KNOW-HOW

bandwidth - Rules of Thumb

NRZ Bandwidth

Maxim Integrated,NRZ Bandwidth - HF Cutoff vs. SNR [https://pdfserv.maximintegrated.com/en/an/AN870.pdf]

\(0.35/T_r\)

32 to 56 Gbps Serial Link Analysis and Optimization Methods for Pathological Channels [https://docs.keysight.com/eesofapps/files/678068240/678068273/1/1629077956000/tutorial-32-to-56-gbps-serial-link-analysis-optimization-methods-pathological-channels.pdf]

Rise Time Sine Wave \[\begin{align} \sin 2\pi f t_{20} &= -1+2\times 0.2 = -0.6 \\ \sin 2\pi f t_{80} &= -1+2\times 0.8 = 0.6 \end{align}\] \[ f = \frac{\arcsin(0.6) - \arcsin(-0.6)}{2\pi (t_{80} - t_{20})} = \frac{0.2}{T_r} \]

Step Response RC Tank

\[\begin{align} 0.8 &= 1- e^{-\frac{t_{80}}{\tau_{RC}}} \to t_{80} = -\ln0.2 \cdot \tau_{RC} \\ 0.2 &= 1- e^{-\frac{t_{80}}{\tau_{RC}}} \to t_{20} = -\ln0.8 \cdot \tau_{RC} \end{align}\]

Then rise time of 20% to 80% is \[ T_{r} = \tau_{RC}\cdot \ln\frac{0.8}{0.2} = 1.3863 \cdot \tau_{RC} \] we have \[ f_{3dB} = \frac{1}{2\pi}\frac{1}{\tau_{RC}} = \frac{1}{2\pi}\frac{ 1.3863}{T_r} = \frac{0.22}{T_r} \]

\(0.5/T_r\)

TODO 📅

CMI & DMI

Ahmed Sada Staff Silicon Validation Engineer, Synopsys. Don't Be Intimidated: Manual Calibration for Stressed Eye Testing

Patrick Kennedy, Impact of Noise Coupler on 128 GT/s Rx Calibration and DUT Testing

Anritsu White paper: PCIe® 6.0: Testing for a New Generation [link]

sinusoidal differential mode interference (DMI), and common mode interference (CMI)

Single-Ended & Differential Signaling

[https://web.stanford.edu/class/archive/ee/ee371/ee371.1066/handouts/markChapt.pdf]

[https://www.allaboutcircuits.com/technical-articles/the-why-and-how-of-differential-signaling/]

Since we have (ideally) no return current, the ground reference becomes less important. The ground potential can even be different at the sender and receiver or moving around within a certain acceptable range. However, you need to be careful because DC-coupled differential signaling (such as USB, RS-485, CAN) generally requires a shared ground potential to ensure that the signals stay within the interface's maximum and minimum allowable common-mode voltage.

Coupling noise & Crosstalk

High-speed Serial Interface Lect. 9 – Noise [http://tera.yonsei.ac.kr/class/2017_2_2/lecture/Lect%209%20Noise.pdf]

TODO 📅

Clocking Structures

Synchronous, Mesochronous, Plesiochronous

Cascaded PLLs

Chembiyan T, A General Theory of Cascaded PLL Design [link]

Nicola Da Dalt, ISSCC 2012 T5: JITTER basic and advanced concepts, statistics and applications [https://www.nishanchettri.com/isscc-slides/2012%20ISSCC/TUTORIALS/ISSCC2012Visuals-T5.pdf]

—. ESSCIRC 2019 Tutorials: Jitter in Wireline and Data Converter Applications [https://youtu.be/aapkfCeHTrQ]

To understand the impact of the clock jitter on the performance of a wireline system, the transfer functions of the PLL in the transmitter side and the CDR loop in the receiver should be taken into consideration

the minimum jitter occurs at the point where the transmit PLL UGB is minimum and the CDR UGB is maximized

