Professional Documents
Culture Documents
By
TAOUFIK BOURDI
Beceem Communications Inc., Santa Clara, California, USA
and
IZZET KALE
Westminster University, London, UK and Eastern Mediterranean University,
Famagusta, North Cyprus
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN-10
ISBN-13
ISBN-10
ISBN-13
1-4020-5927-2 (HB)
978-1-4020-5927-8 (HB)
1-4020-5928-5 (e-book)
978-1-4020-5928-5 (e-book)
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Contents
Preface
Nomenclature
ix
xi
1 INTRODUCTION
1.1 Introduction
1.2 Research Contribution
2 WIRELESS COMMUNICATION SYSTEMS
2.1 Introduction
2.2 The WLAN Standards
2.3 WLAN Transceiver Systems
2.3.1 The Transmitter
2.3.2 The Receiver
2.3.3 The Frequency Synthesizer (Local Oscillator)
3 PHASE-LOCKED LOOP FREQUENCY SYNTHESIZERS
3.1 Introduction
3.2 Phase-Locked Loop Frequency Synthesizer
3.2.1 Phase-Locked Loop Main Blocks
3.2.1.1 Phase-Frequency Detector
3.2.1.2 Charge Pump
3.2.1.3 Voltage-Controlled Oscillator
3.2.1.4 Voltage-Controlled Crystal Oscillator
3.2.1.5 Dividers
3.3 Phase-Locked Loop Parameters
3.3.1 Loop Filter Design
1
2
7
7
8
10
12
12
13
15
15
15
16
16
17
18
19
19
19
20
vi
Contents
3.4 Noise in Phase-Locked Loops
3.4.1 Component Noise Models
3.4.1.1 Reference Oscillator and VCO Phase Noise
3.4.1.2 Charge Pump Current Noise
3.4.1.3 Loop Filter Resistor Noise
3.4.1.4 Main Divider Noise
3.4.1.5 Phase-Frequency Detector Phase Noise
3.4.1.6 Overall Phase Noise Contribution
3.5 Fractional-N Synthesizers
3.5.1 '6 Modulators in Frequency Synthesizers
3.5.1.1 Fractional-N Case Study
3.6 RMS Phase Error (Irms) and Error Vector Magnitude
3.7 Conclusion
27
29
30
30
30
31
31
32
34
36
39
41
42
45
4.1 Introduction
4.2 Phase-Domain Model
4.2.1 A Constituent Blocks Behavioral Models
4.2.1.1 The Reference Oscillator
4.2.1.2 The '6 Modulator/Feedback Integer Divider
4.2.1.3 The VCO
4.2.1.4 The PFD/CP
4.2.1.5 The Loop Filter
4.2.2 Noise Modeling Summary
4.3 Synthesizer Platform Evaluation
4.3.1 Dithering Effect
4.3.2 Close-to-Integer Operation
4.3.3 Noise Folding
4.3.4 Effect of Prescaler Divider
4.4 Conclusion
45
46
50
50
50
51
52
52
52
53
57
60
60
62
65
67
5.1 Introduction
5.2 An overview
5.3 A Multimode Multistandard '6-Based PLL Synthesizer Design
5.3.1 Design Methodology
5.4 The '6 Frequency Synthesizer SubBlocks Implementation
5.4.1 The Phase-Frequency Detector
5.4.2 The Charge Pump
5.4.2.1 Dead-Zone Nonlinearity
5.4.2.2 Linear Range and Cycle Slipping
67
67
69
69
71
71
73
76
78
Contents
vii
111
6.1 Introduction
6.2 Overview
6.3 DeltaSigma-Controlled Adaptive Charge Pump
6.3.1 PLL Gain and Phase Variations
6.3.2 Charge Pump System
6.4 Synthesizer Loop Calibration
6.5 Process Calibration I/C Slew Rate and RC Time Constant
6.6 VCO Tuning Gain Calibration
6.6.1 VCO Calibration Algorithm Description
6.6.1.1 'N Values
6.6.1.2 Summary of Tuning Algorithm Operation
6.7 Improved VCO Band Switching
6.8 Experimental Results
6.9 Comparison with Published Results
6.10 Conclusion
111
111
113
113
116
117
119
121
121
125
125
126
127
128
128
131
131
132
APPENDIX A
PHASE-FREQUENCY DETECTORS AND CHARGE PUMPS
135
135
1 Phase-Frequency Detectors
2 Charge Pump
3 PFD/CP Characteristics
135
138
140
viii
Contents
APPENDIX B
CONTROLLED OSCILLATORS
1 Reference Oscillators
1.1 Voltage-Controlled Crystal Oscillator
1.2 Temperature-Compensated Crystal Oscillator
2 Voltage-Controlled Oscillators
2.1 Voltage-Controlled Oscillators: Phase Noise Analysis
2.2 VCO Design Methodology
2.2.1
VCO Design
2.2.2
Phase Noise Optimization
143
143
143
143
145
146
146
149
150
153
APPENDIX C
PHASE NOISE
157
157
1
2
3
4
157
159
161
163
APPENDIX D
FREQUENCY DIVIDERS
1 Reference Divider
1.1 Synchronous Dividers
1.2 Asynchronous Reference Frequency Divider
2 Feedback Divider
2.1 Specification and Different Architecture Evaluation
2.1.1
Direct Division versus Prescaler Method
3 High-Speed CMOS Divider Design
3.1 Current-Mode Logic Design: An Overview
4 Implemented CML Gates
165
165
165
165
166
168
168
168
177
178
183
APPENDIX E
PROGRAMS AND CODES
1 MathcadTM Program used for the Simulations
of all the Mathcad Figures
2 MatlabTM Program used for the Simulations
of the Fractional-N PLL Noise Spectrum
187
187
INDEX
207
187
198
Preface
ix
Preface
NOMENCLATURE
DC
DCOC
DFF
DGA
Div
DMD
D-S
DSC
DSSS
EVM
F
FDC
FOM
IC
IF
Inv
IP
L(f)
LF
LPF
LTI
LTV
LUT
MC
MLF
abstol
AC
ADC
ASIC
Balun
BER
CCK
CLA
CML
CP
DAC
dBc/Hz
xi
xii
Nomenclature
MMD
NTC
PA
PCB
PFD
PLL
PRBS
PSD
PSDMD
PVT
QAM
QPSK
Ref
Multimodulus Divider
Negative Temperature Coefficient Capacitor
Power Amplifier
Printed Circuit Board
Phase-Frequency Detector
Phase-Locked Loop
Pseudorandom Binary Sequence
Power Spectral Density
Phase Switching Dual-Modulus Divider
Process Voltage Temperature
Quadrature-Amplitude Modulation
Quadrature Phase-Shift Keying
Reference
Reltol
RF
RFIC
Rx
SS
SSB
SSBN
SSBPSD
TCXO
TDC
TFF
TRx
Tx
VCO
VCXO
VGA
Vtune
WLAN
TM
Chapter 1
INTRODUCTION
Outline and Contributions
1.1
INTRODUCTION
Chapter 1
the optimum design and implementation of the designed and implemented frequency synthesizer chips described in this book.
The behavioral modeling for the proposed fractional-N delta
Sigma ('6-based PLL is carried out in chapter 4 to check for
architectural limitations, identify dominant noise sources, automate
loop filter optimization, and generate phase-frquency detector/charge
pump (PFD/CP) linearity specifications. Also, a phase-domain model
of the proposed architecture is constructed using The CadenceTM
Verilog-A Language. The model combines the voltage-controlled
oscillator (VCO), reference, and divider integrators into one resettable
integrator within the PFD. The '6 modulator model is also included.
The divider adds '6 noise to the frequency variable, then divides the
sum by the average divide ratio. The simulation results obtained in
this chapter contribute to the optimum design and implementation of
fractional-N synthesizers presented in this monograph.
In chapter 5, simulation results presented in chapter 4 are used for
the optimum chip design of a multimode frequency synthesizer for the
WLAN standards. Unconditionally stable '6 modulators of the
third-order (namely MASH-1-1-1) are implemented and employed in
a PLL fractional-N synthesizer providing a good average estimate for
fractional-N dividers. Using a deep submicron 0.18 Pm complementary
metal-oxide semiconductor (CMOS) process with a supply voltage of
1.8 V, a '6-based fractional-N synthesizer is designed, simulated,
laid out, fabricated, and tested.
In chapter 6, additional circuit designs are proposed and
incorporated to enhance the performance of the synthesizer at the cost
of increased circuit complexity. Those additions include adaptive
charge pump (CP) architecture to maintain loop gain and phase
transfer functions while operating in fractional mode. Another circuit
proposal is that of an adaptive band switching control to maintain
frequency agility while offering optimum phase noise performance in
the band of interest.
The conclusion in chapter 7 wraps up the research monograph by
describing the achievement of the work presented and offers
suggestions for future work.
1.2
RESEARCH CONTRIBUTION
The main contributions of the work presented in this book are the
research, study, design, and implementation of two fractional-N
Introduction
Frequency pulling of the VCO is the frequency change due to nonideal load., i.e. change in
the load causes frequency change in the VCO (hence the term pulling). This is most severe
when the PA/TX frequency is directly related to the VCO frequency.
Chapter 1
Introduction
REFERENCES
[1] J.A. Weldon, et al., A 1.75 GHz Highly-Integrated Narrow-Band
CMOS Transmitter with Harmonic-Rejection Mixers, IEEE Solid-State
Circuits Conference, 2001, Digest of Technical Papers, 2001 IEEE
International 57 Feb. 2001, pp. 160161, 442.
Chapter 2
WIRELESS COMMUNICATION SYSTEMS
An Overview
2.1
INTRODUCTION
With the turn of the new millennium, there has been an explosion in
the usage of wireless equipment. Wireless cellular devices like mobile
phones are now ubiquitous. Mobile Internet employing wireless
devices is now available in most coffee shops in many countries. Such
devices can achieve short-and medium-range coverage. Devices with
short-range coverage are based on communications standards like
Bluetooth [1], while medium-range coverage devices are based on the
wireless local area networks (WLAN) communications standards [2].
The latter are categorized under two bands 2.4 and 5 GHz. The lower
band is mainly based on the 802.11b standard and employs the
complementary code keying (CCK) modulation [2]. The higher band
(5 GHz) is based on the 802.11a standard and employs the orthogonal
frequency division multiplexing (OFDM) modulation. Recently, the
same OFDM modulation has been ported to the lower frequency of
2.4 GHz of 802.11b. This standard was termed 802.11g.
WLAN devices are now used in most indoor places from homes to
shops to offices. Such devices are incorporated in appliances like televisions, laptops, computers, telephones, PDAs, printers, etc. Figure 2-1
illustrates a typical usage where coverage is also shown.
The synthesizer described in this book is designed and implemented
for direct incorporation in to radio frequency integrated circuit (RFIC)
transceivers for all the WLAN standards (802.11a, b, and g). In this
chapter, a brief overview of those standards is given. An adequate
transceiver system for such standards is also described.
Chapter 2
Printer
Television
Computer
Internet
Laptop
WAN
Access Point
LAN
Computer
Telephone
PDA
Cellular
Tower
2.2
Figure 2-3. OFDM Physical Layer Frequency Channel Plan Implementation for US Standards
10
Chapter 2
The 802.11a standard requires a 16.6 MHz pass band for one
operating network. The modulation technique allowed in 5 GHz
(orthogonal frequency division multiplexing (OFDM) with 64QAM
subcarriers) is more efficient than the spread spectrum techniques
WLAN uses (more bits/second/hertz), and provides up to 54 Mbps of
data rate to network users. For the US-based 802.11a standard, the 5
GHz unlicensed band covers 300 MHz of spectrum and therefore
supports 12 nonoverlapping, simultaneously operating networks, as
shown in Figure 2-3.
The 802.11g occupies similar frequency bands as the 802.11b,
however the single subcarrier is OFDM modulated as in the case of
802.11a. A summary of the main characteristics for all the WLAN
standards is shown in Table 2-1.
Table 2-1. WLAN Characteristics
Parameters
Frequency in GHz
802.11b
2.42.58
802.11g
2.42.58
Modulation
Data rate (Mbits/s)
DSSS CCK
1, 2, 5.5 & 11
Receiver sensitivity
(dBm)
Transmit power
(dBm)
Transmit EVM (dB)
74 @ 11 Mbps
OFDM
6, 9, 12, 18, 36, 48 &
54
65 @ 54 Mbps
802.11a
4.914.99 & 5.03
5.24 Japanese Bands
5.155.25 Lower
5.255.35 Middle
5.7255.825 Upper
OFDM
6, 9, 12, 18, 24, 36,
48 & 54
65 @ 54 Mbps
30
30
16, 23 & 29
25
25
2.3
11
12
2.3.1
Chapter 2
The Transmitter
2.3.2
The Receiver
Figure 2-7 shows the receiver chain block diagram. The shared
receiver base-band section consists of a buffer amplifier, a digitally
controlled VGA, an LPF and an output buffer (A2) whose gain can be
adjusted. The base-band section of the Q-channel is identical, not
shown here for brevity.
The channel selection LPF is designed for its rapid increase of
attenuation outside the pass band, and thus a narrow transition band. A
nominal cutoff frequency of 8.7 MHz is automatically calibrated to
account for process variations. This satisfies the channel selection
requirement for 802.11a, b, and g standards. Due to a significantly
13
2.3.3
14
Chapter 2
REFERENCES
[1] IEEE Bluetooth Drafts Standards, http://grouper.ieee.org/groups/Bluetooth/
[2] IEEE 802.11 Drafts Standards, http://grouper.ieee.org/groups/802/11/
[3] G. Chien et al., A 2.4 GHz CMOS Transceiver and Baseband Processor
Chipset for 802.11b Wireless LAN Application, IEEE International
Solid-State Circuits Conference Digest of Technical Papers, Feb. 2003,
pp. 356357.
[4] B. Razavi and P. Zhang, Mixer Noise Reduction Technique, US
Patent, 6,748,204, June 2004.
[5] J.A. Weldon, et al., A 1.75 GHz Highly-Integrated Narrow-Band
CMOS Transmitter with Harmonic-Rejection Mixers, IEEE Solid-State
Circuits Conference, 2001. Digest of Technical Papers, 2001 IEEE
International 57 Feb. 2001, pp. 160161, 442.
