Iru odujh ru lqqlwh eorfn ohqjwkv wkh huuruv dsshdu dv vlqjoh v|pero huuruv ru
dv d jurxs ri huurqhrxv v|perov1 Wkh odwwhu lv hvshfldoo| wuxh iru orz Grssohu
iuhtxhqflhv zkhuh wkh idglqj glsv lq Ud|ohljk idglqj fkdqqhov kdyh gxudwlrq
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 46
0
10
PEP, additive
-2
10
-4
10
-6
10
0
10
20 50
150 100
Eb/N0 30 250 200
hypothesis
Iljxuh 7= Sdluzlvh huuru suredelolwlhv +SHS, iru doo 588 huurqhrxv k|srwkhvhv
lq dq ;0v|pero eorfn xvlqj wkh dgglwlyh phwulf dqg {hg owhu frh!flhqwv1
ryhu vhyhudo v|perov1 Lq d idglqj gls wkh suredelolw| lv kljk wkdw wkh zurqj
sdwkv lq wkh wuhoolv zloo eh fkrvhq/ zklfk phdqv wkdw wkh fkdqqho hvwlpdwhv duh
ohvv uholdeoh wkhuh1 Krzhyhu/ li dq huurqhrxv sdwk lv fkrvhq wkh phwulf ehfrphv
kljk diwhu d ihz v|perov dqg wkhuhiruh wkh fruuhfw sdwk zloo eh uhdfkhg djdlq
diwhu d fhuwdlq qxpehu ri v|perov1 Wkh uhfhlyhu fdq eh vhhq dv d Pdunry
fkdlq zkhuh wkh wudqvlwlrq suredelolw| dqg wkh suredelolw| ri pdnlqj d fruuhfw
ghflvlrq duh ghshqghqw rq wkh vwdwh/ l1h1 wkh sduwlfxodu v|perov +wkdw pljkw eh
huurqhrxv, xvhg wr irup wkh fkdqqho hvwlpdwh1
Wkhuh lv d kljk fruuhodwlrq ehwzhhq wkh phdq ri wkh ghflvlrq yduldeoh dqg wkh
fruuhvsrqglqj sdluzlvh huuru suredelolw|/ dqg wkhuhiruh vwxg|lqj wkh phdq lv dq
hdv| zd| wr qg wkh prvw suredeoh huuru sdwkv1 Wkh prvw suredeoh huurqhrxv
k|srwkhvhv dovr kdyh wkh orzhvw phdq ri wkh ghflvlrq yduldeoh1 Wkhuh lv dovr
d vwurqj ghshqghqfh ehwzhhq wkh udqn ri wkh pdwul{ UK
j N dqg wkh sdluzlvh
huuru suredelolw|1 Li wkh udqn lv orz/ wkh phdq ri wkh ghflvlrq yduldeoh lv orz dv
zhoo dqg wkhuh lv d kljk suredelolw| iru wkdw vshflf huuru hyhqw1
wkh vlqjoh huuru hyhqwv/ vhh ^55` iru EHU fdofxodwlrqv ri wuhoolv frghg prgxodwlrq1
Iru wkh vxshulpsrvhg slorw whfkqltxh/ krzhyhu/ wkhuh lv d ghshqghqfh ehwzhhq
wkh huuruv dqg d xqlrq erxqg grhv qrw frqyhujh1 Lqvwhdg zh xvh wkh sdluzlvh
huuru suredelolwlhv wr fdofxodwh wkh wudqvlwlrq suredelolwlhv lq wkh wuhoolv/ l1h1 lq
wkh Pdunry fkdlq1 Dvvxph wkdw wkh huuru hyhqw kdv d olplwhg gxudwlrq1 Iru d
jlyhq wudqvplwwhg vhtxhqfh dqg d fhuwdlq vwdwh lq wkh wuhoolv zh uvw lghqwli| wkh
wzr prvw suredeoh k|srwkhvhv zlwk dqg zlwkrxw dq huuru dw srvlwlrq n= Zh wkhq
xvh rqo| wkhvh wzr vhtxhqfhv wr fdofxodwh wkh suredelolw| wkdw wkh ghflvlrq lv
pdgh lq idyru ri wkh fruuhfw v|pero dqg wkhuhe| wkh wudqvlwlrq suredelolwlhv wr
wkh qh{w vwdwhv lq wkh wuhoolv1 Dvvxplqj wkdw wkh gdwd vhtxhqfh lv 4> = = = > 4>
zh fdq xvh wkhvh wudqvlwlrq suredelolwlhv wr fdofxodwh wkh suredelolw| ri ehlqj lq
d fhuwdlq vwdwh lq vwhdg| vwdwh/ l1h1 lq wkh plggoh ri d orqj eorfn1 Wkhvh vwdwh
suredelolwlhv duh wkhq pxowlsolhg e| wkh suredelolw| ri pdnlqj dq huurqhrxv
ghflvlrq iru hdfk ri wkh vwdwhv wr jhw wkh dssur{lpdwlyh EHU1 Iru h{dpsoh/
dvvxph wkdw O @ 6/ @ 3=38> iGpd{ Wv @ 3==34 dqg He @Q3 @ 48 gE1 Ixuwkhu/
dvvxph wkdw wkh wudqvplwwhg slorw vhtxhqfh lv d vhtxhqfh ri rqhv pdsshg wr
s W
s @ +4> s 4> 4> 4> 4> 4> 4, dqg wkh gdwd vhtxhqfh lv wkh }hur vhtxhqfh pdsshg
