• proportional term (P) depends on the present error
  • integral term (I) depends on past errors
  • derivative term (D) depends on anticipated future errors

PID controller makes use of linear extrapolation of the measured output

PI controller does not make use of any prediction of the future state of the system

The prediction by linear extrapolation (D) can generate large undesired control signals because measurement noise is amplified, that's why D is not used widely


TODO 📅

reference

Gene F. Franklin, J. David Powell, and Abbas Emami-Naeini. Feedback Control of Dynamic Systems, Global Edition (8th Edition). Pearson. [pdf]

Åström, K.J. & Murray, Richard. (2021). Feedback Systems: An Introduction for Scientists and Engineers Second Edition [pdf]

Dawson, J. L. (2021). A guide to feedback theory. Cambridge University Press.

Yan Lu, ISSCC2021 T10: Fundamentals of Fully Integrated Voltage Regulators [https://www.nishanchettri.com/isscc-slides/2021%20ISSCC/TUTORIALS/ISSCC2021-T10.pdf]

image-20241004163356709

charge pumps are capacitive DC-DC converters. The two most common switched capacitor voltage converters are the voltage inverter and the voltage doubler circuit


image-20241014211627207


voltage doubler

image-20241019092038444

output buffer capacitor

To achieve a stable DC output voltage

Step-Wise Ramp-Up

\[ V_{in} C_p + V_{out,n-1}C_o = (V_{out,n}-V_{in})C_p + V_{out,n}C_o \]

We derive a recursive equation that describes the output voltage \(V_{out,n}\) after the \(n\)th clock cycle \[ V_{out,n} = \frac{2V_{in}C_p + V_{out,n-1}C_o}{C_p + C_o} \]

Voltage Ripple & Droop

ripple_droop.drawio

\[\begin{align} (V_t - V_h)(C_p + C_o) &= \frac{I_{load}}{2f_{sw}} \\ (V_h - V_b)C_o &= \frac{I_{load}}{2f_{sw}} \end{align}\]

we obtain \[ V_t - V_b = \frac{I_{load}}{f_{sw}C_o}\left(1 - \frac{C_p}{2(C_p + C_o)}\right) \] That is, peak-to-peak ripple \[ \Delta V_{out,p2p} \approx \frac{I_{load}}{f_{sw}C_o} \space\space\space\space \text{if}\space\space C_o \gg C_p \]

Then, with aforementioned Step-Wise Ramp-Up equation, \(V_t = \frac{2V_{in}C_p + V_bC_o}{C_p + C_o}\) \[\begin{align} V_b &= 2V_{in} - \frac{I_{load}}{f_{sw}C_p}\left(1 + \frac{C_p}{2C_o}\right) \\ V_t &= 2V_{in} - \frac{I_{load}}{f_{sw}C_p}\left(1 - \frac{C_p}{2(C_p+C_o)}\right) \end{align}\]

Therefore, average output voltage \(\overline{V}_{out}\) in steady-state is \[ \overline{V}_{out} = \frac{V_t+V_b}{2}=2V_{in} - \frac{I_{load}}{f_{sw}C_p}\left(1 + \frac{C_p^2}{4C_o(C_p+C_o)}\right) \approx 2V_{in} - \frac{I_{load}}{f_{sw}C_p} \] which results in a simple expression for the output voltage droop

\[ \Delta V_{out} = \frac{I_{load}}{f_{sw}C_p} \]

The charge pump can be modeled as a voltage source with a source resistance \(R_\text{out}\). Therefore, \(\Delta V_{out}\) can be seen as the voltage drop across \(R_\text{out}\) due to the load current:

\[ R_{out} = \frac{\Delta V_{out}}{I_{load}} = \frac{1}{f_{sw}C_p} \] image-20241015072846141

multiphase CP

multiphaeCP.drawio

\[ (V_t - V_b) (C_p + C_o) = I_{load}\Delta t \]

Therefore peak-to-peak ripple \[ \Delta V_{out,p2p} = \frac{I_{load}\Delta t}{C_p+C_o} = \frac{I_{load}\Delta t}{C_{tot}} \]

where \(C_{tot} = C_p+C_o\)

with \[ \left\{ \begin{array}{cl} V_b &= 2V_{in} - \frac{I_{load}\Delta t}{C_p} \\ V_t &= 2V_{in} - \frac{I_{load}\Delta t}{C_p} + \frac{I_{load}\Delta t}{C_p+C_o} \end{array} \right. \]

Then \[ \overline{V}_{out} = \frac{V_t+V_b}{2}=2V_{in} - \frac{I_{load}\Delta t}{C_p}\cdot \frac{C_p+2C_o}{2C_p+2C_o} \approx 2V_{in} - \frac{I_{load}\Delta t}{C_p} \] That is output voltage droop \[ \Delta V_{out} = \frac{I_{load}\Delta t}{C_p} \]

reference

Bernhard Wicht, "Design of Power Management Integrated Circuits". 2024 Wiley-IEEE Press

Breussegem, T. v., & Steyaert, M. (2013). CMOS integrated capacitive DC-DC converters. Springer

Zhang, Milin, Zhihua Wang, Jan van der Spiegel and Franco Maloberti. "Advanced Tutorial on Analog Circuit Design." (2023).

Anton Bakker, Tim Piessens., ISSCC2014 T9: Charge Pump and Capacitive DC-DC Converter Design

Wicht, B., ISSCC2020 T2: Analog Building Blocks of DC-DC Converters [https://www.nishanchettri.com/isscc-slides/2020%20ISSCC/TUTORIALS/T2Visuals.pdf]

Hoi Lee, ISSCC2018 T8: Fundamentals of Switched-Mode Power Converter Design [slides,transcript]

image-20241019142915175


alternative view of sampling, assuming DC value is \(A\)

sampling-c2d-d2d.drawio

  • \(x_c(t)\) and \(x_s(t)\)

    \(\overline{x_c} = A\); \(\overline{x_s}=\frac{A}{T}\): therefore \(X_s(j0) = \frac{1}{T}X_c(j0)\)

  • \(x[n]\) and \(x_d[n]\)

    \(\overline{x} = A\); \(\overline{x_d}=\frac{A}{2}\): therefore \(X_d(e^{j0}) = \frac{1}{2}X(e^{j0})\)

expander

sampling-expander.drawio

  • \(x[n]\) and \(x_e[n]\)

    \(\overline{x} = A\); \(\overline{x_e}=A\): therefore \(X_e(e^{j0}) = X(e^{j0})\)

    Fourier transform of the output of the expander is a frequency-scaled version of the Fourier transform of the input


Subsampling or Downsampling

image-20241004151215993

image-20241004151308422

image-20241004151434477

  • Eqs. (4.72)

    the superposition of an infinite set of amplitude-scaled copies of \(X_c(j\Omega)\), frequency scaled through \(\omega = \Omega T_d\) and shifted by integer multiples of \(2\pi\)

  • Eq. (4.77)

    the superposition of \(M\) amplitude-scaled copies of the periodic Fourier transform \(X (e^{j\omega})\), frequency scaled by \(M\) and shifted by integer multiples of \(2\pi\)


downsampled by a factor of \(M = 2\)

image-20241004161805974


image-20241005073349726

image-20241005073534041

Upsampling or Zero Insertion

image-20250701070658641

image-20250616212057960


image-20250616215844032

Assuming \(X(e^{j\omega_1}) = U_f(e^{j\omega_1})\) with \(\omega_1 = \Omega T_1\), upsampled by ratio \(L\), then obtain

\[ Y(e^{j\omega_2})=X(e^{j\omega_2 L}) = U_f(e^{j\omega_2 L}) \]

by EQ. (4.85), i.e. substitute \(\omega_1\) with \(\omega_2 L\), where with \(\omega_2 = \Omega T_2\) and \(T_2 L = T_1\)

Provided that \(\xi = e^{j\omega_1}\) and \(z = e^{j\omega_2}\), we have \(U_f(\xi)\) upsampled to \(U_f(z^L)\)

Interpolation filter

image-20250616214711197


image-20250611205725078

Pavan, Schreier and Temes, "Understanding Delta-Sigma Data Converters, Second Edition"


image-20250618225150839

Markus Nentwig. Polyphase filter / Farrows interpolation [https://www.dsprelated.com/showarticle/22.php]


image-20250630230658621

sampling identities

sampling-ID.drawio


downsampling identity

image-20241007085509889

image-20241007090624888


upsampling identity

image-20241007085527233

image-20241007090939701

Polyphase Decomposition

Polyphase decomposition is a powerful technique used in digital signal processing to efficiently implement multirate systems.

image-20241020122709610

image-20241020122726153

where \(e_k[n]=h[nM+k]\)


Polyphase Implementation of Decimation Filters & Interpolation Filters

Decimation system Interpolation system
image-20241020123035001 image-20241020123043829
image-20241020123027067 image-20241020123101780
sampling identity image-20241020123345371 image-20241020123355113

LPTV Implementation

TODO 📅

The interpolation filter following an up-sampler generally is time varying and cannot be represented by a simple transfer function. The equivalent filter in a zero-order hold is an exception, perhaps unique, that can be represented with a time-invariant transfer function

Dr. Deepa Kundur, Multirate Digital Signal Processing: Part I [pdf, https://www.comm.utoronto.ca/dkundur/course/discrete-time-systems/]

ZOH interpolator

The interpolation filter following an up-sampler generally is time varying and cannot be represented by a simple transfer function. The equivalent filter in a Zero-Order Hold is an exception, perhaps unique, that can be represented with a time-invariant transfer function

image-20250627173816810

image-20250627173926092


zoh.drawio \[ F_1(z) = X(z^{LM})\frac{1-z^{-LM}}{1-z^{-1}} \]

Split the \(1:LM\) hold process into a \(1 : L\) hold followed by a \(1 : M\) hold \[ Y(\eta)=X(\eta^{L})\frac{1-\eta^{-L}}{1-\eta^{-1}} \] then \[\begin{align} F_2(z) &= Y(z^M)\cdot\frac{1-z^{-M}}{1-z^{-1}} \\ &=X(z^{LM})\frac{1-z^{-LM}}{1-z^{-M}}\cdot \frac{1-z^{-M}}{1-z^{-1}} \\ &= X(z^{LM})\frac{1-z^{-LM}}{1-z^{-1}} \end{align}\]

That is \(F_1(z)=F_2(z)\), i.e. they are equivalent


image-20241103180315919

Random Signals & Multirate Systems

Balu Santhanam, Probability Theory & Stochastic Process 2020: Random Signals & Multirate Systems [https://ece-research.unm.edu/bsanthan/ece541/rand.pdf]

Decimation by Summing

proportional path

The loop gain of a proportional path is unchanged

phug_loop.drawio

In (a), the loop gain is \(\frac{\phi_o(z)}{\phi_e(z)}\), which is \[ LG_a(z)=\frac{\phi_o(z)}{\phi_e(z)} = \frac{1}{1-z^{-1}} \]

In (b), Accumulate-And-Dump (AAD) is \(\frac{1-z^{-L}}{1-z^{-1}}\), then \(\phi_m(\eta)\) can be expressed as \[ \phi_m(\eta) = \frac{1-\eta^{-1}}{1-\eta^{-1/L}}\cdot \frac{1}{L} \] Hence \[\begin{align} \phi_o(\eta) &= \phi_m(\eta) \frac{1}{1-\eta^{-1}} \\ &= \frac{1-\eta^{-1}}{1-\eta^{-1/L}}\cdot \frac{1}{L}\cdot \frac{1}{1-\eta^{-1}} \end{align}\]

After zero-order hold process, we obtain \(\phi_f(z)\), which is \[\begin{align} \phi_f(z) &= \phi_o(z^L) \cdot \frac{1-z^{-L}}{1-z^{-1}} \\ &=\frac{1-z^{-L}}{1-z^{-1}}\cdot \frac{1}{L}\cdot \frac{1}{1-z^{-L}}\cdot \frac{1-z^{-L}}{1-z^{-1}} \end{align}\] i.e., \[ LG_b(z) = \frac{1}{1-z^{-1}}\cdot \frac{1}{L}\cdot \frac{1-z^{-L}}{1-z^{-1}} \]

When bandwidth is much less than sampling rate (data rate), \(\frac{1}{L}\cdot \frac{1-z^{-L}}{1-z^{-1}} \approx 1\)

Therefore \[ LG_b(z) \approx \frac{1}{1-z^{-1}} \]

In the end \[ LG_a(z) \approx LG_b(z) \]


Assume PD output is constant

phug_seq.drawio

integral path

integral path gain reduced by \(L\)

frug_loop.drawio

In (a), \(\phi_o(z)=\frac{1}{(1-z^{-1})^2}\), i.e. \[ LG_a(z) = \frac{1}{(1-z^{-1})^2} \]

In (b), after Accumulate-and-dump (AAD), \(\phi_(\eta)\) is \[ \phi_m(\eta) = \frac{1-\eta^{-1}}{1-\eta^{-1/L}}\cdot \frac{1}{L} \]

After frequency integrator and phase integrator \[\begin{align} \phi_o(\eta) &= \phi_m(\eta) \cdot \frac{1}{(1-\eta^{-1})^2} \\ &= \frac{1-\eta^{-1}}{1-\eta^{-1/L}}\cdot \frac{1}{L} \cdot \frac{1}{(1-\eta^{-1})^2} \end{align}\] Then \(\phi_f(z)\) is shown as below \[\begin{align} \phi_f(z) &= \phi_o(z^L)\cdot \frac{1-z^{-L}}{1-z^{-1}} \\ &= \frac{1-z^{-L}}{1-z^{-1}}\cdot \frac{1}{L}\cdot \frac{1}{(1-z^{-L})^2}\cdot \frac{1-z^{-L}}{1-z^{-1}} \\ &= \frac{1}{L} \cdot \frac{1}{(1-z^{-1})^2} \end{align}\]

That is, \[ LG_b(z) = \frac{1}{L} \cdot \frac{1}{(1-z^{-1})^2} = \frac{1}{L}\cdot LG_a(z) \]


Assume PD output is constant

frug_seq.drawio

\[ \lim_{n\to +\infty} \frac{\Delta P_1}{\Delta P_0} = \lim_{n\to +\infty}\frac{n+2L}{nL+\alpha L+\beta L^2} = \frac{1}{L} \]

Decimation by Voting

image-20241126211307012


In above screenshot

  1. \(K_D\) is just relative value
  2. frug shall not be scaled by decimator factor

proved as below

DC gain \(K_B\) of summing (boxcar filter) is decimation factor \(M\) , voting gain \(K_V\) is about \(0.54K_b=0.54M\)

  1. downsampling \(\frac{1}{M}\) and ZOH \(\frac{1-z^{-M}}{1-z^{-1}}\) can be canceled out at low frequency
  2. decimation gain: accumulator \(\frac{1-z^{-M}}{1-z^{-1}}\) replaced with linearizing gain \(K_B\) and majority voting replaced with \(K_V\)

proportional path: \[\begin{align} LG_{ph} &= K_{BB}\cdot \frac{1-z^{-M}}{1-z^{-1}}\cdot \frac{1}{M}\cdot \frac{1}{1-z^{-M}}\cdot \frac{1-z^{-M}}{1-z^{-1}} \\ &\approx K_{BB}\cdot \frac{1-z^{-M}}{1-z^{-1}}\cdot \frac{1}{1-z^{-M}} \\ &= K_{BB}\cdot K_D\cdot \frac{1}{1-z^{-M}} \end{align}\]

integral path: \[\begin{align} LG_{fr} &= K_{BB}\cdot \frac{1-z^{-M}}{1-z^{-1}}\cdot \frac{1}{M}\cdot \frac{1}{(1-z^{-M})^2}\cdot \frac{1-z^{-M}}{1-z^{-1}} \\ &\approx K_{BB}\cdot \frac{1-z^{-M}}{1-z^{-1}}\cdot \frac{1}{(1-z^{-M})^2} \\ &= K_{BB}\cdot K_D\cdot \frac{1}{(1-z^{-M})^2} \end{align}\]

J. Stonick. ISSCC 2011 "DPLL-Based Clock and Data Recovery" [slides,transcript]

J. L. Sonntag and J. Stonick, "A Digital Clock and Data Recovery Architecture for Multi-Gigabit/s Binary Links," in IEEE Journal of Solid-State Circuits, vol. 41, no. 8, pp. 1867-1875, Aug. 2006 [https://sci-hub.se/10.1109/JSSC.2006.875292]

J. Sonntag and J. Stonick, "A digital clock and data recovery architecture for multi-gigabit/s binary links," Proceedings of the IEEE 2005 Custom Integrated Circuits Conference, 2005.. [https://sci-hub.se/10.1109/CICC.2005.1568725]

Y. Xia et al., "A 10-GHz Low-Power Serial Digital Majority Voter Based on Moving Accumulative Sign Filter in a PS-/PI-Based CDR," in IEEE Transactions on Microwave Theory and Techniques, vol. 68, no. 12 [https://sci-hub.se/10.1109/TMTT.2020.3029188]

J. Liang, A. Sheikholeslami, "On-Chip Jitter Measurement and Mitigation Techniques for Clock and Data Recovery Circuits" [https://tspace.library.utoronto.ca/bitstream/1807/91138/3/Liang_Joshua_201706_PhD_thesis.pdf]

J. Liang, A. Sheikholeslami. ISSCC2017. "A 28Gbps Digital CDR with Adaptive Loop Gain for Optimum Jitter Tolerance" [slides,paper]

J. Liang, A. Sheikholeslami,, "Loop Gain Adaptation for Optimum Jitter Tolerance in Digital CDRs," in IEEE Journal of Solid-State Circuits [https://sci-hub.se/10.1109/JSSC.2018.2839038]

