part ii: recent advances on oversampled analog-to-...
TRANSCRIPT
- 1 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Jose Silva-MartinezAmesp02.tamu.edu/~jsilva
Department of ECE
Texas A&M University
Part II: Recent advances on Oversampled Analog-to-
Digital Converters
- 2 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
State of the Art
Fundamentals of Sigma-Delta Modulators Noise and Signal Transfer Function
Error Function
Blockers tolerance
System Limitations Stability
Clock Jitter
DAC issues
Global Tuning
Cases of Study Sigma-Delta Modulator using TDC Techniques
Sigma-Delta Modulator employing Multi-phase approach
Methodologies to improve ADC resolution
Conclusions
Recommended readings
Outline
- 3 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Introduction: ConnectivityWLAN
UWB, 802.15.3a
Connectivity to
EVERYTHING
PAN, 802.15
GPS
WI-FIMAN
Wi-Max,
802.16a,e
BLUETOOTH
Increasing number of wireless standards
Support of multiple-standards on the same chip
Advances in Integrated RF design towards universal devices
Software Radio: easy addition of new standards
- 4 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Direct Conversion Multi-Standard Receiver
System issues in broadband systems:High frequency filtering is especially critical in broadband applicationsRejection of Blockers: ADC filtering must be complemented by LP filteringNeighbor channels are quite relevant even if heavy filtering is used
Trade-offs:Light filtering in front demands an ADC with higher SNR and higher SDRHigher SNDR-ADC implies more power and more circuit complexity
- 5 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
1.E-01
1.E+00
1.E+01
1.E+02
1.E+03
1.E+04
1.E+05
1.E+06
1.E+07
10 20 30 40 50 60 70 80 90 100 110 120
ISSCC 2010
ISSCC 1997-2009
VLSI 1997-2009
FOM=100fJ/conv-step
FOM=10fJ/conv-step
B. Murmann, http://www.stanford.edu/~murmann/adcsurvey.html.SNDR (dB)
State of the art
Video
100mW @10MHz
10mW @10MHz
Po
we
r/fs
(pJo
ule
s)
Challenge 1: Jitter Tolerance
- 6 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Challenge 2: Co-existenceSimultaneous Usage of Radio Bands
RF Interference: Frequency Overlap, Out-of-Band Emissions, Receiver Saturation
RF Interference from Frequency Overlap, Out-of-Band Emissions and Receiver Saturation
-9
0
30
20
Output Power (dBm)
24
33
16
3.1 3.3 4.9 5.1 5.2 5.3 5.47 5.7 5.8
Frequency (GHz)
3.8
WiMAX
802.16e
. . .
Y-Axis
...
UWB
0.8 0.9 2.7
GPS
WWAN
WCDMA
1.58 1.60 1.8 1.9 2.2
WLAN
802.11b/g
2.3 2.4 2.48
WWAN
GPRS/EDGE
0.4 0.7
UHF
WLAN
802.11a
Bluetooth
WiMAX
802.16e
WiMAX
802.16e
WLAN
802.11a
E
u
r
o
p
e
J
a
p
a
n
* Plot courtesy of Camille Chen/IntelFrequency (GHz)
Ou
tpu
tP
ow
er
(dB
m)
- 7 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
1
2
3
LNA
RF Filter
RF
Anti-
Aliasing
Filter
A/D
SCF, Gm
C
OP-RC
Anti-
Aliasing
Filter
A/D
Dig. Filter
DSP
DSPRF
RF Filter
RF Filter
A/D
Dig. Filter
DSPG
Dig. Mod.RF
IF or BB
DR
DR
BB
How much RF processing should be done before the ADC? The front-end must be scalable and configurable to fit multiple standards. Better performance at higher frequency.
Roadmap for high-resolution Receivers: ParadigmSlide courtesy of Prof. Yun Chiu
- 8 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Impact of SNDR on ADC Power Consumption
♦ For 4MHz BW a 10dB increase in DR leads to ADC power increment of
50mW at 70dB or 1W at 90dB
♦ Required: Systematic evaluation of filter-ADC interdependence for broadband receivers
A
B
Slope is around:
x10 in Power/12.5 dB in SNDR
x10 in Power/2 bits in SNDR
- 9 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Conventional (Nyquist) ADC
A/D
Data
Out
Vin
SQNR=6.02*n+1.76
- 10 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Basic concepts in SD Modulators
)s(H
)s(HSTF
1 )s(HNTF
1
1
DAC
Vin
ss
FT
1
S H(s) A/D
Data
Out
Vin
Quantization
Noise
Feedback
DAC
H(s) A/DDout
VDAC
NTF =Dout
Qn
STF =Dout
Vin
ff0
0 dB
ff0
OBG
- 11 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Fundamentals of Oversampled A/D Conversion
22
22
11
2
2221
TsinzH
TsinjezH
eeeeZzH
e
Tj
e
Tj
Tj
Tj
Tj
e
Check the input-output trajectoriesOriginal E(z) is shaped by NTF
Original E(z) is amplified here!
2
2
21
22
1
Tsin
*zE
zXSQNR
zHzE
ZzXSQNR
e
Spot SQNR
+ +1
1
1
Z
Z+
-
qE
ZY ZX
DAC
Filter
- 12 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Oversampled A/D Conversion
Signal to Quantization
Noise Ratio (SQNR)
N=number of bits
OSR=fs/2fb
SQNR improves by 30dB when
OSR increases by 10
Or 9dB SQNR improvement
when doubling OSR
D
D
2
32
2
3
2
2
0
2
2
0
2
2
0
2
21
31251
26
2
12
2
12
2
1
22
1
OSR***.SQNR
f*
f
f*
*
f*
V
V
SQNR
then,fsfbIf
dfT
sin
*zE
zXSQNR
dfzHzE
zX
dfzHzE
ZzXSQNR
N
S
b
S
S
q
q
N
f
f
e
f
e
b
bb
dBOSRlog*.N*..SQNR
OSR***.SQNR N
10
2
32
3025026761
31251
+ +1
1
1
Z
Z+
-
qE
ZY ZX
DAC
Filter
- 13 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
System Design Considerations
0
25
50
75
100
125
150
0 10 20 30 40
Sig
na
l to
Qu
an
tiza
tio
n N
ois
e R
ati
o (d
B)
Oversampling Ratio (OSR)
Filter order=2Quantizer Bits=1
Filter order=3Quantizer Bits=1
Filter order=4Quantizer Bits=1
Filter order=5Quantizer Bits=1
Dashed lines are for 2-bit quantizerDashed=Solid+6.02
- 14 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
4
6
8
10
12
14
16
18
20
4 8 12 16 20 24 28 32 36 40
Eff
ec
tive
Nu
mb
er
of
Bit
s (
EN
OB
)
Oversampling Ratio (OSR)
Design Considerations: 3rd order loop
4-bit quantizer
2-bit Quantizer
3-bit quantizer
1-bit Quantizer
3rd order loop
02.6
76.1)dB(SNDRENOB
- 15 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
STF and NTF are definitely not enough to design a robust system; a1*a2=1; b=2
At the input of the first amplifier we have:
Linearized Model to Analyze SD Modulators: 2nd
order Multiple Feedback
Schreier and Temes
zEazzzUazzz
azzVzU
zEzzUzzVzU
zEzzUzzV
q
q
q
1
11
1
11
1
1
1
212
212
111
11
1
If a1>0.5, hence the in-band output of the 1st integrator is > than the input !