wr x @ 4 +4> 4> 4> 4> 4> 4> 4,W = Iru vwdwh 33 wkh prvw suredeoh
k|srwkhvhv zlwk dqg zlwkrxw dq huuru dw srvlwlrq n @ 8 lv 3333433 dqg 3333333/
uhvshfwlyho|1 Khuh zh kdyh wzr v|perov wr jhw lqwr vwhdg| vwdwh/ wzr v|perov wr
uhdfk wkh dfwxdo vwdwh/ wkh huurqhrxv ru fruuhfw v|pero dqg wkhq wzr v|perov
wr uhdfk wkh fruuhfw vwdwh djdlq1 Iru vwdwh 34 wkh fruuhvsrqglqj k|srwkhvhv duh
3343433 dqg 3343333/ uhvshfwlyho|1 Wkh sdluzlvh huuru suredelolw| ehwzhhq wkh
vhtxhqfhv 3343433 dqg 3343333 zkhq wkh gdwd vhtxhqfh lv 3333333 lv 3135: iru
vwdwh 34 dqg wkhuhiruh wkh suredelolw| ri pdnlqj dq huurqhrxv ghflvlrq lv 3135:
lq wklv vwdwh/ wkh wudqvlwlrq suredelolw| wr vwdwh 43 lv 3135: dqg wkh wudqvlwlrq
suredelolw| wr vwdwh 33 lv 31<:61 Lq Ilj1 8 wkh sdwkv ri wkh vwhdg| vwdwh wuhoolv
lv suhvhqwhg zlwk lwv fruuhvsrqglqj wudqvlwlrq dqg vwdwh suredelolwlhv1 Wkh
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 48
0
10
fDmaxTs=10-1
fDmaxTs=10-3
calculations
-1 perfect CE
10
BER
-2
10
-3
10
-4
10
0 5 10 15 20 25 30
Eb/N0, dB
Iljxuh 9= EHU iru glhuhqw vljqdo wr qrlvh udwlrv1 Wkh grwwhg olqhv duh fdofx0
odwlrqv dvvxplqj wkh }hur gdwd vhtxhqfh dqg dq p0vhtxhqfh dv slorw vhtxhqfh1
Iljxuh := Srvvleoh vwdwh wudqvlwlrqv lq wkh xqfrghg +gdvkhg olqh, dqg frghg
+vrolg olqh, fdvh1
0
10
optimal filter
fixed filter
fDmaxTs=10-1
fDmaxTs=10-3
BER
-1
10
-2
10
0 0.2 0.4 0.6 0.8 1
pilot power
Iljxuh ;= Uhvxowlqj EHU zkhq xvlqj wkh rswlpdo dqg {hg owhu frh!flhqwv1
0
10
fDmaxTs=10-1
fDmaxTs=10-2
fDmaxTs=10-3
BER
-1
10
-2
10 -2 -1 0
10 10 10
pilot power
Iljxuh <= Rswlpl}dwlrq ri wkh srzhu/ > ri wkh vxshulpsrvhg slorw vhtxhqfh1
srzhu lv derxw @ 3=35 = = = 3=38 iru doo Grssohu iuhtxhqflhv ri lqwhuhvw= Wkh
srzhu ri slorw v|perov vkrxog rqo| eh derxw 5 ( wr 8 ( ri wkh srzhu ri wkh
gdwd v|perov1 Lq wkh iroorzlqj/ @ 3=38 lv dssolhg1
Lq d vhfrqg vhw ri vlpxodwlrqv zh lqyhvwljdwhg wkh lq xhqfh ri wkh owhu
ohqjwk O/ dqg wkh uhvxowv duh suhvhqwhg lq Ilj1 431 Djdlq/ He @Q3 @ 48 gE1 Lw
dsshduv iurp wkh fxuyhv wkdw d owhu ohqjwk ri derxw O @ 9 lv vx!flhqw iru doo
Grssohu iuhtxhqflhv ri lqwhuhvw1 Dowkrxjk qrw vkrzq khuh/ wkh phdq vtxduhg
huuru ri wkh owhu/ zklfk fdq eh fdofxodwhg dqdo|wlfdoo|/ vkrzv wkh vdph sulqflsoh
ehkdylru dv wkh EHU fxuyh lq wkh jxuh1
Jlyhq wkhvh rswlpl}dwlrqv/ zh vlpxodwhg wkh EHU yhuvxv He @Q3 iru wkh
vfkhph xqghu lqyhvwljdwlrq1 Wkh uhvxow lv vkrzq lq Ilj1 91 Dv d ehqfkpdun/
wkh EHU shuirupdqfh ri 50SVN rq d dw Ud|ohljk idglqj fkdqqho jlyhq shuihfw
fkdqqho hvwlpdwlrq lv sorwwhg dv zhoo1 Iru idglqj udwhv xs wr 435 wkh orvv lv ohvv
wkdq 5 gE/ dqg hyhq iru idvw idglqj +iGpd{ Wv @ 3=4, qr huuru rru lv ylvleoh lq
wkh lqwhuhvwlqj udqjh1 Wklv lv lq frqwudvw wr frqyhqwlrqdo slorw0v|pero0dvvlvwhg
FH1 Iru kljk Grssohu iuhtxhqflhv slorw0v|pero0dvvlvwhg FH vkrzv dq huuru rru
dqg wkh vxshulpsrvhg slorw vhtxhqfh whfkqltxh vljqlfdqwo| rxwshuirupv wklv
phwkrg1
Lq Ilj1 44 zh frpsduh wkh EHU iru glhuhqw Grssohu ydoxhv zkhq xvlqj
frqyhqwlrqdo slorw0v|pero0dvvlvwhg FH dqg vxshulpsrvhg slorw vhtxhqfhv1 Dv