Chapter 3
PHASE-LOCKED LOOP FREQUENCY
SYNTHESIZERS
Principles, Analyses, and Design
3.1
INTRODUCTION
3.2
15
16
Chapter 3
the phase error between phases )samp and )feed. When passive loop
filter is employed, a CP is used to convert the voltage to current. The
CP phase error )CP then drives the loop filter. Other high-frequency
components also present at the output of the CP are removed by the
loop filter. The phase error at the output of the filter, )filt controls the
input voltage of the oscillator to obtain the frequency of interest fout
with a phase )out. A brief description of the PLL subblocks is listed
below.
VCXO
fref
)ref
f
Reference samp
PFD
Divider )
samp
Charge
Pump
)feed
ffdbk
Icp
fout
VCO
)CP
Z(s)
)filt
Kvco/s
)out
Divider
N
3.2.1
The main blocks used in the PLL are briefly described below.
Detailed description of those blocks is included in the appendices.
3.2.1.1
Phase-Frequency Detector
The PFD [2] compares the divided down reference signal with the
divided down feedback signal to generate a signal proportional to the
phase error. Several types of frequency detectors are used in PLLs
[13]; however, the most commonly used is the PFD as it offers both
phase and frequency comparison. A conventional PFD is shown in
Figure 3-2.
The timing diagram for this PFD for the case of a reference signal
lagging the VCO feedback signal is shown in Figure 3-3. The up and
down pulses shown control the source and sink currents that charge or
discharge the loop filter capacitor as described below. A detailed
description of this PFD as well as other types of PFDs is included in
Appendix A.
17
'1'
Ref
D
R
'1'
VCO
D
R
UP
DN
Ref
VCOf
Dn
Up
Reset
3.2.1.2
Charge Pump
18
Chapter 3
Vsup
Vsup
Iup
Iup
Iup
Up
Up
Up
Icp
Icp
Icp
Dn
Dn
Dn
Zs
Idn
Zs
Zs
Idn
Idn
3.2.1.3
Voltage-Controlled Oscillator
19
Dividers
There are two types of dividers. The ones used as reference dividers;
those are usually low frequency. The others are high- frequency
dividers and those are used as feedback dividers. Detailed circuit
topologies of synchronous and asynchronous, as well as dual-modulus
prescaler dividers are found in the Appendix.
3.3
K d I CP Z ( s )
K VCO 1
s N
(3.1)
AClosedloop ( s )
)out
)samp
AOpenLoop ( s )
1 AOpenLoop ( s )
(3.2)
20
Chapter 3
ICP
R3
R2
C1
C2
R4
C3
Vtune
C4
Figure 3-5. A typical Second-Order Loop Filter with Optimal third-and-fourth order Spur
Cancelation Network
The transfer function for the loop filter shown in Figure 3-5 is
given by:
Z (s)
1 sC 2 R2
s C1 C 2 sR2 C1C 2
(3.3)
AOpenLoop ( s )
fn
(3.4)
(3.5)
Equations (3.4) and (3.5) are the main equations used in the design of
the optimum loop filter.
3.3.1
Using equations (3.4) and (3.5), values for the second-order loop filter
components can be easily derived. Those are shown below after some
algebraic manipulation [14]
C1
2 2
K VCO T1 1 Zp T2
I CP K d
N Zp2T2 1 Zp2T12
21
(3.6)
C2
C1 2 1
T1
(3.7)
R2
T2
C2
(3.8)
and
where
T1
sec(I ) tan(I )
(3.9)
Zp
and
T2
(3.10)
T1Zp2
For a stable loop, a good phase margin must be between 45o and
60 . The damping ratio of the loop is also given here as a function of
the phase margin
o
(tan(I )) 4
2
16(1 (tan(I )) )
0.25
(3.11)
For a 56o phase margin the damping ratio is equal to 0.55. The other
values R3, C3, R4, and C4 can be selected to reject the PFD feed-through
frequency signals. Other equations for third- and fourth- order loop filters
could be found in [14].
22
Chapter 3
CASE STUDY
The case presented here is for a possible usage in the WLAN
standard. The specified parameters are shown in Table 3-1.
Table 3-1. Phase-Locked Loop Specified Parameters
Parameter
Synthesized frequency
Sampling frequency
VCO gain
Charge pump gain
Loop bandwidth
Phase margin
Value
1.72 GHz
40/3 MHz
100 MHz/V
2 mA
100 kHz
56o
Value
11.6 nF
447 :
1.2 nF
129
0.49 Ps
5.2 Ps
Figure 3-6 shows the open-loop gain and phase transfer functions
for the design of the PLL, whereas Figure 3-7 shows the closed-loop
gain and phase transfer functions. It can be seen from Figure 3-6 that
the gain drops to 1 (0 dB) at the specified LBW frequency (100 kHz)
and the phase is at its peak of 124o which corresponds to a phase
margin of 56o (180o124o).
The loop filter transfer function Z(s) is also plotted in Figure 3-8.
The gain of Z(s) shows the change in the 20 dB/decade slope for fp /10
(10 kHz) and 10fp (1 MHz).
23
200
100
20log Aol (f)
100
200
100
1 10
1 10
1 10
1 10
1 10
1 10
100
120
180
arg (Aol (f))
140
S
160
180
100
1 10
1 10
1 10
1 10
24
Chapter 3
Closed-Loop Gain Transfer Function
5
100
1 10
1 10
1 10
1 10
1 10
50
180
arg (Acl (f))
S
100
150
100
1 10
1 10
1 10
1 10
1 10
25
20log(|Z(f)|)
180
arg(Z(f))
50
100
100
1.103
1.104
1.105
1.106
1.107
26
Chapter 3
[ R2C2Zn
f 2 ( f1 f 2 )e [Z n t cos Zn 1 [ 2 t
sin Zn 1 [ 2 t
1[ 2
(3.12)
LockTime
tol
ln
1 [ 2
( f 2 f1 )
[Zn
(3.13)
A classical model for the settling time for the closed-loop PLL
with the second-order loop filter is shown in Figure 3-9.
27
3.4
Ipd
1/R
)r
ICP
_
F(s)
)vco
Kv
1/s
)out
6
1/N
)n
Figure 3-10. Noise Contributions in the Phase-Locked Loop System
Table 3-3. Gain and Noise Terms and their Units
Gain terms
Reference Divider
R
Charge pump
Gain ICP
Loop filter Z(s)
VCO tuning gain
KVCO
PLL feedback
Divider N
Gain units
No units
Amps/rad
:
rad/V
Noise terms
Reference oscillator
Phase noise )r
Charge pump
Current noise Ipd
Loop filter Voltage
noise Vf
VCO phase Noise
)VCO
No units
Feedback Divider
phase Noise )n
Output phase Noise
)out
Noise units
rad-rms or radrms/Hz
Amps or Amps/Hz
Volt or Volt/Hz
rad-rms or radrms/Hz
rad-rms or radrms/Hz
rad-rms or radrms/Hz
The notations used in Figure 3-10 are listed below. Gain terms and
their units, as well as phase noise and their units are included (Table
3-3). From basic control theory, it is easy to determine the transfer
28
Chapter 3
) out
)r
I CP K VCO Z ( s )
s
I CP K VCO Z ( s )
1
sN
(3.14)
K VCO Z ( s )
s
I CP K VCO Z ( s )
1
sN
(3.15)
K VCO
s
I CP K VCO Z ( s )
1
sN
(3.16)
1
I K Z (s)
1 CP VCO
sN
(3.17)
I CP K VCO Z ( s )
s
I CP K VCO Z ( s )
1
sN
(3.18)
The output noise power is the product of the input noise power and
the magnitude squared of the transfer function. The total output noise
power is calculated by summing the output noise power contributed
from each noise source (assuming the noise sources are uncorrelated).
29
3.4.1
L(f)
(dB)
1/f3
1/f2
1/f1
fc32
fc21
1/f0
fc10
log(f)
Figure 3-11. Phase Noise slopes for White Noise (1/f 0 ) , Flicker Noise (1/f 1 ) , Oscillator Noise
in the Thermal Region (1/f 2), and Oscillator Noise in the Upconverted Flicker Noise Region
(1/f 3 ) .
30
Chapter 3
Noise power spectra for some components (such as the loop filter
resistors) can be directly calculated with good accuracy. More complex
noise sources, such as oscillator phase noise, must often be measured
in order to provide accurate results. Measured noise spectra can be
characterized by noting the actual noise power at a certain frequency,
the 1/f n region in which that frequency lies, and the corner
frequencies for the different 1/f n regions.
3.4.1.1
Oscillators generally have 1/f 3, 1/f 2, and white noise (1/f 0) regions.
Phase noise models and simulations (e.g. Leesons equation [15],
impulse sensitivity functions [18], and periodic noise analysis [19])
can be used to estimate the phase noise spectra, but direct
measurements are preferred. Degradations to the VCO and reference
noise from the buffering and divider circuits should be included, as
well. For example, it is difficult to obtain noise floors below 145
dBc/Hz without significant effort at reference frequencies around
2030MHz.
3.4.1.2
The CP current sources have 1/f and white noise (1/f 0 ) regions, which
can be estimated with reasonable accuracy from simulations (e.g.
SPICE). A complication that arises in the CP PLL is the aliasing effect
caused by the periodic switching of the CP current. In lock, the CP has
a duty cycle determined by delays in the phase detector, leakage in the
loop filter, and other systematic design choices. While periodic noise
simulations should be used to accurately estimate the net noise power
spectrum coupled into the loop filter, in some cases a good
approximation is obtained by attenuating the CP current noise by the
(average) duty cycle of the CP pulses. The noise power is attenuated
by multiplying it by the average duty cycle.
3.4.1.3
For the loop filter of Figure 3-5 we calculate the voltage noise present
at the output from resistors R2 and R3 (a third-order loop filter). In
general, a noisy resistor is modeled as an ideal resistor of the same
value in series with a noise voltage generator [20]. The noise voltage
density is given by:
v2
31
(3.19)
4kTR
(4kTR2 )
Vo
VR3
3.4.1.4
C1
(3.20)
C1 ( sR3C3 1) ( sR2C1 1)( sR3C2C3 C2 C3 )
(4kTR3 )
sR2C1C2 C1 C2
(3.21)
C3 ( sR2C1 1) ( sR3C3 1)( sR2C1C2 C1 C2 )
S) ,in_divider ( f ) 1014.7
1016.5
2
N
f
(3.22)
+( f )
4S 2 't 2 f 02
1
SIout ( f ) 10 log10
[dBc/Hz]
2
fs
(3.23)
32
Chapter 3
Symbol
Lref
Value
143
Units
dBc/Hz
Lrefdiv
173
dBc/Hz
Lpd
216
dBc/Hz
Lvco
Ldiv
159
173
dBc/Hz
dBc/Hz
33
loop
bandwidth
logic plateau
noise
Filter roll
off
20log(fout/fref)
20log(fout/fsamp)
logic noise
VCXO phase
noise
1/f3 LBW
1/f2 frequency
offset
The MathcadTM simulation results obtained using the data from the
case study and data from Table 3-4 are shown in Figure 3-13. They
correlate well with the conceptual phase noise contributions of Figure
3-l2.
34
Chapter 3
3.5
FRACTIONAL-N SYNTHESIZERS
From the study of noise in the previous section, it was shown that the
noise improves if a higher sampling frequency is used [8, 9]. That
results in the usage of fractional division ratio to satisfy the output
VCO frequency and frequency step of interest. Frequency synthesizers
employing such fractional dividers are called fractional-N frequency
synthesizers. Figure 3-14 shows a conventional fractional-N frequency
synthesizer. Early Implementation of the fractional dividers employs a
digital accumulator [12] that controls a dual-modulus divider.
The synthesizer shown in Figure 3-14 is termed first-order
fractional-N frequency synthesizer. The fractional divider is composed
of two parts: the integral part N and the fractional part F and is often
35
fref
1/R
fsamp
PFD
VCO
f out
DMD
N/N+1
overflow
Frac
latch
fsamp u N .Frac
fsamp u N
F
(3.26)
36
Chapter 3
fref
Reference
Divider
fsamp
Charge
Pump
PFD
ffdbk
VCO
fout
Divider
'6
Modulator
N.Frac
3.5.1
In this section, '6 modulators of third order are described as they are
unconditionally stable [57, 11]. This type of modulators is often
termed MASH-1-1-1 modulators as they incorporate three first-order
modulators in parallel. Figure 3-16 shows a basic block diagram of
this type of modulator.
Typical time-domain output of this modulator is obtained in
Matlab Simulation for the case of a fractional divisor of 0.835 and is
shown in Figure 3-17.
As can be seen from Figure 3-17, the '6 modulator emulates the
average fractional part of the divider into instantaneous several
integral levels. Those vary in a random manner between 7 and 8 for
this order of modulator. In general, for Norder MASH modulator, the
output would vary between ( 2 Norder 1 ) and 2 Norder [21]. The fractional
divider N.Frac would then be emulated by instantaneous dividers that
vary in a random manner between (N( 2 Norder 1 )) and ( N 2 Norder )
37
such that the average value of the sequence of those divisors is equal
to the desired fractional-N divider value.
X(n)
E 1(n)
Z -1
Y(n)
-E 1(n)
Z -1
E 2 (n)
Z -1
-E 2 (n)
Z -1
E 3 (n)
Z -1
(2S ) 2
12 f samp N .Frac
[2sin(
Sf
fsamp
38
Chapter 3
MASH-1-1-1 Output
-2
-4
5200
5205
5210
5215
5220
5225
5230
5235
39
P OW E R S P E CTRUM (dB )
-50
-100
-150
-200
-250
100
200
300
400
500
600
FREQUENCY (KHz)
700
800
900
Figure 3-18. Typical SSB Power Spectral Density of third- Order MASH1-1-1
'6Modulator
3.5.1.1
40
Chapter 3
Value
1.725 GHz
40 MHz
100 MHz/V
2 mA
100 kHz
56o
Using [14], the values for the third-order loop filter components
are obtained. Those are shown in Table 3-6.