• the net rms jitter that impacts the performance of a wireline transceiver is much lower than the rms jitter of the transmit PLL
• the jitter requirements of the transmit PLL on the wireline system is much more relaxed compared to the wireless transceiver

Low-Latency PCIe

TODO 📅

AC-coupling vs DC-coupling

TODO 📅

Nonlinear & Intermodulations

S. Stegemann, W. Mathis. MOS-AK 2012: Interference and Distortion Analysis for Nonlinear Analog Circuits [https://www.mos-ak.org/dresden_2012/publications/T8_Stegemann_MOS-AK_Desden_12.pdf]

Ali Sheikholeslami. A-SSCC 2024 insight: Noise and Distortion, [https://youtu.be/bvsJgHJ19jI?si=1CKDJvTy5EQdPLB4]

B. Razavi, "Design considerations for direct-conversion receivers," in IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing, vol. 44, no. 6, pp. 428-435, June 1997 [http://www.seas.ucla.edu/brweb/papers/Journals/RTCAS97.pdf]

Even-Order Distortion

Odd-order distortion: symmetry

Even-Order Distortion: non-symmetry (Effect of Mismatch)

[http://cc.ee.ntu.edu.tw/~ecl/Courses/105AIC/lock/Analog_Chapter_09_Nonlinearity%20and%20Mismatch.pdf]

PAM4

Eye-Diagram and Bit-Error-Ratio (BER)

Noman Hai, Synopsys. CICC 2025 Circuit Insights: Basics of Wireline Transmitter Circuits [https://youtu.be/oofViBGlrjM?si=WZnOqtDVG3iDnBHI]

JTOL btw DSP-based vs Analog PAM4 RX

CC Chen, Why Analog PAM4 Receiver? [https://youtu.be/J2ojSMYiuBs?si=7y41W91ciIw_hme2]

challenges in DSP-based SerDes

Parallel implementation

Loop-Unrolling DFE

Corresponding to the three distinct voltage thresholds in the PAM4 systems, it would need 12 slicers, 3 multiplexers, and one thermometer-to-binary decoder in each deserialized data path, even if only one tap of the DFE is unrolled

Look-Ahead Multiplexing DFE

The look-ahead multiplexing technique brings the key benefit that the timing constraint can be significantly relaxed, as the iteration bound is doubled at the expense of extra hardware

Synopsys Italia, Tech Talk: Introduction to DSP-based SerDes [https://youtu.be/puEP0DlVZGI?si=lFiu1Kl4AKsg3O9f]

Chen, Kuan-Chang (2022) Energy-Efficient Receiver Design for High-Speed Interconnects. Dissertation (Ph.D.), California Institute of Technology. [https://thesis.library.caltech.edu/14318/9/chen_kuan-chang_2022_thesis_final.pdf]

Trellis Coding

TODO 📅

Convolutional Code

TODO 📅

Forward Error Correction (FEC)

It is called "forward" error correction because it can correct errors even in the common situations where there is no backward channel

Keysight Technologies. Tutorial – Why Use Forward Error Correction (FEC) [https://youtu.be/56nF4c61KR0?si=jYJSH50q9M3pIvQt]

Paul McLellan, What the FEC is Forward Error Correction? [https://community.cadence.com/cadence_blogs_8/b/breakfast-bytes/posts/fec]

Baseline Wander

TODO 📅

Pete Anslow, Ciena. Baseline wander with FEC [https://www.ieee802.org/3/bs/public/17_05/anslow_3bs_03_0517.pdf]

Vladimir Dmitriev-Zdorov. Baseline Wander, its Time Domain and Statistical Analysis [https://ibis.org/summits/feb19/dmitriev-zdorov.pdf]

Pavel Zivny, Tektronix. Baseline Wander: Systematic Approach to Rapid Simulation and Measurement [pdf]

Update on Performance Studies of 100 Gigabit Ethernet Enabled by Advanced Modulation Formats [https://www.ieee802.org/3/bm/public/sep12/wei_01_0912_optx.pdf]

