Elad Alon, ISSCC 2014, "T6: Analog Front-End Design for Gb/s Wireline
Receivers"
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.
we can get \(V_{i,n}^2 =
\frac{V_{i,sig}^2}{\text{SNR}}\), which is constant also
That is \[
V_{i,n}^2 = \frac{V_{i,sig}^2}{V_{o,sig}^2}V_{o,n}^2 =
\frac{V_{o,n}^2}{A_v^2}
\] where \(V_{i,sig}\) is
constant signal is applied to input of comparator
offset
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.
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]
Rabuske, Taimur & Fernandes, Jorge. (2017), Chapter 5 Noise-Aware
Synthesis and Optimization of Voltage Comparators, "Charge-Sharing SAR
ADCs for Low-Voltage Low-Power Applications"
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
When \(h[n]\) and \(h_c(t)\) are related through the above
equation, i.e., the impulse response of the discrete-time system is a
scaled, sampled version of \(h_c(t)\), the discrete-time
system is said to be an impulse-invariant
version of the continuous-time system
we have \[
H(e^{j\hat{\omega}}) = H_c\left(j\frac{\hat{\omega}}{T}\right)
\]
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.
Given below sequence \[
X[n] =A e^{j\omega T_s n}
\]
\[\begin{align}
A e^{j(\omega_s + \Delta \omega) T_s n} &= A e^{j(k\omega_s + \Delta
\omega) T_s n} \\
A e^{j(\omega_s - \Delta \omega) T_s n} &= A e^{j(k\omega_s - \Delta
\omega) T_s n}
\end{align}\]
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 \(a_k\)
spectral sampling
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 \(D_n =\frac{1}{T_0}X(n\omega_0)\)
alternative method by direct Fourier series
Why DFT ?
We can use DFT to compute DTFT samples and CTFT samples
\[
\overline{x}(t) = \sum_{n=0}^{N_0-1}x(nT)\delta(t-nT)
\] 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
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 \(X_c(j\omega)\)
\[\begin{align}
x_r[n] &= \frac{1}{2\pi} \int_{\infty}X_c(\omega) e^{j\omega T
n}d\omega \\
&= \frac{1}{2\pi} \int_{\infty} \pi[\delta(\omega - \omega_0) +
\delta(\omega + \omega_0)]e^{j\omega T n}d\omega \\
&= \frac{1}{2}(e^{j\omega_0 T n}+e^{-j\omega_0 T n}) \\
&= \cos(\hat{\omega}_0 n)
\end{align}\]
where \(\hat{\omega}_0 = \omega_0
T\)
D/C
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
Transfer function
sampled impulse response
The below equation demonstrates how to obtain continuous
Fourier Transform from DTFT . \[
X_c(\omega) = T \cdot X(\omega)
\]
\(T\) is sample period, follow
previous equation
useful functions
using fft
The outputs of the DFT are samples of the
DTFT
using freqz
modeling as FIR filter, and the impulse response
sequence of an FIR filter is the same as the sequence of filter
coefficients, we can express the frequency response in terms of either
the filter coefficients or the impulse response
fft is used in freqz internally
Example
Question:
How to obtain continuous system transfer function from sampled
impulse
Answer:
using above mentioned functions
First order lowpass filter with 3-dB frequency
1Hz
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 \(x(t)=Ae^{j\phi}e^{j\omega t}\)
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: \[\begin{align}
x(t) &= A cos(\omega t + \phi) \\
&= \frac{1}{2}Ae^{\phi}e^{\omega t} +
\frac{1}{2}Ae^{-\phi}e^{-\omega t}
\end{align}\]
On your intuition, input can be expressed as below \[\begin{align}
V_{N-1} &= \sum_{i=0}^{N-1} B[i] \frac{1}{2}\cdot 2^{i-N+1} \\
&= B[N-1]\frac{1}{2} + B[N-2]\frac{1}{2}\cdot 2^{-1} +
B[N-3]\frac{1}{2}\cdot 2^{-2 }...
\end{align}\]
It divides the process into several comparison stages, the number of
which is proportional to the number of bits
Due to the pipeline structure of both analog and digital signal path,
inter-stage residue amplification is needed which
consumes considerable power and limits high speed operation
reduced residue
amplification gain
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
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
Non-Binary Successive
Approximation
The overlapped search range compensates for wrong decisions
made in earlier stages as long as they are within the error tolerance
range
Kuttner, Franz. “A 1.2V 10b 20MSample/s non-binary successive
approximation ADC in 0.13/spl mu/m CMOS.” 2002 IEEE International
Solid-State Circuits Conference. Digest of Technical Papers (Cat.