Table 3-6. Loop Filter-Designed Parameters
Parameter
Capacitor C2
Resistor R2
Capacitor C1
Main divider
Capacitor C3
Resistor R2
Value
34.8 nF
150 :
3.6 nF
43.125
366 pF
25 :
41
-70
-80
-90
-100
-110
-120
-130
-140
-150
-160
-170
2
10
10
10
10
Offset Frequency (Hz)
10
10
10
3.6
Irms
f2
+( f )
10
f1
10
df
(3.28)
42
Chapter 3
(-1,1)
(1,1)
EVM
Irms
I
(-1-1)
(1,-1)
Figure 3-20. RMS Phase Error and Error Vector Magnitude in a QPSK System
3.7
CONCLUSION
43
REFERENCES
[1] W.F. Egan, Modeling Phase Noise in Frequency Dividers, IEEE
Transactions on Ultrasonics, Ferroelectrics and Frequency Control, 37
(4), pp. 307315, July 1990.
[2] J.A. Crawford, The Phase Frequency Detector, R.F. Design, Feb.
1985, pp. 4657.
[3] I. Thompson and P.V. Brennan, Phase/Frequency Detector Phase
Noise Contribution in PLL Frequency Synthesizer, IEEE Electronics
Letters, July 2001, 37 (15), pp. 939940.
[4] I. Thompson and P.V. Brennan, Phase Noise Contribution of the
Phase/Frequency Detector in a Digital PLL Frequency Synthesizer,
IEE Proceedings on Circuits, Devices and Systems, Feb. 2003, 150 (1),
pp. 15.
[5] T.A.D. Riley, M.A. Copeland, and T.A. Kwasniewski, DeltaSigma
Modulation in Fractional-N Frequency Synthesis, IEEE Journal SolidState Circuits, 28, pp. 553559, May 1993.
[6] B. Miller and B. Conley, A Multiple Modulator Fractional Divider,
Proceedings of IEEE 44th Annual Symposium Frequency Control,
1990, pp. 559567.
[7] B. Miller and B. Conley, A Multiple Modulator Fractional Divider,
IEEE Transactions of Instrumentation Measurement, 40, pp. 578583,
June 1991.
[8] V. Manassewitsch, Frequency Synthesizers, Theory and Design, 3rd
edn., Wiley: New York, 1987.
[9] R.E. Best, Phase Locked-Loop Design Simulation and Applications, 3rd
edn. MacGraw-Hill, New Jersey, 1997.
[10] F.M. Gardner, Charge-Pump Phase Lock Loops. IEEE Transactions
on Communication, COM-28:18491858, Nov. 1980. Description of
the charge pump mechanism in a PLL.
[11] B.-G. Goldberg. Digital Techniques in Frequency Synthesis, MacGrawHill, New Jersey, 1996.
[12] B.-G. Goldberg, Analog and Digital Fractional-n PLL Frequency
Synthesis: A Survey and Update, Applied microwave and wireless,
June 1999. Tutorial presenting fractional-N frequency synthesis.
[13] P.V. Brennan, Phase-Locked Loops, Principles and Practice, McGrawHill, New Jersey, 1996.
[14] D. Banerjee, PLL Performance, Simulation and Design, 3rd edn.,
2003, National Semiconductor, (http://www.national.com/appinfo/wireless/
files/Deansbook3.pdf)
[15] D.B. Leeson, A Simple Model of Feedback Oscillator Noise Spectrum,
Proceedings of the IEEE, MI, 54 (2), pp. 329330, 1966.
44
Chapter 3
[16] C.-H. Chen, et al., Direct Calculation of the MOSFET High Frequency
Noise Parameters, Proceedings of the 14th International Conference
on Noise and Physical Systems and 1/f Fluctuations, pp. 488491, July
1997.
[17] W. Liu, et al., RF MOSFET Modeling Accounting for Distributed
Substrate and Channel Resistance with Emphasis on the BSIM3v3
SPICE Model, IEDM Technical Digest, pp. 309312, Dec. 1997.
[18] A. Hajimiri and T.H. Lee, A General Theory of Phase Noise in
Electrical Oscillators. IEEE Journal of Solid-State Circuits, 33 (2), pp.
179194, Feb. 1998.
[19] Cadence Design Systems, Periodic S-Parameter and Noise Analysis
using SpectreRF PSP/PNOISE Analyses, Application Notes and White
Papers, http://www.cadence.com/whitepapers/pspapn1.pdf
[20] A. Mehrotra, Noise in Radio Frequency Circuits: Analysis and Design
Implications, International Symposium on Quality Electronic Design,
ISQED San Jose, Mar. 2001.
[21] M. Kozak, I. Kale, A. Borjak, and T. Bourdi, A pipelined All-Digital
DeltaSigma Modulator for Fractional-N Frequency Synthesis, IEEE
Instrumentation and Measurement Technology Conference (IMTC
2000), Vol. 2, pp. 11531157, Baltimore, MD, May 2000.
Chapter 4
SYSTEM SIMULATION OF '6-BASED
FRACTIONAL-N SYNTHESIZERS
Efficient Modeling and Characterization
4.1
INTRODUCTION
45
46
Chapter 4
4.2
PHASE-DOMAIN MODEL
Tuning
Curve
PFD/CP
LUT
)
40 MHz
= 40V
Integrator
CP
Divider
'6 MMD
Frequency
VCO
Frequency
VCO
phase
47
48
Chapter 4
49
The PFD/CP linearity curve, whether taking into account the deadzone or not, is provided by a lookup table immediately after the PFD.
One case where the PFD suffers from the dead-zone [4, 5] is presented
in Figure 4-3. The non-linear PFD data could be easily stored in a file
and loaded within the used block. The output of this lookup table is a
voltage driving the subsequent CP that has normalized up and down
currents yielding the desired CP values.
The output of the CP current creates a voltage at the output of the
loop filter that in turn drives the VCO.
The VCO gain characteristic can either be a lookup table obtained
by either simulating or measuring the gain of the VCO or a
polynomial generated by curve fitting the characteristic gain curve [8].
The gain curves of this multiband VCO are shown in Figure 4-4. The
output of the VCO is a frequency which is represented by a voltage
value in this proposed model. To obtain a phase value at the output of
the VCO, it is mandatory to use an integrator as shown in Figure 4-2.
What follows is a brief description of the behavioral models of each
constituent blocks in the synthesizer.
Figure 4-4. Multiband VCO Tuning Characteristics Showing the Frequency (a Voltage in the
Phase-Domain Model) Versus the Tuning Voltage
50
Chapter 4
4.2.1
4.2.1.1
Fref
The noise term has two parts as shown in the equation above. The
first term is taken at each reference sample time. It should be noted
that the noise here is frequency noise. f O is the offset frequency
where the measured PSD Lref ( f O ) is read. The normal distribution
and the white noise functions are Verilog-A built-in proprietary
functions.
4.2.1.2
The '6 modulator and the feedback divider are treated jointly. The
combined model represents the model of the desired fractional
divider. The third-order '6 MASH modulator model is derived by
employing the sampled difference equations of each node. The
modeled '6noise; mainly quantization noise; enters the loop linearly
after passing thorough a digital integrator [6]. Since integration is a
linear operation and since operation is at frequencies where the digital
integrator could be replaced with an analog integrator, both integrators
51
(divider and VCO) could be pulled into the PFD model, and hence the
'6 divider as well as the VCO operate on frequency rather than
phase.
'6quantization noise is straightforward to model in the time
domain; it just implements the difference equations. An equivalent
frequency-domain noise source for the third-order '6 MASH
modulator can be found from the z-domain expression for the output
noise. Consider the linearized noise transfer function and PSD of the
input quantization [2]:
H n ( z)
1 z
1 3
(4.2)
And the PSD = T / 12, where T is the clock period. The output
PSD is:
PSD. H n ( z )
(4.3)
The VCO
K VCO
F1 2 F2 u x 3F3 u x 2 4 F4 u x 3
(4.4)
52
4.2.1.4
Chapter 4
The PFD/CP
The noise sources of the PFD and CP are added as random noise
similar to the reference oscillator case [1]. The PFD/CP linearity curve
could also be included as a data file, showing the duty cycle versus
CP, to gauge the effect of the dead zone as shown in Figure 4-3.
4.2.1.5
The loop filter uses real components rather than a transfer function
and therefore the noise due to the resistors although negligible adds up
to the overall noise of the loop; thus their noise should be taken into
account when phase noise frequency-domain analysis is performed.
4.2.2
53
Figure 4-5. Open Loop of Synthesizer Constituent Blocks Obtained by Direct PSD Transform
4.3
54
Chapter 4
Figure 4-6. Time-Domain Simulation of the Phase-Domain Model of the Platform Showing
the Voltages at each Node in the Synthesizer
55
Value
270p
1000
2.5n
91
120p
91
120p
1 mA
91.65
56
Chapter 4
(a)
(b)
Figure 4-7. Voltages at Each Node in the Synthesizer during Lock (a) Unzoomed,
(b) Zoomed
57
Figure 4-8 also shows the presence of several spurious signals that
can be suppressed if further dithering is applied to the '6 modulator
as will be discussed in the ensuing section. These spurious signals are
due to close-to-integer operation.
Figure 4-8. Phase Noise Waveforms for the LO Synthesizer Obtained by PSD Transform for
Two Loop Bandwidth Cases (a) 300 kHz and (b) 1 MHz
4.3.1
Dithering Effect
58
Chapter 4
Figure 4-9. Power Spectral Density of DeltaSigma Noise Shaper with and without
Dithering
Figure 4-10. Power Spectral Density of DeltaSigma Noise Shaper with and without
Dithering, Log Scale
59
Figure 4-11. Phase Noise Mask for the LO Synthesizer with 300 kHz Loop Bandwidth with
and without Dithering
Figure 4-11 shows the phase noise profile for the frequency
synthesizer local oscillator at 5.5 GHz with a loop bandwidth of 300
60
Chapter 4
kHz. Two cases are superimposed. The first case (bottom trace) is when
no dithering is employed which shows the presence of spurious
fractional content whereas the top trace is the case where dithering is
applied. In this case, the spurious energy is spread across the spectrum
and hence the lifting of the phase noise as illustrated in the figure. This
effect must be taken into account when designing synthesizers to strike
a compromise between deterministic spurious noise and random phase
noise.
4.3.2
Close-to-Integer Operation
4.3.3
Noise Folding
One important factor that deteriorates the close-in phase noise and that
is easily seen in measurement but not proven by simulation is the
effect of noise folding [2]. The platform developed in this research
proves this phenomenon with ease. Figure 4-15 shows the effect of
noise folding due to the PFD/CP nonlinearity. The high-frequency '6
quantization is folded back to within the loop bandwidth. This is
illustrated in the 10 dB deterioration of the phase noise plateau
rendering the fractional-N synthesizer unattractive. Fortunately, the
developed platform outlines this problem and shows how to mitigate
Figure 4-13. Modulator Output for DC Input 0.998 with Dither Applied
61
62
Chapter 4
Figure 4-14. Effect of 1-LSB and 5-LSB Dithering for Close-to-Integer Divide Ratio
63
Figure 4-15. Effect of Noise Folding on phase Noise Profile Due to Noise Folding
Nonlinearity
Figure 4-16. Introduction of Offset Leakage current to Mitigate the Effect of Noise Folding
Due to Delta-Sigma Modulation
64
Chapter 4
)
40 MHz
= 40V
Tuning
Curve
Integrator
CP
Divider
'6 MMD
Frequency
Div-by-4
VCO
Frequency
VCO
phase
Figure 4-17. Phase-Domain Model for the Synthesizer Employing a Divide-by-4 Prescaler
Preceding the Main Feedback Divider
4.4
65
CONCLUSION
REFERENCES
[1] Affirma RF Simulator (SpectreRF) User Guide, An Introduction to the
PLL Library: How the PFD Model Works.
[2] M.H. Perrott, M.D. Trott, and C.G. Sodini, A Modeling Approach for
SigmaDelta Fractional-N Frequency Synthesizer Allowing Straightforward Noise Analysis. IEEE Journal of Solid-State Circuits, 37 (8),
Aug. 2002.
[3] K. Kundert, http://www.designers-guide.com/Analysis/PLLnoise+jitter.pdf
[4] J. Crawford, Frequency Synthesizer Design Handbook. Equation (7.81)
on page 349.
[5] B. De Muer and M. Steyaert. CMOS Fractional-N Synthesizers.
[6] J. van Engelen, R. van de Plassche. Bandpass Sigma Delta Modulators.
[7] N.M. Filiol, T.A.D. Riley, C. Plett, and M.A. Copeland An Agile ISM
Band Frequency Synthesiser with Built-In GMSK Data Modulation,
IEEE Journal of Solid-State Circuits, 33 (7), July 1998.
[8] D. Banerjee, PLL Performance, Simulation and Design, 3rd edn., 2003,
National Semiconductor, (http://www.national.com/appinfo/wireless/files/
Deansbook3.pdf )
Chapter 5
MULTIMODE '6-BASED FRACTIONAL-N
FREQUENCY SYNTHESIZER
5.1
INTRODUCTION
5.2
AN OVERVIEW
67
68
Chapter 5
fref
Reference
Divider
fsamp
Charge
Pump
PFD
ffdbk
VCO
fout
Divider
'6
Modulator
N.K
69
5.3
5.3.1
Design Methodology
The first step is to devise the architecture for the frequency synthesizer
that covers all the frequency ranges shown in the specification table
(Table 5-1). A scheme based on oscillator synthesis that generates
two-thirds of the desired frequency mixed with a divided-by-2 version
of it is employed for the 802.11 standards. This architecture is shown
below in Figure 5-2. As can be seen from Figure 5.2, the required
frequencies of interest are not directly generated by the respective
voltage-controlled oscillators. That was done to avoid frequency
pulling in the transceiver [17].
Table 5-1. Frequency Ranges for the Proposed Synthesized Architecture
Parameters
Frequency
B, g
(GHz) for
a
802.11
Current consumption (mA)
Supply voltage (V)
Locking time (us)
Minimum
2.4
4.8
1.7
Nominal
20
1.8
224
Maximum
2.5
5.805
1.9
70
Chapter 5
f802.11a
f802.11b,g
/2
40MHz
VCXO
PFD
VCO1
VCO2
LF
CP
/2
MMD
n 6
'6
30
5.
6.
7.
8.