d uhihuhqfh zh dovr jlyh wkh EHU iru shuihfw fkdqqho hvwlpdwlrq1 Lq erwk fdvhv
wkh slorw srzhu lv 8(1 Wklv phdqv iru wkh slorw0v|pero0dvvlvwhg vfkhph wkdw
hyhu| 53wk v|pero lv d slorw v|pero dqg wkdw wkh hqhuj| shu lqirupdwlrq elw lv
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 4<
0.1
fDmaxTs=10-1
fDmaxTs=10-3
perfect CE
BER
0.01
2 3 4 5 6 7 8 9 10
filter length
9 Frqfoxvlrqv
Lq wklv sdshu/ zh h{soruhg wkh vxshulpsrvhg slorw vhtxhqfh whfkqltxh iru wkh
sxusrvh ri fkdqqho hvwlpdwlrq1 Gxh wr wkh uhgxqgdqf|/ wuxo| frkhuhqw ghprg0
xodwlrq lv dfklhyhg1 Wkh pdlq frqwulexwlrq zdv wkh ghulydwlrq dqg dqdo|vlv ri d
uhfhlyhu edvhg rq wkh Ylwhuel dojrulwkp1 Zh ghulyhg erwk wkh rswlpdo phwulf/
dqg d vxe0rswlpdo phwulf vxlwdeoh iru uhfxuvlyh ghwhfwruv1 Wkh shuirupdqfh ri
wkh sursrvhg uhfhlyhu zdv wkhq dqdo|}hg erwk dqdo|wlfdoo| dqg e| vlpxodwlrqv1
Iru wkh dqdo|wlfdo dqdo|vlv zh ghulyhg wkh sdluzlvh huuru suredelolwlhv dqg edvhg
rq wkhvh zh pdgh dssur{lpdwlrqv ri wkh ryhudoo EHU1
Frpsduhg wr slorw0v|pero0dvvlvwhg FH/ zklfk lv fxuuhqwo| vwdwh0ri0wkh0duw/
glvwlqfw dgydqwdjhv duh dv iroorzv=
Qr edqgzlgwk h{sdqvlrq
PSA CE
SPP
perfect CE
-1
10
BER
-2 15 dB
10
25 dB
-3
10
-2 -1
10 10
Doppler rate
Iljxuh 44= EHU iru shuihfw fkdqqho hvwlpdwlrq/ vxshulpsrvhg slorw vhtxhqfhv
+VSS, ru slorw0v|pero0dvvlvwhg FH +SVD FH,1 Lq wkh wzr odwwhu fdvhv 8( ri wkh
srzhu duh vshqw rq slorwv1
Slorw0v|pero0dvvlvwhg FH/ zlwk lvrodwhg slorw v|perov/ idlov zkhq wkh slorw
v|perov duh dhfwhg e| iuhtxhqf|0vhohfwlyh idglqj
Gdwd v|perov duh htxdoo| vhqvlwlyh djdlqvw wudqvplvvlrq huuruv zkhq df0
txlvlwlrq lv qrw frqvlghuhg1
Wkh sursrvhg Ylwhuel uhfhlyhu lv/ krzhyhu/ pruh frpsoh{ wkdq d vlqjoh lqwhu0
srodwlrq owhu gxh wr wkh shu0vxuylyru surfhvvlqj1 Wkhuhiruh/ vlpsolhg vwuxf0
wxuhv edvhg rq uhgxfhg0vwdwh vhtxhqfh hvwlpdwlrq zhuh sursrvhg dv zhoo1 Dv
rxwolqhg lq wkh dqdo|vlv/ wkh shuirupdqfh ghshqgv rq wkh eorfn ohqjwk dqg rq
wkh fkrlfh ri wkh SQ vhtxhqfh1
Ixwxuh zrun pd| eh ghyrwhg wr rswlpl}dwlrqv ri wkh slorw vhtxhqfh/ dq
hydoxdwlrq ri P 0du| prgxodwlrq vfkhphv/ pxowl0fduulhu prgxodwlrq vfkhphv
+h1j1/ RIGP zlwk 50G slorw duud|,/ dqg uhfhlyhu vwuxfwxuhv zlwk ryhuvdpsolqj1
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 54
,K !4 +| |
4 4
i +| m F, @ h{s+ +| | ,,> +58,
+5,Q@5 ghw4@5 +!, 5
zkhuh | ghqrwhv wkh phdq ri wkh yhfwru | dqg ! ghqrwhv wkh fryduldqfh pdwul{
ri wkh uhfhlyhg vdpsohv/ ! @H^+| | ,+| |,K `= Wkh phdq |
lv qrw nqrzq d
wkh phdq lv jlyhq e| wkh fkdqqho hvwlpdwhv
sulrul/ exw jlyhq wkh k|srwkhvlv F
dv
@F F 4
|
@ H^| m F` i @ FD |= +59,
F
! @ H^+L FD 4 ,||K +L FD
F 4 ,K ` +5:,
F
@ +L FD 4 ,+FU
K K 5 4 K
i F . q L,+L FDF ,
zkhuh Ui lv wkh fryduldqfh pdwul{ ri wkh fkdqqho1 Vxevwlwxwlqj +59, dqg +5:,
lqwr +58, dqg wdnlqj wkh qdwxudo orjdulwkp |lhogv wkh orj0olnholkrrg ixqfwlrq
@ 4 oq++5,Q ghw+!,, 4 +| FD
oq i+| m F, F 4 +| FD
4 |,K ! F 4 |,
5 5
4 4 K K
@ oq++5, ghw+!,, | +FUi F . 5q L,4 |
Q +5;,
5 5
rsw @ oq+ghw+!,, . +| FD
F 4 |,K !