No.02CH37315) 1 (2002): 176-177 vol.1.
a global clock running at the sample rate is still used for an
uniform sampling
The concept of asynchronous processing is to trigger the internal
comparison from MSB to LSB like dominoes.
reference
S. -W. M. 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
Single-Pole Filter and Complex Conjugate Pole pair in Event-Driven
PWL model
Real number modeling of analog circuits in hardware
description languages (HDLs) has become more common as a part of
mixed-signal SoC validation. Piecewise linear (PWL)
waveform approximation represent analog signals and dynamically
schedule the events for approximating the signal waveform to
PWL segments with a well controlled error bound.
Definition of a piecewise liner (PWL) waveform using struct in
Systemverilog
1 2 3 4 5
typedefstruct { real y; // signal offset real slope; // signal slope real t0; // time offset } pwl; // pwl datatype
When to update piecewise
model
model parameter update once new input come in
error is greater than user-define tolerance \(e_{tol}\), trigger by \(\Delta T\)
Dynamic Time Step Control
When approximating a function \(y(t)\) to a piecewise linear
segment for the interval \(t_0 \le
t_0 + \Delta t\), the approximation error \(err\) is bounded by \[
\left| err \right| \le \frac{1}{8}\cdot \Delta t^2 \cdot \max(\left|
\ddot{y(t)} \right|)
\] Using Rolle's theorem for the interval \(t_0 \le t_0 + \Delta t\), the needed time
step \(\Delta t\) is givend by \[
\Delta t(t=t_0) = \sqrt{\frac{8\cdot e_{tol}}{\max(\left| \ddot{y(t)}
\right|)}}
\]
Single-Pole Filter Model
The ramp input \(X(s)\), the single pole system Laplace
s-domain \(H(s)\) and the output
response \(Y(s)\), \[\begin{align}
X(s) &= \frac{a}{s} +\frac{b}{s^2} \\
H(s) &= \frac{Y(s)}{X(s)} = \frac{1}{1+\frac{s}{\omega_{1}}} \\
Y(s) &= X(s) \cdot H(s)
\end{align}\]
Time domain of ramp input shown as below \[
x(t) = a +b \cdot t
\]
The output transfer function \[\begin{align}
Y(s) &= X(s) \cdot H(s) \\
&= \frac{\omega_1}{\omega_1+s}\cdot X
\end{align}\]
step-1 transfer function in Laplace s-domain, which
don't initial conditon and is only steady response.
step-2 differential equation
step-3 Laplace transform of \(Y(s)\), (the initial conditon of input
\(X(s)\) is zero, that of \(Y(s)\) is explicit)
step-4 inverse Laplace transform, with the help of Laplace transform
table or matlab syms and ilaplace function
\(y(t)\) has a continuous second
derivative \(\ddot{y(t)}\)\[
\ddot{y(t)} =(-a+\frac{b}{\omega_1}+y_0)\cdot \omega_1^2\cdot
e^{-\omega_1t}
\] It's obvious \(\left| \ddot{y(t)}
\right|\) is a decaying function and thus the maximum value is
\(\left| \ddot{y(t_0)} \right|\) for
the interval \(t_0 \le t_0 + \Delta
t\). The time step \(\Delta t\)
for the error tolerance \(e_{tol}\):
\[
\Delta t(t=t_0) = \sqrt{\frac{8\cdot e_{tol}}{\left| \ddot{y(t_0)}
\right|}}
\]
where \(\omega_p\) and \(r\) are complex numbers, \(r=r_r+jr_i\), \(\omega_p=\omega_{pr}+j\omega_{pi}\)
Follow the procedure as above single pole \[
\frac{Y(s)}{X(s)} = \frac{s\cdot r_{cs}+e}{s^2+s\cdot \omega_{p\_cs}+f}
\] where \(r_{cs}=r+r^*\), \(\omega_{p\_cs}=\omega_p+\omega_p^*\) and
\(e=r\omega_p^*+r^*\omega_p\), \(f=\omega_p\omega_p^*\) implies \[
s^2Y(s)+s\omega_{p\_cs}Y(s)+fY(s)=(s\cdot r_{cs}+e)X(s)
\] or a differential equation \[
\frac{d^2y(t)}{dt^2}+\omega_{p\_cs}\frac{dy(t)}{dt}+fy(t)=r_{cs}\frac{dx(t)}{dt}+e\cdot