71
5.4
72
Chapter 5
Ref
'1'
D
R
UP
Q
Q
Mux
delay
VCO
'1'
delay
set
D
R
DN
Q
Q
Up
Ref
delay
Enable
Dn
VCOf
73
5.4.2
The behavioral model of the CP was shown in Figure 3-4 and repeated
here in Figure 5-5. In this section, a detailed circuit design for a CP is
described. CP parameters that affect the performance of the PLL are:
UP and DOWN current mismatches, unequal rise and fall times,
glitches, and feed-through [20]. Several circuit simulations that
characterize the CP are shown.
Vdd
Iup
SwP
Icp
Dn
SwN
Zs
Idn
74
Chapter 5
Vdd
Up
P3
Vdd
N3
SwP2
SwP1
P2
P1
N4
N5
N2 Dn
SwN1
P8
P7
Iout
ip
P5
P6
'I
N9
Vdd
N1
P4
ip='I
Z(s)
N8
N6
N7
(a)
Vdd
Ip
+
Iout
Ib
Vss
(b)
Figure 5-6. Circuit Schematic of the Charge Pump (CP Block) used in the '6-based
Synthesizer (a) CP, (b) PushPull Op-amp in the CP circuit
75
Figure 5-7. UP, DN, and Charge Pump Currents versus Tuning Voltage
'I
I UP I DN
(5.1)
76
Chapter 5
Figure 5-8. PFD/CP Linearity Curve (the Dead Zone Shown inside the Eclipse)
77
result of UP and DN pulses driving reset logic. Those pulses drive the
CP switch transistors that have some time constants related to their
input capacitances. Those switches cannot respond to small pulses and
never fully switch ON depending on the speed of operation, hence the
dead-zone effect.
Figure 5-8 shows the dead-zone nonlinearity plot for an ideal PFD
(with no delay in the reset path) driving a real CMOS CP. The
reference frequency used is 40 MHz.
78
Chapter 5
79
80
Chapter 5
Figure 5-11. Charge Pump Block and Its Offset Current Block
81
In this simulation, the control signals of the circuit are varied and the
output current is changing accordingly from 0 to 100, 200, and 100 PA
for a CP current of 1 mA (CP<0> = 0) and 200, 400, and 200 PA
for a CP current of 2 mA (CP<0> = 1).
82
Chapter 5
Figure 5-14 shows the simulation setup of the PFD/CP circuit. In this
simulation, the output short circuit current of the CP is measured.
Figures 5-155-17 show the transient responses of the PFD and CP
outputs for three different cases in which the two inputs (a) have
almost the same phase, (b) the reference signal from the crystal is
leading, and (c) the output of the divider in the PLL is leading. As it
can be seen, there is always a positive glitch after the output is reset.
This is due to the extra delay in the signal that turns off the PMOS
switch of the CP and also the PMOS switch itself.
83
84
Chapter 5
85
5.4.3
86
Chapter 5
inverters. The ring oscillator usually has worse phase noise performance
[22] and is not suitable for high-performance design as the one
presented in this chapter. Figure 5-18 shows a current supplied LC
VCO used in the design. This VCO provides an in-phase and
quadrature-phase signals (namely I and Q) as it is used to drive an IQ
image rejection mixer. The structure is fully differential as it offers
better power supply rejection. The LC oscillator is constructed using
all PMOS transistors as PMOS provide better noise performance since
they have lower flicker noise compared to their NMOS counterparts
[23] (more on that in Appendix B).
Vdd
X1
X4
X4
Vdd
Vdd
X4
X4
750u
Q- I+
Q+
I+
I-
Q+
I-
Q-
Vcntl
Vss
87
88
Chapter 5
Figure 5-20 shows the simulated phase noise profile VCO. The
phase noise at 100 kHz offset from the carrier is 103 dBc/Hz and at
20 MHz is 156 dBc/Hz.
Figure 5-21 shows the simulated VCO tuning curves covering all
the VCO-synthesized frequencies for 802.11 standards. These are the
synthesized frequencies at point X in Figure 5-2.
89
The gain of the VCO is directly derived from the simulated data in
Figure 5-21 and is shown in Figure 5-22 for VCO1 and VCO2.
5.4.4
90
Chapter 5
5.4.5
Figure 5-24 shows the first-order linear model of the '6 modulator
and its hardware accumulator-based implementation. To establish the
link between the model and the hardware implementation, its time
domain behavior is first analyzed.
Using Figure 5-24, the time-domain equations of the '6modulator
are as follows:
u[n]
X b[n 1]
(5.2)
(5.3)
91
u[n]
X[n]
6
+ 6
-
v[n]
-1
-1
E[n]
b[n]
X(n) m
b(n)
X+Y
m
-E[n]
-E(n)=v(n)-b(n)
Latch
(b) Hardware Implementation
Figure 5-24. The First-Order modulator (a) Linear Model, (b) Hardware Implementation
b[n]
1 if v[n] t 0
1 otherwise
(5.4)
(5.5)
X b[n 1] v[n 1]
(5.6)
X E[n 1]
(5.7)
92
Chapter 5
(5.8)
b[n] 1
(5.9)
Hence E[n]
(5.10)
X Y
1 and
(5.11)
Accumulator content = E[ n] 1
5.4.5.2 The 30-bit Structural MASH Coder Implementation
6
X[n]
LATCH
C1[n]
X+Y
LATCH
C2[n]
-e1[n]
C3[n]
LATCH
X+Y
-e2[n]
LATCH
X+Y
Y
LATCH
93
Si X i Yi x ( X i x Yi ) Ci
Ci1 X i Yi x X i x Yi Ci
(5.12)
Where Si is the ith full sum of the ith input vectors and Ci + 1 is the
carry of the next operation.
94
Chapter 5
C in '0'
C in 8-bit C out
1-bit
Latch
x(23:0)
8-bit
Latch
24
s(23:0)
1-bit
Latch
16-bit
Latch
8-bit C
CLA out
C in
24
C in
8-bit C
CLA out C 1
clk
24 y(23:0)
Cout
(a)
1-bit
Latch
clk
Cin
8-bitC
CLA out C1
y(23:16)
8-bitC
CLA out
x(23:16)
clk
Cin
y(15:8)
1-bit
Latch
x(15:8)
y(7:0)
x(7:0)
8-bitC
CLA out
s(23:16)
s(15:8)
s(7:0)
(b)
Figure 5-26. Pipelined 24-bit Adder (a) for First-stage Accumulator, (b) for Subsequent
Stages
95
k 1
r 1
(1)
r 1
(5.13)
. D r . Ck n r 1
C1
C2
C3
C1 ( n ) C2 ( n ) C2 ( n 1) C3 ( n ) 2C3 ( n 1)
+ C3 ( n )2
Logic1
3-bit
Latch
+
C2(n-1)
C3(n-1)
C3(n-2)
Logic2
(5.14)
Yout(n)
CLK
96
Chapter 5
Table 5-2. Error Correction Network and Logic 1 and Logic 2 Truth Tables
C1
0
0
0
0
1
1
1
1
C2
0
0
1
1
0
0
1
1
C3
0
1
0
1
0
1
0
1
Out1
0
1
1
2
1
2
2
3
C2(n 1)
0
0
0
0
1
1
1
1
C3(n 1)
0
0
1
1
0
0
1
1
C3(n 2)
0
1
0
1
0
1
0
1
Out2
0
1
2
1
1
0
3
2
and Table 5-3 show the special 2s complement representation and the
special handling of number 4.
ML2
Map _ Log1
ML
1
C2 x C1 C3 (C3 x C1 )
C x C x C C x C C x C x C (C x C )
1
3
2
1
3
1
3
2 1 3
(5.15)
ML1
A
B
C
A B C A
B C B C
A B xC
x
x B
1 - b it
Latc h
cl
k
B
C
cl
k
1 - b it
Latc h
C3
C3
1 - b it
Latc h
1 - b it
Latc h
1 - b it
Latc h
C2
C2
cl
k
C1
1 - b it
L atc h
(5.16)
ML1
M a p p ing M L
2
L og ic 2
'0 '
X
Y
M L3
M a p p ing
L og ic 1
M L1
Y
M L2
'0 '
Y
C1
C o ut
A d d er
y3
C in
C o ut
A d d er
y2
C in
C o ut
A d d er
C in
y1
97
Yout(1)
0
0
1
1
0
0
1
1
Yout(0)
0
1
0
1
0
1
0
1
Mapping
000000
000001
000010
000011
000100
111101
111110
111111
98
Chapter 5
Figure 5-30. Power Spectral Density of DeltaSigma Noise Shaper with and without
Dithering (Linear Plot)
Figure 5-31. Power Spectral Density of DeltaSigma Noise Shaper with and without
Dithering (Log Plot)
99
One of the practical issues that are often overlooked in fractional-N PLL
designs is the problem of having to synthesize frequencies that are integer
multiples of the reference frequency, i.e. the divider value is an integer.
This becomes a problem if not catered for in advance in the design.
If the input is an integer, the '6 noise shaper input is zero.
Previously, we introduced an error to remove any fractional spurs
that may arise from limit cycles. That error will propagate in the
modulator and cause the accumulators to overflow in a determined
and cyclic manner causing spurious tones for integer frequencies.
Figures 5-32 and 5-33 show that the modulator behaves like a secondorder '6 with small DC input that eventually produces fractional spurs.
It is advisable to bypass the noise shaper when selecting to operate
in integer division in the synthesizer. In other words, only use the '6
noise shaper when operating the synthesizer in fractional mode.
Figures 5-34 and 5-35 show the spectral densities of the modulator
when the input is a DC with values of 0.998 and 0.005, respectively.
It can be clearly seen that for both cases, the modulator exhibits lowfrequency spurs due to insufficient dither. This problem is rectified
by effectively increasing the dither via the introduction of an initial
seed [1] in the modulator as illustrated in Figure 5-36.
100
Chapter 5
101
Figure 5-35. Modulator Output for DC input 0.005 Showing Low-Frequency Spurs due to
Insufficient Dither
Figure 5-36. Increasing Dither Suppresses the Spurs to Some Extent in the Fractional-N
Modulator for Close-to-Integer Divider Values
102
5.5
Chapter 5
MEASURED PERFORMANCE
OF THE IMPLEMENTED SYNTHESIZER
103
The measured band switching curves are shown in Figure 5-39 for
the VCO1 part. They correlate pretty well with those simulated and
shown in Figure 5-21.
104
Chapter 5
Figure 5-39. Measured Switching Curves for the Implemented VCO1 at LO Frequency
105
Figure 5-41. Phase Noise Profile of the Synthesizer for 802.11a, b, and g
106
Chapter 5
Figure 5-43. 64-QAM Constellation and EVM at the Output of the Transmitter
107
5.6
108
Chapter 5
Process
PLL
type
Int-N
Freq.
Rategh
2000
0.24
um
CMOS
Liu
2000
0.25
um
CMOS
Int-N
5
GHz
Zargari
2002
0.25
um
CMOS
Int-N
5
GHz
8
MHz
250
kHz
This
work
0.18
um
CMOS
FracN
56
GHz
40
MHz
400
kHz
4.8
5
GHz
PFD
freq.
11
MHz
Loop
BW
280
kHz
Lock
time
4 uS
PN @
Offset
101
dBc/Hz
@1
MHz
112
dBc/Hz
@1
MHz
112
dBc/Hz
@1
MHz
113
dBc/Hz
@1
MHz
Power
Vsup
25
mW
3V
180
mW
2.5 V
36
mW
1.8 V
109
high and the PFD/CP becomes active. In essence, noise cross talk is
avoided by time-domain isolation.
As a conclusion, in this chapter, the author has detailed the design
and performance of a multimode fractional-N synthesizer. The
synthesizer was designed as a local oscillator and constructed as part
of a complete direct conversion transceiver. The measured results
obtained for this synthesizer supersede most published results (see
Table 5-4). The developed platform of chapter 4 has helped the
designed synthesizer in achieving the best performance to date. In
chapter 6, we propose a new adaptive and enhanced synthesizer
architecture that offers optimum performance.
REFERENCES
[1] N.M. Filiol, T.A.D. Riley, C. Plett, and M.A. Copeland An Agile ISM
Band Frequency Synthesiser with Built-In GMSK Data Modulation,
IEEE Journal of Solid-State Circuits, 33 (7), July 1998.
[2] B. De Muer, M.S.J. Steyaert A CMOS Monolithic -Controlled
Fractional-N Frequency Synthesizer for DCS-1800, IEEE Journal of
Solid-States Circuits, 37 (7), pp. 835844, July 2002.
[3] Z. Shu, Ka Lok Lee, B.H. Leung A 2.4-GHz Ring-Oscillator-Based
CMOS Frequency Synthesizer with a Fractional Divider Dual-PLL
Architecture, IEEE Journal of Solid-States, 39 (3), pp. 452462, Mar.
2004.
[4] P. Zhang, T. Nguyen, C. Lam, D. Gambetta, T. Soorapanth, B. Cheng,
S. Hart, I. Sever, T. Bourdi, A. Tham, and B. Razavi, A Direct-Conversion
CMOS Transceiver for IEEE 802.11a WLANs, ISSCC Digest of
Technical Papers, pp. 354355, Feb. 2003.
[5] P. Zhang, T. Nguyen, C. Lam, D. Gambetta, T. Soorapanth, B. Cheng,
S. Hart, I. Sever, T. Bourdi, A. Tham, and B. Razavi, A 5-GHz DirectConversion CMOS Transceiver, IEEE Journal of Solid-State Circuits,
38 (12), pp. 22322238, Dec. 2003.
[6] P. Zhang, L. Der, D. Guo, I. Sever, T. Bourdi, C. Lam, A. Zolfaghari,
J. Chen, D. Gambetta, B. Cheng, S. Gower, S. Hart, L. Huynh, T. Nguyen,
and B. Razavi, A CMOS Direct-Conversion Transceiver for IEEE
802.11a/b/g WLANs, IEEE Custom Integrated Circuits Conference,
Digest of Technical Papers, Sect. 18, No. 4, Oct. 2004.
[7] T.A.D. Riley, M.A. Copeland, and T.A. Kwasniewski, DeltaSigma
Modulation in Fractional-N Frequency Synthesis, IEEE Journal of
Solid-State Circuits, 28, pp. 553559, May 1993.
[8] B. Miller and B. Conley, A Multiple Modulator Fractional Divider,
Proceedings of IEEE 44th Annual Symposium Frequency Control, 1990,
pp. 559567.
[9] B. Miller and B. Conley, A Multiple Modulator Fractional Divider,
IEEE Transactions on Instrumentation Measurements, 40, pp. 578583,
June 1991.