4 +| FD
F 4 |, +5<,
. |K +FU
@ oq+ghw+!,, K K 5 4
i F . q L, |
@ +| FD
F 4 |,K +| FD
F 4 |,= +63,
55 FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV
Uhihuhqfhv
^4` M1 J1 Surdnlv1 Gljlwdo frppxqlfdwlrqv1 Suhqwlfh0Kdoo/ 6ug hglwlrq/ 4<<81
^7` P1 Prkhu dqg M1 Orgjh1 D wlph glyhuvlw| prgxodwlrq vwudwhj| iru wkh
vdwhoolwh0preloh fkdqqho1 Lq 46wk Elhqqldo V|ps1 Frppxq1/ Nlqjvwrq/
Fdqdgd/ Mxqh 4<;91
^;` W1 Kroghq dqg N1 Ihkhu1 D vsuhdg vshfwuxp edvhg v|vwhp whfkqltxh iru
v|qfkurql}dwlrq ri gljlwdo preloh frppxqlfdwlrq v|vwhpv1 LHHH Wudqv1
Eurdgf1/ sdjhv 4;84<7/ Vhsw1 4<<31
^45` H1 Ehmmdql/ M10F1 Ehoruh/ dqg S1 Ohfodlu1 Frkhuhqw ghwhfwlrq iru wudqvplv0
vlrq ryhu vhyhuo| wlph dqg iuhtxhqf| glvshuvlyh pxowlsdwk idglqj fkdqqhov1
Lq Surf1 LHHH Lqwhuq1 V|ps1 Lqi1 Wkhru|/ sdjh 545/ Zklvwohu/ Fdqdgd/
Vhsw1 4<<81
FKDQQHO HVWLPDWLRQ XVLQJ VXSHULPSRVHG SLORWV 56
Paper G
This part is a revised version of:
Fredrik Tufvesson and Ove Edfors, "A Comparaitve Analysis Between Different
Detection Schemes in Flat Rayleigh-fading Channels", Proceedings of IEEE Vehicu-
lar Technology Conference, Tokyo, Japan, pp. 1205-1209, May 2000.
IEEE 2000, reprinted with permission.
D Frpsdudwlyh Dqdo|vlv Ehwzhhq Glhuhqw
Ghwhfwlrq Vfkhphv lq Iodw Ud|ohljk0idglqj
Fkdqqhov
Devwudfw
4 Lqwurgxfwlrq
Lw lv zhoo nqrzq wkdw frkhuhqw ghwhfwlrq uhvxowv lq jrrg shuirupdqfh1 Lw dovr
hqdeohv wkh xvh ri pxowlohyho vljqdoolqj exw uhtxluhv uholdeoh skdvh dqg2ru fkdq0
qho hvwlpdwlrq1 Glhuhqwldo ghwhfwlrq/ rq wkh rwkhu kdqg/ lv urexvw dqg uhtxluhv
qr fkdqqho hvwlpdwlrq/ exw wkhuh lv d shuirupdqfh orvv frpsduhg wr frkhuhqw
ghwhfwlrq dqg lw lv prvwo| xvhg iru skdvh vkliw nh|lqj +SVN,1 Pdq| qhz vfkhphv
kdyh ehhq suhvhqwhg wr hqkdqfh wkh shuirupdqfh ri erwk frkhuhqw dqg glhuhq0
wldo phwkrgv1 Lq wklv sdshu zh pdnh dq dqdo|wlfdo frpsdulvrq ehwzhhq vrph ri
wkhvh ghwhfwlrq vfkhphv1 Wkh dqdo|vlv lv edvhg rq d xqlhg iudphzrun zkhuh
zh fdofxodwh wkh sdluzlvh suredelolw| wkdw dq huurqhrxv vhtxhqfh lv ghwhfwhg
lq idyru ri wkh fruuhfw rqh1 Wkh dqdo|}hg phwkrgv lqfoxgh= frqyhqwlrqdo gli0
ihuhqwldo ghwhfwlrq +Gl,/ pxowlsoh0v|pero glhuhqwldo ghwhfwlrq +PVGG,/ slorw
v|pero dvvlvwhg prgxodwlrq +SVDP,/ ghwhfwlrq xvlqj vxshusrvhg slorw v|perov
+VSS,/ dqg d qryho frkhuhqw k|eulg ghwhfwlrq vfkhph +K|eu,1
Wkh PVGG zdv lqwurgxfhg iru DZJQ fkdqqhov lq ^4` dqg h{whqghg wr idg0
lqj fkdqqhov lq ^5` dqg ^6`1 Wkh lghd ehklqg wklv whfkqltxh lv wr vlpxowdqhrxvo|
4
5 D FRPSDUDWLYH DQDO\VLV
ghwhfw d jurxs ri skdvh vkliwv lq rughu wr kdyh pruh wkdq rqh v|pero dv d uhi0
huhqfh1 Lq wklv zd| wkh uhihuhqfh lv ohvv dhfwhg e| qrlvh dqg wkh shuirupdqfh
lv wkhuhe| hqkdqfhg1 Wkh dqdo|vlv lq ^6` xvhv h{suhvvlrqv iru wkh sdluzlvh hu0
uru suredelolwlhv dqg d xqlrq erxqg wr fdofxodwh dq dssur{lpdwlyh elw0huuru
udwh1 Lq ^7` d ghwhfwru edvhg rq wkh Ylwhuel dojrulwkp zdv suhvhqwhg/ zklfk zh
xvh lq rxu dqdo|vlv1 Wklv Ylwhuel0edvhg ghwhfwru lv hvvhqwldoo| wkh vdph dv wkh