M. M. Khanghah, K. D. Sadeghipour, D. Kelly, C. Antony, P. Ossieur and P. D. Townsend, "A 7-Bit 7-GHz Multiphase Interpolator-Based DPC for CDR Applications," in IEEE Transactions on Circuits and Systems I: Regular Papers [https://cora.ucc.ie/bitstreams/7ae5bfaa-8dd9-45a7-8276-99676b7b6078/download]

[CDR CIRCUIT-BLOCKS: DESIGN AND VERIFICATION USING VERILOG - 2.6. DECIMATOR]

Michael H. Perrott, Tutorial on Digital Phase-Locked Loops, CICC 2009, San Jose, CA, Sept. 13, 2009 [https://www.cppsim.com/PLL_Lectures/digital_pll_cicc_tutorial_perrott.pdf]

Liu, Tao, Tiejun Li, Fangxu Lv, Bin Liang, Xuqiang Zheng, Heming Wang, Miaomiao Wu, Dechao Lu, and Feng Zhao. 2021. "Analysis and Modeling of Mueller-Muller Clock and Data Recovery Circuits" Electronics 10 [https://www.mdpi.com/2079-9292/10/16/1888/pdf?version=1628492599]

Gu, Youzhi & Feng, Xinjie & Chi, Runze & Chen, Yongzhen & Wu, Jiangfeng. (2022). Analysis of Mueller-Muller Clock and Data Recovery Circuits with a Linearized Model. 10.21203/rs.3.rs-1817774/v1. [https://assets-eu.researchsquare.com/files/rs-1817774/v1_covered.pdf?c=1664188179]

Chen, Junkun, Youzhi Gu, Xinjie Feng, Runze Chi, Jiangfeng Wu, and Yongzhen Chen. 2024. "Analysis of Mueller–Muller Clock and Data Recovery Circuits with a Linearized Model" Electronics [https://mdpi-res.com/electronics/electronics-13-04218/article_deploy/electronics-13-04218-v2.pdf?version=1730106095]

K. Yadav, P. -H. Hsieh and A. C. Carusone, "Loop Dynamics Analysis of PAM-4 Mueller–Muller Clock and Data Recovery System," in IEEE Open Journal of Circuits and Systems [https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9910561]

TODO 📅

Tristate: \(\alpha=1\)

XOR: \(\alpha=1\)

\(\frac{1}{T}\) in Divider

image-20240928004526381

image-20240928004308700

Michael H. Perrott, PLL Design Using the PLL Design Assistant Program. [https://designers-guide.org/forum/Attachments/pll_manual.pdf]


\(\frac{1}{T}\) & \(T\) come from CT-DT & DT-CT

image-20240928203714450

H. Kang et al., "A 42.7Gb/s Optical Receiver With Digital Clock and Data Recovery in 28nm CMOS," in IEEE Access, vol. 12, pp. 109900-109911, 2024 [https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10630516]

Sonntag JSSC 2006

image-20241129222258061

image-20241129223706720

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
clear;
close all;
clc;


Tb = 200e-12;
Ts = Tb*8; % the decimation factor was 8
z = tf('z', Ts);

Kdpc = 1/2^9;
Kv = 8*0.54;
Kpd = 10.6;
phug = 2^-3;
frug = 2^-12;
Nel = 18;

options = bodeoptions;
options.FreqUnits = 'MHz';
options.XLim = [1e-2, 1e1];
options.YLim = [-10, 5];

L = Kpd*Kv*Kdpc/(1-z^-1)*(phug + frug/(1-z^-1))*z^-Nel;
TF = L/(1+L);
bodemag(TF,options);

hold on;
frug = 2^-11;
L = Kpd*Kv*Kdpc/(1-z^-1)*(phug + frug/(1-z^-1))*z^-Nel;
TF = L/(1+L);
bodemag(TF,options);

hold on;
frug = 2^-10;
L = Kpd*Kv*Kdpc/(1-z^-1)*(phug + frug/(1-z^-1))*z^-Nel;
TF = L/(1+L);
bodemag(TF,options);

legend('frug=2^{-12}','frug=2^{-11}', 'frug=2^{-10}', 'FontSize',10)
grid on;
title('phase transfer function', 'FontSize', 12)
xlabel('frequency', 'FontSize',10)
ylabel('frequency response', 'FontSize',10)

Full View

image-20241129223734870

Kpd, Kb, Kv

Both decimation factor and factor for voting are 4

image-20241130162850467

  • Kpd formula: 12.467; Kpd_bb_0 12.465
  • Kpd_Kb: 49.860; Kpd_Kv 27.265
  • Kb: 4.00; Kv 2.19

That is

  1. gain of BoxCar is the decimation factor
  2. Voting across 4 inputs had a 54% reduced gain relative to boxcar filter
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
import numpy as np
from scipy.stats import norm
import itertools
from collections import defaultdict
import matplotlib.pyplot as plt

sigmai = 0.032 #UI, input jitter
Ptrans = 0.5 # Transition density
deci_factor = 4

phase_error = np.linspace(-0.1, 0.1, 201) #UI, phase offset
pd_late = norm.cdf(phase_error/sigmai)
pd_early = 1.0 - pd_late
pd_avg = pd_late*1.0 - 1.0*pd_early

Kpd_bb = (pd_avg[1:] - pd_avg[:-1])/(phase_error[1:] - phase_error[:-1])*Ptrans
Kpd_bb_0 = np.max(Kpd_bb)

## by formula
Kpd_calc = 1.0/(sigmai*np.sqrt(2*np.pi))

print(f'Kpd formula: {Kpd_calc:.3f}; Kpd_bb_0 {Kpd_bb_0:.3f}') # Kpd formula: 12.467; Kpd_bb_0 12.465

plt.figure()
plt.plot(phase_error, pd_avg, color='r', linewidth=3)
plt.title('!! PD average output vs timing offset(UI)')
plt.grid()
plt.show()


prob = np.zeros((phase_error.shape[0],3))
prob[:,0] = pd_early*Ptrans # -1
prob[:,1] = 1.0 - Ptrans # 0
prob[:,2] = pd_late*Ptrans # 1

pd_out = np.array([-1.0,0.0,1.0])
idxs = list([[0,1,2] for _ in range(deci_factor)])
boxcar_avg = []
voting_avg = []
for i in range(phase_error.shape[0]):
prob_i = prob[i,:]
boxcar_tmp = 0.0
voting_tmp = 0.0
for idxs_tmp in itertools.product(*idxs):
pd_list = pd_out[[idxs_tmp]]
prob_list = prob_i[[idxs_tmp]]
pd_sum = np.sum(pd_list)
pd_vote = 1.0 if pd_sum > 0.0 else -1.0 if pd_sum <0.0 else 0.0
prob_prod = np.prod(prob_list)
boxcar_tmp += pd_sum*prob_prod
voting_tmp += pd_vote*prob_prod
boxcar_avg.append(boxcar_tmp)
voting_avg.append(voting_tmp)

boxcar_avg = np.array(boxcar_avg)
voting_avg = np.array(voting_avg)

plt.figure()
plt.plot(phase_error,boxcar_avg, label='FIR BoxCar', color='r', linewidth=3)
plt.plot(phase_error,voting_avg, label='Voting', color='b', linewidth=3, linestyle='--')
plt.legend()
plt.title('!!PD+BoxCar / !!PD+Voting vs timing offset(UI)')
plt.grid()
plt.show()


Kpd_Kb = (boxcar_avg[1:] - boxcar_avg[:-1])/(phase_error[1:] - phase_error[:-1])
Kpd_Kv = (voting_avg[1:] - voting_avg[:-1])/(phase_error[1:] - phase_error[:-1])
Kpd_kb_0 = np.max(Kpd_Kb)
Kpd_kv_0 = np.max(Kpd_Kv)
print(f'Kpd_Kb: {Kpd_kb_0:.3f}; Kpd_Kv {Kpd_kv_0:.3f}') # Kpd_Kb: 49.860; Kpd_Kv 27.265

plt.figure()
plt.plot(phase_error[:-1], Kpd_Kb, color='r', linewidth=3)
plt.plot(phase_error[:-1], Kpd_Kv, color='b', linewidth=3, linestyle='--')
plt.legend(['Kpd_Kb', 'Kpd_Kv'])
plt.title('Kpd*Kb / Kpd*Kv vs timing offset(UI)')
plt.grid()
plt.show()

Kb = Kpd_kb_0 / Kpd_bb_0
Kv = Kpd_kv_0 / Kpd_bb_0
print(f'Kb: {Kb:.2f}; Kv {Kv:.2f}') # Kb: 4.00; Kv 2.19

reference

Alan V Oppenheim, Ronald W. Schafer. 2010. Discrete-Time Signal Processing, 3rd edition

R. E. Crochiere and L. R. Rabiner, "Multirate Digital Signal Processing", Prentice Hall, 1983.

John G. Proakis and Dimitris G. Manolakis, Digital Signal Processing: Principles, Algorithms, and Applications, 4th edition, 2007.

D. Sundararajan. 2024. Digital Signal Processing: An Introduction 2nd Edition

F. M. Gardner, "Phaselock Techniques", 3rd Edition, Wiley Interscience, Hoboken, NJ, 2005 [https://picture.iczhiku.com/resource/eetop/WyIgwGtkDSWGSxnm.pdf]

Rhee, W. (2020). Phase-locked frequency generation and clocking : architectures and circuits for modern wireless and wireline systems. The Institution of Engineering and Technology


Qasim Chaudhari. Sample Rate Conversion [https://wirelesspi.com/sample-rate-conversion/]

Push-Pull

TODO 📅

Rinaldo Castello, "LINEARIZATION TECHNIQUES FOR PUSH-PULL AMPLIFIERS" [https://www.ieeetoronto.ca/wp-content/uploads/2020/06/AMPLIFIERS_Stanf_Tor_2016_Last.pdf]

Noise and Distortion

TODO 📅

Ali Sheikholeslami, University of Toronto, A-SSCC 2024 Circuit Insights:FT1 Noise and Distortion [link]

Response Speed in Analog Circuits

Hyun-Sik Kim, KAIST, A-SSCC 2024 Circuit Insights: FT3 Accelerating Response Speed in Analog Circuits [link]

image-20250105085449759


image-20250105072433452

Bandwidth limitation

image-20250105072748483

image-20250105073322162

image-20250105090203938image-20250105090204188

slew rate limitation

image-20250105083039901

Assuming linear response \[ V_o(t) = 1 - e^{-\omega_T t} \]

\[ \frac{dV_o}{dt} = \omega_Te^{-\omega_T t} = \frac{g_m}{C_L}e^{-\omega_T t} = \frac{g_m}{I_B}\cdot \frac{I_B}{C_L}\cdot e^{-\omega_T t} \gt \frac{I_B}{C_L} \]

where \(\frac{g_m}{I_B} e^{-\omega_T t} \gt 1\) at initial response

Therefore, initial response speed is dominated by SR, rather than \(G_m\) (or bandwidth)

image-20250105090105095

MOS parasitic Rd&Rs, Cd&Cs

Decrease the parasitic R&C

priority: \(R_s \gt R_d\), \(C_s \gt C_d\)

XCP as Negative Impedance Converter (NIC)

The Cross-Coupled Pair (XCP) can operate as an impedance negator [a.k.a. a negative impedance converter (NIC)]

A common application is to create a negative capacitance that can cancel the positive capacitance seen at a port, thereby improving the speed

image-20240922174319496 \[ I_{NIC} =\frac{V_{im} - V_{ip}}{\frac{2}{g_m}+\frac{1}{sC_c}} = \frac{-2V_{ip}}{\frac{2}{g_m}+\frac{1}{sC_c}} \] Therefore \[ Z_{NIC} = \frac{V_{ip} - V_{im}}{I_{NIC}}=\frac{2V_{ip}}{I_{NIC}} =- \frac{2}{g_m}-\frac{1}{sC_c} \] half-circuit

If \(C_{gd}\) is considered, and apply miller effect. half equivalent circuit is shown as below

nic.drawio

B. Razavi, "The Cross-Coupled Pair - Part III [A Circuit for All Seasons]," IEEE Solid-State Circuits Magazine, Issue. 1, pp. 10-13, Winter 2015. [https://www.seas.ucla.edu/brweb/papers/Journals/BR_Magzine3.pdf]

S. Galal and B. Razavi, "10-Gb/s Limiting Amplifier and Laser/Modulator Driver in 0.18um CMOS Technology,” IEEE Journal of Solid-State Circuits, vol. 38, pp. 2138-2146, Dec. 2003. [https://www.seas.ucla.edu/brweb/papers/Journals/G&RDec03_2.pdf]

source follower

A. Sheikholeslami, "Voltage Follower, Part III [Circuit Intuitions]," in IEEE Solid-State Circuits Magazine, vol. 15, no. 2, pp. 14-26, Spring 2023, doi: 10.1109/MSSC.2023.3269457

—, ESSCIRC2023 Circuit Insights [https://youtu.be/2xFIZM5_FPw?si=536cMdIXyIny27Uk]

—, CICC2025 Circuit Insights: From Simple to Super Source Follower [https://youtu.be/CWfMKltPIQ8?si=s0npv2GSQKYBv513]

Paul R. Gray. 2009. Analysis and Design of Analog Integrated Circuits (5th. ed.). Wiley Publishing. [pdf]

Super-source follower (SSF)

image-20240924213742877

image-20240924213845608

image-20240924213853954

Flipped Voltage Follower (FVF)

image-20240921110019881

image-20240921113630249

T&H buffer in ADC

image-20240923200147070

[https://www.linkedin.com/posts/chembiyan-t-0b34b910_flipped-voltage-follower-fvf-basics-activity-7118482840803020800-qwyX?utm_source=share&utm_medium=member_desktop]

Z. Guo et al., "A 112.5Gb/s ADC-DSP-Based PAM-4 Long-Reach Transceiver with >50dB Channel Loss in 5nm FinFET," 2022 IEEE International Solid-State Circuits Conference (ISSCC), San Francisco, CA, USA, 2022, pp. 116-118, doi: 10.1109/ISSCC42614.2022.9731650.

Double differential Pair

\(V_\text{ip}\) and \(V_\text{im}\) are input, \(V_\text{rp}\) and \(V_\text{rm}\) are reference voltage \[ V_o = A_v(\overline{V_\text{ip} - V_\text{im}} - \overline{V_\text{rp} - V_\text{rm}}) \]

2diffpair.drawio

In differential comparison mode, the feedback loop ensure \(V_\text{ip} = V_\text{rp}\), \(V_\text{im} = V_\text{rm}\) in the end

assume input and reference common voltage are same

Pros of (b)

  • larger input range i.e., \(\gt \pm \sqrt{2}V_\text{ov}\) of (a), it works even one differential is off due to lower voltage
  • larger \(g_m\) (smaller input difference of pair)

Cons of (b)

  • sensitive to the difference of common voltage between \(V_\text{ip}\), \(V_\text{im}\) and \(V_\text{rp}\), \(V_\text{rm}\)

common-mode voltage difference

doublepair_cm.drawio

copy aforementioned formula here for convenience \[ V_o = A_v(\overline{V_\text{ip} - V_\text{im}} - \overline{V_\text{rp} - V_\text{rm}}) \]

at sample phase \(V_\text{ip}= V_\text{im}= V_\text{cmi}\) and \(V_\text{rp}= V_\text{rm}= V_\text{cmr}\)

  • \(I_\text{ip0}= I_\text{im0} = I_\text{i0}\)
  • \(I_\text{rp0}= I_\text{rm0} = I_\text{r0}\)

i.e. \(\overline{I_\text{ip} + I_\text{rm}} - \overline{I_\text{im} + I_\text{rp}} = 0\)

at compare start

  • \(V_\text{ip}= V_\text{im}= V_\text{cmi}\) and \(V_\text{rp}= V_\text{cmr}+\Delta\), \(V_\text{rp}= V_\text{cmr}-\Delta\)

  • \(I_\text{ip}\lt I_\text{ip0}\), \(I_\text{rp} \gt I_\text{rp0}\)

  • \(I_\text{im}\gt I_\text{im0}\), \(I_\text{rm} \lt I_\text{rm0}\)

i.e. \(\overline{I_\text{ip} + I_\text{rm}} - \overline{I_\text{im} + I_\text{rp}} \lt 0\), we need to increase \(V_\text{ip}\) and decrease \(V_\text{im}\).

at the compare finish

\[\begin{align} V_\text{ip}= V_\text{cmi} + \Delta \\ V_\text{im}= V_\text{cmi} - \Delta \end{align}\]

and \(I_\text{ip0}= I_\text{im0} = I_\text{i0}\), \(I_\text{rp0}= I_\text{rm0} = I_\text{r0}\)

i.e. \(\overline{I_\text{ip} + I_\text{rm}} - \overline{I_\text{im} + I_\text{rp}} = 0\)


If \(V_\text{cmr} - V_\text{cmi} = \sqrt{2}V_{OV} + \delta\), and \(\delta \gt 0\). one transistor carries the entire tail current

  • \(I_\text{ip} =0\) and \(I_\text{rp} = I_{SS}\), all the time

At the end, \(V_\text{im} = V_\text{cmi} - (\Delta - \delta)\), the error is \(\delta\)

In closing, \(V_\text{cmr} - V_\text{cmi} \lt \sqrt{2}V_{OV}\) for normal work

Furthermore, the difference between \(V_\text{cmr}\) and \(V_\text{cmi}\) should be minimized due to limited impedance of current source and input pair offset