110
Chapter 5
Chapter 6
IMPROVED PERFORMANCE FRACTIONAL-N
FREQUENCY SYNTHESIZER
6.1
INTRODUCTION
6.2
OVERVIEW
Figure 6-1 shows the basic fractional-N architecture that was implemented in chapter 5. Several circuit techniques could be added to this
synthesizer architecture to enhance its performance.
111
112
VCX
O
Chapter 6
fref
Reference
Divider
fsamp
Charge
Pump
PFD
ffdbk
fout
VCO
Divider
'6
Modulator
N.K
Half-Band
Improved Band
Switching
VCXO
fref
/
R
fsamp
PFD
Noise Shaped
Charge
Pump
ffeedback
Loop
Filter
Vtune
VCO
Mult
i
Modulus
Divider
ffeedback
Noise
Shaper
Nfrac
fref
+
-
A/D
Controller
Reg1
Reg2
fout
113
6.3
DELTASIGMA-CONTROLLED ADAPTIVE
CHARGE PUMP
6.3.1
114
Chapter 6
The open loop gain of the PLL synthesizer of Figure 6-1 at the
averaged fractional divider ratio is given by:
Aolaverage
I cp F ( s )
K vco
N frac
(6.1)
Aolinst
I cp F ( s )
K vco
s N inst
N 2 m 1 1 N inst N 2 m 1
(6.2)
(6.3)
N 3 N inst N 4
(6.4)
The closed-loop equation for the PLL with the average case is
given by:
Aclaverage
Aolaverage
1 Aolaverage
(6.5)
and the closed loop equation for the PLL with the instantaneous case
is given by:
Aclinst
Aolinst
1 Aolinst
(6.6)
115
Notice here that since equations (6.1) and (6.2) yield different results,
equations (6.5) and (6.6) also yield different results. So the higher the
order of the '6 modulator, the larger the divider variations and
consequently the more loop gain and phase disturbances, as shown in
Figure 6-3.
3.5
Nmin
3
2.5
Ncenter
2
1.5
Nmax
1
0.5
10
-0.6
-0.8
Nmin
-1
Ncenter
-1.2
Nmax
-1.4
-1.6
10
Frequency (Hz)
116
Chapter 6
Aolinst
I cp(inst) F ( s )
K vco
s N inst
(6.7)
Aol
I cp F ( s )
K vco 1
s N
(6.8)
I N 4 I cp(inst) I N 3
(6.9)
6.3.2
Figure 6-4 shows the system overview of the control of the implemented
adaptive CP architecture. The coding method employed on the b bits
is thermometer coding as used in current-segmented digital-to-analog
converters [8]. The outputs of the thermometer encoders control two
banks of UP and DOWN CP circuitry.
117
Encoder 3to 7
bits
D
ck
UP
Q
Thermometer L<7:1>
Encoder 3to 7
b<2:0>
bits
IupMatrix
Encoder
Control Logic
& Unit Cells
Iup
ck
IdownMatrix
Encoder
Control Logic
& Unit Cells
DOWN
Idown
ck
6.4
118
Chapter 6
Aol
(6.10)
(6.11)
(6.12)
and,
RC
(6.13)
Aol
fI
I CP .K VCO
,
N nom.Cnom
119
(6.14)
1
(1 ' R ).(1 ' C )
(6.15)
fI
f IN . f IKvco . f IIC
(6.16)
Since N variations have been taken care of, equation (6.16) can be
written as:
fI
1 'I
(1 ' Kvco ).(
)
1 'C
(6.17)
6.5
120
Chapter 6
VBG
R
I ref
(6.18)
I ref .R
T
VBG
T
I ref
C
(6.19)
Therefore,
T
and,
RC
(6.20)
VBG
T
I
C
121
(6.21)
6.6
The VCO tuning gain (KVCO) differentiates the sensitivity of the VCO
output frequency (FVCO) to changes in its tuning voltage (Vtune). It is
defined as:
K VCO
dFvco
dVtune
(6.22)
6.6.1
The used tuning gain calibration system approximates the tuning gain
by measuring the difference in tuning voltage for a predetermined
difference in VCO frequency. This approximates the tuning gain with
finite number differences:
K VCO
dFvco
( F2 F1 )
#
dVtune (Vtune2 Vtune1 )
'F
'Vtune
(6.23)
122
Chapter 6
K vco_nom
'F
'Vtune_nom
(6.24)
and
K vco
'F
'Vtune
(6.25)
Hence,
K vco
K vco_nom
'Vtune_nom
(6.26)
'Vtune
'Vtune
ADBits2 ADBits1
(6.27)
where ADBits represents the binary output word of the ADC and the
'ADBits is the binary difference of final and first ADC readings.
For the nominal tuning gain, there is a nominal difference in ADC
output values, 'ADBitsnom. The ratio of the measured to the nominal
ADC value is then equal to the ratio of the nominal to the measured
tuning gain.
K vco_nom
K vco
'Vtune
'Vtune_nom
'ADBits
'ADBitsnom
(6.28)
123
Aol
I cp K vco
NC
'ADBits
I cpnom
K vco
'ADBitsnom
NC
(6.29)
After calibrating the loop, the effect of the loop gain Aol on the
loop should be the same as the golden nominal loop gain Aolnom. This
can be verified by examining the loop gains ratio, equation (6.30):
Aol
Aolnom
I cp .K vco
N .C
N .C
I cp_nom .K vco_nom
'ADBits
I cp_nom .
.K vco
'ADBitsnom
I cp o K v o
K vco_nom
I cp_nom .
.K vco
K vco
I cp_nom .K vco_nom
(6.30)
I cp_nom .K vco_nom
I cp_nom .K vco_nom
1
124
Chapter 6
$<1>
$<0>
$<n-1>
2n
To CP
switch
Icp_nom
Icpp
$nom
$<n-1>
$<n-1>
$<0>
Vbp
Vbp
2n
Icp_nom
UP
VDD
DN
Charge
Pump
Switches
Icp_nom
Iref
$<0>
Vbn
$<0>
Vbn
$<n-1> 2
$<n-1>
$nom
VSS
Figure 6-7. Charge Pump Employing Mirroring as in Figure 6-6 for VCO Gain Calibration
Figure 6-7 shows the CP circuit employing the mirror ratio. The
CP comprises a typical current reference cell, with the first current
branches mirrors weighted according to the expected 'ADBitsnom
125
(denoted ' nom in the figure), and the output mirrors binary weighted,
and switched according to the measured 'ADBits (denoted '<> for
simplicity).
The ' N values are chosen based on the nominal VCO gain KVCO_nom
and the nominal ADC output, 'ADBitsnom. Based on the ADC size
and the nominal KVCO, the nominal tuning voltage, 'Vtune_nom is
determined by:
'Vtune_nom
'ADBitsnom .VDD
2n
(6.31)
VDD is the supply voltage and n is the number of bits of the ADC
used in the algorithm. The nominal VCO frequency change is
determined by 'Fnom:
'Fnom
'Vtune_nom .K VCO_nom
(6.32)
'Fo
Fref
(6.33)
'N
(6.34)
To summarize the operation of the PLL synthesizer VCO gain calibration system, the calibration algorithm is outlined as follows:
x The channel frequency is set and hence determines the final divider
value, N.
x A 'N value is then chosen based on the ADC size and the nominal
VCO gain that is determined by design.
126
Chapter 6
Figure 6-8. Half-Band Tuning Shown in Terms of third Band from the Bottom
6.7
127
6.8
EXPERIMENTAL RESULTS
Figure 6-9. Phase Noise for the Enhanced Fractional-N PLL @ 2.4 GHz, Spurs are Shown
Transposed on Phase Noise Curve
128
Chapter 6
6.9
A summary of the achieved results obtained by measuring the fractional-N synthesizer is shown in the Table 6-1.
Table 6-1. Comparison with other Work
Ref.
Process
(CMOS)
PLL
Type
Freq.
(GHz)
PFD
Freq.
(MHz)
Loop
BW
(KHz)
[9]
[10]
[11]
This
work
0.24
0.25
0.25
0.18
Integer
5
5
5
2.45
11
280
8
40
250
280
6.10
Integer
Fractional
L (f) @
offset
(d Bc / Hz
@ 1 MHz)
101
112
112
119
Power
Usage
(mW)
Supply
Voltage
(V)
25
180
40
3
2.5
1.8
CONCLUSION
REFERENCES
[1] T. Riley, M. Copeland, and T. Kwasniewski, Delta-Sigma Modulation
in Fractional-N Frequency Synthesis, IEEE Journal of Solid-State
Circuits, 28, pp. 553559, May 1993.
129
Chapter 7
CONCLUSION AND FURTHER WORK
7.1
CONCLUSION
131
132
Chapter 7
7.2
FURTHER WORK
133
Data
Ref.
TDC
m-bits +
-
error
n-bits Filter
C-bits
Synthesized
Freq
NCO
TDC
Time to digital
Converter
Figure 7-1. An all Digital Frequency Synthesizer Example
134
Chapter 7
REFERENCES
[1] J.A. Weldon, et al., A 1.75 GHz Highly-integrated Narrow-band CMOS
Transmitter with Harmonic-rejection Mixers, IEEE Solid-State Circuits
Conference, 2001. Digest of Technical Papers, 2001 IEEE International
57 Feb. 2001, pp. 160161, 442.
[2] H.R. Rategh, H. Samavati, and T.H. Lee, A CMOS Frequency
Synthesizer with an Injection-Locked Frequency Divider for a 5-GHz
Wireless LAN Receiver, IEEE Journal on Solid-State Circuits, 35 (5),
May 2000.
[3] T.-P. Liu and E. Westerwick, 2- 5-GHz CMOS Radio Transceiver
Front-End Chipset, IEEE Journal Solid States Circuits, 35 (12), Dec. 2000.
[4] M. Zargari, D.K. Su, C.P. Yue, S. Rabii, D. Weber, B.J. Kaczynski,
S. Mehta, K. Singh, S. Mendis, and B.A. Wooley, A 5-GHz CMOS
Transceiver for IEEE 802.11a Wireless LAN Systems, ISSCC Digest of
Technical Papers, Feb. 2002.
[5] A. Mantyneemi, An Integrated CMOS High Precision Time-To-Digital
Converter Based on Stabilised Three-Stage Delay Line Interpolation,
Academic Dissertation, Faculty of Technology, University of Oulu, Dec. 3,
2004.
Appendix A
PHASE-FREQUENCY DETECTORS
AND CHARGE PUMPS
PHASE-FREQUENCY DETECTORS
'1'
UP
Q
Q
RESET
IN2
'1'
D
R
DN
135
136
Appendix A
IN2
DN
UP
RESET
Figure A-3. Timing Diagram of the PFD, Case IN2 Leads IN1
IN2
IN1
UP
DN
RESET
IN2
IN1
UP
DN
RESET
137
138
Appendix A
In practice, the UP and DN signals do not reset to low instantaneously but go to zero after a certain delay. This is illustrated in Figure
A-6 for the case of Figure A-3.
IN1
IN2
UP
DN
RESET
Delay in Reset Path
Figure A-6. Practical Timing Diagram of the PFD for IN2 Leading IN1. The Figure is not to Scale
as in Reality the RESET Pulse is a Full Pulse and not an Impulse as the Figure might Suggest
CHARGE PUMP
139
Vsup
Vsup
Iup
Iup
Iup
Up
Up
Up
Icp
Icp
Icp
Dn
Dn
Dn
Zs
Zs
Zs
Idn
Idn
Idn
Vdd
Vdd
UP
vp
vp
UP
V dd
V dd
Icp
vn
Vss
Icp
DN
Vss
vn
DN
(a)
(b)
140
Appendix A
Output impedance of CP
Current mismatch between UP and DN parts
Switching time and switch feed-forward
Leakage when UP and DN are off (CP Icp = 0)
Example charge pump UP, DN, and net currents are shown in
Figure A-9. It can easily be seen from the net current that there is a
mismatch between the UP and DN currents for tuning voltage values
less than 0.3 V and greater than 1.2 V.
PFD/CP CHARACTERISTICS
141
Appendix B
CONTROLLED OSCILLATORS
REFERENCE OSCILLATORS
1.1
VCXOs are the most accurate, clean, and stable frequency sources
that are used as frequency references in the PLL. The crystal is made
of a piezoelectric resonator that is electromechanical in nature. Figure
B-1 shows a basic crystal model and a configuration that forms the
basis of a VCXO.
The motional capacitance Cs and the motional inductance Ls determine the series resonance of the crystal equation (B.1) as their
impedances cancel out. The crystal impedance at the series resonance is
approximated to Rs, the motional resistance. This resonant frequency is
the operation frequency.
143
144
Appendix B
Ls
Rs
Cs
Cp
(a)
(b)
Figure B-1. Crystal Oscillator (a) Equivalent Circuit, (b) Simplified VCXO Circuit
Implementation
fs
1
2 LsCs
(B.1)
Controlled Oscillators
145
fp
1 Cp Cs
'f
2S LsCpCs
(B.2)
'f
1.2
Cs
2(CL Cp )
(B.3)
VCXOs have a superior phase noise characteristic and hence are very
suitable for PLL synthesizers with stringent phase noise requirements.
However, crystal oscillators exhibit a fundamental drift in frequency
with temperature. Equation B.4 [B1] shows frequency drifts due to
ambient temperature variations of an AT-cut1 crystal.
'f
[D1 (T T0 ) D 2 (T T0 ) 2 D 3 (T T0 )3 ] f 0
(B.4)
146
Appendix B
VCC VDD
R b1 Rg1
M1 M1
Re2
Crystal
Crystal
Cc
RT
RT
CNPO1CNPO1CNPO2CNPO2CNPO3CNPO3
R b2 Rg2 C
1
C2
C1
C2
Cc
Output
M2
Output
R e1 RS
VOLTAGE-CONTROLLED OSCILLATORS
2.1
Equation B5 shows the SSB power spectral density (PSD) of the total
phase noise.