ghflvlrq0gluhfwhg frkhuhqw ghwhfwru lqwurgxfhg lq ^8`/ h{fhsw wkdw lq wkh odwwhu
rqh v|pero0e|0v|pero ghflvlrqv duh pdgh lqvwhdg ri vhtxhqfh ghflvlrqv1
SVDP zdv ruljlqdoo| sursrvhg lq ^9` dqg odwhu dqdo|}hg lq ^:`1 Lq wklv
vfkhph nqrzq slorw v|perov duh pxowlsoh{hg lqwr wkh gdwd vhtxhqfh1 Wkh uh0
fhlyhg slorw v|perov pdnhv lw srvvleoh wr revhuyh wkh lq xhqfh ri wkh fkdqqho
dw wkh slorw srvlwlrqv1 Wkh fkdqqho hvwlpdwhv iru wkh gdwd v|perov duh wkhq
dfklhyhg e| lqwhusrodwlrq1 Vlqfh wkh slorw v|perov frqwdlq qr gdwd wkhuh lv d
vpdoo ryhukhdg fdxvlqj d edqgzlgwk h{sdqvlrq dqg dq hqhuj| orvv1 Krzhyhu/
wkh phwkrg kdv vkrzq wr eh urexvw lq idglqj hqylurqphqwv zlwk rqo| d vpdoo
orvv frpsduhg wr lghdo frkhuhqw ghwhfwlrq1 Lq dgglwlrq/ lw dovr doorzv pxowlohyho
vljqdoolqj1
Wkh xvh ri vxshusrvhg slorw vhtxhqfhv zdv lqwurgxfhg lq ^;` iru fduulhu
skdvh wudfnlqj1 Dv rssrvhg wr SVDP/ wkh slorw vhtxhqfh lv dgghg wr wkh
gdwd vhtxhqfh +vxshusrvhg, dqg wkhuhe| wudqvplwwhg lq wkh vdph iuhtxhqf|
edqg dqg dw wkh vdph wlph1 Vlqfh wkh slorw vhtxhqfh dqg wkh gdwd vhtxhqfh
duh vxshusrvhg wkhuh lv qr lqfuhdvh lq edqgzlgwk1 Lq ^<` wklv whfkqltxh zdv
sursrvhg iru frkhuhqw ghwhfwlrq ri gdwd v|perov1 Wkh VSS vfkhph lv urexvw
djdlqvw kljk Grssohu iuhtxhqflhv dqg lw hqdeohv pxowlohyho vljqdoolqj1 Krzhyhu/
wkh frpsoh{lw| jurzv zlwk wkh qxpehu ri vljqdoolqj dowhuqdwlyhv1
Wkh vwuhqjwkv ri SVDP dqg vxshusrvhg slorw v|perov fdq eh frpelqhg
lqwr d qhz k|eulg ghwhfwlrq vfkhph zkhuh qrw rqo| nqrzq slorw v|perov/ exw
dovr k|srwkhvhv ri wkh xqnqrzq gdwd v|perov/ duh xvhg iru fkdqqho hvwlpdwlrq1
Lq ^43` d vlplodu vfkhph zdv suhvhqwhg xvlqj ghflvlrqv ri wkh xqnqrzq gdwd
v|perov iru frqvhfxwlyh skdvh fruuhfwlrqv ri wkh fkdqqho hvwlpdwh1 Rxu qhz
vfkhph xvhv k|srwkhvhv ri wkh gdwd v|perov dv yluwxdo slorw v|perov dqg lw lv
wkhuhiruh ohvv vhqvlwlyh wr ghflvlrq huuruv1
Wkh frqwulexwlrqv ri wklv sdshu lqfoxgh erwk wkh qhz k|eulg ghwhfwlrq
vfkhph dqg d xqlhg dqdo|vlv iudphzrun iru sdluzlvh huuru hyhqwv/ wkdw fdq
eh dssolhg wr doo ri wkh deryh ghwhfwlrq vfkhphv1 Wkrxjk eh|rqg wkh vfrsh ri
wklv sdshu/ wkh sdluzlvh huuru suredelolwlhv suhvhqwhg fdq eh xvhg wr fdofxodwh
erxqgv iru wkh uhvxowlqj elw huuru udwh +EHU, xvlqj wkh prvw suredeoh huuru
vhtxhqfhv lq wkh vdph zd| dv lq ^6`1
5 Ghwhfwlrq Vfkhphv
Wkh dqdo|}hg ghwhfwlrq phwkrgv/ lqfoxglqj wkh glhuhqwldo rqhv/ uhtxluh vrph
nlqg ri fkdqqho hvwlpdwh iru ghwhfwlrq1 Frqyhqwlrqdo glhuhqwldo ghwhfwlrq
D FRPSDUDWLYH DQDO\VLV 6
~
Metric Decision
calculation
~
f
Channel A
estimator
^
f ~
C
Remodulation
y
xvhv h1j1 wkh odvw uhfhlyhg v|pero dv d fkdqqho hvwlpdwh1 Lq SVDP wkh uh0
fhlyhg slorw v|perov duh xvhg wr fdofxodwh fkdqqho hvwlpdwhv1 Zkhq xvlqj VSS
wkh fkdqqho hvwlpdwhv duh edvhg rq wkh nqrzq slorw vhtxhqfh wrjhwkhu zlwk
k|srwkhvhv ri wkh xqnqrzq gdwd vhtxhqfh1 Wkh jhqhudo uhfhlyhu xvhg iru wkh
dqdo|vlv ri doo ghwhfwlrq phwkrgv lv suhvhqwhg lq Ilj141