At the output of the first integrator we have (saturation?)
Small for inband signals but what about the blockers?
+ +1
1
1
1
Z
Za+
-
qE
ZY ZU
DAC
1
1
2
1
Z
Za+
+
-
ZV
b
- 17 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Eq stands for the quantization noise
Ed stands for DAC non-idealities (jitter + thermal noise)
Filter’s thermal noise is accounted in Eh
The modulator’s output becomes
The error signal (Filter’s input) is
Oversampled A/D ConversionFeedforward Architecture
qhd E*NTFEEX*STFY
1
1
1
1
1
Z*sZOH*sH
NTF
Z*sZOH*sH
sZOH*sHSTF
H(s) ZOH
Z-1
X Ve Vy Y
Ef
Eq
S
Ed
SS
Eh
- 18 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
ZOH parameters are defined as
The lowpass transfer function H(s) provides large loop gain (Noise shaping)
DC gain >>0 dB
Enough phase margin at 0 dB gain
Z is a complex number (phase)
Loop stability is fundamental
Stability Issues: 2nd Order System
2
S
Tj
TcsinzZOH
eZ S
H(s) ZOH
Z-1
X Ve Vy Y
Ef
Eq
S
Ed
SS
Eh
1
1
1
1
1
Z*sZOH*sH
NTF
Z*sZOH*sH
sZOH*sHSTF
Magnitude of loop transfer function
- 19 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
-30
-25
-20
-15
-10
-5
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Notice that
Oversampled A/D Conversion: Effects of ZOH
S
Tj
TzPhase
eZ S
1
H(s) ZOH
Z-1
X Ve Vy Y
Ef
Eq
S
Ed
SS
Eh
-30
-25
-20
-15
-10
-5
0
0.1 1
S
S
S
f
f
f
fsin
f
fcsin
Sf
f
Sf
fcsinfZOH
dB
1 dB
4 dB
20 dB
- 20 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
L(f) is defined as the loop gain
Phase of L(f) is quite important for stability
Z is quite relevant for phase of L(f)
Oversampled A/D Conversion: Stability Issues
fLNTF
fL
Z*fLSTF
1
11
1
2
Z*zZOH*zH)f(L
Tcsin*TzZOH
Tsin*jTcoseZ
SS
SS
Tj S
H(s) ZOH
Z-1
X Ve Vy Y
Ef
Eq
S
Ed
SS
Eh
- 21 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Check the phase contribution of the filter and the delay element!
Check the phase of the loopat the unity gain frequency!
Can’t stabilize the loop employing the conventional filter design approach
Additional parasitic poles!
fLNTF
fL
Z*fLSTF
1
1
1
ADC
Freq0
L(f) H(f)
Fs
Sf
f2
fH
2
2
3
Stability Issues: SD Modulators
- 22 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Stability Issues: SD Modulators
Not very difficult to stabilize the loop if the unity gain frequency is below Fs/4! (not very Low OSR); e.g. around 100Mhz if clock frequency is 400MHz
Notice that larger OSR allow you to reduce the filter gain at high-frequency
Out-of-Band noise is distributed in wider bandwidth, hence smaller noise density but same integrated noise
ADC
Freq
0
L(f)
Sf
f2
fH
2
2
3
H(f) fS0.5fS
Loop
Phase
Hard to implement H(s)
Split the filter in 2 parts
Loop filter and fast path
- 23 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Remarks on Filter’s operation: Filter’s input signal
Effect of the blockers on loop operation
Inband signal is usually very small: Nice property
Transition band is more critical for filter’s linearity (neighbor channels);
Filter must be designed for the blockers
Blocker tolerance is a major issue in broadband applications (LTE)
sL
E*ZEEXV
qhd
e
1
1
H(s) ZOH
Z-1
X Ve Vy Y
Ef
Eq
S
Ed
SS
Eh
Fs Freq
InbandTransition
bandOut of
bandVe(f)
ADC
0 dB
H(f)
ZOH(f)
FsNTF Freq
1/(1+L)
X(f)
- 24 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Typical Quantizer: Flash Architecture S/H operates at clock rate
Huge input capacitance if N>6
Kick back noise
Requires a precise low-
impedance resistive ladder:
Power-accuracy-Speed
tradeoff
Limited by comparator
Speed and accuracy
Offset voltage
Hard to improve its resolution
State of the art: ~ >2.4 GS/s 6 bits resolution
Digital
Signal
Processor
and
MemoryTh
erm
om
etr
ic
to B
ina
ry
+Vref
-Vref
R/2
R/2
R
R
Vin
+
-
+
-
+
-
+
-
clock
Dig
ita
l d
ata
- 25 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Kick-back noise and coupling capacitors
Preamp delay is signal-dependent (Metastability)
Power of the latch could excessive during the resetting phase (M9-ON)
A major challenge: distributing clock signals across 2N-1 comparators
with minimum clock skew (routing, PVT effects)
Input capacitance could be excessive for multi-bit applications
M1 M2
Vi+
M5 M6
M9
Φ
Vo+
Vo-
Fully-diff. PA Latch
M3 M4
Vi-
VR+
VR-
M7 M8Φ
- 26 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Excess loop delay : Constant delay between ideal and implemented DAC
feedback pulse
Decision time required by quantizer affecting latches used for
synchronizing DAC inputs
Finite response time of DAC to its clock and inputs
Filter’s delay is sensitive to PV variations
Excess Loop Delay can be incorporated into: Z-1Z-1-D
Inter-symbol interference : Finite slew rate of DAC outputs with unequal
rise and fall times
Additional noise and tones fold into baseband
Clock jitter in DAC
DAC and Filter Non-linearity
ADC Non-idealities
Processed by STF
Not noise-shaped
- 27 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Remarks on DAC non-idealities. Same analysis apply to the input referred filter’s noise
Oversampled A/D Conversion
dDAC E*STFY
11
Z*sZOH*sH
sZOH*sHSTF
DAC non-idealities are reflected at the ADC output similarly to the signal
Baseband noise can be directly mapped to ADC input (Ed and Eh)
Non-linearities in DAC that translate HF noise into baseband affects directly SQNR
DAC is a critical component
H(s) ZOH
Z-1
X Ve Vy Y
Ef
Eq
S
Ed
SS
Eh
- 28 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Binary-Weighted Current Steering DAC
Current switching is simple and fast.