Sampling Front-End (SFE) Pulse Response

sweep the setup time between ideal pulse input and clock, sample the output of SFE at falling edge

ISI & DDJ filtering

Modulation and SNR

Data and noise mutually uncorrelated

\[ x_{RX,n}[p] = d[p]h_{RX}[0] +\sum \text{ISI} + n[p] \]

"ISI cancellation" based equalization is conceptually more straightforward but suffers from SNR penalty or error propagation

Jitter Amplification by Passive Channels

Enhancing Resolution with a \(\Delta \Sigma\) Modulator

Sub-Resolution Time Averaging

\(\Delta \Sigma\) modulator effectively dithers the LSB bit between zero and one, such that you can get the effective resolution of a much higher resolution DAC in the number of bits

Decimation

how they affect sampling phase

DLF's input bit-width can be reduced by decimating BBPD's output. Decimation is typically performed by realizing either majority voting (MV) or boxcar filtering.

Note that deserialization is inherent to both MV and boxcar filtering

• Decimation is commonly employed to alleviate the high-speed requirement. However, decimation increases loop-latency which causes excessive dither jitter.
• Decimation is basically, widen the data and slowing it down
• Decimating by \(L\) means frequency register only added once every \(L\) UI, thus integral path gain reduced by \(L\) in linear model
• proportional path gain is unchanged

CDR Linear Model

condition:

Linear model of the CDR is used in a frequency lock condition and is approaching to achieve phase lock

Using this model, the power spectral density (PSD) of jitter in the recovered clock \(S_{out}(f)\) is \[ S_{out}(f)=|H_T(f)|^2S_{in}(f)+|H_G(f)|^2S_{VCO}(f) \] Here, we assume \(\varphi_{in}\) and \(\varphi_{VCO}\) are uncorrelated as they come from independent sources.

Jitter Transfer Function (JTF)

\[ H_T(s) = \frac{\varphi_{out}(s)}{\varphi_{in}(s)}|_{\varphi_{vco}=0}=\frac{K_{PD}K_{VCO}R_s+\frac{K_{PD}K_{VCO}}{C}}{s^2+K_{PD}K_{VCO}R_s+\frac{K_{PD}K_{VCO}}{C}} \]

Using below notation \[\begin{align} \omega_n^2=\frac{K_{PD}K_{VCO}}{C} \\ \xi=\frac{K_{PD}K_{VCO}}{2\omega_n^2} \end{align}\]

We can rewrite transfer function as follows \[ H_T(s)=\frac{2\xi\omega_n s+\omega_n^2}{s^2+2\xi \omega_n s+\omega_n^2} \]

The jitter transfer represents a low-pass filter whose magnitude is around 1 (0 dB) for low jitter frequencies and drops at 20 dB/decade for frequencies above \(\omega_n\)

• the recovered clock track the low-frequency jitter of the input data
• the recovered clock DONT track the high-frequency jitter of the input data

The recovered clock does not suffer from high-frequency jitter even though the input signal may contain high-frequency jitter, which will limit the CDR tolerance to high-frequency jitter.

Jitter Peaking in Jitter Transfer Function

The peak, slightly larger than 1 (0dB) implies that jitter will be amplified at some frequencies in the CDR, producing a jitter amplitude in the recovered clock, and thus also in the recovered data, that is slightly larger than the jitter amplitude in the input data.

This is certainly undesirable, especially in applications such as repeaters.

Jitter Generation

If the input data to the CDR is clean with no jitter, i.e., \(\varphi_{in}=0\), the jitter of the recovered clock comes directly from the VCO jitter. The transfer function that relates the VCO jitter to the recovered clock jitter is known as jitter generation. \[ H_G(s)=\frac{\varphi_{out}}{\varphi_{VCO}}|_{\varphi_{in}=0}=\frac{s^2}{s^2+2\xi \omega_n s+\omega_n^2} \] Jitter generation is high-pass filter with two zeros, at zero frequency, and two poles identical to those of the jitter transfer function

Jitter Tolerance (JTOL)

To quantify jitter tolerance, we often apply a sinusoidal jitter of a fixed frequency to the CDR input data and observe the BER of the CDR

The jitter tolerance curve DONT capture a CDR's true tolerance to random jitter. Because we are applying "sinusoidal" jitter, which is deterministic signal.