In the end \[ V_\text{cmr} - V_\text{cmi} \lt \sqrt{2}V_{OV} - V_{OS} \]

Under the condition, every transistor of pairs are on in equilibrium

pair mismatch

diff_mismatch_connect.drawio

\[\begin{align} I_{SE} &= g_m(\sigma_{vth,0} + \sigma_{vth,1}) \\ I_{DE} &= g_m(\sigma_{vth,0} + \sigma_{vth,1}) \end{align}\]

The input equivalient offset voltage \[\begin{align} V_{os,SE} &= \frac{I_{SE}}{2g_m} = \frac{\sigma_{vth,0} + \sigma_{vth,1}}{2} \\ V_{os,DE} &= \frac{I_{DE}}{g_m} = \sigma_{vth,0} + \sigma_{vth,1} \end{align}\]

Then \[\begin{align} \sigma_{vos,SE} &= \sqrt{\frac{2\sigma_{vth}^2}{4}} = \frac{\sigma_{vth}}{\sqrt{2}} \\ \sigma_{vos,DE} &= \sqrt{2\sigma_{vth}^2} = \sqrt{2}\sigma_{vth} \end{align}\]

We obtain \[ \sigma_{vos,DE} = 2\sigma_{vos,SE} \]

peaking without inductor

TODO 📅

How to generate complex poles without inductor? [https://a2d2ic.wordpress.com/2020/02/19/basics-on-active-rc-low-pass-filters/]

Input Diff-Pair

DM Distortion

image-20241027095213326

CM Distortion

image-20241027095248946

Resistive Degeneration

Resistive degeneration in differential pairs serves as one major technique for linear amplifier

image-20240824132739726

The linear region for CMOS differential pair would be extended by \(±I_{SS}R/2\) as all of \(I_{SS}/2\) flows through \(R\). \[\begin{align} V_{in}^+ -V_{in}^- &= V_{OV} + V_{TH}+\frac{I_{SS}}{2}R - V_{TH} \\ &= \sqrt{\frac{2I_{SS}}{\mu_nC_{OX}\frac{W}{L}}} + \frac{I_{SS}R}{2} \end{align}\]

Jri Lee, "Communication Integrated Circuits." https://cc.ee.ntu.edu.tw/~jrilee/publications/Comm_IC.pdf

Figure 14.12, Design of Analog CMOS Integrated Circuits, Second Edition [https://electrovolt.ir/wp-content/uploads/2014/08/Design-of-Analog-CMOS-Integrated-Circuit-2nd-Edition-ElectroVolt.ir_.pdf]

Biasing Tradeoffs in Resistive-Degenerated Diff Pair

image-20241027095520556

Todd Brooks, Broadcom "Input Programmable Gain Amplifier (PGA) Design for ADC Signal Conditioning" [https://classes.engr.oregonstate.edu/eecs/spring2021/ece627/Lecture%20Notes/OSU%20Classroom%20Presentaton%20042511.ppt]

Source-Degenerated Differential Pairs

TODO 📅

reference

Elad Alon, ISSCC 2014, "T6: Analog Front-End Design for Gb/s Wireline Receivers" [https://picture.iczhiku.com/resource/eetop/wHKfZPYpAleAKXBV.pdf]

Byungsub Kim, ISSCC 2022, "T11: Basics of Equalization Techniques: Channels, Equalization, and Circuits"

Minsoo Choi et al., "An Approximate Closed-Form Channel Model for Diverse Interconnect Applications," IEEE Transactions on Circuits and Systems-I: Regular Papers, vol. 61, no. 10, pp. 3034-3043, Oct. 2014.

K. Yadav, P. -H. Hsieh and A. Chan Carusone, "Linearity Analysis of Source-Degenerated Differential Pairs for Wireline Applications," in IEEE Open Journal of Circuits and Systems [https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10769573]

image-20250906072230725

image-20250906072050727

image-20250906171710938

linearized model

image-20250824092757793 \[\begin{align} v[n] = \{0,1,2,...,M-1\} &\space\Rightarrow\space y[n] = 0 \space\Rightarrow\space e_q[n] = \{0, -\frac{1}{M},-\frac{2}{M},...,-\frac{M-1}{M}\} \\ v[n] = \{M,M+1,M2,...,2M-1\} &\space\Rightarrow\space y[n] = 1 \space\Rightarrow\space e_q[n] = \{0, -\frac{1}{M},-\frac{2}{M},...,-\frac{M-1}{M}\} \end{align}\]

image-20250823232924985

For the three stages of the MASH 1-1-1 DDSM

image-20250823232212295


1st order DDSM (digital accumulator)

image-20250604000323199

assuming \(n_0=2\)

\(x[n]+s[n]\) \(v\) \(e[n]\) \(c[n]\), y
0|00 0 0 0
0|01 1 1 0
0|10 2 2 0
0|11 3 3 0
1|00 4 0 1
1|01 5 1 1
1|10 6 2 1
1|11 7 3 1

yield \(M=2^{n_0}=4\)

2nd order DDSM

image-20250601170123635

In \(z\)-domain \[ \left\{(A + D - Y)\frac{z^{-1}}{1-z^{-1}} - 2Y \right\}\frac{z^{-1}}{1-z^{-1}} + Q = Y \] That is \[ Y = A z^{-2} + Dz^{-2} + Q(1-z^{-1})^2 \] In time domain \[\begin{align} y[n] &= \alpha[n-2] + d[n-2] + q[n]-2q[n-1]+q[n-2] \\ &= \alpha + d[n-2] + q[n]-2q[n-1]+q[n-2] \end{align}\]

LSB Dither

image-20250905064118796

image-20250905064139686


image-20250905065402176

?? integer valued impulse responses

S. Pamarti, J. Welz and I. Galton, "Statistics of the Quantization Noise in 1-Bit Dithered Single-Quantizer Digital Delta–Sigma Modulators," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 54, no. 3, pp. 492-503, March 2007 [pdf]

stability of DSM

image-20250908213730155

image-20250908213849546

image-20250913161933338

accumulator wordlength

Z. Ye and M. P. Kennedy, "Hardware Reduction in Digital Delta–Sigma Modulators Via Error Masking—Part II: SQ-DDSM," in IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 56, no. 2, pp. 112-116, Feb. 2009 [https://sci-hub.se/10.1109/TCSII.2008.2010188]

—, "Hardware Reduction in Digital Delta-Sigma Modulators Via Error Masking - Part I: MASH DDSM," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 56, no. 4, pp. 714-726, April 2009 [https://sci-hub.se/10.1109/TCSI.2008.2003383]

image-20250906134655253

Truncation DAC

accumulator is implicit quantizer

image-20241022204239594

with \(\frac{y}{2^{m_2}} + q= v\), where \(v = \lfloor\frac{y}{2^{m_2}}\rfloor\)

\[ \left\{ \begin{array}{cl} Y + 2^{m_2} Q &= 2^{m_2}V \\ U - z^{-1}2^{m_2}Q &= Y \end{array} \right. \]

The STF & NTF is shown as below \[ V = \frac{1}{2^{m_2}}U + (1-z^{-1})Q \]

To avoid accumulator overflow, stable input range is only of a fraction of the full scale ( \(2^{m_1+m_2}-1\)) \[ u \leq = 2^{m_1+m_2} - 2^{m_2} \]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
m1 = 2  # MSBs
m2 = 4 # LSBs

ymax = 2**(m1 + m2)
umax = 2**(m1 + m2) - 2**m2 # int(m1*'1'+m2*'0', 2)
# format(48, '06b')
# Out[4]: '110000'

u = 48
assert u <= umax

ylist = [0]; vlist = [0]
elist = []; outlist = []

Niter = 2**10
for _ in range(Niter):
ecur = vlist[-1] - ylist[-1]
elist.append(ecur)
ycur = (u - ecur)
assert ycur < ymax, print(ycur)
ylist.append(ycur)
ycur_bin = format(ycur, f'0{m1+m2}b')
vcur = int(ycur_bin[:-m2]+'0'*m2, 2)
vlist.append(vcur)
outlist.append(int(ycur_bin[:-m2], 2))

print(vlist); print(ylist)
print(sum(vlist)/len(vlist)); print(sum(outlist)/len(outlist)*2**m2)

Fractional-N PLL

image-20250824103717743

image-20250824103933652

Divider Model

image-20250824221530772 \[ (N+\alpha)T_{PLL} - \tau[n-1] +\tau[n] = (N+y[n])T_{PLL} \]

i.e. \[ \tau[n] = \tau[n-1] + (y[n] - \alpha)T_{PLL} \]

where \(\tau[n] = t_{v_{DIV}} - t_{v_{DIV}, desired}\)

image-20250824221741018

\(\Delta\Sigma\) noise in PLL

image-20250824162417584

image-20250824183123922

image-20250824210526248

Impulse Train Modulator (ITM)

M. H. Perrott, M. D. Trott and C. G. Sodini, "A modeling approach for /spl Sigma/-/spl Delta/ fractional-N frequency synthesizers allowing straightforward noise analysis," in IEEE Journal of Solid-State Circuits, vol. 37, no. 8, pp. 1028-1038, Aug. 2002 [https://www.cppsim.com/Publications/JNL/perrott_jssc02.pdf]

image-20250913130430564


image-20250913130708018

image-20250913130847600

Sigma-Delta DAC

The spectrum of the high resolution digital signal \(u_1\) contains the original baseband portion and its replicas located at integer multiples of \(f_{s1}\), plus a small amount of quantization noise shown as a solid line

image-20250906170436567


Sigma-delta digital-to-analog converters (SD DAC’s) are often used for discrete-time signals with sample rate much higher than their bandwidth

  • Because of the high sample rate relative to signal bandwidth, a very simple DAC reconstruction filter (Analog lowpass filter) suffices, often just a one-pole RC lowpass

image-20250616000829208

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
R= 4.7e3;                 % ohms resistor value
C= .01e-6; % F capacitor value
fs= 1e6; % Hz DAC sample rate
% input signal
x= [zeros(1,20) .9*ones(1,200) .1*ones(1,200)];
% find output y of SD DAC and output y_filt of RC filter
[y,y_filt]= sd_dacRC(x,R,C,fs);

t = linspace(0,length(x)-1, length(x))*1/fs*1e3;
subplot(3,1,1)
plot(t, x, '.'); title('x'); grid on
subplot(3,1,2)
plot(t, y, '.'); title('y'); grid on
subplot(3,1,3)
plot(t, y_filt); title('y_{filt}'); xlabel('t(ms)'); grid on

image-20250621223451691


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
% https://www.dsprelated.com/showarticle/1642.php
% Neil Robertson, Model a Sigma-Delta DAC Plus RC Filter

% function [y,y_filt] = sd_dacRC(x,R,C,fs) 2/5/24 Neil Robertson
% 1-bit sigma-delta DAC with RC filter
% Model does not include a zero-order hold.
%
% x = input signal vector, 0 <= x < 1
% R = series resistor value, Ohms. Normally R > 1000 for 3.3 V logic.
% C = shunt capacitor value, Farads
% fs = sample frequency, Hz
% y = DAC output signal vector, y(n) = 0 or 1
% y_filt = RC filter output signal vector
%
function [y,y_filt] = sd_dacRC(x,R,C,fs)
N= length(x);
x= fix(x*2^16)/2^16; % quantize x to 16 bits
%I 1-bit Sigma-delta DAC
s= [x(1) zeros(1,N-1)];
for n= 2:N
u= x(n) + s(n-1);
s(n)= mod(u,1); % sum
y(n)= fix(u); % carry
end

%II One-pole RC filter model
% Matched z-Transform https://ocw.mit.edu/courses/2-161-signal-processing-continuous-and-discrete-fall-2008/cc00ac6d468dc9dcf2238fc1d1a194d4_lecture_19.pdf
Ts= 1/fs;
Wc= 1/(R*C); % rad -3 dB frequency
fc= Wc/(2*pi); % Hz -3 dB frequency
a1= -exp(-Wc*Ts);
b0= 1 + a1; % numerator coefficient
a= [1 a1]; % denominator coeffs
y_filt= filter(b0,a,y); % filter the DAC's output signal y

DAC ZOH

image-20250628204404959

The last D2C is in human vision, which connect discrete time \(y(m)\) with line, implicitly

image-20250628203216965

oversampling & noise shaping

maximum output signal 22kHz

image-20250906195627446

\[SNR = 16\times 6.02 + 1.76 = 98.08\]

image-20250906200230380


image-20250906205517957

1
2
3
4
5
6
7
8
9
10
11
12
OSR = 5.65e6/(2*22e3);
Nin = 16;
Nout = 1;

SNR_in = 6.02*Nin + 1.76;

SNR_ds = 6.02*Nout + 1.76 - 10*log10(pi^4/5) + 50*log10(OSR);

QN_in = 1/10^(SNR_in/10);
QN_ds = 1/10^(SNR_ds/10);

SNR_out = 10*log10(1/(QN_in + QN_ds));

image-20250906205622949

reference

Michael Peter Kennedy. scv-cas 2014: Digital Delta-Sigma Modulators [pdf,recording]

—, Recent advances in the analysis, design and optimization of Digital Delta-Sigma Modulators [pdf]

Kaveh Hosseini and Peter Kennedy. 2006 Hardware Efficient Maximum Sequence Length Digital MASH Delta Sigma Modulator [pdf]

Jason Sachs. Return of the Delta-Sigma Modulators, Part 1: Modulation [https://www.dsprelated.com/showarticle/1517/return-of-the-delta-sigma-modulators-part-1-modulation]


Neil Robertson, Model a Sigma-Delta DAC Plus RC Filter [https://www.dsprelated.com/showarticle/1642.php]

—, Modeling a Continuous-Time System with Matlab [https://www.dsprelated.com/showarticle/1055.php]

—, DAC Zero-Order Hold Models [https://www.dsprelated.com/showarticle/1627.php]

—, “A Simplified Matlab Function for Power Spectral Density”, DSPRelated.com, March, 2020, [https://www.dsprelated.com/showarticle/1333.php]

Dan Boschen. sigma delta modulator for DAC [https://dsp.stackexchange.com/a/88357/59253]

Woogeun Rhee. ISCAS 2019 Mini Tutorials: Single-Bit Delta-Sigma Modulation Techniques for Robust Wireless Systems [https://youtu.be/OEyTM4-_OyA?si=vllJ5Pe8I3lqb_Vl]

—, 2001 Phd Thesis: Multi-Bit Delta -Sigma Modulation Technique for Fractional-N Frequency Synthesizers [https://www.ime.tsinghua.edu.cn/Thesis_rhee.pdf]


S. Pamarti, J. Welz and I. Galton, "Statistics of the Quantization Noise in 1-Bit Dithered Single-Quantizer Digital Delta–Sigma Modulators," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 54, no. 3, pp. 492-503, March 2007 [https://ispg.ucsd.edu/wordpress/wp-content/uploads/2017/05/2007-TCASI-S.-Pamarti-Statistics-of-the-Quantization-Noise-in-1-Bit-Dithered-Single-Quantizer-Digital-Delta-Sigma-Modulators.pdf]

—. "LSB Dithering in MASH Delta–Sigma D/A Converters," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 54, no. 4, pp. 779-790, April 2007 [https://sci-hub.se/10.1109/TCSI.2006.888780]

—. CICC 2020 ES2-2: Basics of Closed- and Open-Loop Fractional Frequency Synthesis [https://youtu.be/t1TY-D95CY8?si=tbav3J2yag38HyZx]

Ian Galton. Delta-Sigma Fractional-N Phase-Locked Loops [https://ispg.ucsd.edu/wordpress/wp-content/uploads/2022/10/fnpll_ieee_tutorial_2003_corrected.pdf]

—. ISSCC 2010 SC3: Fractional-N PLLs [https://www.nishanchettri.com/isscc-slides/2010%20ISSCC/Short%20Course/SC3.pdf]

—. “Delta-Sigma Fractional-N Phase-Locked Loops.” (2003).

Mike Shuo-Wei Chen, ISSCC 2020 T6: Digital Fractional-N Phase Locked Loop Design [https://www.nishanchettri.com/isscc-slides/2020%20ISSCC/TUTORIALS/T6Visuals.pdf]


Pavan, Shanthi, Richard Schreier, and Gabor Temes. (2016) 2016. Understanding Delta-Sigma Data Converters. 2nd ed. Wiley.