INL and DNL depend on matching, not inherently monotonic.
Thermometric arrays usually lead to better INL figures
Large component spread (2N-1:1)
VX
Vo
b3 b2 b1 b0 A
I/2 I/4 I/8 I/16
R
N
1jj
j-N
o2
bIV
- 29 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
This is a major research area in multi-bit sigma-delta modulators. Several alternatives have been reported; main trends are:
Randomize the errors such that the non-linear errors are converted in noise instead of tones that degrades SNDR Dynamic element matching techniques
Pseudo Randomizers such as Rotators
Digital signal processors
Drawbacks?
Calibration Usage of large overdrive voltages and large dimensions
Pre-calibration of current cells
Other options?
DAC Calibration (Signal-to-Distortion Ratio)
- 30 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Dynamic Element Matching: Simpified scheme
Representing DAC input using a thermometer DAC
I1
I2
I3
I4
I5
I6
I7
I8
I1
I2
I3
I4
I5
I6
I7
I8
I1
I2
I3
I4
I5
I6
I7
I8
I1
I2
I3
I4
I5
I6
I7
I8
Data -out= 1
I1
I2
I3
I4
I5
I6
I7
I8
I1
I2
I3
I4
I5
I6
I7
I8
I1
I2
I3
I4
I5
I6
I7
I8
I1
I2
I3
I4
I5
I6
I7
I8
Data-out= 3
Data-Out= 1 can be represented by any one of I1-8
Averaging all possible combinations produces the ideal output
Use different combinations to represent a given code
Errors are randomized, and do not generate harmonics but noise
t
0 Ts 2Ts 3Ts
t
0 Ts 2Ts 3Ts
- 31 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Implementation of DEM Scheme
Shifter performs a Rotate-right shift on its inputs
PN-sequence generator indicates number of shifts
DEM is operated at higher frequency to minimize its delay
Routing and extra cells increase the bill of silicon and power
Excessive delay for high frequency applications
Noise level increases
- 33 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
A reference current source is used At the end of a calibration cycle, CCAL attains the correct VGS
required to generate Iref
Analog Self-Calibration of Current Sources
R. T. Baird and T. S. Fiez, IEEE TCAS-II, December 1995.
- 35 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Simulation Results: DEM increases noise level
100
101
102
103
-140
-120
-100
-80
-60
-40
-20
0
Frequency (MHz)
Po
we
r (d
B)
Output Spectrum
Ideal
1% mismatch
1% mismatch +DEM + Calibration
Mismatch DEM Calibration SNR (dB)
N N N 73.8
Y N N 55
Y Y N 63.6
Y N Y 70
Y Y Y 72.7
We found thatCalibration (Digital Memory) looks better for wideband applications, but..
OFFLINE!
- 36 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Clock Jitter Sensitivity (SJNR)
The main effects of clock jitter is present at the input of the quantizer and DAC.
Jitter induced noise at DAC output is processed according to the STF, which is a serious problem for continuous-time sigma-delta modulators
Clock Jitter in quantizer
is not very relevant:
processed by NTF.
Clock jitter introduce
uncertainty at the DAC
output (jitter induced
noise) processed by STF.
Vin
Loop Filter Quantizer
Dout
VDAC
Jittered
Clock
Re-timed
Data
Multi-Bit
DAC
Flip-Flops
Jittered
Clock
- 38 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Jitter Insensitivity of Switched-Capacitor DACs
•So far, SC-DAC is the most jitter tolerant approach•Quite precise and may not require further calibration procedures
• Drawbacks of the SC DAC•Peak current in the SC-DAC can be as high as 5 times IDAC used in the current mode DAC!•Very demanding SR and GBW for the OPAMP: Power consumption; RC discharge is a good trade-off for medium speed architectures!•Class AB OPAMP may help while dealing with this issue
Jitter error
- 39 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Phase Modulation: wide frequency spectrum
Phase Modulation:
Then
Typical phase noise measured in the spectrum analyzer
RMS value of total jitter
22
2
n
no
T
T
tcos
D
dffT
JRMS nper
2
2
0
2
2
dffLT
JRMS per
L (in dBc)
1010
dBcL
fL
Clock Jitter Sensitivity (SJNR)
- 40 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
The DAC error due to clock jitter:
In the frequency domain: Differential of Dout convolves with Jn()
In-band signal is shaped by , then it is not very criticalOut-of-Band quantization noise and blockers convolve with the clock jitter
T
tTnDnDnE outoutj
D1
Jitter Issues: High Clock Frequency
noutS
noutj JDT
sinJDZE2
21 1
11 Z
Error Function
Vin
Loop Filter
Quantizer
Dout
VDAC
Jittered
Clock
Re-timed
Data
Multi-Bit
DAC
Flip-Flops
Jittered
Clock
- 41 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
ZJZV*NTFZ
ZJZV*STFZ)Z(E
Therefore
ZJZVZNTFZV*ZSTFZ)Z(E
Or
ZJZVZ)Z(E
thenT
ntnVnV)n(e
q
inj
qinj
outj
outoutj
1
1
1
1
1
1
1
1
1D
dJVNTFZP q
OSR/
Q,j
21
21
012
dEP j
OSR/
S,j
221
02
Quantization noise
Clock NoiseDensity
SJNR is function of the convolution of NTF and J(Z)
Effect of clock jitter and Quantization noise on SNR
2
21 1 sTsinZ
NTF*Vq
Jn
f
f
Jn
fs-fs
- 42 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
2
2
1021
10
4
2sinc
101
10
s
j
soin
BW
onQuantizatin
inNRZ
T
T*P*OSR
logdfVZJ
PlogSJNR
Continuous-Time BP-ΣΔ ADC: Jitter effects
In absence of blockers, assuming white jitter spectrum, the jitter induced noise has been usually approximated as:
LPF(s) HoldVin
Data
out
CLK
N-bit
Quantizer
Multi-bit
DAC
S Sampler
Z-1
S
Qn
ffck
+
-
0 fckf00 fckf0
0 fckf0
H(s) HoldVin
Data
out
CLK
N-bit
Quantizer
Multi-bit
DAC
S Sampler
S Z-1
JnX
S
Qn
ffck
+
-
fck
1-Z-1
Quantization
and Jitter
Jitter induced noise
f0
0 fckf00 fckf0
0 fckf0
0 fck
f0
0 fckf0
2sTsinc
sj
j
sin
NRZ
T
TPOSR
dBSJNR
3
6
2
10
10
4
*
60
- 43 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
BW
onQuantizatin
BW
kerBlocn
kerBlocno
BW
kerBlocnonQuantizatin
in
dfVZJ
dfVZJ
logSJNR
dfVZJVZJ
PlogSJNR
21
21
10
21
21
10
1
1
110
11
10
Continuous-Time ΣΔ ADC: Jitter & Blockers
High frequency blockers convolve with Jn and are folded back in band.