We can deal only with the jitter's amplitude and frequency instead of the PSD of the jitter thanks to deterministic sinusoidal jitter signal. \[ JTOL(f) = \left | \varphi_{in}(f) \right |_{\text{pp-max}} \quad \text{for a fixed BER} \] Where the subscript \(\text{pp-max}\) indicates the maximum peak-to-peak amplitude. We can further expand this equation as follows \[ JTOL(f)=\left| \frac{\varphi_{in}(f)}{\varphi_{e}(f)} \right| \cdot |\varphi_e(f)|_\text{pp-max} \]

Relative jitter, \(\varphi_e\) must be less than 1UIpp for error-free operation

In an ideal CDR, the maximum peak-to-peak amplitude of \(|\varphi_e(f)|\) is 1UI, i.e.,\(|\varphi_e(f)|_\text{pp-max}=1UI\)

Accordingly, jitter tolerance can be expressed in terms of the number of UIs as \[ JTOL(f)=\left| \frac{\varphi_{in}(f)}{\varphi_{e}(f)} \right|\quad \text{[UI]} \] Given the linear CDR model, we can write \[ JTOL(f)=\left| 1+\frac{K_{PD}K_{VCO}H_{LF}(f)}{j2\pi f} \right|\quad \text{[UI]} \] Expand \(H_{LF}(f)\) for the CDR, we can write \[ JTOL(f)=\left| 1-2\xi j \left(\frac{f_n}{f}\right) - \left(\frac{f_n}{f}\right)^2 \right|\quad \text{[UI]} \] At frequencies far below and above the natural frequency, the jitter tolerance can be approximated by the following \[ JTOL(f) = \left\{ \begin{array}{cl} \left(\frac{f_n}{f}\right)^2 & : \ f\ll f_n \\ 1 & : \ f\gg f_n \end{array} \right. \]

the jitter tolerance at very high jitter frequencies is limited to 1UIpp

1
2
3
4
5
6
7
8
9
10
11
12
13
14

clc;
clear all;

f_fn = logspace(-1, 2, 60);
for xi = [2, 1, 0.5, 0.2]
    jtol = abs(1- 1i*2*xi.*(1./f_fn)- (1./f_fn).^2);
    loglog(f_fn, jtol,LineWidth=2)
    disp(["min(JTpp)=", min(jtol),"@\xi=",xi])
    hold on
end    
grid on;
xlabel("f/f_n")
ylabel('JT_{pp}')
legend('\xi=2', '\xi=1', '\xi=0.5', '\xi=0.2')

CC Chen, Circuit Images: Why JTOL in a CDR? [https://youtu.be/kZExm9wy0G8?si=5ULT0t_oHNkp7c2v]

—. Why Pseudo-JTOL in a CDR Design or Verification? [https://youtu.be/DZyzLhk59aY?si=xMcN2Xo3hnCeL3RX]

OJTF

Concepts of JTF and OJTF

Simplified Block Diagram of a Clock-Recovery PLL

Jitter Transfer Function (JTF)

• Input Signal Versus Recovered Clock
• JTF, by jitter frequency, compares how much input signal jitter is transferred to the output of a clock-recovery's PLL (recovered clock)
  • Input signal jitter that is within the clock recovery PLL's loop bandwidth results in jitter that is faithfully transferred (closed-loop gain) to the clock recovery PLL's output signal. JTF in this situation is approximately 1.
  • Input signal jitter that is outside the clock recovery PLL's loop bandwidth results in decreasing jitter (open-loop gain) on the clock recovery PLL's output, because the jitter is filtered out and no longer reaches the PLL's VCO