P
(Z 'Z ,1Hz )
Ltotal {'Z} 10 log sideband 0
(dBc/Hz)
Pcarrier
(B.5)
147
Controlled Oscillators
L('Z)
(dB)
Z 2 'Z1/ f 3
0
1
1
Z
2
Q
'
'Z
(B.6)
1/f3
1/f2
1/f1
'Z1
f3
1/f0
'Z
Figure B-3. Typical Curve of the Phase Noise of an Oscillator Versus Offset from Carrier
148
Appendix B
X(jZ) +
H(jZ)
Y(jZ)
Y
> j(Z0 'Z )@
X
1
dH
('Z )
dZ
(B.7)
h) (t ,W )
*(Z0t )
u (t W )
qmax
(B.8)
where *(Z0t) is the ISF function of the output and qmax is the
maximum charge offset across the capacitor. The total excess phase
due to a noise current can be described by the following equation:
t
f
T (t )
f
h) (t ,L )i (L )dL
*(Z0L )
i (L )dL
qmax
f
(B.9)
149
Controlled Oscillators
in2 f 2
cn
'f
n 0
L{'Z } 10 log 2
2
8q max 'Z
(B.10)
in2
is the power spectral density of the input noise current, cn
'f
is the coefficient of the Fourier transform of the ISF function, and 'Z
is the frequency shift from the carrier frequency.
where,
2.2
Figure B-5 shows the steady state parallel LC oscillator model. The
tank loss is represented by gtank, and the effective negative conductance
of the active devices, required to compensate for the tank losses is
represented by gacitve.
Typically, LC oscillators operate in two different modes: namely
current- and voltage-limited modes [B7]. In the current-limited mode,
the tank amplitude, Vtank, linearly increases with the bias current
according to Vtank = Ibias/gtank until the oscillator enters the voltagelimited mode where the amplitude is limited to Vlimit, which is
determined by the supply voltage. Vtank can be expressed in Equation
(B.11) where Ibias is the independent variable.
150
Appendix B
Vtank
I bias
g tank
V
limit
( I Limited )
(B.11)
(V Limited )
2 Etank
C
2 EtankZ02 L
(B.12)
Etank
2
CVtank
2
(B.13)
Vtank
2.2.1
2E tankZ02 L
2
Vlimit
(L Limited)
(V Limited)
(B.14)
VCO Design
Figure B-6 shows the building block of the used LC VCO core. In
addition to the inductor dimensions, Figure B-7, the core can be
optimized by optimizing the MOS W/L dimensions, the varactor
tuning range, and the bias current.
Using the parallel oscillator model on the Figure B-7, the tank loss
gtank, the negative conductance -gactive, the tank inductance Ltank, and
the capacitance Ctank are given by:
151
Controlled Oscillators
2 g tank
2 g active
Ltank
2Ctank
g op g v g L
(B.15)
(B.16)
g mp
(B.17)
2L
CPMOS CL Cv C load
(B.18)
152
Appendix B
Figure B-7. Equivalent Oscillator Model and Symmetrical Spiral Inductor Model
I bias d I max
Vtank
I bias
g tank,max
t Vtank,min
(B.20)
1
2
Zmax
(B.21)
153
Controlled Oscillators
1
2
Zmin
(B.22)
(B.23)
The final spec. is size or area which needs to be kept small. Being
the most area-consuming component in the VCO core, the inductor
has its size specified so as not to exceed a certain value and hence its
diameter is specified as dmax, i.e.
(B.24)
d d d max
L ^ f off `
in2
1
1
2
rms,n
2
8S 2 f off2 qmax
n 'f
(B.25)
in2
'f
represents the equivalent differential noise PSD due to drain current
noise, inductor noise, and varactor noise, expressed as:
i M2 ,d
'f
2 kTJ ( g d 0 ,n g d 0 , p )
(B.26)
154
Appendix B
2
iind
'f
2kTg L
(B.27)
2
ivar
'f
2kTg v,max
(B.28)
where J is approximately
2
3
and
5
2
transistors, respectively. It has been proven [B8] that drain current noise
is predominantly amongst the three noise sources. Taking only the drain
current noise term into consideration in B.25, and replacing qmax by
Vtank
2 I drain
, and g d0
for short-channel transistors, (* 2rms = 1/2
2
LtankZ
Lchannel Esat
is used for pure sinusoidal waveform), the phase noise can be expressed
as:
L2 g L2 / I bias
L{ f off } v 2
2
L I bias / Vsupply
(L Limited)
(V Limited)
(B.29)
B.29 states that for a given bias current, the phase noise rises with
increasing L in the voltage-limited mode, hence once the V-limit is
reached any excess in inductance, L, will worsen the phase noise.
Additionally, for a given inductance, increasing the bias current
translates to an increase of the phase noise in the voltage-limited
mode, inducing power wastage. For a typical on-chip spiral inductor,
the minimum effective parallel conductance gL decreases with an
increasing inductance; the factor L2g2L also increases. Thus, for a
given Ibias, the phase noise increases with the inductance in the
inductance-limited mode and hence a smaller inductance results in a
better phase noise. So the design needs to be based on finding the
smallest inductor that satisfies both the tanks amplitude and start-up
condition for the maximum allowable bias current allowed by the
current budget.
Controlled Oscillators
155
REFERENCES
[B1] Frerking and E. Marvin, Crystal Oscillator Design and Temperature
Compensation, Van Nostrand, New York, 1978.
[B2] M.A. Haney, Design Technique for Analog Temperature Compensation of Crystal Oscillators, Master thesis, Faculty of the Virginia
Polytechnic Institute and State University, Blacksburg, VI, 2001.
156
Appendix B
[B3] T.C. Weigandt, B. Kim, and P.R. Gray, Analysis of Timing Jitter in
CMOS Ring Oscillators, Proceedings of ISCAS, June 1994.
[B4] J. McNeill, Jitter in Ring Oscillators, IEEE Journal of Solid-State
Circuits, 32, pp. 870879, June 1997.
[B5] B. Razavi, A Study of Phase Noise in CMOS oscillators, IEEE
Journal of Solid-State Circuit, 31, pp. 331343, Mar. 1996.
[B6] A. Hajimiri and T.H. Lee, A General Theory of Phase Noise in Electrical
Oscillators, IEEE Journal of Solid-State Circuits, 33, pp. 179194, Feb.
1998.
[B7] D.B. Leeson, A Simple Model of Feedback Oscillator Noises Spectrum, Proceedings on IEEE, 54, pp. 329330, Feb. 1966.
[B8] D. Ham and A. Hajimiri, Concepts and Methods in Optimization of
Integrated LC VCOs, IEEE Journal of Solid State Circuits, 36,
pp. 896909, June 2001.
[B9] F. Fang and K. Phang, Phase Noise Analysis of VCO and Design
Approach to LC VCOs, Term Paper University of Toronto, pp. 314,
2001.
Appendix C
PHASE NOISE
Analysis
s (t )
(C.1)
s (t )
If the phase modulation index is small, 'Ipk << 1, then the second
cosine term approaches unity and the second sine term approaches its
argument.
s (t ) | A cos(Zct ) A'Ipk sin(Zc t) sin(Zm t )
A'Ipk
[cos(Zc Zm )t cos(Zc Zm )t ] ,
A cos(Zct )
2
157
(C.3)
158
Appendix C
which is a single carrier tone having two sideband tones with relative
amplitude ('Ipk/2), spaced by Z m from the carrier.
Note that the peak phase error is related to the amplitude and
power of the phase modulating signal. Similarly, the RMS phase error
is given by the RMS amplitude of the phase modulating signal,
'Irms
'Ipk
(C.4)
power in sideband
power in carrier
A'Ipk
2
A
2
'Ipk
2
'Irms
2
(C.5)
(C.6)
f2
f1
2
'Irms
( f )df
f2
2 10L(f ) /10 df
f1
(C.7)
159
Phase Noise
L( f ) 10 log10
2 BW
(C.8)
A
Resultant
T
160
Appendix C
B
B
A
T
(C.9)
If we assume that 2B/A << S/2, then we can arrive at a value for e(t),
representative of the phase modulation of this carrier at this particular
frequency offset, m:
e(t ) A cos(Zct 2AB sin(Zm t ))
where 2B/A sin(Z mt) is the phase modulation.
(C.10)
161
Phase Noise
The output of a PLL is a sinusoid with phase noise. The ultimate goal
is to calculate from this phase noise spectrum the RMS phase error of
this sinusoid. First considering a pure sinusoid with white noise, such
as that associated with a resistor, in communications terminology
there is an associated signal-to-noise ratio, (S/N), and bandwidth.
Here C will be used to represent the carrier replacing S, the sinusoid,
in the commonly accepted 1 Hz bandwidth of SSB phase noise, L(f ).
If we look at a 1 Hz bandwidth of this noise power at an offset
frequency fm, from the carrier, and calculate the effect of this small
amount of noise, the effect of all the noise in the total bandwidth can
be found. The relationship between the vector and spectral representations of the carrier with noise is shown in Figure C-3;
Vc
C
N
fo + fm
fo
fo
Vn
fo + fm
T pk tan 1 ( VV ) | VV , in power
n
N
C
(C.11)
T pk
2
N
2C
(C.12)
162
Appendix C
C
Vc
Vn
T
fo
fo - fm
fo + fm
VnTotal
Thus T pk
Vn21 Vn22
2 Vn
Vc
2N
C
2Vn2
2 Vn
in RMSI
T pk
2
(C.13)
N
C
rad RMS
(C.14)
1800
N
C
df
1800
N
b 2C
df
(C.15)
1800
b
NP ( f )
C
df Deg RMS
(C.16)
163
Phase Noise
RESIDUAL FM
S 'f ( f )
2
'f rms
(f)
B
res FM
f 2 SI ( f )
S 'f ( f )
(C.17)
f 2 SI ( f )
B
[Hz]
(C.18)
(C.19)
L( f )
SI ( f )
2
(C.20)
Appendix D
FREQUENCY DIVIDERS
REFERENCE DIVIDER
1.1
Synchronous Dividers
165
166
Appendix D
terminal count as a code which will be zero. This makes the bit which is
1 a dont care bit. With this arrangement, terminal count is detected
at the defined value and for the forbidden all-zeros state. The terminal
count forces a parallel load on the shift register as part of the normal
divider operation, so it will load a permissible code if the forbidden
state occurs.
q6
q5
L<5:0>
S Q
q1
D1
D0
C
S Q
q2
D1
D0
C
S Q
q3
D1
D0
C
S Q
q5
D1
D0
C
S Q
q4
D1
D0
C
S Q
q6
D1
D0
C
fin
q5
q4
q3
fout
q2
q1
These dividers are fast, simple, and fully programmable. They are
nearly as fast and simple for big division ratios as for small ones. The
disadvantages are that they count in pseudorandom fashion (not binary)
so a lookup table is required for division ratios versus parallel load
value. They also consume more power than ripple counters because all
the D-type flip-flops used are clocked at the input clock-rate. If
designed carefully however, the maximum frequency of operation is not
much less than that of a single divide by two using the same D-type
flip-flop.
1.2
167
Frequency Dividers
Rout
SN
Rsyn
R<5>
R<4>
R<3>
R<2>
Q
QN
X
Y
SN
D
C
R<1>
Q
QN
RN
Vdd
R<5:0>
QN
L
QN
L
QN
L
Q
QN
R<5>
R<0>
Q
QN
R<4>
R<3>
Q
QN
R<2>
R<1>
Rin
Q
QN
SN
D
QN
RN
RN SN
The clock drives only the first TFF and each subsequent stage is
driven by the previous output stage. Figure D-4 shows the timing
diagram of the reference divider programmed to divide by 16.
Additional flip-flops and gates are used as decision logic to detect when
the count reaches final count and to reload the divider value.
168
Appendix D
FEEDBACK DIVIDER
2.1
High speed, power consumption, and low noise are the main
requirements of the VCO frequency divider. The divider needs to
operate at the VCO frequency making it unavoidably power hungry.
However, some architectures have shown to consume less power than
others. A good design is one that can be low noise and consume low
power. These requirements are important as most wireless communication devices are portable and battery operated.
2.1.1
Ideally a MMD such as the one in Figure D-5 driven directly by the
RF VCO frequency (RFin) will give minimal noise as the ratio
169
Frequency Dividers
between RFin and the clock frequency of the fractional '6 modulator
is at a maximum. However, the MMD is implemented using
synchronous CML and hence each stage is driven and operated at the
RF VCO frequency. This implementation yields maximum power
consumption and is difficult to design especially as the divider
modulus is increased from 6 to 7 bits to cover frequency bands such as
802.11b, 802.11g, and BluetoothTM (2.42.5 GHz) or higher
frequency bands such as 802.11a (4.95.805 GHz). Additionally, if
the divider is designed to marginally meet its highest frequency of
operation, this leads to worst noise performance.
The following example illustrates the speed constraint of the
feedback divider: taking the 802.11b band, if the RF channel
frequency is:
2.46 GHz
f RF
(D.1)
f VCXO
(D.2)
f RF
f VCXO
61.5
(D.3)
Div
Out
AND
MUX1
MUX2
Y B
S
div0
Y B
S
MUX3
MUX4
Y B
S
div1
MUX5
Y B
S
div2
MUX6
Y B
S
div3
A
Y B
S
div4
DFF1
DFF2
DFF3
DFF4
DFF5
DFF6
C Q
C Q
C Q
C Q
C Q
div5
C Q
RF in
XNOR
NOR
XORa
XOR
NAND
XNORa
NORa
NANDa
XNORb
170
Appendix D
Ref.
VCO
PFD/CP
fcomp
MMD
fp
RF
2/4
Prescaler
Feedback Divider
Figure D-6. Divide-by-2/4 Prescaler and MMD in the Feedback Divider
(D.4)
171
Frequency Dividers
INP
INN
/2
0
90
/2
180
270
MUX
4 to 1
/16
OUT
Control
Vdd
INP
Vdd
Vdd
Vdd
INN
OUTP
OUTN
This could send the wrong information to the modulus control and
hence divide by the wrong divisor. Also the input sensitivity of the
divide-by-2 is much worse than its CML counterpart [D3].
The low frequency CMOS divide-by-16 counter needs to operate at
a quarter of VCO frequency which might be fine for a 2 GHz range
VCO frequency. However, as we move higher in the spectrum to
cover the 802.11a high band of (5.805 GHz), that might be difficult to
achieve. From a noise standpoint, this architecture is similar to that of
the divide-by-4 prescaler-driven MMD as it exhibits similar quantization noise amplification of 12 dB, equation (D.4).