Lq wkh iroorzlqj/ zh xvh frpsoh{ edvhedqg qrwdwlrq dqg dvvxph shuihfw
wlph/ iuhtxhqf| dqg iudph v|qfkurql}dwlrq1 Wkh wudqvplwwhg gdwd v|perov
duh ghqrwhg dv fn 5 F1 Wkh v|pero fn lvshlwkhu d gdwd v|pero/ xn > d slorw
s
v|pero/ sn > ru d frpelqdwlrq ri erwk/ fn @ 4 n xn . n sn > zkhuh n lv
wkh dprxqw ri hqhuj| vshqw rq slorw lqirupdwlrq dw wlph n= Ohw |n ghqrwh wkh
uhfhlyhg v|pero vdpsohv/
|n @ fn in . qn > +4,
zkhuh in 5 F lv d pxowlsolfdwlyh +wlph0vhohfwlyh, idglqj surfhvv zlwk H^min m5 ` @
4/ dqg qn 5 F duh }hur0phdq zklwh Jdxvvldq qrlvh vdpsohv zlwk rqh0vlghg
srzhu vshfwudo ghqvlw| Q3 1 Zlwkrxw orvv ri jhqhudolw|/ wkh dyhudjh hqhuj| ri wkh
uhfhlyhg v|perov lv Hv @ H^mfn in m5 ` @ 4= Lw lv frqyhqlhqw wr xvh pdwul{ qrwdwlrq
dqg zh ghqrwh wkh uhfhlyhg vdpsoh yhfwru e| | @ +|4 > |5 > ===> |Q ,W @ Fi . q/
zkhuh i dqg q duh yhfwruv ri wkh idglqj surfhvv dqg wkh qrlvh/ uhvshfwlyho|/ dqg
F lv d gldjrqdo pdwul{ zlwk wkh wudqvplwwhg v|perov rq lwv gldjrqdo1
Wkh jrdo lv wr frpsxwh wkh pd{lpxp0olnholkrrg hvwlpdwh ri wkh wudqvplw0
whg vhtxhqfh1 Wkh uhfhlyhu whvwv doo srvvleoh k|srwkhvhv dqg ghflghv iru wkh
vhtxhqfh uhvxowlqj lq wkh orzhvw phwulf1 Iru Frqvwdqw0hqyhorsh prgxodwlrq/
wkh rswlpdo phwulf lv jlyhq ^6` dv
@ |K F+
. L,4 F K |> +5,
lv d gldjrqdo pdwul{ ri wkh k|srwkhvlv wr eh whvwhg +ghqhg odwhu rq,
zkhuh F
dqg lv wkh fryduldqfh pdwul{ ri wkh fkdqqho1 Vlqfh lw lv dgydqwdjhrxv wr kdyh
7 D FRPSDUDWLYH DQDO\VLV
dq dgglwlyh phwulf lqfuhphqw iru hdfk qhz v|pero rqh riwhq xvh dssur{lpdwlrqv
lqvwhdg1 Wkh qrlvh lv dvvxphg wr eh Jdxvvldq/ dqg wkhuhiruh d vxlwdeoh phwulf
lqfuhphqw lv
fn > fnS > = = = > fnS , @ m|n fn in m5 >
n +
3 O +6,
zkhuh fn lv d k|srwkhvlv iru wkh n0wk gdwd v|pero dqg in lv d fkdqqho hvwlpdwh1
Wkh odwwhu lv ghulyhg iurp wkh gdwd k|srwkhvlv ^ xnS3 > = = = > x
nSO ` dqg2ru slorw
v|perov ^snS3 > = = = > snSO `1 Wkh sdudphwhu O lv wkh qxpehu ri v|perov xvhg
iru wkh fkdqqho hvwlpdwh dqg So ghqrwhv wkh wlph lqglfhv ri wkh uhfhlyhg v|perov
xvhg iru wkh hvwlpdwh1
Lq rughu wr pdnh ghflvlrqv/ wkh uhfhlyhu ghulyhv fkdqqho hvwlpdwhv dqg xvlqj
wkhvh lw whvwv doo srvvleoh k|srwkhvhv djdlqvw wkh uhfhlyhg v|perov1 Wkh fkdqqho
hvwlpdwhv duh fdofxodwhg dv iroorzv= D whqwdwlyh fkdqqho hvwlpdwh lv rewdlqhg
e| uh0prgxodwlrq/ ianSo @ |nSo @ fnSo = Jlyhq wkh k|srwkhvlv dqg wkh slorw
v|perov/ wkh fkdqqho hvwlpdwh in fdq eh frpsxwhg e| olqhdu owhulqj ^44`=
O[4 O
[4 |
nSo
in @ dn>So ianSo @ do = +7,
fnSo
o@3 o@3
Wkh owhu frh!flhqwv dn>p duh ghshqghqw rq wkh ghwhfwlrq phwkrg xvhg1 Iru
frqyhqwlrqdo glhuhqwldo ghwhfwlrq rqo| wkh odvw v|pero lv xvhg iru wkh fkdqqho
hvwlpdwh vr O @ 4> S3 @ 4> d4 @ 4> dqg doo rwkhu owhu frh!flhqwv duh }hur1 Li
d Zlhqhu hvwlpdwru lv xvhg/ dv lq h1j1 SVDP ru VSS/ wkh owhu frh!flhqwv duh
rewdlqhg dv wkh vroxwlrq ri wkh Zlhqhu0Krsi htxdwlrqv1
Wkh fkdqqho hvwlpdwh dw wlph n> edvhg rq d fhuwdlq k|srwkhvlv/ lv jlyhq e|