Rhee, W. (2020). Phase-locked frequency generation and clocking : architectures and circuits for modern wireless and wireline systems. The Institution of Engineering and Technology

Kaveh Hosseini, Michael Peter Kennedy. Springer 2011. Minimizing Spurious Tones in Digital Delta-Sigma Modulators

Classification

  • Continuous-Time (CT) Analog Modulator
    • Sampled Quantizer: Synchronous Modulator
    • Unsampled Quantizer: Asynchronous Modulator
  • Discrete-Time (DT) Analog Modulator
    • The main application is in \(\Delta\Sigma\) ADCs
  • Discrete-Time Digital Modulators (DDSM)
    • Single Quantizer DDSMs - output feedback
    • Error feedback modulators (EFM) - error feedback
    • Multi stAge noise SHaping (MASH)

image-20250903210531360


image-20250611074830238

"Quantizers" and "truncators", and "integrators" and "accumulators" are used in delta-sigma ADCs and DACs, respectively

P. Kiss, J. Arias and Dandan Li, "Stable high-order delta-sigma DACS," 2003 IEEE International Symposium on Circuits and Systems (ISCAS), Bangkok, 2003 [https://www.ele.uva.es/~jesus/analog/tcasi2003.pdf]

Oversampling

image-20250611232612319

David Johns and Ken Martin. Oversampling Converters [https://www.eecg.toronto.edu/~johns/ece1371/slides/14_oversampling.pdf]


Over Sampling

Nyquist sampling theorem @signal: no aliasing, signal remain the same

noise folding @noise: same total noise power spread over a wider frequency

[https://dsp.stackexchange.com/a/40261/59253]


image-20250629215715378

image-20250629215830077

Noise Shaping

image-20250629232343017

image-20250629232453811

1
2
3
4
5
6
7
8
9
10
11
[h1, w1] = freqz([1 -1], 1);
[h2, w2] = freqz([1 -2 1], 1);

plot(w1/2/pi, abs(h1), LineWidth=3)
hold on
plot(w2/2/pi, abs(h2), LineWidth=3)
grid on
legend('MOD1', 'MOD2')
xlabel('fs')
ylabel('mag')
title('NTF of MOD1 & MOD2')

image-20250824151828263

SQNR improvement

In general, for an \(l\)th order modulator with \(\text{NTF}(z) = (1 − z^{−1})^l\), the SQNR increases by \((6l + 3)\) dB for every doubling of the OSR, which provides \(l+0.5\) extra bits resolution

without the delta-sigma loop

image-20250823220900699

\(10\log (2) \approx 3\)dB

first order delta-sigma modulator

image-20250823220842529

\(30\log (2) \approx 9\)dB

second order delta-sigma modulator

image-20250823220922480

\(50\log (2) \approx 15\)dB


image-20250823224500839

1
2
3
4
5
6
7
8
9
10
11
12
13
14
OSR= linspace(1,16,16);

SQNR_ovonly_delta = 10*log10(OSR);
SQNR_1st_delta = -10*log10(pi^2/3) + 30*log10(OSR);
SQNR_2st_delta = -10*log10(pi^4/5) + 50*log10(OSR);

plot(OSR, SQNR_ovonly_delta,'ro-', LineWidth=4);
hold on
plot(OSR, SQNR_1st_delta,'bo-', LineWidth=4);
plot(OSR, SQNR_2st_delta,'mo-', LineWidth=4);
grid on; grid minor;
xlim([1 16]); ylim([-20 50]);
xlabel('OSR', FontSize=16); ylabel('\DeltaSQNR (dB)', FontSize=16);
legend('Oversampling Only', '1st \Delta\Sigma', '2nd \Delta\Sigma', fontsize=16)

TI. ADC12EU050 Continuous-Time Sigma-Delta ADCs [https://www.ti.com/lit/an/snaa098/snaa098.pdf]

image-20250906192735041

where \(N\) is the number of bits in the output, \(M\) is known as the over-sampling ratio, \(L\) is loop orders

quantizer levels

The greater the number of quantizer levels, the smaller quantization error

image-20250824081318669

quantizer overload

image-20250905062939783

\(\Delta \Sigma\) vs. \(\Delta\) modulation

  • \(\Delta \Sigma\) modulators, and other noise-shaping modulators, change the spectrum of the noise but leave the signal unchanged

  • \(\Delta\) modulators and other signal-predicting modulators shape the spectrum of the modulated signal but leave the quantization noise unchanged at the receiver

output vs. error-feedback

The error-feedback architecture is problematic for analog implementation, since it is sensitive to variations of its parameters (subtractor realization)

  • The error-feedback structure is thus of limited utility in \(\Delta \Sigma\) ADCs
  • The error-feedback structure is very useful and applied in digital loops required in \(\Delta \Sigma\) DACs

ADC

image-20250618203604863

image-20250618203636417

DAC

image-20250617223537672

P. Kiss, J. Arias and Dandan Li, "Stable high-order delta-sigma DACS," 2003 IEEE International Symposium on Circuits and Systems (ISCAS), Bangkok, 2003 [https://www.ele.uva.es/~jesus/analog/tcasi2003.pdf]


output-feedback

img

[https://www.linkedin.com/posts/danboschen_signalprocessing-dsp-pythonforengineers-activity-7345777588746788866-SprG?utm_source=share&utm_medium=member_desktop&rcm=ACoAAD-cuiIBDJ62eh9q3qTSSdslYXr-XMd8TGw]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
// https://github.com/hamsternz/second_order_sigma_delta_DAC

`timescale 1ns / 1ps
module second_order_dac(
input wire i_clk,
input wire i_res,
input wire i_ce,
input wire [15:0] i_func,
output wire o_DAC
);

reg this_bit;

reg [19:0] DAC_acc_1st;
reg [19:0] DAC_acc_2nd;
reg [19:0] i_func_extended;

assign o_DAC = this_bit;

always @(*)
i_func_extended = {i_func[15],i_func[15],i_func[15],i_func[15],i_func};

always @(posedge i_clk or negedge i_res)
begin
if (i_res==0)
begin
DAC_acc_1st<=16'd0;
DAC_acc_2nd<=16'd0;
this_bit = 1'b0;
end
else if(i_ce == 1'b1)
begin
if(this_bit == 1'b1)
begin
DAC_acc_1st = DAC_acc_1st + i_func_extended - (2**15);
DAC_acc_2nd = DAC_acc_2nd + DAC_acc_1st - (2**15);
end
else
begin
DAC_acc_1st = DAC_acc_1st + i_func_extended + (2**15);
DAC_acc_2nd = DAC_acc_2nd + DAC_acc_1st + (2**15);
end
// When the high bit is set (a negative value) we need to output a 0 and when it is clear we need to output a 1.
this_bit = ~DAC_acc_2nd[19];
end
end
endmodule

Time and Frequency Domain

image-20250627193435726

\(M \gt N\)

[https://web.engr.oregonstate.edu/~temes/ece627/Lecture_Notes/Intro_to_Delta_Sigma_Data_Converters.pdf]


Chun-Hsien Su ( 蘇純賢). Fundamentals of Sigma-Delta Data Converters,July, 2006 [pdf]

image-20250809235244362

image-20250809235311542

ADC

image-20250611234653738

image-20250612000925089

hackaday. Tearing Into Delta Sigma ADC’s [https://hackaday.com/2016/07/07/tearing-into-delta-sigma-adcs-part-1/]


image-20250617234727838

DAC

an interpolation filter effectively up-samples its low-rate input and lowpass-filters the resulting high-rate data to produce a high-rate output devoid of images

image-20250612000423191

P.E. Allen -CMOS Analog Circuit Design: Lecture 39 – Oversampling ADCs – Part I (6/26/14) [https://aicdesign.org/wp-content/uploads/2018/08/lecture39-140626.pdf]

P.E. Allen -CMOS Analog Circuit Design: Lecture 40 – Oversampling ADCs – Part II (7/17/15) [https://aicdesign.org/wp-content/uploads/2018/08/lecture40-150717.pdf]


image-20250720201944707

David Johns and Ken Martin. Oversampling Converters [https://www.eecg.toronto.edu/~johns/ece1371/slides/14_oversampling.pdf]


image-20250627194351778

[https://web.engr.oregonstate.edu/~temes/ece627/Lecture_Notes/Intro_to_Delta_Sigma_Data_Converters.pdf]

No delay-free loops

Any such physically feasible device will take a finite time to operate – in other words, the quantized output will only be available a small time after the quantizer has "looked" at the input - insert a one-sample delay

image-20250617231014547

there cannot be a "delay free loop" is a common idea in sequential digital state machine design


image-20241128232040924

Both integrator and quantizer are delay free

NTF realizability criterion: No delay-free loops in the modulator

image-20241128233022231

linear settling & GBW of amplifier

TODO 📅

Switched capacitor has been the common realization technique of discrete-time (DT) modulators, and in order to achieve a linear settling, the sampling frequency used in these converters needs to be significantly lower than the gain bandwidth product (GBW) of the amplifiers.

MOD1 & MOD2

MOD1: first-order noise-shaped converter (\(\Delta\Sigma\) modulator)

MOD2: second-order noise-shaped converter (\(\Delta\Sigma\) modulator)

MOD1

image-20241005120659945 \[ V(z) = U(z) +(1-z^{-1})E(z) \]

  • A binary DAC (and hence a binary modulator) is inherently linear
  • With a CT loop filter, MOD1 has inherent anti-alising

image-20241005202024498 \[\begin{align} v[1] &= u - (0) + e[1] \\ v[2] &= 2u - (v[1]) + e[2] \\ v[3] &= 3u - (v[1]+v[2]) + e[3] \\ v[4] &= 4u - (v[1]+v[2]+v[3]) + e[4] \end{align}\]

That is \[ v[n] = nu - \sum_{k=1}^{n-1}v[k] + e[n] \] Therefore, we have \(v[n-1] = (n-1)u - \sum_{k=1}^{n-2}v[k] + e[n-1]\), then \[\begin{align} v[n] &= nu - \sum_{k=1}^{n-1}v[k] + e[n] \\ &= u + \left((n-1)u - \sum_{k=1}^{n-2}v[k]\right) - v[n-1] + e[n] \\ &= u + v[n-1] - e[n-1] -v[n-1] + e[n] \\ &= u + e[n] - e[n-1] \end{align}\]


image-20250524215712688

Dout, the low frequency component of ADC out is same with Vin

MOD2

[https://web.engr.oregonstate.edu/~temes/ece627/Lecture_Notes/2nd_Higher_Order.pdf]

image-20241005160203074

MOD1 with DC Excitation

TODO 📅

Mismatch Shaping

image-20241112220458335

Data-Weighted Averaging (DWA)

image-20241113000942025 \[\begin{align} \sum_{i=0}^{n}v[i] + e_\text{DAC}[n] &= y[n] \\ \sum_{i=0}^{n-1}v[i] + e_\text{DAC}[n-1] &= y[n-1] \end{align}\]

and we have \(w[n] = y[n] - y[n-1]\), then \[ w[n] = v[n] + e_\text{DAC}[n] - e_\text{DAC}[n-1] \] i.e. \[ W = V + (1-z^{-1})e_\text{DAC} \]

Element Rotation:

image-20241112233059745

[http://individual.utoronto.ca/schreier/lectures/12-2.pdf], [http://individual.utoronto.ca/trevorcaldwell/course/Mismatch.pdf]

integrator leakage

When the integrator includes leakage (\(\alpha\))

\[ x[n-1] + \alpha y[n-1] = y[n] \]

then, \[ \frac{Y}{X} = \frac{z^{_1}}{1-\alpha z^{-1}} \]

reference

Pavan, Shanthi, Richard Schreier, and Gabor Temes. (2016). Understanding Delta-Sigma Data Converters. 2nd ed. Wiley.

Norsworthy, Steven R., Richard Schreier, Gábor C. Temes and Ieee Circuits. “Delta-sigma data converters : theory, design, and simulation.” (1997).

Horowitz, P., & Hill, W. (2015). The art of electronics (3rd ed.). Cambridge University Press. [pdf]


R. Schreier, ISSCC2006 tutorial: Understanding Delta-Sigma Data Converters

Shanthi Pavan, ISSCC2013 T5: Simulation Techniques in Data Converter Design [https://www.nishanchettri.com/isscc-slides/2013%20ISSCC/TUTORIALS/ISSCC2013Visuals-T5.pdf]

Bruce A. Wooley , 2012, "The Evolution of Oversampling Analog-to-Digital Converters" [https://r6.ieee.org/scv-sscs/wp-content/uploads/sites/80/2012/06/Oversampling-Wooley_SCV-ver2.pdf]

Venkatesh Srinivasan, ISSCC 2019 T5: Noise Shaping in Data Converters

B. Razavi, "The Delta-Sigma Modulator [A Circuit for All Seasons]," IEEE Solid-State Circuits Magazine, Volume. 8, Issue. 20, pp. 10-15, Spring 2016. [http://www.seas.ucla.edu/brweb/papers/Journals/BRSpring16DeltaSigma.pdf]

P. M. Aziz, H. V. Sorensen and J. vn der Spiegel, "An overview of sigma-delta converters," in IEEE Signal Processing Magazine, vol. 13, no. 1, pp. 61-84, Jan. 1996 [https://sci-hub.st/10.1109/79.482138]


Richard E. Schreier, ECE 1371 Advanced Analog Circuits - 2015 [http://individual.utoronto.ca/schreier/ece1371-2015.html]

Gabor C. Temes. ECE 627-Oversampled Delta-Sigma Data Converters [https://classes.engr.oregonstate.edu/eecs/spring2017/ece627/lecturenotes.html]

Joshua Reiss. Understanding sigma delta modulation: the solved and unsolved issues

[https://www.eecs.qmul.ac.uk/~josh/documents/2008/Reiss-JAES-UnderstandingSigmaDeltaModulation-SolvedandUnsolvedIssues.pdf]

V. Medina, P. Rombouts and L. Hernandez-Corporales, "A Different View of Sigma-Delta Modulators Under the Lens of Pulse Frequency Modulation [Feature]," in IEEE Circuits and Systems Magazine, vol. 24, no. 2, pp. 80-97, Secondquarter 2024

FFE Coeff. Selection

Jose E. Schutt-Aine, Spring 2024 ECE 546 Lecture - 27 Equalization [http://emlab.uiuc.edu/ece546/Lect_27.pdf]

Sam Palermo. Lecture 7 - Equalization Intro & TX FIR EQ [https://people.engr.tamu.edu/spalermo/ecen689/lecture7_ee720_eq_intro_txeq.pdf]

David Johns. ECE1392H - Integrated Circuits for Digital Communications - Fall 2001: Equalization [https://www.eecg.utoronto.ca/~johns/ece1392/slides/equalization.pdf]

TODO 📅

TX FIR

RX FIR

Multipliers

TODO 📅

Adders

TODO 📅

overlapped tuning range

TODO 📅

Mueller-Muller PD

Mueller-Muller type A timing function

image-20241019163636292

Mueller-Muller type B timing function

image-20241019163813449

LMS (Least-Mean-Square)

minimum mean square error (MMSE)

This simplified version of LMS algorithm is identical to the zero-forcing algorithm which minimizes the ISI at data samples

Sign-Sign LMS (SS-LMS)

T11: Basics of Equalization Techniques: Channels, Equalization, and Circuits, 2022 IEEE International Solid-State Circuits Conference

V. Stojanovic et al., "Autonomous dual-mode (PAM2/4) serial link transceiver with adaptive equalization and data recovery," in IEEE Journal of Solid-State Circuits, vol. 40, no. 4, pp. 1012-1026, April 2005, doi: 10.1109/JSSC.2004.842863.

Jinhyung Lee, Design of High-Speed Receiver for Video Interface with Adaptive Equalization; Phd thesis, August 2019. thesis link

Paulo S. R. Diniz, Adaptive Filtering: Algorithms and Practical Implementation, 5th edition

E. -H. Chen et al., "Near-Optimal Equalizer and Timing Adaptation for I/O Links Using a BER-Based Metric," in IEEE Journal of Solid-State Circuits, vol. 43, no. 9, pp. 2144-2156, Sept. 2008

DFE h0 Estimator

summer output \[ r_k = a_kh_0+\left(\sum_{n=-\infty,n\neq0}^{+\infty}a_{k-n}h_n-\sum_{n=1}^{\text{ntap}}\hat{a}_{k-n}\hat{h}_n\right) \] error slicer analog output \[ e_k=r_k-\hat{a}_k \hat{h}_0 \] error slicer digital output \[ \hat{e}_k=|e_k| \] It's NOT possible to implement \(e_k\), which need to determine \(\hat{a}_k=|r_k|\) in no time. One method to approach this problem is calculate \(e_k^{a_k=1}=r_k-\hat{a}_k \hat{h}_0\) and \(e_k^{a_k=-1}=r_k+\hat{a}_k \hat{h}_0\), then select the right one based on \(\hat{a}_k\)

The update equation based on Sign-Sign-Least Mean square (SS-LMS) and loss function \(L(\hat{h}_{\text{0~ntap}})=E(e_k^2)\) \[ \hat{h}_n(k+1) = \hat{h}_n(k)+\mu \cdot |e_k|\cdot \hat{a}_{k-n} \] Where \(n \in [0,...,\text{ntap}]\). This way, we can obtain \(\hat{h}_0\), \(\hat{h}_1\), \(\hat{h}_2\), ...

\(\hat{h}_0\) is used in AFE adaptation

We may encounter difficulty if the first tap of DFE is unrolled, its \(e_k\) is modified as follow \[ r_k = a_kh_0+\left(\sum_{n=-\infty,n\neq0}^{+\infty}a_{k-n}h_n-\sum_{n=2}^{\text{ntap}}\hat{a}_{k-n}\hat{h}_n\right) \] Where there is NO \(\hat{h}_1\)

To find \(\hat{h}_1\), we shall use different pattern for even and odd error slicer

MLSD (Maximum Likelihood Sequence Detection)

The process is also referred to as Maximum Likelihood Sequence Estimator (MLSE)

image-20240807233152154

image-20240812205534753

image-20240812205613467

[IBIS-AMI Modeling and Correlation Methodology for ADC-Based SerDes Beyond 100 Gb/s https://static1.squarespace.com/static/5fb343ad64be791dab79a44f/t/63d807441bcd266de258b975/1675102025481/SLIDES_Track02_IBIS_AMI_Modeling_and_Correlation_Tyshchenko.pdf]

M. Emami Meybodi, H. Gomez, Y. -C. Lu, H. Shakiba and A. Sheikholeslami, "Design and Implementation of an On-Demand Maximum-Likelihood Sequence Estimation (MLSE)," in IEEE Open Journal of Circuits and Systems, vol. 3, pp. 97-108, 2022, doi: 10.1109/OJCAS.2022.3173686.