Observed Jitter Transfer Function

• Input Signal Versus Sampled Signal
• OJTF compares how much input signal jitter is transferred to the output of a receiver's decision making circuit as effected by a clock recovery's PLL. As the recovered clock is the reference for detecting the input signal
  • Input signal jitter that is within the clock recovery PLL's loop bandwidth results in jitter on the recovered clock which reduces the amount of jitter that can be detected. The input signal and clock signal are closer in phase
  • Input signal jitter that is outside the clock recovery PLL's loop bandwidth results in reduced jitter on the recovered clock which increases the amount of jitter that can be detected. The input signal and clock signal are more out of phase. Jitter that is on both the input and clock signals can not detected or is reduced

JTF and OJTF for 1st Order PLLs

The observed jitter is a complement to the PLL jitter transfer response OJTF=1-JTF (Phase matters!)

OTJF gives the amount of jitter which is tracked and therefore not observed at the output of the CDR as a function of the jitter rate applied to the input.

Jitter Measurement

\[ J_{\text{measured}} = JTF_{\text{DUT}} \cdot OJTF_{\text{instrument}} \]

The combination of the OJTF of a jitter measurement device and the JTF of the clock generator under test gives the measured jitter as a function of frequency.

For example, a clock generator with a type 1, 1st order PLL measured with a jitter measurement device employing a golden PLL is \[ J_{\text{measured}} = \frac{\omega_1}{s+\omega_1}\frac{s}{s+\omega_2} \]

Accurate measurement of the clock JTF requires that the OJTF cutoff of the jitter measurement be significantly below that of the clock JTF and that the measurement is compensated for the instrument's OJTF.

The overall response is a band pass filter because the clock JTF is low pass and the jitter measurement device OJTF is high pass.

The compensation for the instrument OJTF is performed by measuring the jitter of the reference clock at each jitter rate being tested and comparing the reference jitter with the jitter measured at the output of the DUT.

The lower the cutoff frequency of the jitter measurement device the better the accuracy of the measurement will be.

The cutoff frequency is limited by several factors including the phase noise of the DUT and measurement time.

Digital Sampling Oscilloscope

How to analyze jitter:

• TIE (Time Interval Error) track
• histogram
• FFT

TIE track provides a direct view of how the phase of the clock evolves over time.

histogram provides valuable information about the long term variations in the timing.

FFT allows jitter at specific rates to be measured down to the femto-second range.

Maintaining the record length at a minimum of \(1/10\) of the inverse of the PLL loop bandwidth minimizes the response error

reference

Dalt, Nicola Da and Ali Sheikholeslami. “Understanding Jitter and Phase Noise: A Circuits and Systems Perspective.” (2018).

neuhelium, 抖动、眼图和高速数字链路分析基础 URL: http://www.neuhelium.com/ueditor/net/upload/file/20200826/DSOS254A/03.pdf

Keysight JTF & OJTF Concepts, https://rfmw.em.keysight.com/DigitalPhotonics/flexdca/FlexPLL-UG/Content/Topics/Quick-Start/jtf-pll-theory.htm?TocPath=Quick%20Start%7C_____4

Complementary Transmitter and Receiver Jitter Test Methodlogy, URL: https://www.ieee802.org/3/bm/public/mar14/ghiasi_01_0314_optx.pdf

SerDesDesign.com CDR_BangBang_Model URL: https://www.serdesdesign.com/home/web_documents/models/CDR_BangBang_Model.pdf

M. Schnecker, Jitter Transfer Measurement in Clock Circuits, LeCroy Corporation, DesignCon 2009. URL: http://cdn.teledynelecroy.com/files/whitepapers/designcon2009_lecroy_jitter_transfer_measurement_in_clock_circuits.pdf

Link Budgets

VCO model

TODO 📅

respone to vctrl focus on phase

[https://designers-guide.org/verilog-ams/functional-blocks/vco/vco.va]