wkh suhylrxv uhfhlyhg v|perov dv
in @ dn F 4 |> +8,
zkhuh dn lv d yhfwru ri wkh fkdqqho hvwlpdwlrq frh!flhqwv dn>p / h1j1 wkh Zlhqhu
owhu frh!flhqwv/ dqg F lv d gldjrqdo pdwul{ frqwdlqlqj wkh k|srwkhvlv ri wkh
s s
wudqvplwwhg v|perov rq lwv gldjrqdo/ l1h1 fn>n @ 4 n xn . n sn x n =
Wkh fruuhvsrqglqj yhfwru ri fkdqqho hvwlpdwhv iru d fhuwdlq k|srwkhvlv fdq
frqvhtxhqwo| eh fdofxodwhg dv
i @ DF 4 |= +9,
Wkh hohphqwv lq wkh fkdqqho hvwlpdwlrq pdwul{ D dqg wkh uhprgxodwlrq pdwul{
4 duh ghwhuplqhg e| wkh ghwhfwlrq phwkrg1 Wkhvh duh ghvfulehg ehorz1
F
Li v|pero0e|0v|pero ghwhfwlrq lv xvhg wkhq wkh phwulf lqfuhphqw +6, lv xvhg
iru wkh ghflvlrqv dqg li vhtxhqfh ghwhfwlrq lv xvhg wkhq d vxp ri wkhvh lv xvhg
lqvwhdg1 Dgglqj wkh phwulf lqfuhphqwv dqg vxevwlwxwlqj wkh fkdqqho hvwlpdwhv
e| +9, ohdgv wr wkh fxpxodwlyh phwulf
@ |K +L FD
F 4 ,K +L FD
F 4 ,|= +:,
Ehorz zh vkruwo| ghvfuleh wkh dqdo|}hg ghwhfwlrq phwkrgv dqg jlyh wkhlu uh0
vshfwlyh fkdqqho hvwlpdwlrq pdwulfhv dqg k|srwkhvlv pdwulfhv1
D FRPSDUDWLYH DQDO\VLV 8
Wkh k|srwkhvlv pdwul{ lv gldjrqdo dqg kdv wkh nqrzq slorw v|pero ydoxhv dw
wkhlu fruuhvsrqglqj srvlwlrqv dqg wkh ydoxhv iru wkh gdwd k|srwkhvlv wr eh whvwhg
dw wkh rwkhu srvlwlrqv/ l1h1
Dv rssrvhg wr wkh rwkhu phwkrgv/ wkh ghflvlrqv duh pdgh v|pero e| v|pero
zkhq xvlqj frqyhqwlrqdo glhuhqwldo ghwhfwlrq1 Wkhuhiruh/ wkh dqdo|vlv suh0
vhqwhg lq qh{w vhfwlrq lv xvhg wr vwxg| rqo| vlqjoh0huuru hyhqwv iru wklv vfkhph1
6 Ghwhfwru Shuirupdqfh
Lq wklv vhfwlrq zh ghulyh dqdo|wlfdo h{suhvvlrqv iru wkh suredelolw| ri sdluzlvh
huuru hyhqwv1 Wkh dqdo|vlv lv wr vrph h{whqw vlplodu wr wkh rqh lq ^6`/ exw zh dovr
surylgh forvhg irup h{suhvvlrqv iru wkh glvwulexwlrq ri wkh ghflvlrq yduldeoh1
Wkh uhfhlyhu pdnhv d fruuhfw ghflvlrq li wkh phwulf iru wkh fruuhfw sdwk lv
vpdoohu wkdq doo rwkhu sdwk phwulfv1 Ghqh wkh ghflvlrq yduldeoh a @
>zkhuh dqg duh wkh fxpxodwlyh phwulfv ri wkh k|srwkhvlv dqg fruuhfw
sdwk/ uhvshfwlyho|1 Li a A 3 iru doo k|srwkhvhv wkhq wkh fruuhfw sdwk zloo eh
fkrvhq1 E| uhzulwlqj wkh uhfhlyhg vdpsohv dv | @ Gj zlwk G @ F L dqg
W
j@ i q > wkh ghflvlrq yduldeoh fdq eh h{suhvvhg
a @ |K ^+L FD
F 4 ,K +L FD
F 4 , +;,
+L FDF ,K +L FDF 4 ,`|
4
@ jK Nj=
zkhuh Ui lv wkh fryduldqfh pdwul{ ri wkh idglqj surfhvv1 Iru 50G lvrwurslf
vfdwwhulqj Ui kdv hohphqwv ul>m @ M3 +5iG pd{ Wv +l m,,/ zkhuh M3 +, lv wkh
}hurwk rughu Ehvvho ixqfwlrq/ iG pd{ lv wkh pd{lpxp Grssohu iuhtxhqf| dqg
Wv lv wkh v|pero wlph1
Wkh sdluzlvh huuru suredelolw|/ l1h1 wkh suredelolw| wkdw wkh ghflvlrq yduldeoh
ehfrphv qhjdwlyh vr wkdw wkh uhfhlyhu pdnhv d ghflvlrq lq idyru ri wkh huurqhrxv
k|srwkhvlv v lqvwhdg ri wkh fruuhfw rqh/ lv wkhq jlyhq e|
+q ,Qv 4
Qv
[
Sv @ Q = +45,
q@p.4
Tv
+n q ,
n@4>n9@q
vhqwhg1 Lq wklv uhylvhg yhuvlrq wkh fxuyh lv rplwwhg dqg wkh wh{w kdv ehhq fkdqjhg dffruglqjo|1
D FRPSDUDWLYH DQDO\VLV <
04
43
SVDP
VSS
K|eu
PVGG
1
e
r
u
s
u
r
u
u 05
h
h
43
v
l
z
u
l
d
S