Zaman, Arshad Kamruz (2019). A Maximum Likelihood Sequence Equalizing Architecture Using Viterbi Algorithm for ADC-Based Serial Link. Undergraduate Research Scholars Program. Available electronically from [https://hdl.handle.net/1969.1/166485]

There are several variants of MLSD (Maximum Likelihood Sequence Detection), including:

  • Viterbi Algorithm
  • Decision Feedback Sequence Estimation (DFSE)
  • Soft-Output MLSD

[Evolution Of Equalization Techniques In High-Speed SerDes For Extended Reaches. https://semiengineering.com/evolution-of-equalization-techniques-in-high-speed-serdes-for-extended-reaches/]

S. Song, K. D. Choo, T. Chen, S. Jang, M. P. Flynn and Z. Zhang, "A Maximum-Likelihood Sequence Detection Powered ADC-Based Serial Link," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 65, no. 7, pp. 2269-2278, July 2018

[http://contents.kocw.or.kr/document/lec/2012/Korea/KoYoungChai/33.pdf]

image-20240824193839108

Mueller-Muller CDR

image-20240812222307061

MMPD infers the channel response from baud-rate samples of the received data, the adaptation aligns the sampling clock such that pre-cursor is equal to the post-cursor in the pulse response

image-20240807230029591

Faisal A. Musa. "HIGH-SPEED BAUD-RATE CLOCK RECOVERY" [https://www.eecg.utoronto.ca/~tcc/thesis-musa-final.pdf]

Faisal A. Musa."CLOCK RECOVERY IN HIGH-SPEED MULTILEVEL SERIAL LINKS" [https://www.eecg.utoronto.ca/~tcc/faisal_iscas03.pdf]

Eduardo Fuentetaja. "Analysis of the M&M Clock Recovery Algorithm" [https://edfuentetaja.github.io/sdr/m_m_analysis/]

Liu, Tao & Li, Tiejun & Lv, Fangxu & Liang, Bin & Zheng, Xuqiang & Wang, Heming & Wu, Miaomiao & Lu, Dechao & Zhao, Feng. (2021). Analysis and Modeling of Mueller-Muller Clock and Data Recovery Circuits. Electronics. 10. 1888. 10.3390/electronics10161888.

Gu, Youzhi & Feng, Xinjie & Chi, Runze & Chen, Yongzhen & Wu, Jiangfeng. (2022). Analysis of Mueller-Muller Clock and Data Recovery Circuits with a Linearized Model. 10.21203/rs.3.rs-1817774/v1.

Baud-Rate CDRs [https://ocw.snu.ac.kr/sites/default/files/NOTE/Lec%206%20-%20Clock%20and%20Data%20Recovery.pdf]

F. Spagna et al., "A 78mW 11.8Gb/s serial link transceiver with adaptive RX equalization and baud-rate CDR in 32nm CMOS," 2010 IEEE International Solid-State Circuits Conference - (ISSCC), San Francisco, CA, USA, 2010, pp. 366-367, doi: 10.1109/ISSCC.2010.5433823.

K. Yadav, P. -H. Hsieh and A. C. Carusone, "Loop Dynamics Analysis of PAM-4 Mueller–Muller Clock and Data Recovery System," in IEEE Open Journal of Circuits and Systems, vol. 3, pp. 216-227, 2022 [https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9910561]

Jaeduk Han, "Design and Automatic Generation of 60Gb/s Wireline Transceivers" [https://www2.eecs.berkeley.edu/Pubs/TechRpts/2019/EECS-2019-143.pdf]

image-20240810095006113

image-20240808001201612

image-20240808001256515

image-20240808001449664

image-20240808001501485

SS-MM CDR

image-20240807232814202

\(h_1\) is necessary

  • without DFE

    SS-MMPD locks at the point (\(h_1=h_{-1}\)​)

  • With a 1-tap DFE

    1-tap adaptive DFE that forces the \(h_1\) to be zero, the SS-MMPD locks wherever the \(h_{-1}\)​ is zero and drifts eventually.

    Consequently, it suffers from a severe multiple-locking problem with an adaptive DFE

image-20240812232618238

Kwangho Lee, "Design of Receiver with Offset Cancellation of Adaptive Equalizer and Multi-Level Baud-Rate Phase Detector" [https://s-space.snu.ac.kr/bitstream/10371/177584/1/000000167211.pdf]

Pattern filter

pattern main cursor
011 \(s_{011}=-h_1+h_0+h_{-1}\)
110 \(s_{110}=h_1+h_0-h_{-1}\)
100 \(s_{100}=h_1-h_0-h_{-1}\)
001 \(s_{001}=-h_1-h_0+h_{-1}\)

During adapting, we make

  • \(s_{011}\) & \(s_{110}\) are approaching to each other
  • \(s_{100}\) & \(s_{001}\) are approaching to each other

Then, \(h_{-1}\) and \(h_1\) are same, which is desired

Bang-Bang CDR

alexander PD or !!PD

The alexander PD locks that edge clock (clkedge) is located at zero crossings of the data. The \(h_{-0.5}\) and \(h_{0.5}\) are equal at the lock point, where the \(h_{-0.5}\) and \(h_{0.5}\) are the cursors located at -0.5 UI and 0.5 UI.

Kwangho Lee, "Design of Receiver with Offset Cancellation of Adaptive Equalizer and Multi-Level Baud-Rate Phase Detector" [https://s-space.snu.ac.kr/bitstream/10371/177584/1/000000167211.pdf]

Shahramian, Shayan, "Adaptive Decision Feedback Equalization With Continuous-time Infinite Impulse Response Filters" [https://tspace.library.utoronto.ca/bitstream/1807/77861/3/Shahramian_Shayan_201606_PhD_thesis.pdf]

MENIN, DAVIDE, "Modelling and Design of High-Speed Wireline Transceivers with Fully-Adaptive Equalization" [https://air.uniud.it/retrieve/e27ce0ca-15f7-055e-e053-6605fe0a7873/Modelling%20and%20Design%20of%20High-Speed%20Wireline%20Transceivers%20with%20Fully-Adaptive%20Equalization.pdf]

reference

Stojanovic, Vladimir & Ho, A. & Garlepp, B. & Chen, Fred & Wei, J. & Alon, Elad & Werner, C. & Zerbe, J. & Horowitz, M.A.. (2004). Adaptive equalization and data recovery in a dual-mode (PAM2/4) serial link transceiver. IEEE Symposium on VLSI Circuits, Digest of Technical Papers. 348 - 351. 10.1109/VLSIC.2004.1346611.

A. A. Bazargani, H. Shakiba and D. A. Johns, "MMSE Equalizer Design Optimization for Wireline SerDes Applications," in IEEE Transactions on Circuits and Systems I: Regular Papers, doi: 10.1109/TCSI.2023.3328807.

Masum Hossain, ISSCC2023 T11: "Digital Equalization and Timing Recovery Techniques for ADC-DSP-based Highspeed Links" [https://www.nishanchettri.com/isscc-slides/2023%20ISSCC/TUTORIALS/T11.pdf]

—, "LOW POWER DIGITAL EQUALIZATION FOR HIGH SPEED SERDES" [https://www.ieeetoronto.ca/wp-content/uploads/2020/06/SSCS_invited_talk.pdf]

A. Sharif-Bakhtiar, A. Chan Carusone, "A Methodology for Accurate DFE Characterization," IEEE RFIC Symposium, Philadelphia, Pennsylvania, June 2018. [PDF] [Slides – PDF]

Tony Chan Carusone. High Speed Communications Part 11 – SerDes DSP Interactions [https://youtu.be/YIAwLskuVPc?si=MYIbXLwFqQj0EElU]

—, 2022 Optimization Tools for Future Wireline Transceivers [https://www.ieeetoronto.ca/wp-content/uploads/2022/12/UofT-Future-of-Wireline-Workshop-2022.pdf]

Alphawave IP CEO. How DSP is Killing the Analog in SerDes [https://youtu.be/OY2Dn4EDPiA?si=czIYfFrHpY4F-lKK]


S. Kiran, S. Cai, Y. Zhu, S. Hoyos and S. Palermo, "Digital Equalization With ADC-Based Receivers: Two Important Roles Played by Digital Signal Processingin Designing Analog-to-Digital-Converter-Based Wireline Communication Receivers," in IEEE Microwave Magazine, vol. 20, no. 5, pp. 62-79, May 2019 [https://sci-hub.se/10.1109/MMM.2019.2898025]

K. K. Parhi, "Design of multigigabit multiplexer-loop-based decision feedback equalizers," in IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 13, no. 4, pp. 489-493, April 2005 [http://sci-hub.se/10.1109/TVLSI.2004.842935]

T. Toifl et al., "A 3.5pJ/bit 8-tap-feed-forward 8-tap-decision feedback digital equalizer for 16Gb/s I/Os," ESSCIRC 2014 - 40th European Solid State Circuits Conference (ESSCIRC), Venice Lido, Italy, 2014 [https://sci-hub.se/10.1109/ESSCIRC.2014.6942120]


Daniel Friedman, 2018 Considerations and Implementations for High data Rate Serial Link Design [https://www.ieeetoronto.ca/wp-content/uploads/2020/06/DL-Toronto-Nov-2018.pdf]


Keshab K. Parhi [http://www.ece.umn.edu/users/parhi/]

Tinoosh Mohsenin. CMPE 691: Digital Signal Processing Hardware Implementation [https://userpages.cs.umbc.edu/tinoosh/cmpe691/]


Cathy Ye Liu, Broadcom Inc. DesignCon 2019: 100+ Gb/s Ethernet Forward Error Correction (FEC) Analysis

—, Broadcom Inc. DesignCon 2024: 200+ Gbps Ethernet Forward Error Correction (FEC) Analysis


Tony Chan Carusone Integrated Systems Laboratory, University of Toronto [https://isl.utoronto.ca/publications/]

Tony Chan Carusone 2022. Optimization Tools for Future Wireline Transceivers [https://www.ieeetoronto.ca/wp-content/uploads/2022/12/UofT-Future-of-Wireline-Workshop-2022.pdf]

Aleksey Tyshchenko, SeriaLink Systems Clinton Walker, Alphawave IP. DesignCon 2022. IBIS-AMI Modeling and Correlation Methodology for ADC-Based SerDes Beyond 100 Gb/s [https://static1.squarespace.com/static/5fb343ad64be791dab79a44f/t/63d807441bcd266de258b975/1675102025481/SLIDES_Track02_IBIS_AMI_Modeling_and_Correlation_Tyshchenko.pdf]

[https://ibis.org/summits/apr22/tyshchenko.pdf]

[https://www.mathworks.com/content/dam/mathworks/conference-or-academic-paper/ibis-ami-modeling-and-correlation.pdf]


Ali Sheikholeslami Electronics Group, University of Toronto [https://www.eecg.utoronto.ca/~ali/]


Barry, John R., Edward A. Lee, and David G. Messerschmitt. Digital communication. Springer, 2003.


Qasim Chaudhari. Maximum Likelihood Estimation of Clock Offset [https://wirelesspi.com/maximum-likelihood-estimation-of-clock-offset/]

—. Channel Estimation in Wireless Communication. [https://wirelesspi.com/channel-estimation-in-wireless-communication/]

—. Phase Locked Loop (PLL) in a Software Defined Radio (SDR) [https://wirelesspi.com/phase-locked-loop-pll-in-a-software-defined-radio-sdr/]

—. Phase Locked Loop (PLL) for Symbol Timing Recovery [https://wirelesspi.com/phase-locked-loop-pll-for-symbol-timing-recovery/]

—. How Decision Feedback Equalizers (DFE) Work [https://wirelesspi.com/how-decision-feedback-equalizers-dfe-work/]

—. Maximum Likelihood Sequence Estimation (MLSE Equalizer) [https://wirelesspi.com/maximum-likelihood-sequence-estimation-mlse-equalizer/]

—. Least Mean Square (LMS) Equalizer – A Tutorial [https://wirelesspi.com/least-mean-square-lms-equalizer-a-tutorial/]

—. Early-Late Bit Synchronizer in Digital Communication [https://wirelesspi.com/early-late-bit-synchronizer-in-digital-communication/]

—. Gardner Timing Error Detector: A Non-Data-Aided Version of Zero-Crossing Timing Error Detectors [https://wirelesspi.com/gardner-timing-error-detector-a-non-data-aided-version-of-zero-crossing-timing-error-detectors/]

—. Mueller and Muller Timing Synchronization Algorithm [https://wirelesspi.com/mueller-and-muller-timing-synchronization-algorithm/]

—. Digital Filter and Square Timing Recovery [https://wirelesspi.com/digital-filter-and-square-timing-recovery/]

—. What is a Symbol Timing Offset and How It Distorts the Rx Signal [https://wirelesspi.com/what-is-a-symbol-timing-offset-and-how-it-distorts-the-rx-signal/]

—. How Automatic Gain Control (AGC) Works [https://wirelesspi.com/how-automatic-gain-control-agc-works/]


Noise Analysis

image-20250526201936387


image-20250526195323660

sampling (amplification) phase

image-20250526195656447

Noise Simulation

PSS + Pnoise Method

Comparator Output SNR during sampling region and decision region go up

Comparator Output SNR during regeneration region is constant, where noise is critical

image-20250526221529514

image-20241109163928889

Transient Noise Method

Noise Fmax sets the bandwidth of the random noise sources that are injected at each time point in the transient analysis


image-20241109154249513

We can identify the RMS noise value easily by looking at 15.9% or 84.1% of CDF (\(1\sigma\)), the input-referred noise in the RMS is 0.9mV

image-20241109160311684

Thus, if \(V_S\) is chosen so as to reduce the probability of zeros to 16%, then \(V_S = 1\sigma\), which is also the total root-mean square (rms) noise referred to the input.

Comparison of two methods

image-20250526225952590

here, fundamental frequency = fclk; integrated noise (0 ~ 0.5fclk)

image-20250526230126010

E. Gillen, G. Panchanan, B. Lawton and D. O'Hare, "Comparison of transient and PNOISE simulation techniques for the design of a dynamic comparator," 2022 33rd Irish Signals and Systems Conference (ISSC), Cork, Ireland, 2022, pp. 1-5

Chenguang Yang, "Comparator Design for High Speed ADC" [https://lup.lub.lu.se/luur/download?func=downloadFile&recordOId=9164380&fileOId=9164388]

J. Conrad, J. Kauffman, S. Wilhelmstatter, R. Asthana, V. Belagiannis and M. Ortmanns, "Confidence Estimation and Boosting for Dynamic-Comparator Transient-Noise Analysis," 2024 22nd IEEE Interregional NEWCAS Conference (NEWCAS), Sherbrooke, QC, Canada, 2024, pp. 1-5

There are some ambiguity in formula in ADC Verification Rapid Adoption Kit (RAK)(Product Version: IC 6.1.8, SPECTRE 18.1 March, 2019)

  • Transient Noise Analysis: \(\sqrt{2}\sigma\), why ratio \(\sqrt{2}\) ???
  • PSS+Pnoise: why two fundamental tones fclk/2 ???

Common-Mode (Vcmi) Variation Effects

image-20240925225059596

image-20240925225823184


image-20250527202331008


image-20250609224554118

Zhaokai Liu. Time-interleaved SAR ADC Design Using Berkeley Analog Generator [https://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/EECS-2020-109.pdf]

offset simulation

T. Caldwell. ECE 1371S Advanced Analog Circuits [http://individual.utoronto.ca/trevorcaldwell/course/comparators.pdf]

Eric Chang. EECS240-s18 Discussion 9


image-20241109092310123

Graupner, Achim & Sobe, Udo. (2007). Offset-Simulation of Comparators. [https://designers-guide.org/analysis/comparator.pdf]

1
2
3
4
Comment on "Offset-Simulation of Comparators"

If the input referred offset follows a normal distribution than it is sufficient to apply a single offset voltage to calculate the offset voltage.
See details in Razavi, B., The StrongARM Latch [A Circuit for All Seasons], IEEE Solid-State Circuits Magazine, Volume:7, Issue: 2, Spring 2015

Omran, Hesham. (2019). Fast and accurate technique for comparator offset voltage simulation. Microelectronics Journal. 89. 10.1016/j.mejo.2019.05.004.