x(t)
\] Taking Laplace transform with initial conditions \(y_0\), \(\dot{y_0}\) and \(x_0=0\), \[
s^2-sy_0-\dot{y_0}+\omega_{p\_cs}(sY(s)-y_0)+f\cdot y(t) = r_{cs}\cdot
(sX(s)-0)+e\cdot X(s)
\] Solving for \(Y(s)\)\[
Y(s)=\frac{s\cdot
y_0+\dot{y_0}+\omega_{p\_cs}y_0}{s^2+s\cdot{\omega_{p\_cs}}+f}+\frac{s\cdot{r_{cs}}+e}{s^2+s\cdot{\omega_{p\_cs}}+f}X(s)
\] With an ramp input, height \(a\), slope \(b\), i.e. \(X(s)=\frac{a}{s}+\frac{b}{s^2}\)\[
Y(s)=\frac{s\cdot
y_0+\dot{y_0}+\omega_{p\_cs}y_0}{s^2+s\cdot{\omega_{p\_cs}}+f}+\frac{s\cdot{r_{cs}}+e}{s^2+s\cdot{\omega_{p\_cs}}+f}(\frac{a}{s}+\frac{b}{s^2})
\] After inverse Laplace transform, we can get total response
\[
y(t)=e^{-\omega_{pr}t}\cdot \left[ y_0\cdot
\cos(\omega_{pi}t)+\frac{\dot{y_0}+y_0\omega_{pr}}{\omega_{pi}}\sin(\omega_{pi}t)+D\cdot
\cos(\omega_{pi}t)+\frac{C-D\cdot{\omega_{pr}}}{\omega_{pi}}\sin(\omega_{pi}t)
\right]+B+A\cdot{t}
\] where \[\begin{align}
A &= \frac{e\cdot{b}}{f} \\
B &= \frac{r_{cs}\cdot{b}+a\cdot{e}-A\cdot{\omega_{p\_{cs}}}}{f} \\
C &= a\cdot{r_{cs}}-A-B\cdot{\omega_{p\_cs}} \\
D &= -B
\end{align}\]
As a double check, note that at \(t=0\), \[
y(0)=\left[ y_0 + D \right]+B=y_0
\]
To derive derivative, we first assume \[
y_0\cdot
\cos(\omega_{pi}t)+\frac{\dot{y_0}+y_0\omega_{pr}}{\omega_{pi}}\sin(\omega_{pi}t)+D\cdot
\cos(\omega_{pi}t)+\frac{C-D\cdot{\omega_{pr}}}{\omega_{pi}}\sin(\omega_{pi}t)
= \alpha \cdot{\cos(\omega_{pi}t+\phi)}
\] The above equation implies \[\begin{align}
y_0+D &= \alpha\cdot{\cos(\phi)} \\
\frac{\dot{y_0}+y_0\omega_{pr}}{\omega_{pi}}+\frac{C-D\cdot{\omega_{pr}}}{\omega_{pi}}
&= -\alpha\cdot{\sin(\phi)}
\end{align}\] Then \[
\alpha^2=(y_0+D)^2+\left(\frac{\dot{y_0}+y_0\omega_{pr}}{\omega_{pi}}+\frac{C-D\cdot{\omega_{pr}}}{\omega_{pi}}
\right)^2
\] And \(\alpha\) can be used to
estimate time step size. The total response is \[
y(t)=e^{-\omega_{pr}t}\cdot \alpha
\cdot{\cos(\omega_{pi}t+\phi)}+B+A\cdot{t}
\] It's second derivative is \[
\ddot{y(t)} = \alpha\left[
(\omega_{pr}^2-\omega_{pi}^2)e^{-\omega_{pr}t}\cos(\omega_{pi}t+\phi)+2\cdot
\omega_{pr}\omega_{pi}e^{-\omega_{pr}t}\sin(\omega_{pi}t+\phi) \right]
\] Absolute value \[
\left| \ddot{y(t)} \right| = \left| \alpha \right| \left|
(\omega_{pr}^2-\omega_{pi}^2)e^{-\omega_{pr}t}\cos(\omega_{pi}t+\phi)+2\cdot
\omega_{pr}\omega_{pi}e^{-\omega_{pr}t}\sin(\omega_{pi}t+\phi) \right|
\] Define new function \(g_0(t)\)\[
g_0(t) = \left| \alpha \right| \left|
(\omega_{pr}^2-\omega_{pi}^2)e^{-\omega_{pr}t}\cos(\omega_{pi}t+\phi)
\right|+2\cdot |\alpha| \left|
\omega_{pr}\omega_{pi}e^{-\omega_{pr}t}\sin(\omega_{pi}t+\phi) \right|
\] another new funtion \(g_1(t)\), by equating \(\sin(\omega_{pi}t+\phi)\) and \(\cos(\omega_{pi}t+\phi)\) to one \[
g_1(t) = \left| \alpha \right| \left|
(\omega_{pr}^2-\omega_{pi}^2)e^{-\omega_{pr}t} \right|+2\cdot |\alpha|
\left| \omega_{pr}\omega_{pi}e^{-\omega_{pr}t} \right|
\]
By triangular inequality, \(g_0(t)\)
is the upper bound of \(\left| \ddot{y(t)}
\right|\), and \(g_1(t)\) is the
upper bound of \(g_0(t)\)
Because \(g_1(t)\) is a decaying
exponential function, Therefore, a conservative time step can be
obtained, for inteval \(t_0 \le t_0 + \Delta
t\), \[
\Delta t(t=t_0) = \sqrt{\frac{8\cdot e_{tol}}{\left| g_1(t_0) \right|}}
\]
My colleague, Zhang Wenpian help me a lot in understanding this
modeling method. Lots of content here are copied from Zhang's note.