ADC Spec

TODO 📅

ENOB - Not sufficient & not accurate enough

• Based on SNDR
• Assume unbounded Gaussian distribution

quantization noise is ~ bounded uniform distribution

Using unbounded Gaussian -> pessimistic BER prediction

AFE Nonlinearity

"total harmonic distortion" (THD) in AFE

Relative to NRZ-based systems, PAM4 transceivers require more stringent circuit linearity, equalizers which can implement multi-level inter-symbol interference (ISI) cancellation, and improved sensitivity

Because if it compresses, it turns out you have to use a much more complicated feedback filter. As long as it behaves linearly, the feedback filter itself can remain a linear FIR

Linearity can actually be a critical constraint in these signal paths, and you really want to stay as linear as you can all the way up until the point where you've canceled all of the ISI

A. Roshan-Zamir, O. Elhadidy, H. -W. Yang and S. Palermo, "A Reconfigurable 16/32 Gb/s Dual-Mode NRZ/PAM4 SerDes in 65-nm CMOS," in IEEE Journal of Solid-State Circuits, vol. 52, no. 9, pp. 2430-2447, Sept. 2017 [https://people.engr.tamu.edu/spalermo/ecen689/2017_reconfigurable_16_32Gbps_NRZ_PAM4_SERDES_roshanzamir_jssc.pdf]

Hongtao Zhang, designcon2016. "PAM4 Signaling for 56G Serial Link Applications − A Tutorial"[https://www.xilinx.com/publications/events/designcon/2016/slides-pam4signalingfor56gserial-zhang-designcon.pdf]

Elad Alon, ISSCC 2014, "T6: Analog Front-End Design for Gb/s Wireline Receivers"

BER with Quantization Noise

\[ \text{Var}(X) = E[X^2] - E[X]^2 \]

Impulse Response or Pulse Response

TX FFE

TX FFE suffers from the peak power constraint, which in effect attenuates the average power of the outgoing signal - the low-frequency signal content has been attenuated down to the high-frequency level

[https://www.signalintegrityjournal.com/articles/1228-feedforward-equalizer-location-study-for-high-speed-serial-systems]

S. Palermo, "CMOS Nanoelectronics Analog and RF VLSI Circuits," Chapter 9: High-Speed Serial I/O Design for Channel-Limited and Power-Constrained Systems, McGraw-Hill, 2011.

Eye-Opening Monitor (EOM)

An architecture that evaluates the received signal quality

data slicers, phase slicers, error slicers, scope slicers

Analui, Behnam & Rylyakov, Alexander & Rylov, Sergey & Meghelli, Mounir & Hajimiri, Ali. (2006). A 10-Gb/s two-dimensional eye-opening monitor in 0.13-??m standard CMOS. Solid-State Circuits, IEEE Journal of. 40. 2689 - 2699, [https://chic.caltech.edu/wp-content/uploads/2013/05/B-Analui_JSSC_10-Gbs_05.pdf]

reference

G. Balamurugan, A. Balankutty and C. -M. Hsu, "56G/112G Link Foundations Standards, Link Budgets & Models," 2019 IEEE Custom Integrated Circuits Conference (CICC), Austin, TX, USA, 2019, pp. 1-95 [https://youtu.be/OABG3u2H2J4?si=CxryBSGbxrUpZNBT] [https://picture.iczhiku.com/resource/ieee/SHKhwYfGotkIymBx.pdf]

Paul Muller Yusuf Leblebici École Polytechnique Fédérale de Lausanne (EPFL). Pattern generator model for jitter-tolerance simulation; VHDL-AMS models

Anritsu Company, "Measuring Channel Operating Margin," 2016. [https://dl.cdn-anritsu.com/en-us/test-measurement/files/Technical-Notes/White-Paper/11410-00989A.pdf]

Kiran Gunnam, Selected Topics in RF, Analog and Mixed Signal Circuits and Systems

H. Shakiba, D. Tonietto and A. Sheikholeslami, "High-Speed Wireline Links-Part I: Modeling," in IEEE Open Journal of the Solid-State Circuits Society, vol. 4, pp. 97-109, 2024 [https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10608184]