Matthews, Thomas W. and Perry L. Heedley. “A simulation method for accurately determining DC and dynamic offsets in comparators.” 48th Midwest Symposium on Circuits and Systems, 2005. (2005): 1815-1818 Vol. 2. [https://athena.ecs.csus.edu/~pheedley/MSDL/MSDL_DOTB_cmp_test_bench_MWSCAS05.pdf]

Hysteresis

P. Bruschi: Notes on Mixed Signal Design http://www2.ing.unipi.it/~a008309/mat_stud/MIXED/archive/2019/Optional_notes/Chap_3_4_Comparators.pdf

TODO 📅

Kickback Noise

Kickback noise trades with the dimensions of the input transistors and hence with the offset voltage

  • affects the comparator's own decision
  • corrupts the input voltage while it is sensed by other circuits

image-20241110004944542

Tetsuya Iizuka,VLSI2021_Workshop3 "Nyquist A/D Converter Design in Four Days"

Figueiredo, Pedro & Vital, João. (2006). Kickback noise reduction techniques for CMOS latched comparators. Circuits and Systems II: Express Briefs, IEEE Transactions on. 53. 541 - 545. 10.1109/TCSII.2006.875308. [https://sci-hub.se/10.1109/TCSII.2006.875308]

P. M. Figueiredo and J. C. Vital, "Low kickback noise techniques for CMOS latched comparators," 2004 IEEE International Symposium on Circuits and Systems (ISCAS), Vancouver, BC, Canada, 2004, pp. I-537 [https://sci-hub.se/10.1109/ISCAS.2004.1328250]

Lei, Ka Meng & Mak, Pui-In & Martins, R.P.. (2013). Systematic analysis and cancellation of kickback noise in a dynamic latched comparator. Analog Integrated Circuits and Signal Processing. 77. 277-284. 10.1007/s10470-013-0156-1. [https://rto.um.edu.mo/wp-content/uploads/docs/ruimartins_cv/publications/journalpapers/57.pdf]

O. M. Ívarsson, "Comparator Kickback Reduction Techniques for High-Speed ADCs," Dissertation, 2024. [https://liu.diva-portal.org/smash/get/diva2:1872476/FULLTEXT01.pdf]


Current mirrors are used between stages to reduce charge kick back from the logic level swing of the latch onto the small comparator input capacitors

Mike Shuo-Wei Chen and R. W. Brodersen, "A 6-bit 600-MS/s 5.3-mW Asynchronous ADC in 0.13-μm CMOS," in IEEE Journal of Solid-State Circuits, vol. 41, no. 12, pp. 2669-2680, Dec. 2006 [pdf, slides]

K. Bult and A. Buchwald, "An embedded 240-mW 10-b 50-MS/s CMOS ADC in 1-mm/sup 2/," in IEEE Journal of Solid-State Circuits, vol. 32, no. 12, pp. 1887-1895, Dec. 1997 [https://sci-hub.st/10.1109/4.643647]

CMOS Latch

TODO 📅

image-20241215162321832 \[ V_{o,fb}^+ - V_{o,fb}^- = \frac{g_m}{sC_L}(V_o^+ - V_o^-) = A(s)\cdot(V_o^+ - V_o^-) \]

We have \[ A(s)\cdot (V_{i} + V_o) = V_o \]

that is \[ V_o = \frac{A(s)}{1-A(s)}V_{i} = \frac{1}{s - g_m/C_L}\cdot \frac{g_mV_i}{C_L} \]

therefore \[ V_o(t) = \frac{g_mV_i}{C_L}\cdot\exp\left({\frac{g_m}{C_L}t}\right) = V_o(t=0)\cdot\exp\left({\frac{g_m}{C_L}t}\right) \] image-20241215173645188

Asad Abidi, ISSCC 2023: Circuit Insights "The CMOS Latch" [https://youtu.be/sVe3VUTNb4Q?si=Pl75jWiA0kNPOlOs]

Metastability

TODO 📅

If the comparator can not generate a well-defined logical output in half of the clock period, we say the circuit is "metastable"

image-20241215162430509

Pre-amp (preamplifier)

preampSong202412181018

Vishal Saxena "CMOS Comparator Design Extra Slides" [https://www.eecis.udel.edu/~vsaxena/courses/ece614/Handouts/Comparator%20Slides.pdf]


image-20250805230555464

W. Liu, P. Huang and Y. Chiu, "A 12b 22.5/45MS/s 3.0mW 0.059mm2 CMOS SAR ADC achieving over 90dB SFDR," 2010 IEEE International Solid-State Circuits Conference - (ISSCC), San Francisco, CA, USA, 2010 [https://sci-hub.se/10.1109/ISSCC.2010.5433830]

Math Background

Relating \(\Phi\) and erf

Error Function (Erf) of the standard Normal distribution \[ \text{Erf}(x) = \frac{2}{\sqrt{\pi}}\int_0^x e^{-t^2} \mathrm{d}t. \] Cumulative Distribution Function (CDF) of the standard Normal distribution \[ \Phi(x) = \frac{1}{\sqrt{2\pi}}\int_{-\infty}^x e^{-z^2/2} \mathrm{d}z. \]

Figure

\[\begin{align} \Phi(x) &= \frac{\text{Erf}(x/\sqrt{2})+1}{2}. \\ \Phi(x\sqrt{2}) &= \frac{\text{Erf}(x) + 1}{2} \end{align}\]

Considering the mean and standard deviation \[ \Phi(x,\mu,\sigma)=\frac{1}{2}\left( 1+\text{Erf} \left( \frac{x-\mu}{\sigma\sqrt{2}} \right)\right) \]


image-20241109135425126

John D. Cook. Relating Φ and erf [https://www.johndcook.com/erf_and_normal_cdf.pdf]

reference

Xu, H. (2018). Mixed-Signal Circuit Design Driven by Analysis: ADCs, Comparators, and PLLs. UCLA. ProQuest ID: Xu_ucla_0031D_17380. Merritt ID: ark:/13030/m5f52m8x. Retrieved from [https://escholarship.org/uc/item/88h8b5t3]

A. Abidi and H. Xu, "Understanding the Regenerative Comparator Circuit," Proceedings of the IEEE 2014 Custom Integrated Circuits Conference, San Jose, CA, 2014, pp. 1-8. [https://picture.iczhiku.com/resource/ieee/WHiYwoUjPHwZPXmv.pdf]

T. Sepke, P. Holloway, C. G. Sodini and H. -S. Lee, "Noise Analysis for Comparator-Based Circuits," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 56, no. 3, pp. 541-553, March 2009 [https://dspace.mit.edu/bitstream/handle/1721.1/61660/Speke-2009-Noise%20Analysis%20for%20Comparator-Based%20Circuits.pdf]

Sepke, Todd. "Comparator design and analysis for comparator-based switched-capacitor circuits." (2006). [https://dspace.mit.edu/handle/1721.1/38925]

P. Nuzzo, F. De Bernardinis, P. Terreni and G. Van der Plas, "Noise Analysis of Regenerative Comparators for Reconfigurable ADC Architectures," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 55, no. 6, pp. 1441-1454, July 2008 [https://picture.iczhiku.com/resource/eetop/SYirpPPPaAQzsNXn.pdf]


J. Kim, B. S. Leibowitz, J. Ren and C. J. Madden, "Simulation and Analysis of Random Decision Errors in Clocked Comparators," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 56, no. 8, pp. 1844-1857, Aug. 2009, doi: 10.1109/TCSI.2009.2028449. URL:https://people.engr.tamu.edu/spalermo/ecen689/simulation_analysis_clocked_comparators_kim_tcas1_2009.pdf

J. Kim, B. S. Leibowitz and M. Jeeradit, "Impulse sensitivity function analysis of periodic circuits," 2008 IEEE/ACM International Conference on Computer-Aided Design, 2008, pp. 386-391, doi: 10.1109/ICCAD.2008.4681602. [https://websrv.cecs.uci.edu/~papers/iccad08/PDFs/Papers/05C.2.pdf]

Jaeha Kim, Lecture 12. Aperture and Noise Analysis of Clocked Comparators URL:https://ocw.snu.ac.kr/sites/default/files/NOTE/7038.pdf

Sam Palermo. ECEN720: High-Speed Links Circuits and Systems Spring 2023 Lecture 6: RX Circuits [https://people.engr.tamu.edu/spalermo/ecen689/lecture6_ee720_rx_circuits.pdf]


Y. Luo, A. Jain, J. Wagner and M. Ortmanns, "Input Referred Comparator Noise in SAR ADCs," in IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 66, no. 5, pp. 718-722, May 2019. [https://sci-hub.se/10.1109/TCSII.2019.2909429]

X. Tang et al., "An Energy-Efficient Comparator With Dynamic Floating Inverter Amplifier," in IEEE Journal of Solid-State Circuits, vol. 55, no. 4, pp. 1011-1022, April 2020 [https://sci-hub.se/10.1109/JSSC.2019.2960485]

Chen, Long & Sanyal, Arindam & Ma, Ji & Xiyuan, Tang & Sun, Nan. (2016). Comparator Common-Mode Variation Effects Analysis and its Application in SAR ADCs. 10.1109/ISCAS.2016.7538972. [https://labs.engineering.asu.edu/mixedsignals/wp-content/uploads/sites/58/2017/08/ISCAS_comp_long_2016.pdf]

V. Stojanovic, and V. G. Oklobdzija, "Comparative Analysis of Master–Slave Latches and Flip-Flops for High-Performance and Low-Power Systems," IEEE J. Solid-State Circuits, vol. 34, pp. 536–548, April 1999. [https://www.ece.ucdavis.edu/~vojin/CLASSES/EEC280/Web-page/papers/Clocking/Vlada-Latches-JoSSC-Apr-1999.pdf]

C. Mangelsdorf, "Metastability: Deeply misunderstood [Shop Talk: What You Didn’t Learn in School]," in IEEE Solid-State Circuits Magazine, vol. 16, no. 2, pp. 8-15, Spring 2024

Rabuske, Taimur & Fernandes, Jorge. (2014). Noise-aware simulation-based sizing and optimization of clocked comparators. Analog Integr. Circuits Signal Process.. 81. 723-728. 10.1007/s10470-014-0428-4. [https://sci-hub.se/10.1007/s10470-014-0428-4]

Rabuske, Taimur & Fernandes, Jorge. (2016). Charge-Sharing SAR ADCs for Low-Voltage Low-Power Applications. 10.1007/978-3-319-39624-8.


Masaya Miyahara, Yusuke Asada, Daehwa Paik and Akira Matsuzawa, "A low-noise self-calibrating dynamic comparator for high-speed ADCs," 2008 IEEE Asian Solid-State Circuits Conference, Fukuoka, Japan, 2008 [slides, paper]

Art Schaldenbrand, Senior Product Manager, Keeping Things Quiet: A New Methodology for Dynamic Comparator Noise Analysis URL:https://www.cadence.com/content/dam/cadence-www/global/en_US/videos/tools/custom-_ic_analog_rf_design/NoiseAnalyisposting201612Chalk%20Talk.pdf


B. Razavi, "The Design of a Comparator [The Analog Mind]," IEEE Solid-State Circuits Magazine, Volume. 12, Issue. 4, pp. 8-14, Fall 2020. https://www.seas.ucla.edu/brweb/papers/Journals/BR_SSCM_4_2020.pdf

B. Razavi, "The StrongARM Latch [A Circuit for All Seasons]," IEEE Solid-State Circuits Magazine, Issue. 2, pp. 12-17, Spring 2015. https://www.seas.ucla.edu/brweb/papers/Journals/BR_Magzine4.pdf

CHUNG-CHUN (CC) CHEN. Why A Dedicated Noise Analysis for A Strong-arm Latch / Comparator? [https://youtu.be/S5GnvFxuxUA?si=w38iLvzjr0azhu43]

—. Why Transient Noise (Trannoise) Analysis for A Strong-arm Latch / Comparator? [https://youtu.be/gpQggSM9_PE?si=apMd6yWVO1JHOHm_]

—. Why A Periodic Steady-State (PSS), Periodic Noise (Pnoise), and Hand Calculation for A Sampler? [https://youtu.be/lGqCfg5R-rY?si=nQ8QBwW2x8QUMryV]

Tony Chan Carusone,. 28 Comparator Specs and Characterization [https://youtu.be/mRfWM1bpr3k?si=WJXz0r3MJ7WvZ6h2]

Prof. Seung-Tak Ryu (KAIST) "Advanced ADC Design Techniques" Online Course (2022) : Dynamic Latch [https://youtu.be/zE1ZdG_XzWk?si=rk-DGUqMnQyjDiuU]

discrete-time frequency: \(\hat{\omega}=\omega T_s\), units are radians per sample


Below diagram show the windowing effect and sampling

NinDFT.drawio

For general window function, we know \(W(e^{j\hat{\omega}})=\frac{1}{T_s}W_c(j\frac{\hat\omega}{T_s})\),

\[ \frac{W_c(j\frac{\hat{\omega}}{T_s})X_c(j\frac{\hat{\omega}}{T_s})}{T_s}\cdot \frac{1}{2\pi} = \frac{T_sW(e^{j\hat{\omega}})X_c(j\frac{\hat\omega}{T_s})}{T_s}\cdot \frac{1}{2\pi}=W(e^{j\hat{\omega}})X_c(j\frac{\hat\omega}{T_s})\cdot \frac{1}{2\pi} \overset{\hat{\omega}=0}{\Longrightarrow} \sum_{n=-N_w}^{+N_w}w[n] \cdot X_c(j\omega)\cdot \frac{1}{2\pi} \]

e.g. \(\frac{W_c(j\omega|\omega=0)}{T_s} = N\) for Rectangular Window, shown in above figure

warmup

Continuous-time signals \(x_c(t)\) Discrete-time signals \(x[n]\)
Aperiodic signals Continuous Fourier transform Discrete-time Fourier transform
Periodic signals Fourier series Discrete Fourier transform

Continuous Time Fourier Series (CTFS)

\[\begin{align} a_k &= \frac{1}{T}\int_T x(t)e^{-jk(2\pi/T)) t}dt \\ x(t) &= \sum_{k=-\infty}^{+\infty}a_ke^{jk(2\pi/T) t} \end{align}\]

Continuous-Time Fourier transform (CTFT)

\[\begin{align} X(j\omega) &=\int_{-\infty}^{+\infty}x(t)e^{-j\omega t}dt \\ x(t)&= \frac{1}{2\pi}\int_{-\infty}^{+\infty}X(j\omega)e^{j\omega t}d\omega \end{align}\]

[https://www.rose-hulman.edu/class/ee/yoder/ece380/Handouts/Fourier%20Transform%20Tables%20w.pdf]

image-20240831104459715

Discrete-Time Fourier Transform (DTFT)

\[\begin{align} X(e^{j\hat{\omega}}) &=\sum_{n=-\infty}^{+\infty}x[n]e^{-j\hat{\omega} n} \\ x[n] &= \frac{1}{2\pi}\int_{2\pi}X(e^{j\hat{\omega}})e^{j\hat{\omega} n}d\hat{\omega} \end{align}\]

DTFT is defined for infinitely long signals as well as finite-length signal

DTFT is continuous in the frequency domain

We could verify that is the correct inverse DTFT relation by substituting the definition of the DTFT and rearranging terms


image-20240831152155093

Discrete-Time Fourier Series (DTFS)

TODO 📅

Discrete Fourier Series (DFS)

TODO 📅

Discrete Fourier Transform (DFT)

Two steps are needed to change the DTFT sum into a computable form:

  1. the continuous frequency variable \(\hat{\omega}\) must be sampled
  2. the limits on the DTFT sum must be finite

\[\begin{align} X[k] &= \sum_{n=0}^{N-1}x[n]e^{-j(2\pi/N)kn}\space\space\space k=0,1,...,N-1 \\ x[n] &= \frac{1}{N}\sum_{k=0}^{N-1}X[k]e^{j(2\pi/N)kn} \space\space\space n=0,1,...,N-1 \end{align}\]

Part of the proof is given by the following step:

image-20240830222204470

DFT \(X[k]\) is a sampled version of the DTFT \(X(e^{j\hat{\omega}})\), where \(\hat{\omega_k} = \frac{2\pi k}{N}\)

impulse train

CTFT:

image-20240830224755336

image-20240911221811991

using time-sampling property

impulse_train.drawio


DTFT:

Given \(x[n]=\sum_{k=-\infty}^{\infty}\delta(n-k)\)

\[\begin{align} X(e^{j\hat{\omega}}) &= X_s(j\frac{\hat{\omega}}{T}) \\ &= \frac{2\pi}{T}\sum_{k=-\infty}^{\infty}\delta(\frac{\hat{\omega}}{T}-\frac{2\pi k}{T}) \\ &= \frac{2\pi}{T}\sum_{k=-\infty}^{\infty}T\delta(\hat{\omega}-2\pi k) \\ &= 2\pi\sum_{k=-\infty}^{\infty}\delta(\hat{\omega}-2\pi k) \end{align}\]

[http://courses.ece.ubc.ca/359/notes/notes_part1_set4.pdf]


Fourier series of impulse train

image-20241106232432131

Dirac delta (impulse) function

image-20241013174738030

image-20241013174801954

[https://bingweb.binghamton.edu/~suzuki/Math-Physics/LN-7_Dirac_delta_function.pdf]

Topic 3 The \(\delta\)-function & convolution. Impulse response & Transfer function [https://www.robots.ox.ac.uk/~dwm/Courses/2TF_2011/2TF-N3.pdf]

image-20241122231208806


impulse scaling

\[ \delta(\alpha t)= \frac{1}{\alpha}\delta( t) \]

where \(\alpha\) is scaling ratio

Multiplication

aka Modulation or Windowing Theorem

CTFT: \[ x_1(t)x_2(t)\overset{FT}{\longrightarrow}\frac{1}{2\pi}X_1(\omega)*X_2(\omega) \]


DTFT:

image-20240909215833750

Duality

image-20240921181908992

image-20240921182105935

Conjugate Symmetry

image-20240921181015717

image-20240921181258063

Parseval's Relation

CTFT:

image-20240830230835764


DTFT:

image-20230516022936168


DFT:

image-20241214002405992

image-20241214002606672

Eigenfunctions & frequency response

Complex exponentials are eigenfunctions of LTI systems, that is,

continuous time: \(e^{j\omega t}\to H(j\omega)e^{j\omega t}\)

discrete time: \(e^{j\hat{\omega}n} \to H(e^{j\hat{\omega}})e^{j\hat{\omega}n}\)

where \(H(j\omega)\), \(H(e^{j\hat{\omega}})\) is frequency response of continuous-time systems and discrete-time systems, which is the function of \(\omega\) and \(\hat{\omega}\) \[\begin{align} H(j\omega) &= \int_{-\infty}^{+\infty}h(t)e^{-j\omega t}dt \\ \\ H(e^{j\hat{\omega}}) &= \sum_{n=-\infty}^{+\infty}h[n]e^{-j\hat{\omega} n} \end{align}\]