RF clock filtering is an method to partially overcome this issue!
Considering
Pblocker and
Pquantization
LPF(s) HoldVin
Data
out
CLK
2-bit
Quantizer
Multi-bit
DAC
S Sampler
S Z-1
JnX
S
Qn
ffck
+
-
fck
1-Z-1
Quantization
and Jitter
Jitter induced noise
f0
0 fckf00 fckf0
0 fck
f0
0 fckf0
0 fckf0
0 f0 fck
- 44 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
H(s)Vin
D
Multi-bit
ADC
Multi-bit
DAC
DAC
ADCS
S Z-1
Jn
ffck
ffck
Lowpass
Filter
Blocker
Desired
Images
ffck
ffck
ffck
fck
fck
D
DY
Y
Y
a
b
*
*
*
1-Z-1
10-1
100
101
102
-140
-120
-100
-80
-60
-40
-20
0
SNRIdeal
= 71.2dB
SNRJitter
= 67.2dB
Frequency (MHz)
Po
we
r (d
BF
S)
No jitter tone
Jitter tone at 497.43Mhz
Jitter tone around Fsconvolve with images ofinband signal and producein-band noise tones
10-1
100
101
102
-140
-120
-100
-80
-60
-40
-20
0
SNR = 39.3dB
SNR = 71.2dB
Frequency (MHz)
Po
we
r (d
BF
S)
fj =80.26Mhz, f
b =75.27Mhz
Ideal
Jitter and blocker
convolves, producing in-
band noise tones.
Jitter also convolves with
OOB quantization noise
and downconverts into in-
band noise.
- 45 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Another Major Issue: Loop SaturationSignal Bandwidth 20MHz
Oversampling Ratio 10
Order 5
No. of Quantizer levels 9
Peak SQNR 80dB
Max. NTF Gain 10dB
Max. Stable Input Amplitude -3dBFS
BIQUAD1
LP1
BP1
VIN
RIN
NR
Z
DA
C1
BIQUAD2
LP2
BP2
LPF1
LP3
kLP3
kLP2
kBP2
kBP1
Quantizer
_
NR
Z
DA
C2
DAC
Controller
Clock
Generator
VOUT
DAC Control Signals
kLP1
107
108
-60
-50
-40
-30
-20
-10
0
10
20
30
Frequency (Hz)
Ga
in (
dB
)
BP1
LP1
BP2
LP2
105
106
107
108
-60
-40
-20
0
20
Ma
gn
itu
de
(d
B)
Frequency (Hz)
STF FB
STF FF
NTF(z)
Feed-forward and feedback architectures
Internal gain: BP1, BP2 and LP1 are critical nodes
- 46 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
“Blocker and Jitter Tolerant Wideband SD Modulators” Silva-
Karşılayan-Geddada, MWCAS-12
BP2
LP1
Quantizer
Z-1
Clock
Generator
YOUT
NR
Z
DA
C1
NR
Z
DA
C2BP1
RIN
VIN
Active-RC
5th
order Filter
LP2
LP3
I DA
C2
I DA
C1
PGA
SATURATION
DETECTORVCONTROL
+
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10-6
0.4
0.5
0.6
0.7
0.8
0.9
Time (sec)
VO
UT
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10-6
0
0.5
1
VIN
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10-6
0
0.5
1
VD
ET
=0.5u sec
10-1
100
101
102
-120
-100
-80
-60
-40
-20
0
Frequency (MHz)
Po
we
r (d
BF
S)
SNR = 75.1dB
10-1
100
101
102
-120
-100
-80
-60
-40
-20
0
Frequency (MHz)
Po
we
r (d
BF
S)
SNR = 46.5dB
Vin = -6dBFS @ 5MHz
No blocker
Vin = -6dBFS @ 5MHz
Blocker = -9dBFS @ 40MHz
Attenuation factor = 12dB
Transient response of the
ADC to the 8.7dBFS blocker
at 100MHz in the presence
of a -6dBFS in-band signal
at 5MHz
Effectiveness of the
saturation detector:
Output spectrum of the
ADC with and without
the -9dBFS blocker at
40Mhz
Stable Operation
- 47 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
System-level Optimization
7
8
9
10
11
2.5
3
3.5
472
74
76
78
80
82
84
86
88
90
No. of Quantizer levels
Peak SQNR vs NTFmax
, No. of Quantizer levels
NTFmax
Pe
ak
SQ
NR
(d
B)
Order of modulator (L), and OSR set by target SQNR (~80dB) and
settling time requirements on 1st integrator stage
L = 5, OSR = 10
Order of modulator (L),
and OSR set by target
SQNR (~80dB)
L = 5, OSR = 10
A number of tradeoffs
involved
Just picking
values is not a
good approach
EDUCATED
GUESS!