The ordinary DMD in Figure D-8 works somewhat differently
from the PSDMD. A and B counters are loaded with the desired
divider ratio and the prescaler divides by P + 1 as long as the modulus
control is low. When the content of the B counter reaches 0, the MC
goes high and the prescaler divides by P until the content of the
counter A reaches 0. For proper division, counter A has to be greater
172
Appendix D
STATIC CMOS
A
A,B Value
RFin
P/P+1
Logic
Fdiv
Modulus Control
AP B
(D.5)
RFin
B
Fdiv
and B
P
RFin
AP
Fdiv
(D.6)
173
Frequency Dividers
NOR
DFF
D
Output
NOR
C Q
R
DFF
D
Control
Input
Reset
174
Appendix D
Divide by 2/3
NOR
DFF
D
NOR
C Q
R
DFF
D
C Q
R
Input
Reset
Output
Divide by 2
DFF
D
NOR
C Q
R
Control
175
Frequency Dividers
Divide by 2/3
NOR
DFF
D
DFF
NOR
C Q
R
D
C
Q
R
Input
Reset
Divide by 4
Output
OR
DFF
DFF
D
C Q
R
D
C
Q
R
Control
176
Appendix D
RFmin .
Re f
2
3 n
min
4800.
40
2
3 3 77
(D.7)
2
3 n
max
5805.
40
2
3 4 100
(D.8)
N frac
8 >8 N 5 : 3 ! ] N 2 : 0 ! @
(D.9)
177
Frequency Dividers
Decision Logic
SN
Q
C QN
Reset
SN
Q
C QN
RN
Mod Control
RFin
8/9 Prescaler
0
SN
Q
C QN
RN
Nout
Vdd
<3:0>
Q<3:0>
Counter A Clk
D<6:3>
L<3:0>
L<2:0>
N<5:0>
N <2:0>
N <5:3>
Vdd
Implementing the high-speed prescaler in CMOS requires careful attention. The high-frequency operation means high power consumption.
178
Appendix D
However, using the architecture outlined in Figure D-13, only the flipflops of the front divide-by-2/3 stage operate at full VCO speed. The
rest of the logic operates at half speed. The flip-flops of the divide-by-4
stages operate at half and quarter speeds. All 8/9 prescaler blocks are
designed in high-speed CML.
3.1
Unlike static CMOS, CML gates can operate at high frequency in the
GHz region. To help us understand the various elements that set the
speed and hence the power consumption of a CML gate, let us
examine a simple differential CMOS amplifier also a CML inverter.
Figure D-16 shows a basic CML inverter gate.
The speed of the gate is determined by the transient characteristics,
the current drive capability of its load and the propagation delay.
Vdd
Vdd
R
Vdd
Vswing
CL
A
CL
vin
vin
Vb
(a)
Rs
(b)
Figure D-16. Basic CML Gate (a) Biased Gate (b) Simplified Gate with Capacitive Load
179
Frequency Dividers
Rg
vg
C gd
Cgs
vd
gmVgs
Csb
gmVbs
Ro
vs
CL
Cdb
Rs
Using the half circuit of Figure D-16b, we can derive the propagation
delay of the CML inverter gate. Figure D-17 shows half the CML gate
circuit model.
We can treat the delay as four components for simplicity:
W1 W 2 W 3 W 4
(D.10)
vT
iT
W1
R1Cgs
Rg Rs
1 g m Rs
Rg Rs
Cgs
1 g m Rs
(D.11)
(D.12)
180
Appendix D
Rg
vd
iT
gmVs
gmVT
VT
Ro
vs
Rs
vT
iT
Rg R
vg
Rg
gm
Rg R
1 g m Rs
vT
(D.13)
vd
iT
gmvgs
gmbvbs
Ro
vs
Rs
181
Frequency Dividers
W2
gm
Rg R Cgd
Rg R
1 g m Rs
(D.14)
-g mv T
-g mb v T
Ro
iT
vT
Rs
R3
Rs
1 g m Rs
W3
R3Csb
Rs
Csb
1 g m Rs
(D.15)
(D.16)
Similarly for the fourth delay term, W 3 due to the capacitive load
CL and drainbulk junction capacitance, Figure D-21 shows the
simplified circuit model, where we can ignore Ro:
W4
R Cdb CL
(D.17)
182
Appendix D
Ro
iT
vT
Rs
Rg Rs
RC
gm
Cgs Rg R
Rg R Cgd s sb
1 g m Rs
1 g m Rs
1 g m Rs
Cdb CL R
(D.18)
R Cgd Cdb CL
Vswing
I
WC
ov
WLd C jd CL
(D.19)
Where Cov is the oxide capacitance, Cjd the drain junction capacitor, W
and L are transistor dimensions. Vswing is the required output voltage
swing. Knowing the required swing and the required speed we can
choose the transistor dimensions and the differential tail current
required to drive the CML gate.
183
Frequency Dividers
NOR
Vdd
R
Vdd
R
Q
Or
B
Nor
B
A
Vb2
A
V b1
Vb
(a)
(b)
Vdd
NAND/AND
R
R
Q
Vb
(c)
Figure D-22. Basic CML Gates Implementation (a), (b) NOR/OR, and (c) NAND/AND
184
Appendix D
MUX
XOR/XNOR
Vdd
Vdd
A B
A
A
Vb
Vb
(a)
(b)
D-Latch
Vdd
Q
D
Clk
Clk
Vb
(c)
Figure D-23. CML gates (a) Multiplexer, (b) XOR/XNOR, and (c) Latch
185
Frequency Dividers
Vdd
Vdd
R
Nor
Or
Vb2
Clk
Clk
Clk
Vb1
Vb1
Slave
Master
(a)
Vdd
Vdd
Vdd
Q
Q
Clk
Clk
Master
Clk
Clk
Vb
Vb
Vb
MUX
Slave
(b)
Figure D-24. Combination CML Logic (a) NOR/OR Mater/Slave D-Type Flip-Flop
(b) MUX/D-Type Flip-Flop
REFERENCES
[D1] J. Craninckx and M. Steyaert A 1.75-GHz/3-V Dual-Modulus Divideby-128/129 Prescaler in 0.7-m CMOS, JSSC, 31, 7, pp. 890897,
July 1996.
[D2] R. Ahola and K. Halonen, A 4 GHz CMOS Multiple Modulus Prescaler,
Proceedings of IEEE ICECS 1998, Lisbon, Portugal, Sept. 1998,
pp. 2.3232.326.
[D3] T. Seneff et al., A Sub-1 mA 1.6 GHz Silicon Bipolar Dual Modulus
Prescaler, IEEE Journal of Solid-State Circuits, 29, pp. 12061211,
Oct. 1994.
Appendix E
PROGRAMS AND CODES
15
M { 10
p { 10
G { 10
12
n { 10
T { 10
12
9
P { 10
R 10
6
m { 10
Set Resolution:-
3
Hz
MHz/V
VCXO_Ref_Freq 40 MHz
KI 2.0 m
mA
Fsamp 40 M Hz
Start_Freq 10
Stop_Freq 10 M
Duty Cycle 2 S
Fmax
Hz
XScale (F ) 10
Duty Cycle
1725 M Hz
Fmin 1725 M Hz
187
10
k { 10
188
Appendix E
10
R
10log( Stop_Freq )
K fm K vco
K fm 200k
Enter values for calculating the loop filter components:-
LBW 100k Hz
Ip 56 Degrees
ATTEN 15 dB
(Make this value0to remove the
additional R3 C3 low pass filter)
RF Noise Source Values:VCO Noise
dBc/Hz
189
VCODiv_Plateau 173
RF Amplifier Noise
Ref_9dB 100
Amp_NF 8
Ref_12dB 10
dBc/Hz
VCODiv_Noise_3dB 0.3
dBm
Power_in 15
dB
Hz
R_Div_6dB 10
RFAmp_noise_3dB 100
)_6dB 50
)_9dB 20 )_12dB 5
Zp 2 S LBW
8
sec Ip
T1
Zp
rad
LBW
1 u 10
1 u 10
C1T22O
C1T22O
Fsamp
tan Ip S
180
180
43.125
T2
Zp
FmaxFmin
N
Z p T1
2
1 2 S LBW T2
1 2 S LBW T1
K
KI
T1 vco
Z p rad
N
T2 2 S LBW
9
3.591536 u 10
7
T1
4.866 u 10
T2
5.206 u 10
6
T2
C2T22O C1T22O
T1
C2T22O
8
3.483242 u 10
ATTEN
T3
10
20
1
2 S Fsamp 2
1
R2T22O
R2T22O
T2
C2T22O
149.451
190
Appendix E
1
T2
( T1 T3) tan Ip
180
( T1 T3)
( T1 T3) 2 T1 T3
C1T23O
( T1 T3) T1 T3
2
S
2
( T1 T3) tan Ip
180
T1
T2
S
( T1 T3) 2 T1 T3
1 1
1
2
( T1 T3) T1 T3
2
S
2
( T1 T3) tan Ip
180
1 1
K vco KI
N
( T1 T3) tan Ip S
2
180
T2
2
( T1 T3) T1 T3
( T1 T3) T1 T3
11 1
2
S
2
( T1 T3) tan Ip
180
( T1 T3) 2 T1 T3
2
1 1 T1
S
2
( T1 T3) tan Ip
180
( T1 T3) tan Ip S
180
1
( T1 T3) 2 T1 T3
1
2
( T1 T3) 2 T1 T3
2
1 1 T3
S
2
( T1 T3) tan Ip
180
( T1 T3) tan Ip S
180
1
( T1 T3) 2 T1 T3
C2T23O C1T23O
T2
1
T1
C1T 23O
9
3.665 u 10
C3 T23O
T2
R2 T23O
C2T23O
10
3.665 u 10
C2 T23O
C3 T23O
C1T23O
8
3.635 u 10
R3 T23O
10
R2 T23O
146.14
T3
C3 T23O
R3 T23O
23.342
C3 if ATTEN 0 0 C3 T23O
191
R3 if ATTEN 0 0 R3T23O
R
VCXO_Ref_Freq
Fsamp
This line calculates the R divider logic noise floor including the 1/f corners for this R divider.
R_Div_3dB
FrqPoint
10log 1 j R_Div_6dB
FrqPoint
This line calculates the ideal divided down (due to the R term) of the incoming reference VCXO source.
Note that this is a direct curve fit on the specification and makes no attempt to correct for the usual oscillator
.
1/f corners, giving 10Log breakpoints.
Ref_3dB
Ref_6dB
10log 1 j
Ref_9dB
FrqPoint
FrqPoint
This VXCO SSB Phase Noise Plot represents the free running VCXO which is considerably
cleaner than the 16 MHz clean upused. This is because the 16 MHz cleanup loop has
a loop bandwidth of the order of 40 Hz, there after the phase noise follows the logic noise and
Op-Amp noise profile of the other components in that loop. These values are not modeled here,
giving an optimistic account of this loops output. This will reflect in a better than measured value in
the main synthesizer output for this part of the phase noise profile.
Hence, the actual phase noise profile of the signal at the sampling frequency is:-
Ref_Div ( FrqPoint)
10
R_Div_Noise ( FrqPoint)
10
80
10
90
100
110
120
130
140
150
160
170
180
190
200
210
10
100
1 10
1 10
1 10
Frequency, f(Hz)
1 10
1 10
192
Appendix E
The effect of the R divider logic noise plateau becomes particuraly important when considering low
sampling frequencies and very large VCO frequencies and hence multiplication of this reference noise.
The clear distinction has to be made between the sampling frequency phase noise and the phase frequency
logic noise.
VCO Noise Calculation
VCONoise ( FrqPoint) VCONoise_Plateau 10log 1 j
10 log 1 j
VCO_9dB
FrqPoint
VCO_3dB
FrqPoint
10 log 1 j VCO_6dB
FrqPoint
40
50
10
10
VCONoise 10 120.937 VCONoise 10 100.902
30
10
VCONoise 10 78.789
L(800 kHz)=
10 log( 800 k)
10
VCONoise 10
138.957
Free Running VCO SSB Phase Noise
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
10
100
1 10
1 10
1 10
Frequency, f(Hz)
1 10
1 10
)_9dB
10log 1
FrqPoint
)_3dB
FrqPoint
10log 1 )_6dB
FrqPoint
10log 1 )_12dB
rqPoint
10log 1
PD_Noise ( FrqPoint) PD_Noise_Norm( FrqPoint) 10log( Fsamp) 20log FmaxFmin
PD_Noise 10
60
10
99.499
Fsamp
)n3dB
193
180
188
196
204
212
220
10
3
4
5
6
7
100 1 10 1 10 1 10 1 10 1 10
Frequency, (Hz)
60
68
76
84
92
100
3
4
5
6
7
100 1 10 1 10 1 10 1 10 1 10
Frequency, f(Hz)
10
VCODiv_Noise_3dB
FrqPoint
23 J/K
K { 1.38066210
T { 273.14
B{1
Hz
giving...