The frequency response of discrete-time LTI systems is always a periodic function of the frequency variable \(\hat{\omega}\) with period \(2\pi\)

Sampling Theorem

time-sampling theorem: applies to bandlimited signals

spectral sampling theorem: applies to timelimited signals

Aliasing

The frequencies \(f_{\text{sig}}\) and \(Nf_s \pm f_{\text{sig}}\) (\(N\) integer), are indistinguishable in the discrete time domain.

image-20220626000016184

Given below sequence \[ X[n] =A e^{j\omega T_s n} \]

  1. \(kf_s + \Delta f\)

\[\begin{align} x[n] &= Ae^{j\left( kf_s+\Delta f \right)2\pi T_sn} + Ae^{j\left( -kf_s-\Delta f \right)2\pi T_sn} \\ &= Ae^{j\Delta f\cdot 2\pi T_sn} + Ae^{-j\Delta f\cdot 2\pi T_sn} \end{align}\]

  1. \(kf_s - \Delta f\)

\[\begin{align} x[n] &= Ae^{j\left( kf_s-\Delta f \right)2\pi T_sn} + Ae^{j\left( -kf_s+\Delta f \right)2\pi T_sn} \\ &= Ae^{-j\Delta f\cdot 2\pi T_sn} + Ae^{j\Delta f\cdot 2\pi T_sn} \end{align}\]

complex signal

\[\begin{align} A e^{j(\omega_s + \Delta \omega) T_s n} &= A e^{j(k\omega_s + \Delta \omega) T_s n} \\ A e^{j(\omega_s - \Delta \omega) T_s n} &= A e^{j(k\omega_s - \Delta \omega) T_s n} \end{align}\]

sampling_aliasing.drawio

CTFS & CTFT

Fourier transform of a periodic signal with Fourier series coefficients \(\{a_k\}\) can be interpreted as a train of impulses occurring at the harmonically related frequencies and for which the area of the impulse at the \(k\)th harmonic frequency \(k\omega_0\) is \(2\pi\) times the \(k\)th Fourier series coefficient \(a_k\)

image-20240830225453601

inverse CTFT & inverse DTFT

time domain frequency domain
inverse CTFT \(\delta(t)\) \(\int_{\infty}d\omega\)
inverse DTFT \(\delta[n]\) \(\int_{2\pi}d\hat{\omega}\)

inverse CTFT shall integral from \(-\infty\) to \(+\infty\) to obtain \(\delta(t)\) in time domain, e.g., \(x_s(t)\) impulse train

spectral sampling

image-20240831185532202

spectral sampling by \(\omega_0\), and \(\frac{2\pi}{\omega_0} \gt \tau\) \[ X_{n\omega_0}(\omega) = \sum_{n=-\infty}^{\infty}X(n\omega_0)\delta(\omega - n\omega_0) \] Periodic repetition of \(x(t)\) is \[ x_{n\omega_0}(t) = \frac{1}{\omega_0}\sum_{n=-\infty}^{\infty}x(t -n\frac{2\pi}{\omega_0})=\frac{T_0}{2\pi}\sum_{n=-\infty}^{\infty}x(t -nT_0) \]

Then, if \(x_{T_0} (t)\), a periodic signal formed by repeating \(x(t)\) every \(T_0\) seconds (\(T_0 \gt \tau\)​), its CTFT is \[ X_{T_0}(\omega) = \frac{2\pi}{T_0} \cdot X_{n\omega_0}(\omega) = \frac{2\pi}{T_0}\sum_{n=-\infty}^{\infty}X(n\omega_0)\delta(\omega - n\omega_0) \] Then \(x_{T_0} (t)\) can be expressed with inverse CTFT as \[\begin{align} x_{T_0} (t) &= \frac{1}{2\pi}\int_{-\infty}^{\infty}X_{T_0}(\omega)e^{j\omega t}d\omega \\ &= \frac{1}{T_0}\sum_{n=-\infty}^{\infty}X(n\omega_0)e^{jn\omega_0 t} =\sum_{n=-\infty}^{\infty}\frac{1}{T_0}X(n\omega_0)e^{jn\omega_0 t} \end{align}\]

i.e. the coefficients of the Fourier series for \(x_{T_0} (t)\) is \(D_n =\frac{1}{T_0}X(n\omega_0)\)

image-20240831190258683

alternative method by direct Fourier series

image-20240831193912709

Why DFT ?

We can use DFT to compute DTFT samples and CTFT samples

image-20240831201335531

\[ \overline{x}(t) = \sum_{n=0}^{N_0-1}x(nT)\delta(t-nT) \] applying the Fourier transform yieds \[ \overline{X}(\omega) = \sum_{n=0}^{N_0-1}x[n]e^{-jn\omega T} \] But \(\overline{X}(\omega)\), the Fourier transform of \(\overline{x}(t)\) is \(X(\omega)/T\), assuming negligible aliasing. Hence, \[ X(\omega) = T\overline{X}(\omega) = T\sum_{n=0}^{N_0-1}x[n]e^{-jn\omega T} \] and \[ X(k\omega_0) = T\sum_{n=0}^{N_0-1}x[n]e^{-jn k\omega_0 T} \] with \(\hat{\omega}_0 = \omega_0 T\) \[ X(k\omega_0) = T\sum_{n=0}^{N_0-1}x[n]e^{-jn k\hat{\omega}_0} \] i.e. the relationship between CTFT and DFT is \(X(k\omega_0) = T\cdot X[k]\), DFT is a tool for computing the samples of CTFT

C/D

Sampling with a periodic impulse train, followed by conversion to a discrete-time sequence

image-20240901155629500

image-20240830231619897

The periodic impulse train is \[ s(t) = \sum_{n=-\infty}^{\infty}\delta(t-nT) \] \(x_s(t)\) can be expressed as \[ x_s(t) = \sum_{n=-\infty}^{\infty}x_c(nT)\delta(t-nT) \] i.e., the size (area) of the impulse at sample time \(nT\) is equal to the value of the continuous-time signal at that time.

\(x_s(t)\)​ is, in a sense, a continuous-time signal (specifically, an impulse train)

samples of \(x_c(t)\) are represented by finite numbers in \(x[n]\) rather than as the areas of impulses, as with \(x_s(t)\)

Frequency-Domain Representation of Sampling

The relationship between the Fourier transforms of the input and the output of the impulse train modulator \[ X_s(j\omega) = \frac{1}{T}\sum_{k=-\infty}^{\infty}X_c(j(\omega -k\omega_s)) \] where \(\omega_s\) is the sampling frequency in radians/s


\(X(e^{j\hat{\omega}})\), the discrete-time Fourier transform (DTFT) of the sequence \(x[n]\), in terms of \(X_s(j\omega)\) and \(X_c(j\omega)\)

continuous-time Fourier transform discrete-time Fourier transform
\(x_s(t) = \sum_{n=-\infty}^{\infty}x_c(nT)\delta(t-nT)\) \(x[n]=x_c(nT)\)
\(X_s(j\omega)=\sum_{n=-\infty}^{\infty}x_c(nT)e^{-j\omega Tn}\) \(X(e^{j\hat{\omega}})=\sum_{n=-\infty}^{\infty}x_c(nT)e^{-j\hat{\omega} n}\)

\[ X(e^{j\omega T}) = \frac{1}{T}\sum_{k=-\infty}^{\infty}X_c(j(\omega-k\omega_s)) \] or equivalently, \[ X(e^{j\hat{\omega}}) = \frac{1}{T}\sum_{k=-\infty}^{\infty}X_c(j(\frac{\hat{\omega}}{T}-\frac{2\pi k}{T})) \]

\(X(e^{j\hat{\omega}})\) is a frequency-scaled version of \(X_s(j\omega)\) with the frequency scaling specified by \(\hat{\omega} =\omega T\)

Ref. 9.5 DTFT connection with the CTFT

image-20240831154638540

Here, \(\Omega = \omega T\)

The factor \(\frac{1}{T}\) in \(X(e^{j\hat{\omega}})\) is misleading, actually \(x[n]\) is not scaled by \(\frac{1}{T}\) once taking \(\hat{\omega}\) variable of integration into account \[\begin{align} x_r[n] &= \frac{1}{2\pi} \int_{2\pi}X(e^{j\hat{\omega}})e^{j\hat{\omega} n}d\hat{\omega} \\ &= \frac{1}{2\pi}\int_{2\pi}\frac{1}{T}\sum_{k=-\infty}^{+\infty}X_c \left[ j\left(\frac{\hat{\omega}}{T} - \frac{2\pi k}{T}\right)\right] e^{j\hat{\omega} n}d\hat{\omega} \\ &\approx \frac{1}{2\pi}\frac{1}{T}\int_{2\pi}X_c (\frac{\hat{\omega}}{T} ) e^{j\hat{\omega} n} d\hat{\omega} \\ &=\frac{1}{2\pi} \frac{1}{T}\int_{2\pi} \left[ \int_{\infty}X_c(\Phi)\delta (\Phi - \frac{\hat{\omega}}{T} )d\Phi \right] e^{j\hat{\omega} n} d\hat{\omega} \\ &=\frac{1}{2\pi} \frac{1}{T} \int_{\infty}X_c(\Phi)d\Phi \int_{2\pi}\delta (\Phi - \frac{\hat{\omega}}{T} )e^{j\hat{\omega} n} d\hat{\omega} \\ &=\frac{1}{2\pi} \frac{1}{T} \int_{\infty}X_c(\Phi)d\Phi \int_{2\pi}T\cdot \delta (\Phi T - \hat{\omega} )e^{j\hat{\omega} n} d\hat{\omega} \\ &=\frac{1}{2\pi} \int_{\infty}X_c(\Phi) e^{j\Phi T n}d\Phi \end{align}\]

That is \[\begin{align} x_r[n] &= \frac{1}{2\pi}\int_{2\pi} \frac{1}{T}X_c (\frac{\hat{\omega}}{T} ) e^{j\hat{\omega} n} d\hat{\omega} \\ &= \frac{1}{2\pi} \int_{\infty}X_c(\omega) e^{j\omega T n}d\omega \tag{31} \end{align}\]

assuming Nyquist–Shannon sampling theorem is met

\[\begin{align} x_r[n] &= \frac{1}{2\pi} \int_{\infty}X_c(\omega) e^{j\omega T n}d\omega \\ &= \frac{1}{2\pi} \int_{\infty}X_c(\omega) e^{j\omega t_n}d\omega \\ &= x_c(t_n) \end{align}\]

where \(t_n = T n\), then \(x_r[n] = x_c(nT)\)


Assuming \(x_c(t) = \cos(\omega_0 t)\), \(x_s(t)= \sum_{n=-\infty}^{\infty}x_c(nT)\delta(t-nT)\) and \(x[n]=x_c(nT)\), that is \[\begin{align} x_c(t) & = \cos(\omega_0 t) \\ x_s(t) &= \sum_{n=-\infty}^{\infty}\cos(\omega_0 nT)\delta(t-nT) \\ x[n] &= \cos(\omega_0 nT) \end{align}\]

  • \(X_c(j\omega)\), the Fourier Transform of \(x_c(t)\) \[ X_c(j\omega) = \pi[\delta(\omega - \omega_0) + \delta(\omega + \omega_0)] \]

  • \(X(e^{j\hat{\omega}})\), the the discrete-time Fourier transform (DTFT) of the sequence \(x[n]\) \[ X(e^{j\hat{\omega}}) =\sum_{k=-\infty}^{+\infty}\pi[\delta(\hat{\omega} - \hat{\omega}_0-2\pi k) + \delta(\hat{\omega} + \hat{\omega}_0-2\pi k)] \]

  • \(X_s(j\omega)\), the Fourier Transform of \(x_s(t)\) \[ X_s(j\omega)= \frac{1}{T}\sum_{k=-\infty}^{+\infty}\pi[\delta(\omega - \omega_0-k\omega_s) + \delta(\omega + \omega_0-k\omega_s)] \]

Express \(X(e^{j\hat{\omega}})\) in terms of \(X_s(j\omega)\) and \(X_c(j\omega)\) \[ X(e^{j\hat{\omega}}) = \frac{1}{T}\sum_{k=-\infty}^{+\infty}\pi[\delta(\frac{\hat{\omega}}{T} - \omega_0-k\omega_s) + \delta(\frac{\hat{\omega}}{T} + \omega_0-k\omega_s)] \] Inverse \(X(e^{j\hat{\omega}})\) \[\begin{align} x_r[n] &= \frac{1}{2\pi} \int_{2\pi}X(e^{j\hat{\omega}}) e^{j\hat{\omega} n} d\hat{\omega} \\ &= \frac{1}{2\pi}\int_{2\pi} \pi[\delta(\frac{\hat{\omega}}{T} - \omega_0) + \delta(\frac{\hat{\omega}}{T} + \omega_0)]e^{j\hat{\omega} n} d\frac{\hat{\omega}}{T} \\ &= \frac{1}{2\pi}\int_{2\pi} \pi[\delta(\frac{\hat{\omega}}{T} - \omega_0)e^{j\hat{\omega}_0 n} + \delta(\frac{\hat{\omega}}{T} + \omega_0)e^{-j\hat{\omega}_0 n}] d\frac{\hat{\omega}}{T} \\ &= \frac{1}{2}[ e^{j\hat{\omega}_0 n}\int_{2\pi} [\delta(\frac{\hat{\omega}}{T} - \omega_0)d\frac{\hat{\omega}}{T} + e^{-j\hat{\omega}_0 n}\int_{2\pi} [\delta(\frac{\hat{\omega}}{T} + \omega_0)d\frac{\hat{\omega}}{T}] \\ &= \frac{1}{2}[ e^{j\hat{\omega}_0 n} + e^{-j\hat{\omega}_0 n} ] \\ &= \cos(\hat{\omega}_0 n) \end{align}\]

or follow EQ.(31)

\[\begin{align} x_r[n] &= \frac{1}{2\pi} \int_{\infty}X_c(\omega) e^{j\omega T n}d\omega \\ &= \frac{1}{2\pi} \int_{\infty} \pi[\delta(\omega - \omega_0) + \delta(\omega + \omega_0)]e^{j\omega T n}d\omega \\ &= \frac{1}{2}(e^{j\omega_0 T n}+e^{-j\omega_0 T n}) \\ &= \cos(\hat{\omega}_0 n) \end{align}\]

where \(\hat{\omega}_0 = \omega_0 T\)

impulse train sampling & impulse sequence

image-20250910204320327

image-20250910204428950

if \(x_c(t) = e^{j\Omega_0t}\), thus \(X_c (j\Omega) = A\delta(\Omega - \Omega_0)\)

Then \[ X_s (j\Omega) = \frac{A}{T_s}\sum_k \delta(\Omega -\Omega_0 - k\Omega_s) \]

DTFT of \(x[n]\) \[\begin{align} X(e^{j\omega}) &= \frac{1}{T_s} \sum_k X_c\left[j(\frac{\omega}{T_s}-\frac{2\pi k}{T_s})\right] \\ &= \frac{A}{T_s} \sum_k \delta(\frac{\omega}{T_s} -\Omega_0- \frac{2\pi k}{T_s}) \\ &= A \sum_k \delta(\omega -\omega_0 - 2\pi k) \end{align}\]

yield \[ x[n] = A e^{j\omega_0 n} = A e^{j\Omega_0 nT_s} \]

1
2
3
4
5
6
7
8
9
import numpy as np
x = np.linspace(0,1,10000)
y = np.cos(2*np.pi*1*x)
rms = np.sqrt(np.power(y, 2).sum()/x.size)
print(rms)
print(1/2**0.5)

# 0.7071421356417675
# 0.7071067811865475

Example 4.1 impulse scaling \(\delta(\omega/T)=T\delta(\omega)\)

\[ \int \delta(\frac{\omega}{T})d\omega = \int T \delta(\omega)d\omega = \int T\delta(\frac{\omega}{T})d\frac{\omega}{T} = T \]

D/C

image-20240831161852787

image-20240831162625943

image-20240831162559492

image-20241024220244992

zero padding

This option increases \(N_0\), the number of samples of \(x(t)\), by adding dummy samples of 0 value. This addition of dummy samples is known as zero padding

We should keep in mind that even if the fence were transparent, we would see a reality distorted by aliasing.

Zero padding only allows us to look at more samples of that imperfect reality

Balu Santhanam. ECE-539: Digital Signal Processing: Zero padding and Resolution [http://ece-research.unm.edu/bsanthan/ece539/zero_pad.pdf]

Gotcha

A remarkable fact of linear systems is that the complex exponentials are eigenfunctions of a linear system, as the system output to these inputs equals the input multiplied by a constant factor.