06
43 4 5 6 7 orqj
Ohqjwkrihuuruhyhqw
1 3
e
r43
u
s 04
u43
r
u
u 05
h
43
h
v
l 06 SVDP
z43 VSS
u
l
d 07 K|eu
S43
1 3
e
r43
u
s 04
u43
r
u
u 05
h43
h
v
l 06
z43 Glii
u
l
d 07 PVGG
S43
3 8 43 48 53 58 63
He2Q3+gE,
Iljxuh 6= Sdluzlvh huuru suredelolw| iru glhuhqw VQU ydoxhv1 Vrolg olqhv
uhsuhvhqw qr Grssohu +iG Wv @ 3, zkhuhdv gdvkhg olqhv uhsuhvhqw idvw idglqj
+iG Wv @ 3=38,=
43 D FRPSDUDWLYH DQDO\VLV
8 Frqfoxvlrqv
Zh kdyh dqdo|}hg wkh shuirupdqfh dqg ehkdylru ri glhuhqw ghwhfwlrq vfkhphv
lq Ud|ohljk idglqj fkdqqhov1 Edvhg rq d xqlhg dqdo|wlfdo iudphzrun zh kdyh
fdofxodwhg wkh sdluzlvh huuru suredelolw| iru fhuwdlq frkhuhqw dqg qrq0frkhuhqw
ghwhfwlrq phwkrgv1 E| vwxg|lqj wkh yduldwlrq ri wkh sdluzlvh huuru suredelolwlhv
zh kdyh frpsduhg wkh urexvwqhvv ri wkhvh phwkrgv djdlqvw yduldwlrqv lq dyhudjh
VQU dqg Grssohu hhfwv1 Dv h{shfwhg/ wkh vfkhphv xvlqj d jurxs ri v|perov
forvh wr wkh v|pero wr eh ghwhfwhg wr ghulyh d fkdqqho hvwlpdwh duh vkrzq wr
eh yhu| urexvw lq idvw idglqj hqylurqphqwv dqg vkrz qr huuru rru1 Frqyhq0
wlrqdo glhuhqwldo ghwhfwlrq/ rq wkh rwkhu kdqg/ vxhuv iurp d qrlv| fkdqqho
hvwlpdwh zkhuhdv SVDP pd| vxhu iurp dq revrohwh fkdqqho hvwlpdwh lq idvw
idglqj fkdqqhov1 Wkh odwwhu fdq eh dyrlghg e| lqfuhdvlqj wkh qxpehu ri slorw
v|perov/ exw wkhq wkh ryhukhdg lqfuhdvhv1 Lq wklv sdshu zh kdyh dovr lqwurgxfhg
d qhz ghwhfwlrq vfkhph wkdw frpelqhv wkh vwuhqjwkv ri xvlqj slorw0 dqg gdwd
v|perov wr ghulyh fkdqqho hvwlpdwhv1 Wkh slorw v|perov hqvxuhv wkdw wkhuh lv
ohvv skdvh dpeljxlw| zkloh wkh xvh ri xqnqrzq gdwd v|perov pdnhv lw srvvleoh
wr wudfn fkdqqho yduldwlrqv ehwzhhq wkh slorw v|perov1
D FRPSDUDWLYH DQDO\VLV 44
Wkh suhvhqwhg dqdo|vlv lv jhqhudo dqg fdq hdvlo| eh dssolhg wr rwkhu ghwhfwlrq
vfkhphv1 Iru d ixuwkhu dqdo|vlv/ wkh h{suhvvlrqv ri wkh sdluzlvh huuru suredelolw|
fdq eh xvhg wr jhw dssur{lpdwlyh h{suhvvlrqv ri wkh uhvxowlqj elw dqg iudph huuru
udwhv e| xvlqj xqlrq erxqg whfkqltxhv1
Dfnqrzohgjhphqw
Wkh dxwkruv zrxog olnh wr wkdqn Suri1 Shwhu Krhkhu dw Xqlyhuvlw| ri Nlho/
Jhupdq|/ iru lqvslulqj wklv zrun1
Uhihuhqfhv
^4` Gdulxvk Glyvdodu dqg Pduylq N1 Vlprq1 Pxowlsoh0v|pero glhuhqwldo gh0
whfwlrq ri PSVN1 LHHH Wudqv1 Frppxq1/ 6;+6,=63363;/ Pdufk 4<<31
^8` Duqh Vyhqvvrq1 Frkhuhqw ghwhfwru edvhg rq olqhdu suhglfwlrq dqg ghflvlrq
ihhgedfn iru GTSVN1 Hohfwurq1 Ohww1/ 63+53,=49754976/ Vhswhpehu 4<<71
^9` Plfkdho O1 Prkhu dqg Mrkq K1 Orgjh1 WFPS D prgxodwlrq dqg frg0
lqj vwudwhj| iru Ulfldq0idglqj fkdqqhov1 LHHH M1 Vhohfw1 Duhdv Frppxq1/
:+<,=467:4688/ Ghfhpehu 4<;<1
^;` W1S1 Kroghq dqg N1 Ihkhu1 D vsuhdg vshfwuxp edvhg v|vwhp whfkqltxh iru
v|qfkurql}dwlrq ri gljlwdo preloh frppxqlfdwlrq v|vwhpv1 LHHH Wudqv1
Eurdgf1/ sdjhv 4;84<7/ Vhswhpehu 4<<31
^<` Shwhu Krhkhu dqg Iuhguln Wxiyhvvrq1 Fkdqqho hvwlpdwlrq zlwk vxshulp0
srvhg slorw vhtxhqfh1 Lq Surf1 Jorehfrp/ sdjhv 54955499/ Ulr gh Mdqhlur/
Eud}lo/ Ghfhpehu 4<<<1
45 D FRPSDUDWLYH DQDO\VLV