Reference
B. C. Lim and M. Horowitz, "Error Control and Limit Cycle Elimination
in Event-Driven Piecewise Linear Analog Functional Models," in IEEE
Transactions on Circuits and Systems I: Regular Papers, vol. 63, no. 1,
pp. 23-33, Jan. 2016, doi: 10.1109/TCSI.2015.2512699.
S. Liao and M. Horowitz, "A Verilog piecewise-linear analog behavior
model for mixed-signal validation," Proceedings of the IEEE 2013 Custom
Integrated Circuits Conference, 2013, pp. 1-5, doi:
10.1109/CICC.2013.6658461.
DaVE - tools regarding on analog modeling,validation, and generation,
https://github.com/StanfordVLSI/DaVE](https://github.com/StanfordVLSI/DaVE)
B. C. Lim, J. -E. Jang, J. Mao, J. Kim and M. Horowitz, "Digital
Analog Design: Enabling Mixed-Signal System Validation," in IEEE
Design & Test, vol. 32, no. 1, pp. 44-52, Feb. 2015 [http://iot.stanford.edu/pubs/lim-mixed-design15.pdf]
P. Schvan et al., "A 24GS/s 6b ADC in 90nm CMOS," 2008 IEEE
International Solid-State Circuits Conference - Digest of Technical
Papers, San Francisco, CA, USA, 2008, pp. 544-634
B. Sedighi, A. T. Huynh and E. Skafidas, "A CMOS track-and-hold
circuit with beyond 30 GHz input bandwidth," 2012 19th IEEE
International Conference on Electronics, Circuits, and Systems (ICECS
2012), Seville, Spain, 2012, pp. 113-116
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
That make sense, charge redistribution consume energy
Comparator input capacitance
\[
-V_{in}\cdot 2^N C = V_c (2^N C + C_p)
\] 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 \[\begin{align}
V_c &= -\frac{2^N C}{2^N C + C_p}V_{in} +V_{ref}\sum_{n=0}^{N-1}
\frac{b_n\cdot2^n C}{2^N C + C_p} \\
&= \frac{2^N C}{2^N C + C_p}\left(-V_{in} +
V_{ref}\sum_{n=0}^{N-1}\frac{b_n }{2^{N-n}} \right)
\end{align}\]
K. Tyagi and B. Razavi, "Performance Bounds of ADC-Based Receivers
Due to Clock Jitter," in IEEE Transactions on Circuits and Systems
II: Express Briefs, vol. 70, no. 5, pp. 1749-1753, May 2023 [https://www.seas.ucla.edu/brweb/papers/Journals/KT_TCAS_2023.pdf]
N. Da Dalt, M. Harteneck, C. Sandner and A. Wiesbauer, "On the jitter
requirements of the sampling clock for analog-to-digital converters," in
IEEE Transactions on Circuits and Systems I: Fundamental Theory and
Applications, vol. 49, no. 9, pp. 1354-1360, Sept. 2002 [https://sci-hub.se/10.1109/TCSI.2002.802353]
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
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:
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
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
SS-MM CDR
\(h_1\) 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
\(s_{011}\) & \(s_{110}\) 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.
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.