  • Both amplitude and phase may change
  • but the frequency does not change

For an input \(x(t)\), we can determine the output through the use of the convolution integral, so that with \(x(t) = e^{st}\) \[\begin{align} y(t) &= \int_{-\infty}^{+\infty}h(\tau)x(t-\tau)d\tau \\ &= \int_{-\infty}^{+\infty} h(\tau) e^{s(t-\tau)}d\tau \\ &= e^{st}\int_{-\infty}^{+\infty} h(\tau) e^{-s\tau}d\tau \\ &= e^{st}H(s) \end{align}\]

Take the input signal to be a complex exponential of the form \(x(t)=Ae^{j\phi}e^{j\omega t}\)

\[\begin{align} y(t) &= h(t)*x(t) \\ &= H(j\omega)Ae^{j\phi}e^{j\omega t} \end{align}\]

The frequency response at \(-\omega\) is the complex conjugate of the frequency response at \(+\omega\), given \(h(t)\) is real

\[\begin{align} H^*(t) &= \left(\int_{-\infty}^{+\infty}h(t)e^{-j\omega t}dt\right)^* \\ &= \int_{-\infty}^{+\infty}h^*(t)e^{+j\omega t}dt \\ &= \int_{-\infty}^{+\infty}h(t)e^{-j(-\omega t)}dt \\ &= H(-j\omega) \end{align}\]

The real cosine signal is actually composed of two complex exponential signals: one with positive frequency and the other with negative \[ cos(\omega t + \phi) = \frac{e^{j(\omega t + \phi)} + e^{-j(\omega t + \phi)}}{2} \]

The sinusoidal response is the sum of the complex-exponential response at the positive frequency \(\omega\) and the response at the corresponding negative frequency \(-\omega\) because of LTI systems's superposition property

  • input: \[\begin{align} x(t) &= A cos(\omega t + \phi) \\ &= \frac{1}{2}Ae^{\phi}e^{\omega t} + \frac{1}{2}Ae^{-\phi}e^{-\omega t} \end{align}\]

  • output with \(H(j\omega)=Ge^{j\theta}\): \[\begin{align} y(t) &= H(j\omega)\frac{1}{2}Ae^{\phi}e^{\omega t} + H(-j\omega)\frac{1}{2}Ae^{-\phi}e^{-\omega t} \\ &= Ge^{j\theta}\frac{1}{2}Ae^{\phi}e^{\omega t} + Ge^{-j\theta}\frac{1}{2}Ae^{-\phi}e^{-\omega t} \\ &= GAcos(\omega t + \phi + \theta) \end{align}\]

Its phase shift is \(\theta\) and gain is \(G\), which is same with \(H(j\omega)\).

reference

Alan V Oppenheim, Ronald W. Schafer. Discrete-Time Signal Processing, 3rd edition [pdf]

B.P. Lathi, Roger Green. Linear Systems and Signals (The Oxford Series in Electrical and Computer Engineering) 3rd Edition [pdf]

Alan V. Oppenheim, Alan S. Willsky, and S. Hamid Nawab. 1996. Signals & systems (2nd ed.) [pdf]

James H. McClellan, Ronald Schafer, and Mark Yoder. 2015. DSP First (2nd. ed.). Prentice Hall Press, USA

Reference Ripple

C-H Chan (U. of Macau) "Extreme SAR ADCs - Exploring New Frontiers" Online Course (2024) : Reference Buffer in SAR ADC [https://youtu.be/vj98B7AaC9E?si=hMt0PM07CdkHN5Qn]

C. Li, C. -H. Chan, Y. Zhu and R. P. Martins, "Analysis of Reference Error in High-Speed SAR ADCs With Capacitive DAC," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 66, no. 1, pp. 82-93, Jan. 2019 [https://ime.um.edu.mo/wp-content/uploads/magazines/961546494e705f6fd16b9f785a121030.pdf]

J. Zhong, Y. Zhu, S. -W. Sin, S. -P. U and R. P. Martins, "Thermal and Reference Noise Analysis of Time-Interleaving SAR and Partial-Interleaving Pipelined-SAR ADCs," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 62, no. 9, pp. 2196-2206, Sept. 2015 [https://sci-hub.st/10.1109/TCSI.2015.2452331]

C. -H. Chan et al., "60-dB SNDR 100-MS/s SAR ADCs With Threshold Reconfigurable Reference Error Calibration," in IEEE Journal of Solid-State Circuits, vol. 52, no. 10, pp. 2576-2588, Oct. 2017 [https://ime.um.edu.mo/wp-content/uploads/magazines/407e580ac0218605bcf9b9bbd0ea1109.pdf]

TODO 📅

sample-by-sample

3rd harmonic

sample2sample-gain-distortion.drawio

bit-by-bit

The amplitude of the reference ripple is code-dependent as it is correlated with switching energy in each bit cycling

quantization error & quantization noise

image-20250910210909363

image-20250910211207655

image-20250910211034914


Notice \(e_q\in (0, \Delta)\) and its average is \(\Delta/2\). To calculate SNDR, DC component shall be excluded

Don't confuse resolution \(\Delta\) with Bounded Quantization Noise \(-\Delta/2 \sim \Delta/2\)

image-20250909233010702

Redundancy

decision level

final digital output for an \(N\)-bit \(M\)-step ADC can be calculated \[ D_{out} = s(M) + \sum_{i=1}^{M-1}(2\cdot b[i] - 1)\times s(i) + (b[0] -1)\cdot s(1) \]

i M M-1 M-2 ... 2 1 0
b[i] b[M-1] b[M-2] ... b[2] b[1] b[0]
s[i] s(M) s(M-1) s(M-2) ... s(2) s(1)

image-20250909211030234

track the decision level

For \(N\)-bit binary weighted algorithm,\(N=M\) and \(s(i)=2^{i-1}\), where \(i\in \{N, N-1,...,2,1 \}\)

\[\begin{align} D_{out} &= s(M) + \sum_{i=1}^{M-1}(2\cdot b[i] - 1)\times s(i) + (b[0] -1) \\ &= 2^{N-1} + \sum_{i=1}^{N-1}2^i\cdot b[i] - \sum_{i=0}^{N-2}2^{i} + (b[0] -1) \\ &= \sum_{i=0}^{N-1} b[i] \cdot 2^i \end{align}\]

Error Tolerance Window

\[ \varepsilon_t(n) = \sum_{i=1}^{n-2} s(i) - s(n-1) \]

where \(n\in [1, N]\), and \(N\)-bit SAR

etw.drawio

For the \(n\)th output bit, once a decision is made, the next decision level will either move up or down by the step size of \(s(n − 1)\)

If this decision is erroneous, then the sum of the follow-on step sizes, \(s(n − 2)\), \(s(n − 3)\), ..., \(s(1)\), must be large enough and exceed the value of the current step size to counteract this mistake

The exceeded amount is the tolerance window for that decision level

image-20250909222730303

image-20250909222310476

image-20250909231804142

image-20250909222622340

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
import numpy as np
import matplotlib.pyplot as plt


def sar(xi, ss):
M = ss.size
th = ss[0]
oob = []
for i in range(M):
ocur = 1 if xi >= th else 0
oob.append(ocur)
if i + 1 < M:
th += (2 * ocur - 1) * ss[i + 1]
else:
break

binstr = ''.join([str(s) for s in oob])
decval = int(binstr, 2)
return binstr, decval


def sar_plot(ss, Npts=10000):
ss = np.asarray(ss)
ssum = np.sum(ss)
xilist = np.linspace(0, ssum + 1, Npts)
outlist = []
for i in range(Npts):
_, decval = sar(xilist[i], ss)
outlist.append(decval)
outmax = np.max(outlist)
plt.figure(figsize=(16,8))
plt.plot(xilist, outlist, '-', linewidth=4)
plt.xticks(range(0, ssum + 2))
plt.yticks(range(0, outmax + 2, 2))
plt.title('search step: {}'.format(ss), fontsize=20)
plt.xlabel('analog out', fontsize=20); plt.ylabel('digital out', fontsize=20)

plt.grid(True)

ss = [8, 4, 2, 1]
sar_plot(ss)

ss = [8, 2, 2, 2, 1]
sar_plot(ss)

plt.show()

ENOB vs. fixed radix

When the ADC is designed with a fixed radix, \(\alpha\) and the required number of conversion steps, \(M\)

the sum of all the step sizes \(s_{tot}\) \[ s_{tot} = \sum_{k=0}^{M-1} s_0 \alpha^k = s_0\frac{\alpha^M-1}{\alpha-1} \]

where \(s(i)\) is step size and \(i \in [0, 1, 2, M-1]\)

The effective number of bits, \(N\), can be calculated \[ N \leq \log 2\left(\frac{s_{tot} + s_0}{s_0}\right) = \frac{\alpha^M+\alpha-2}{\alpha-1} \]

Speed Benefit

TODO 📅

CDAC

The charge redistribution capacitor network is used to sample the input signal and serves as a digital-to-analog converter (DAC) for creating and subtracting reference voltages

sampling charge \[ Q = V_{in} C_{tot} \] conversion charge \[ Q = -C_{tot}V_c + V_{ref}C_\Delta \] That is \[ V_c = \frac{C_\Delta}{C_{tot}}V_{ref} - V_{in} \]


CDAC is actually working as a capacitive divider during conversion phase, the charge of internal node retain (charge conservation law)

assuming \(\Delta V_i\) is applied to series capacitor \(C_1\) and \(C_2\)

cap_divider.drawio \[ (\Delta V_i - \Delta V_x) C_1 = \Delta V_x \cdot C_2 \] Then \[ \Delta V_x = \frac{C_1}{C_1+C_2}\Delta V_i \]

\(V_x= V_{x,0} + \Delta V_x\)

CDAC settling accuracy

cdac-tau.drawio \[\begin{align} V_x(s) &= \frac{C_1+C_2}{RC_1C_2}\cdot \frac{1}{s+\frac{C_1+C_2}{RC_1C_2}}\cdot V_i(s) \\ &= \frac{1}{\tau}\cdot \frac{1}{s+\frac{1}{\tau}}\cdot \frac{1}{s}\\ &= \frac{1}{\tau}\cdot \tau(\frac{1}{s} - \frac{1}{s+\frac{1}{\tau}})=\frac{1}{s} - \frac{1}{s+\frac{1}{\tau}} \end{align}\]

inverse Laplace Transform is \(V_x(t) = 1 - e^{-t/\tau}\)

\[\begin{align} V_y(s) &= V_x\frac{C_1}{C_1+C_2} \\ &= \frac{C_1}{C_1+C_2} \left(\frac{1}{s} - \frac{1}{s+\frac{1}{\tau}}\right)\\ \end{align}\]

inverse Laplace Transform is \(V_y(t) = \frac{C_1}{C_1+C_2}\left(1 - e^{-t/\tau}\right)\)

\(V_x(t)\) and \(V_y(t)\) prove that the settling time is same

\(\tau = R\frac{C_1C_2}{C_1+C_2}\), which means usually worst for MSB capacitor (largest)

both \(\tau\) and \(\Delta V\) are the maximum

A popular way to improve the settling behavior, again, is to employ unit-element DACs that statistically reduce the switching activities, which, unfortunately, exhibits unnecessary complications to the power, area and speed tradeoffs of the design

CDAC Energy Consumption

\[ E_{Vref} = \int P(t)dt = \int V_{ref} I(t) dt = V_{ref}\int I(t)dt = V_{ref}\cdot \Delta Q \]

image-20240922093524720

Given \(V_{c,0}=\frac{1}{2}V_{ref}-V_{in}\) and \(V_{c,1}=\frac{3}{4}V_{ref}-V_{in}\) \[\begin{align} Q_{b0,0} &= \left(V_{ref} - V_{c,0} \right)\cdot 2C = \left(\frac{1}{2}V_{ref}+V_{in} \right)\cdot 2C \\ Q_{b1,0} &= (0 - V_{c,0})\cdot C = \left(-\frac{1}{2}V_{ref}+V_{in} \right)\cdot C \\ Q_{b0,1} &= \left(V_{ref} - V_{c,1} \right)\cdot 2C = \left(\frac{1}{4}V_{ref}+V_{in} \right)\cdot 2C \\ Q_{b1,1} &= \left(V_{ref} - V_{c,1} \right)\cdot C = \left(\frac{1}{4}V_{ref}+V_{in} \right)\cdot C \end{align}\]

Therefore \[ E_{Vref} = V_{ref}\cdot (Q_{b0,1}+Q_{b1,1} - Q_{b0,0}-Q_{b1,0}) = \frac{1}{4}C V_{ref}^2 \]


CDAC total energy change \[\begin{align} \Delta E_{tot} &= \frac{1}{2}\cdot 2C \cdot (U_{2c,1}^2 - U_{2c,0}^2) + \frac{1}{2}\cdot C \cdot (U_{c,1}^2 - U_{c,0}^2) + \frac{1}{2}\cdot C \cdot (U_{c1,1}^2 - U_{c1,0}^2) \\ &= \left(-\frac{3}{16}V_{ref}^2 - \frac{1}{2}V_{ref}V_{in} - \frac{3}{32}V_{ref}^2+\frac{3}{4}V_{ref}V_{vin} + \frac{5}{32}V_{ref}^2-\frac{1}{4}V_{ref}V_{in}\right)C \\ &= -\frac{1}{8}CV_{ref}^2 \end{align}\]

alternative method

CapEnergy.drawio \[ \Delta E_{tot} = \frac{1}{2}\cdot\frac{3}{4}C\cdot V_{ref}^2 - \frac{1}{2}\cdot C\cdot V_{ref}^2 = -\frac{1}{8}CV_{ref}^2 \]

The total energy decreases by \(-\frac{1}{8}CV_{ref}^2\), though \(V_{ref}\) provides \(\frac{1}{4}C V_{ref}^2\)


The charge redistribution change the CDAC energy

cap_redis_energy.drawio

\[ E_{c,0} = \frac{1}{2}CV^2 \] After charge redistribution \[ E_{c,1} = \frac{1}{2}\cdot 2C\cdot \left(\frac{1}{2}V\right)^2 = \frac{1}{4}CV^2 \]

That make sense, charge redistribution consume energy

Comparator

Comparator input cap effect

image-20240907194621524 \[ -V_{in}\cdot 2^N C = V_c (2^N C + C_p) \] Then \(V_c = -\frac{2^N C}{2^N C + C_p}V_{in}\), i.e. this capacitance reduce the voltage amplitude by the factor

During conversion \[\begin{align} V_c &= -\frac{2^N C}{2^N C + C_p}V_{in} +V_{ref}\sum_{n=0}^{N-1} \frac{b_n\cdot2^n C}{2^N C + C_p} \\ &= \frac{2^N C}{2^N C + C_p}\left(-V_{in} + V_{ref}\sum_{n=0}^{N-1}\frac{b_n }{2^{N-n}} \right) \end{align}\]

That is, it does not change the sign

Comparator offset effect

image-20240825204030645

Synchronous SAR ADC

It also divides a full conversion into several comparison stages in a way similar to the pipeline ADC, except the algorithm is executed sequentially rather than in parallel as in the pipeline case.

However, the sequential operation of the SA algorithm has traditionally been a limitation in achieving high-speed operation

image-20241021214958488

  • a clock running at least \((N + 1) \cdot F_s\) is required for an \(N\)-bit converter with conversion rate of \(F_s\)
  • every clock cycle has to tolerate the worst case comparison time
  • every clock cycle requires margin for the clock jitter

The power and speed limitations of a synchronous SA design comes largely from the high-speed internal clock

Split Arrary CDAC

Split capacitor, double-array cap

attenuation capacitance \(C_a\)

image-20240917192957721

image-20240918213856504

splitArray.drawio

\[\begin{align} \Delta V_{dac} &= \frac{1}{2}b_3+\frac{1}{4}b_2+\frac{1}{4}\left(\frac{1}{2}b_1+\frac{1}{4}b_0 \right) \\ &= \frac{1}{2}b_3+\frac{1}{4}b_2 + \frac{1}{8}b_1+\frac{1}{16}b_0 \end{align}\]

Asynchronous SAR ADC

The comparator itself trigger the next bit-conversion cycle as soon as the present bit decision has been taken

image-20241021214922564

image-20250102225355547

The maximum resolving time reduction between synchronous and asynchronous case is two fold

comparator metastable state

when the input is sufficiently small. The time needed for the comparator outputs to fully resolve may take arbitrarily long

In this case, the ready signal generator should still set the flag and the decision result is simply taken from the previous value stored in the SR latch

image-20250701231051158

both outputs (\(Q_p\) and \(Q_n\)) will drop together, NAND is inverter actually

The transition point of this NAND gate is skewed to eliminate metastability issues arising when the input differential voltage level is small (comparator)

reference

Andrea Baschirotto, "T6: SAR ADCs" ISSCC2009

Pieter Harpe, ISSCC 2016 Tutorial: "Basics of SAR ADCs Circuits & Architectures"


Mike Shuo-Wei Chen and R. W. Brodersen, "A 6-bit 600-MS/s 5.3-mW Asynchronous ADC in 0.13-μm CMOS," in IEEE Journal of Solid-State Circuits, vol. 41, no. 12, pp. 2669-2680, Dec. 2006 [pdf, slides]

—. "Power Efficient System and A/D Converter Design for Ultra-Wideband Radio" [http://www2.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-71.pdf]

—. "Asynchronous SAR ADC: Past, Present and Beyond" [https://viterbi-web.usc.edu/~swchen/index_files/async_sar_tutorial_chen_final.pdf]

C. -C. Liu, S. -J. Chang, G. -Y. Huang and Y. -Z. Lin, "A 10-bit 50-MS/s SAR ADC With a Monotonic Capacitor Switching Procedure," in IEEE Journal of Solid-State Circuits, vol. 45, no. 4, pp. 731-740, April 2010 [https://sci-hub.se/10.1109/JSSC.2010.2042254]

L. Jie et al., "An Overview of Noise-Shaping SAR ADC: From Fundamentals to the Frontier," in IEEE Open Journal of the Solid-State Circuits Society, vol. 1, pp. 149-161, 2021 [pdf]

W. Liu, P. Huang and Y. Chiu, "A 12-bit, 45-MS/s, 3-mW Redundant Successive-Approximation-Register Analog-to-Digital Converter With Digital Calibration," in IEEE Journal of Solid-State Circuits, vol. 46, no. 11, pp. 2661-2672, Nov. 2011 [https://sci-hub.st/10.1109/JSSC.2011.2163556]

0%