- 48 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
System-level Optimization
0
2
4
6
8
0
2
4
6
868
70
72
74
76
78
80
82
84
86
Q2 (
0 = 10.8MHz)
Peak SQNR vs Biquad Quality FactorsL = 5, OSR = 10, N = 9
Q1 (
0 = 18.1MHz)
Pe
ak
SQ
NR
(d
B)
Order of modulator
(L), and OSR set by
target SQNR (~80dB)
High-Q filters give
you better SQNR
(due to picking in
the gain; better in-
band noise)
Implications on
filter’s signal
swing and out-
of-band noise?
- 49 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
System-level Optimization
7
8
9
10
11
2.5
3
3.5
4
-5
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
NTFmax
MSA vs NTFmax
, No. of Quantizer levels
No. of Quantizer levels
MS
A (
dB
)
Maximum Signal
Amplitude Increase
with number of levels
and with small NTF
values
Signal at internal
nodes could limited
even more MSA
This is a multi-
dimensional
problem: several
tradeoffs
involved
- 50 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Data
Out
High-Order
LP Filter
ADC
DAC
Input
SignatureAmplitude/Phase
Comparators Filter tuning
schemeFilter tuning
scheme
VinVin
Calibration
Algorithm
cal
op
cal
System Optimization: Tuning Filter Parameters
Calibration of (standalone) building blocks:Does not guarantee loop stability (excess loop delay) Does not guarantee best NTF (Coefficient adjustments)
0 f0
f
Effect of PVT Variations Filter Tuning; reconfiguration orMaster slave techniques could be used
- 51 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
System Optimization: Global Tuning Scheme
Linear model used to findNTF
Model used to calibrate NTF
Loop response to calibration tones is analyzed in digital domain
F. Silva-Rivas, et. al., “Digital Based Calibration Technique for Continuous-Time Bandpass Sigma-Delta Analog-to-Digital Converters,” Analog Integrated Circuits and Signal Processing, April-09.
C.Y. Lu, et. al., “A Sixth-Order 200MHz IF Bandpass Sigma-Delta Modulator With over 68dB SNDRin 10MHz Bandwidth,” IEEE J. Solid-State Circuits, June 2010.
- 52 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Global Tuning Scheme for NTF Parameters: Digital
A set of strategic tones are used to optimize:Loop StabilityBest NTFBandwidth
Frequency
Testing tones Output tones
Vin
QnQuantizer
Dout
β
Feedback DAC
ZOH
Zero-Order
Hold
Calibration
Tones
Software
Platform
FFT Transform, Power
Detector and
Calibration Algorithms
Controller: C,
Coefficients and loop
delay
Loop Filter
- 53 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Filter's
control
Software
Platform
DSP
N-channel
digital
receiver
Wideband
Programmable Filter
Dout
Quantizer
NRZ DAC
z-1/2
Programmable
z-1/2
Array of
DACs
Vin
Calibration
ff0
Test Tones
DACs
Control
Calibration
Algorithm
After
DAC & Freq
calibration
Digitally assisted calibration scheme
Freq calibration
Data control
A 6th-Order 200MHz IF Bandpass Sigma-Delta Modulator With over 68dB SNDR in 10MHz Bandwidth, C.Y. Lu, et.al., IEEE J. Solid-State Circuits, June 2010.
TSMC 1P6M 0.18um
- 55 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Strategy for nanometric technologies
Make use of time domain signal representation
Small supply voltage limits the dynamic range of analog signals
Fast switching gates provide fine time precision, which provide alternate means for high DR signal representation
Also takes advantage of CV2f
0
Vdd
0 Ts
Am
plit
ud
etime timeA
mp
litu
de
Vdd
VQ
TQ
0
2T
s
4T
s
0
2T
s
4T
s
Time domain signal representation
A 20MHz Signal Bandwidth 68dB Dynamic Range Continuous Time DS ADC Based On Multi-bit Time
Domain Quantizer and Feedback Element, V. Dhanasekaran, et.al. ISSCC-2009; JSSC March 2011
- 56 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Proposed ADC Architecture
• Multi-level quantizer and Digital to Analog Converter (DAC) are
replaced by PWM generator and Time to Digtal Converter (TDC)
• Width of p(t) is proportional to the amplitude of the signal in a given
clock period
• Output code (Dout) represents “quantized pulse” edges with a
quantization step size = TQ
Conventional DS Time domain Quantizer-DAC
S+
-
Vin Dout
Multi-bit
DAC
CLK
Multi-bit QuantizerLoop Filter
1-Level
DAC
Loop Filter
Time Quantized PWM
Digital
OutputS
+
-
Vin
TDC &
Digital Logic
CLK
CLK
PWM Generator
Decimation
Filter
THA
P(t)
Dout
- 57 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
PWM Generator
PWM signal generated by standard method - comparing signal with triangle waveform
Double-sampled or “asymmetric sampled” PWM is used to minimize distortion and improve OSR
Performance is relaxed due to noise shaping
S+
-
Vin
pq(t)
Loop Filter
CLK
TDC
PWM Gen
Dout
p(t)
- 58 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
TDC Functionality
Time-to-Digital Converter
Converts pulse timing information to digital codes (Drout & Dfout)
Drout & Dfout can be used to reconstruct signal in clocked digital domain