dBc/Hz
154.235
RFAmp_noise ( FrqPoint )
VCONoise ( FrqPoint )
10
10
VCOEffective_Noise( FrqPoint ) 10log 10
10
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
10
100
1 10
1 10
1 10
Frequency, f(Hz)
1 10
1 10
194
Appendix E
LOOP GAIN PARAMETER CALCULATIONS
Nm
K pd
Fmin Fmax
Fsamp Fsamp
N m 43.125
Fvco
KI
K pd
DutyCycle
K v Kvco 2 S
Kv
Fmax Fmin
4 A/Rad
3.183 u 10
8
6.283 u 10
R2
Rad/V
146.14
Fs ( FrqPoint) if ATTEN
1 j 2 S FrqPoint R2 C2
0
C1 C2
C1 C2
j 2 S FrqPoint C3
1
C1 C2
200
180
160
140
120
100
80
60
40
20
0
20
40
60
80
100
10
100
3
4
5
1 10
1 10
1 10
Frequency, f(Hz)
1 10
1 10
180
160
140
120
100
80
60
40
20
0
20
40
60
80
100
120
140
160
180
10
100
3
4
5
1 10
1 10
1 10
Frequency, f(Hz)
1 10
1 10
195
Kv
j 2 S FrqPoint
K pd
FB
4
K pd
Nm
3.183 u 10
10
5
F
10 180
d
arg Fs 10
5
10
Magnitude, (dB)
10
Magnitude, (dB)
100
1 10
1 10
1 10
XScale( F)
1 10
1 10
1 10
1 10
1 10
20
0
20
40
60
80
100
120
140
160
180
200
10
100
3
4
5
1 10
1 10
1 10
Frequency, f(Hz)
1 10
20
10
0
10
20
30
40
50
60
70
80
90
100
10
Phase, (Deg)
dF
100
3
4
5
1 10
1 10
1 10
Frequency, f(Hz)
1 10
180
150
120
90
60
30
0
30
60
90
120
150
180
10
100
3
4
5
1 10
1 10
1 10
Frequency, f(Hz)
1 10
196
Appendix E
1 Aol( FrqPoint) FB
Aol( FrqPoint)
1 Aol(FrqPoint) FB
Aol( FrqPoint) FB
1 Aol(FrqPoint) FB
Aol( FrqPoint)
1 Aol(FrqPoint) FB
10
10
10
Total_Noise( FrqPoint) 10log 10
VCODiv_Loop_Noise_dB( FrqPoint)
PD_Loop_Noise_dB( FrqPoint)
10
10
10log 10
10
10 log ( 1.7 k)
10
Total_Noise 10
96
Small_Angle ( FrqPoint) 10log( FrqPoint) 30
Marker1
log( Marker1)
Marker1 10
577 P
L(1.5 kHz)=
10
Filter Values:-
Marker2 10
L(800 kHz)=
10log (800k)
10 log(1.5 k)
Total_Noise 10
log( Marker2)
Marker2 135k
dBc/Hz
95.647
Total_Noise
10
125.955
10
dBc/Hz
9
C1 3.665 u 10
8
C2 3.635 u 10
K vco
1 u 10
Hz/V
KI 0.002
Amps
7 Hz
Fsamp 4 u 10
Fvco
1.725 u 10
R2 146.14
Ohms
R3 23.342
Ohms
Phase Margin:-
LBW 1 u 10
Ip 56
10
Hz
C3 3.665 u 10
Hz
Degrees
197
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
10
100
1 10
1 10
Frequency, f(Hz)
1 10
1 10
1 10
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
10
100
1 10
1 10
Frequency, f(Hz)
1 10
1 10
1 10
198
Appendix E
199
200
Appendix E
subplot(2,1,1),semilogx(f,10*log10(s_phi_vco1_f),'b',f,10*
log10(s_phi_vco2_f),'r')
title('different Phase noise characteristic for the crystal oscillator')
% contribution from the crystal frequency reference
% p 93 poly Goldberg, state of the art Xtal
s_phi_state1(1)=10^(10);
s_phi_state1(2)=10^(13);
s_phi_state1(3)=10^(14.3);
s_phi_state1(4)=10^(15.8);
s_phi_state1(5)=10^(16.4);
s_phi_state1(6)=10^(17);
s_phi_state1(7)=10^(17);
f1=logspace(0,7,8);
s_phi_in1_f=s_phi_state1(1)./(f.^3).*((f/10) + 1).*
((f/100) + 1).*((f/5e03) + 1);
% s_phi_in1_f is the linear approximation from the vco p93
% Microwave and RF 1994
s_phi_in2_f=1e-04./((f).^3).*((f/10^3).^2+1).*((f/(4*10^4))+1);
% (Pretty noise at low frequencies !!)
% too ideal xtal below !! (from example)
s_phi_in3_f=1e-011./((f).^3).*((f/10).^2 + 1).*((f/100) + 1) ;
% phase noise of Hy-Q oscillator
s_phi_in4_f=10^(-5.3)./(f.^4).*((f.^3/5e04) + 1).*((f/8e03) + 1);
figure(1)
subplot(2,1,2),semilogx(f,10*log10(s_phi_in1_f),'b',...
f,10*log10(s_phi_in2_f),'r',f,10*log10(s_phi_in3_f),'g',f,10*
log10(s_phi_in4_f),'m')
title('different Phase noise characteristic for the crystal oscillator')
s_phi_in_f=s_phi_in1_f;
% s_phi_in3 is too ideal and s_phi_in2 is too bad (too much noise
% with in the loop bandwidth, one should achieve 80 dBc for DECT
% contribution from the loop filter
s_phi_loop=10^(-11)./((f).^3).*((f/(30)).^2+1).*((f/1e02)+1);
% contribution from the frequency detector
201
202
Appendix E
203
204
Appendix E
figure(3)
subplot(2,1,2),semilogx(f,10*log10(s_phi_dn_frac*N^2.
*(abs(Hcl)).^2),'b',...
f,10*log10(s_phi_pd_f*N^2.*(abs(Hcl)).^2),'r',...
f,10*log10(s_phi_loop./f.^2 .*(abs(one_min_H_cl)).^2),'g',...
f,10*log10(s_phi_in_f*N^2.*(abs(Hcl)).^2),'m',...
f,10*log10(s_phi_vco_f.*(abs(one_min_H_cl)).^2),'c',
f,10*log10(S_phi_tot_frac),'b-.');
grid on
set(gcf,'DefaultTextColor','k')
xlabel('frequency (Hz)')
ylabel('S_\Phi (dB)')
title('Phase noise source after filtering')
set(gcf,'DefaultTextColor','c')
text(1e06,-130,'(1-H)^2 S_\Phi_{ vco}')
set(gcf,'DefaultTextColor','m')
text(1e01,-70,'N^2 H^2 S_\Phi_{ ref}')
set(gcf,'DefaultTextColor','b')
text(1e07,-165,'N^2 H^2 S_\Phi_{ dn}')
set(gcf,'DefaultTextColor','r')
text(1e05,-200,'N^2 H^2 S_\Phi_{ pfd}')
set(gcf,'DefaultTextColor','g')
text(1e02,-250,'(1-H)^2 f^{-2} S_\Phi_{ loop filter}')
set(gcf,'DefaultTextColor','k')
figure(4)
subplot(2,1,2),semilogx(f,10*log10(S_phi_tot_frac),'b',f,10*
log10(s_phi_vco_f),'c',...
[10,1e05-1,1e05,1.2000e+06,f_l],10*log10(dect_mask),'k')
title('Overall phase noise and phase mask');
grid
The following M-file reads in the captured output of the simulated HDL
of the delta sigma block as applied to the MMD.
clear all;
Ts=25; %sampling period is 24nS
fid=fopen('HDL_deltasigma_out_data70p5.dat','r');
out=fscanf(fid,'%i %i %i\n',[3 inf]);
fclose(fid);
205
%M=1000044;
%M=length(out(1,:));
frac=out(2,:);
int=out(1,:);
inst_div=out(3,:);
%M=length(frac);
M=2^19;
y2=inst_div;
y2=y2(20001:M);
mean(y2)
M=length(y2);
fs=1/(Ts*1e-9);
f=(1:M)*fs/M/1e3;
win=hanning(M);
win=win';
a2=abs(fft(y2.*win));
a2=(a2.*a2)/M;
a2=a2(1:M/2);
f=f(1:M/2);
a2=10*log10(a2/max(a2));
figure(1),plot(f,a2);
xlabel('FREQUENCY (KHz)');
ylabel('MAGNITUDE (dB)');
axis([min(f) max(f) -350 0]),grid;
title('Output spectrum of Nemo Delta Sigma Modulator (VHDL)');
delta_f=fs/M;
place=floor(1000e3/delta_f);
a3=a2(1:place);
f3=f(1:place);
figure(2),plot(f3,a3);
xlabel('FREQUENCY (KHz)');
ylabel('POWER SPECTRUM (dB)');
axis([min(f3) max(f3) -350 0]),grid;
title('Baseband output spectrum of Nemo Delta Sigma Modulator (VHDL)');
fB=200e03;
% The cutoff frequency = fpass
fstop=2000e03;
% The stop frequency
rp=3;
% pass band attenuation in dB
rs=30;
% stop band attenuation in dB
wp=fB*2/fs;
% pass normalized frequency
ws=fstop*2/fs; % stop normalized freuqnecy
[filt_order,wn]=buttord(wp,ws,rp,rs);
[b1,a1]=butter(filt_order,wn); % calculate coefficents of butterworth filter
filtered=filter(b1,a1,y2);
206
Appendix E
a4=abs(fft(filtered.*win));
a4=(a4.*a4)/M;
a4=a4(1:M/2);
a4=10*log10(a4/max(a4));
figure(3),plot(f,a4);
xlabel('FREQUENCY (KHz)');
ylabel('MAGNITUDE (dB)');
axis([min(f) max(f) -350 0]),grid;
title('Filtered spectrum of MASH output');
% to be able to print the results you have to decimate the results before
presenting
a5=a2(1:64:length(a2));
f5=f(1:64:length(f));
figure(4),plot(f5,a5);
xlabel('FREQUENCY (KHz)');
ylabel('MAGNITUDE (dB)');
axis([min(f5) max(f5) -350 0]),grid;
title('Decimated output spectrum of MASH');
a6=a3(1:64:length(a3));
f6=f3(1:64:length(f3));
figure(5),plot(f6,a6);
xlabel('FREQUENCY (KHz)');
ylabel('MAGNITUDE (dB)');
axis([min(f6) max(f6) -350 0]),grid;
title('Decimated baseband output spectrum of MASH');
a7=a4(1:64:length(a4));
figure(6),plot(f5,a7);
xlabel('FREQUENCY (KHz)');
ylabel('MAGNITUDE (dB)');
axis([min(f5) max(f5) -350 0]),grid;
title('Decimated filtered spectrum of MASH output');
Index
DC offset compensation 13
dead zone 48, 71, 76, 77, 132, 138
deltasigma modulators ix, 2, 67
deltasigma noise 50, 58, 98
design methodology ix, x, 155
differential to single-ended conversion 12
differentiators 95
digital accumulator 34, 35, 36, 90
digital domain 13, 133
digital-to-analog converters 12, 116
direct frequency synthesis 1, 131
direct modulation 133
direct-sequence spread spectrum 8
direct conversion 107, 109
dithering 57, 98, 132
divide-by-2 13, 107, 170
dividers 19, 36, 62, 89, 131, 143, 165
feedback 19
reference 19
dual-type flip-flop 71
dual-modulus prescaler 19, 35
duty cycle 30, 47, 52, 108
2s complement 96
802.11 a, b and g 3, 10
accumulator-based implementation 95
adaptive band switching 2, 111, 132
adaptive charge pump 2, 111, 113, 116,
117, 128, 132
adjacent channel rejection 13
attenuation 12, 205
automatically calibrated 12
average divide ratio 2, 45
balanced to unbalanced transformer 12
band switching 103, 126, 127, 128
bandgap reference 119
bandwidth 8, 32, 47, 105, 117, 133, 147,
158, 202
base band 10, 38
behavioral modelling 2, 45, 47
Bluetooth 7, 14
calibration 117, 122, 125, 128
calibration algorithm 125
carrier frequency 149, 153, 163
carry look ahead 93
channel selection 12
charge pump 17, 71, 113, 138
closed loop 1
CML see current mode logic
CMOS process 2, 67, 73, 102, 113, 131
complementary code keying 7
CP linearization 127
current mismatches 73
current-mode logic 89
cutoff frequency 12
207
208
fractional-N 1, 34
integer-N 1
front-end signal paths 12
GSM system 159
guard rings 108
impulse sensitivity functions 30
in-band phase noise 38, 72, 170
indirect frequency synthesis 3
integer multiples 60, 99
inter-subcarrier interferences 107
IQ filters 12
IQ image rejection mixer 86
limit cycles 57, 59, 60, 97, 99
linear time invariant xi
local oscillator 11, 53, 105, 131
local oscillators 1, 11, 131
locking time 26, 107, 128
lookup table 13
loop filter 20
loop filter design 15, 42, 131
low-noise amplifier 10
MASH-1-1-1 36, 67, 90, 131
mismatch cancellation 75, 76
mixers 5, 13, 107, 133
mobile phones 7
motional capacitance 143
motional inductance 143
negative-edge triggered 108
NMOS 74, 79, 86
noise contributions 1, 15, 27, 40, 131, 147
noise shaping 90, 93, 108, 113
noise spectrum 29, 57, 97, 161, 162, 198
nonoverlapping 10
normal distribution 50
OFDM 7, 9, 10
offset current 62, 78, 79, 81
offset frequency 50, 153, 158, 160, 161
open loop 1, 42, 45, 53
PFD/CP linearity 2, 45, 47, 49, 52
phase-domain model 2, 45, 46, 47, 50,
53
phase error 16, 35, 78, 105, 157
phase margin 21
phase noise 15, 27, 41
phase-frequency detector 16, 28, 71, 135, 65
phase-locked loop 15, 27, 67, 131
pipelined adder 93
PMOS 74, 75, 79, 82, 86
Index
power amplifier 10, 133
power spectral density 29, 38, 50, 55, 149
process tolerances 117
pseudorandom binary sequence 165
QPSK system 42
quadrature-phase signals 86
quantization noise 50, 51, 55, 64, 65,
132, 133, 172
rms phase error 41
range coverage 7
receiver ix, 1, 10, 11, 12, 13, 42, 70, 102,
105, 107, 131
reference frequency 50, 53, 60, 71, 77,
78, 99, 107, 140, 165, 176
rejection ratio 13
RF front-end circuits 13
RFIC transceivers 7
sampling frequency 31, 34, 35, 39, 40,
71, 125, 165
short-channel transistors 154
single-sideband phase noise 158, 163
single subcarrier 10
slew rate 118
spectral densities 52, 99
spurious performance x, 3
spurious signals 35, 57, 163
substrate noise rejection 18
superimposed noise vectors 162
tank amplitude 149, 150, 152, 153
time to digital converter xii, 134
time domain 26, 36, 55, 90
transceiver 1, 7, 69, 102, 109, 131
transfer functions 1
transmitter ix, 1, 5, 12, 13, 14, 70, 102,
105, 107, 131, 133, 176
tunable varactors 86
tuning curves 51, 88
tuning voltage 18, 49, 122, 140
unlicensed band 10
upconverted differential signal 12
variable-gain amplifiers 10
VCO 18, 40, 86, 132, 162, 201
wireless cellular devices 7
wireless LAN 7, 8, 10, 11, 13, 39, 45, 69,
70, 88, 107, 111, 127, 131, 132
wireless local area networks 1, 7
'6 36, 51, 69, 90, 113, 128, 169
'6:modulators 50