Also generates a feedback pulse aligned to the timing edges represented by the dotted lines
0 2 4 6 8 1210 14
TCLK
TQ
TDCDrout
Dfout
(6)
(11)
CLK
- 59 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
TDC & Time quantized feedback pulse
Digital output code
CT PWM output p(t) is latched by delayed versions of the clock
Flipflop outputs provide “thermometer” coded pulse width
D input of the flipflops are driven by p(t) to facilitate easy generation of f/b pulse
Feedback pulse - pq(t)
Must match output code
Edges of pq(t) are always aligned to the delayed clock edges
D
D
D
D
CK0
CK1
CK2
CK3
p(t)
D
D
D
D
CK4
CK5
CK6
CK7
p(t)
RST
RST
CK1
CK2
CK3
CK4
CK5
CK6
CK7
CK0
RST
RST
p(t)
TQ
TCLK
62
Binary code
pq(t)
- 60 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Time quantized pulse generator
Pq(t) can be generated using a OR gates and SR latch
Pq(t) turns ‘High’ when the earliest of CK0-3 goes High after p(t) is High
Pq(t) turns ‘Low’ when the earliest of CK4-7 goes High after p(t) is Low
D
D
D
D
CK0
CK1
CK2
CK3
p(t)
D
D
D
D
CK4
CK5
CK6
CK7
p(t)
+
+
S
Rp
q(t)
RST
RST
Thermo-to-Bin
DRout
DFout
Thermo-to-Bin
Wired-NOR/NAND gate can be
used to ensure uniform delay
- 61 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Full TDC Schematics
50 levels of time steps are used in the proposed design for 20MHz ADC
Clock phases are generated using cascade of delay elements that are tuned using a phase detector
D D D D D D D D D D D D D D D
1 2 3 4 5 6 7 8 9 10 46 47 48 49 50
P(t)P(t)
CK1 CK2 CK3 CK4 CK5 CK6 CK7 CK8 CK9 CK10 CK46 CK47 CK48 CK49 CK50
CK4 CK5 CK6 CK7CK3 CK9 CK10 CK11 CK12CK8
D
51
D
52
CK50 CK51 CK52CK48 CK49
CK51 CK52
25 Input OR Gate25 Input OR Gate
CLKIN
S R
Pq(t)
- 62 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Active-RC Topology Loop Filter
The gain of the filter in the signal bandwidth (20MHz) serves to
suppress the quantization noise of the TDC
The filter has a minimum inband gain of 37dB
CH
CB
Rq
R1
R3
C1 C3
RF
R2
C2
V_bp
V_lp
Vin
Vout
_
+
_
+
_
+A1
A2A3
Noise Transfer
Function (NTF))16e279.1s7e414.1s(s
25e223.4s17e312.2s312.7
)s(Vin
)s(Vout)s(H
2
2
- 63 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Mitigating Excess loop delay
KFB provides a ‘fast path’ through INT3 => Make AMP3 a minimum phase structure
PWM
T
D
C
DAC
KFB path
FILTER
Characterstics Amplifier 1-2 Amplifier 3
DC Gain High (>30dB) Moderate-low (20dB)
Gain-Bandwidth Moderate (150-250MHz) High (>500MHz)
Supply Noise Rejection
High (>40dB) Moderate
- 64 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Schematic: Amplifier1 / Amplifier2
Two Stage ‘Typical’ configuration
maximizes gain of the initial stages => Better Q control
Output pole ~ 4*GBW
Overall noise limited by input devices
Rf reduces (degenerate) noise due to NMOS current mirrors
Limitations:
• Parasitic poles result in additional loop delay/excess phase in
the filter (could be > 10 deg at 500MHz)
•Limits loop stability and affects the overall performance
Pbias
VDD
VSS
Vinp
cmbias
M1 M2
M9 M8
M6
Vinm
M3 M4
Pbias
cmbias
Voutm
M7
M5
Voutm
Rf
Cc Rc Rc Cc
Rf
- 65 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Simplified schematic: Amplifier 3
A simple single stage pseudo-differential ‘inverter based’ amplifier
Complementary structure with class A-B action
CM noise (supply etc) is suppressed by biquad gain of 18dB
Limitations:
•Bias point not accurately controlled + % variation across
statistical corner & Temp
•Need for ‘Non-conventional’ common-mode control
Single Ended Schematic of Amp3
VDD
GND
Vinp Vinm Vout
- 66 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Amplifier 3 Common Mode control
Current sources at virtual ground control amplifier’s common
mode
Icm are: small current source, sized to meet transconductance
and noise specifications
CH
R3
CB
R3
C3
C3 Cmref
Icm
Icm
EA
Amp3From
Biquad
CB
- 67 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Chip Micrograph
Folded layout to minimize delay in feedback
Digital approach minimizes area requirement
Occupies 0.15mm2
- 68 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Output Spectrum: -5dB Input
Output spectrum for -5dB 4MHzInput signal
• SNR=62dB
• THD=65dB
- 69 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Dynamic Range and SNDR
DR = 68DB
Peak SNDR=60dB @ -5dB input
Peak THD=67dB @ -6dB input
FoM = 319 fJ/step
Dynamic Range (DR) is defined as the amplitude
range with SNR>0dB
- 70 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
tntmCosnmSin
nm
MnmJV
tmCosmSinMmJm
V
tnCosnSinMnJ
n
Vtv
SR
m n
R
S
R
Sn
a
R
m
oa
S
R
Sn
n
R
S
ap
2
22
22
12
22
12
1
1
1
The PWM spectrum consists of tones at
Signal frequency (S) and its odd harmonics,
Reference tones due to the ramp’s fundamental frequency (R) and its harmonics,
Intermodulation products of the signal and the reference tones.
- 71 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Second ADC:System Architecture: Injection Locking
OUT
b1
b2
b4
b5
Σb3
Level-to-PWM
Converter
VCO
3-Bit Two-Step
Quantizer
Vin
Digital
Output
3-Bit NRZ
DAC
1-B
it P
WM
DA
C
BP
LP
2nd
-Order
Filter
Complementary Injection-
Locked Frequency Divider
Φ1
Φ2
Φ3
Φ4
Φ5
Φ6
Φ7
Ts
0
Ts/7
2T
s/7
3T
s/7
4T
s/7
5T
s/7
6T
s/7
+
-
+
-
+
-
RF
RF
RF
RQ
RQ
CC
C C
+-
+-
RF
BP
LP
2nd
-Order
Filter
Rin
Rin
Programmable Delay
4
11
OUTBP
OUTLP
+
-
Rg
C
C
+-
Rg
RC
RC
IN IN
LP
1st
-Order
CILFD
• 5th-order 3-bit feedforward architecture
• Local feedback is to compensate the excess loop delay
• LC-VCO+CILFD are used to generate clean reference clocks
“25MHz Bandwidth (BW) Continuous-Time Lowpass ΣΔ Modulator with Time-Domain3-bit Quantizer and DAC” Cho-Ying Lu, et.al., Sept 2010, JSSC
- 72 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Rq
R
1
C1
RF
R2
C2
V_bp
Vin_
+
_
+A1
A2
-1
RL1 CL1RL2 CL2
gm1 gm2
gmf
in out
V_lp
Two stage amplifier with
feedforward compensation
1st stageVIN+ VIN-
VOUT- VOUT+
VCMFB
VOUT+
VOUT-
VREF
VCMFB
2nd stage
feedforward
stage
CMFB
stage
MN1 MN1
MP1MP1
IB1
MN2 MN2
MN3 MN3
MP2A
MP2B
R1R1
MP4 MP4
MN4 MN4
IB2
IB3
IB4
MP2AMP2B
R2
R2
C2
C2
VIN-VIN+
First stage Second stage
No extra CMFB (current re-use)
needed
Large resistors are used
Feed-forward stage
Biquadratic Filter and Amplifier
Quiescent current ~ 1.0 mA
GBW > 800MHz
- 73 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
2IF
MFMF
Feed-Forward Path
2I3
M3M3
Vcmfb
-Vout ++
Vin
-
2IL
MLML
Linearity
Compensation
AV2AV1
VCM
2I1
R1 R1
-30 -28 -26 -24 -22 -20 -18 -16 -12 -10
45
50
55
60
65
70
75
80
85
90
95
IM3
(d
B)
Output Signal (dBV)
17dB improvement
Convetional
Linearized
Very linear OPAMP for >100MHz applications
C.-Y. Lu, “A 6th-order 200MHz IF
Bandpass Sigma-Delta Modulator
With over 68dB SNDR in 10MHz
Bandwidth,” JSSC, June 2010
- 74 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
VOUT
R1
R1
Gm
VBP1-
VLP1-
VBP2-
VLP2-
VBP1+
VLP1+
VBP2+
VLP2+
R2
R3
R4
R5
R2
R3
R4
R5
VLP3-
VLP3+
RA
RB RC
RA
RCRB
CT
CT
RDAC2
VDAC2-
RDAC2
VDAC2+
CL
CL
Wide-band Adder with delay compensation
)(tan)(tan)( 11
pz
))((
))(()(2222
2
pz
pzpz
d
ddelay
The load resistor is split into 2 pieces and
one section replaced by an RC T-network
Rfeedback = Ra||(Rb+Rc) at low frequencies
Rfeedback = Ra at low frequencies
- 75 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Pulse Amplitude Modulation
Traditional Multi-level Digital Signal
Unit element current mismatch generates DAC non-linearity.
Ts
I
I/7
4I/7
sinj TIQ )(
]1,....7
2,
7
1,0[
where:
Ts
I
I/7
4I/7
Ts
I
I/7
4I/7
Ts
I
I/7
4I/7
- 76 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Time-Domain Pulse Width Modulation
Only inherently linear 1-level DAC achieve multi-bit feedback
PWM scheme employing multi-phase controller
Ts/7
)( sinj TIQ
]1,....7
2,
7
1,0[
where:
I
2Ts/7
I
4Ts/7
I
Ts
I
- 77 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Systematic (Nonlinearity) errors
Error Charge from Device Mismatch
4Ts/7
I
serr ITT
TQ )(2
D
Ts
I
I/7
4I/7
serr ITI
INQ )(
D
Conventional Multi-Level DAC
1-level PWM DAC
Current errors accumulate
Timing errors at the pulse edges
- 78 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
INL, DNL and Linearity
Robustness to device mismatch
only affect the edges of the pulse
%5.0
%16.0
7
2
3
3
%
%
I
T
alconvention
PWM
HD
HD
PWM Linearity could be >10dB better!
%16.0% D
T
TT
%5.0% D
I
II
- 79 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Time-Domain Pulse Width Modulation
Pulse Arrangement To ease the design of interface circuitry
between quantizer and level-to-PWM converter
001
010
011
000 100
101
110
111
Ts
0
Ts/7
2T
s/7
3T
s/7
4T
s/7
5T
s/7
6T
s/7Ts
0
Ts/7
2T
s/7
3T
s/7
4T
s/7
5T
s/7
6T
s/7
-I
-I
-I
-I
+I
+I
+I
+I
22
2
102
10 y
speak
OSRTlogSJNR
- 80 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
LC-VCO + CILFD
Phase noise of VCO is -119dBc/Hz @ 1MHz
CILFD phase noise is -136dBc/Hz @ 1MHz
b1
b2
b4
b5
Σb3
Level-to-PWM
Converter
VCO
3-Bit Two-Step
Quantizer
Vin
Digital
Output
3-Bit NRZ
DAC
1-B
it P
WM
DA
C
BP
LP
2nd
-Order
Filter
Complementary Injection-
Locked Frequency Divider
+
-BP
LP
2nd
-Order
Filter
Programmable Delay
4
11
LP
1st
-Order
CILFD
Φ1P
7-stage chain
phase order:
Φ1,Φ5,Φ2,Φ6,Φ3,Φ7,Φ4
Φ1N
Φ5P Φ5N
Φ4P Φ4N
VCOP
Vbn
VBP VBPVCON
Mpt
Mnt
Mp
Mn
7-stage chain
Vbn
(off-chip)
(off-chip) (off-chip)
(off-chip)
Jitter among phases is highly correlated
- 81 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
The output is composed by 1 MSB + 3LSB
The MSB is determined first
Vin S/H
CLK
Vref1
Vref2
Vref3
Iin
Iref1
Iref2
Iref3
4:1 MUX
MSB
IrefCTRL
Rcmp
MSB
B2
B1
B0
Gm
Gm
Gm
Gm
Gm
VrefDC
Vcmp
IrefDC
(τ1)
(τ2)
(τ3)
(τ4)
τ1
τ4
τ3
τ2
b1
b2
b4
b5
Σb3
Level-to-PWM
Converter
VCO
3-Bit Two-Step
Quantizer
Vin
Digital
Output
3-Bit NRZ
DAC
1-B
it P
WM
DA
C
BP
LP
2nd
-Order
Filter
Complementary Injection-
Locked Frequency Divider
+
-BP
LP
2nd
-Order
Filter
Programmable Delay
4
11
LP
1st
-Order
CILFD
+Vref3 = +150mV
VrefMSB = 0V
0 Ts/7 2Ts/7 3Ts/7 4Ts/7 5Ts/7 6Ts/7 7Ts/7
B2 B1 B0
successive
comparisons to Vref (Iref)
S/H tracking
(for next cycle)
MSBsampling
MSB > VrefDC
MSB < VrefDC
Time
τ1
Differential Voltage Range= decision (latching) instant for a bit
sampling
+200mV
-200mVτ2 τ3 τ4
+Vref2 = +100mV
+Vref1 = +50mV
-Vref1 = -50mV
-Vref2 = -100mV
-Vref3 = -150mV
3-bit Algorithmic Quantizer
- 82 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Proposed Level-to-PWM Converter
SR latches generate the pulses
• AND gates + OR gates decide the required pulses
• Programmable Delay is used to minimize excess loop delay5-Input OR Gate
MSB B2
5-Input OR Gate
MSB B2
to 1-Bit
PWM DAC
Φ1
Φ3
ΔT
S
R
Q S
R
Q S
R
Q S
R
Q S
R
QΦ6
Φ1
Φ4
Φ5
Φ3
Φ4
Φ5
Φ6
B1B1
B0B0
S
R
Q
Q
Φ6 Φ4 Φ2 Φ1 Φ3
D Q
D QD Q
D Q
D Q
D Q
D Q
D Q
D Q
D Q
Programmable
Delay
b1
b2
b4
b5
Σb3
Level-to-PWM
Converter
VCO
3-Bit Two-Step
Quantizer
Vin
Digital
Output
3-Bit NRZ
DAC
1-B
it P
WM
DA
C
BP
LP
2nd
-Order
Filter
Complementary Injection-
Locked Frequency Divider
+
-BP
LP
2nd
-Order
Filter
Programmable Delay
4
11
LP
1st
-Order
CILFD
- 83 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Loop Filter +
Summing Amplifier
3-Bit Two-Step
Quantizer
VCO & Complementary
Injection-Locked
Frequency Divider
Level-to-PWM
Converter
DAC1 &DAC2
Buffers
Core area = 2.6mm2RMS jitter= 0.27ps => SJNR > 75dB
Chip Microphotograph & Clock Phase Noise Fabricated in Jazz Semiconductor, 180nm technology
- 84 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Output Spectrum of the Modulator
104 10
510
610
710
8
-140
-120
-100
-80
-60
-40
-20
0
Frequency (Hz)
Magn
itu
de
(dB
FS
)
78d
B
-2.2dBFS @ 5.08MHz
-70 -60 -50 -40 -30 -20 -10 0
0
10
20
30
40
50
60
70
Input Signal (dBFS)
SN
R &
SN
DR
(d
B)
-4.5 -4 -3.5 -3 -2.564
65
66
67
68
69
SNR
SNDR
DR=69dB
-5-5.5
Peak SNR= 68.5dB @25MHz BW
Peak SNDR= 67.7dB @25MHz BW
SFDR>70dB
- 85 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Signal to Distortion Ratio
Two tones with 2MHz apart and -2dBFS overall power
-72
-73
-74
-75
-76
0 5 10 15 20 25
Two-Tone Average Frequency (MHz)
IM
3 (
dB
)
-120
-100
-80
-60
-40
-20
0
107
Frequency (Hz)
Ma
gn
itu
de
(dB
FS
)
IM3
= -
76
.3d
B
- 86 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Quite relevant topology:
Quantizationnoise E1 gets cancelled
Notice that the auxiliary ADC processes the quantization error E1
What about Eq2?
2241241313
221241113
22122
1111
2413
qqq
q
ENTF*HESTF*HNTF*HXSTF*HzY
E*NTFE*STFHE*NTFX*STFHzY
Then
EzNTFEzSTFzY
EzNTFXzSTFzY
VzHYzHzY
Can we do better?Fundamentals on MASH architectures
Digital
+
1qE
DAC
Loop
Filter
ZY1 ZX ZH1
ZH2
+
2qE
ZY2+
H3
H4
+
ZY
ADC
+
-
+
-
Leslie and SinghISCAS-1990
- 87 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Schreier-Temes
Analog Intensive MASH topology
2
11
1
2
2
2
2
2
1
22413
2413 0
q
q
E*NTF
X*STF
NTF
STF
E
ESQNR
ENTF*HXSTF*HzY
Then
STF*HNTF*HIf
Fundamentals on MASH architectures
+
1qE
DAC
Loop
Filter
ZY1 ZX ZH1
ZH2
+
2qE
ZY2+
H3
H4
+
ZY
ADC
+
-
+
-
- 88 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Analog Intensive Pseudo-MASH topology
Fundamentals on MASH architectures
RIN
kLP2
kBP2
kBP1
4-Bits
_
DELAYDigital signals (D1-D7)
Active-RC
4th
order
Filter
NR
Z
DA
C2
Clock
Generator
DA
C
Software
Platform
High-Resolution
Pipeline ADCN-bits
kLP1
CLK
Vin
Todd Brooks, et. al., JSSC, Dec 1997
- 90 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Target: SNDR Improvement by 3 bits
SQNR improvement should be around 18dB (3 bits)
Power overhead, area overhead < 20%
105
106
107
108
-160
-140
-120
-100
-80
-60
-40
-20
0
X: 5.005e+006
Y: -6.02
Power Spectrum of Output1
Frequency (Hz)
Po
we
r (d
BF
S)
SNR = 79.516dB
SNR = 105.803dB
Output With Cancelation
Output Without Cancelation
- 91 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Conclusions
Minimization of filtering functions in the receiver chain demands innovative ADC solutions;
SNDR > 80 dB will be frequently needed in wireless applications;
Jitter and Blocker tolerant architectures are needed;
DAC calibration techniques for fast and high-resolution applications;
Global tuning strategies (for stability)
Advanced LTE applications require high-resolution for signal BW as high as 200MHz
- 92 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
Recommended Readings
S. Norsworthy, R. Schreier and G. Temes, Delta-Sigma Data Converters: Theory, Design and Simulation, IEEE Press
R. Schreier and G. Temes, Understanding Delta-Sigma Data Converters, IEEE Press
J. Cherry, M. Snelgrove,Theory, Practice, and Fundamental Performance Limits of High-Speed Data Conversion Using Continuous-Time Delta-Sigma Modulators, Kluwer Academic Press.
J. M. de la Rosa, Sigma-Delta Modulators: Tutorial Overview, Design Guide, and State-of-the-Art Survey, TCAS-I, Jan 2011.
Other tutorials: IECE-2012
- 98 -J. Silva-Martinez
TAMU-ECE MWCAS-2012
98
Nice feature: Robust against blockers!
The DEM is usually slow and may lead to significant loop delay!
May require loop compensation