H. Shakiba, D. Tonietto and A. Sheikholeslami, "High-Speed Wireline Links-Part II: Optimization and Performance Assessment," in IEEE Open Journal of the Solid-State Circuits Society, vol. 4, pp. 110-121, 2024 [https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10579874]

G. Souliotis, A. Tsimpos and S. Vlassis, "Phase Interpolator-Based Clock and Data Recovery With Jitter Optimization," in IEEE Open Journal of Circuits and Systems, vol. 4, pp. 203-217, 2023 [https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10184121]

loop dynamic

Hanumolu, Pavan Kumar. 2006. Design Techniques for Clocking High Performance Signaling Systems. : Oregon State University. https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/1v53k219r]

Hae-Chang Lee, "An Estimation Approach To Clock And Data Recovery" [https://www-vlsi.stanford.edu/people/alum/pdf/0611_HaechangLee_Phase_Estimation.pdf]

R. Walker, “Designing Bang-Bang PLLs for Clock and Data Recovery in Serial Data Transmission Systems,” in Phase-Locking in High-Performance Systems, B. Razavi, Ed. New Jersey: IEEE Press, 2003, pp. 34-45. [http://www.omnisterra.com/walker/pdfs.papers/BBPLL.pdf]

J. Kim, Design of CMOS Adaptive-Supply Serial Links, Ph.D. Thesis, Stanford University, December 2002. [https://www-vlsi.stanford.edu/people/alum/pdf/0212_Kim_______Design_Of_CMOS_AdaptiveSu.pdf]

P. K. Hanumolu, M. G. Kim, G. -y. Wei and U. -k. Moon, "A 1.6Gbps Digital Clock and Data Recovery Circuit," IEEE Custom Integrated Circuits Conference 2006, San Jose, CA, USA, 2006, pp. 603-606 [https://sci-hub.se/10.1109/CICC.2006.320829]

Da Dalt N. A design-oriented study of the nonlinear dynamics of digital bang-bang PLLs. IEEE Transactions on Circuits and Systems I: Regular Papers. 2005;52(1):21–31. [https://sci-hub.se/10.1109/TCSI.2004.840089]

Jang S, Kim S, Chu SH, Jeong GS, Kim Y, Jeong DK. An optimum loop gain tracking all-digital PLL using autocorrelation of bang–bang phase frequency detection. IEEE Transactions on Circuits and Systems II: Express Briefs. 2015;62(9):836–840. [https://sci-hub.se/10.1109/TCSII.2015.2435691]

CDR Loop Latency

Denoting the CDR loop latency by \(\Delta T\) , we note that the loop transmission is multiplied by \(exp(-s\Delta T)\simeq 1-s\Delta T\).The resulting right-half-plane zero, \(f_z\) degrades the phase margin and must remain about one decade beyond the BW \[ f_z\simeq \frac{1}{2\pi \Delta T} \]

This assumption is true in practice since the bandwidth of the CDR (few mega Hertz) is much smaller than the data rate (multi giga bits/second).

Fernando , Marvell Italy."Considerations for CDR Bandwidth Proposal" [https://www.ieee802.org/3/bs/public/16_03/debernardinis_3bs_01_0316.pdf]

Loop Bandwidth

The closed-loop −3-dB bandwidth is sometimes called the “loop bandwidth”

Continuous-Time Approximation Limitations

A rule of thumb often used to ensure slow changes in the loop is to select the loop bandwidth approximately equal to one-tenth of the input frequency.

Gardner, F.M. (1980). Charge-Pump Phase-Lock Loops. IEEE Trans. Commun., 28, 1849-1858.

Homayoun, Aliakbar and Behzad Razavi. “On the Stability of Charge-Pump Phase-Locked Loops.” IEEE Transactions on Circuits and Systems I: Regular Papers 63 (2016): 741-750.

N. Kuznetsov, A. Matveev, M. Yuldashev and R. Yuldashev, "Nonlinear Analysis of Charge-Pump Phase-Locked Loop: The Hold-In and Pull-In Ranges," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 68, no. 10, pp. 4049-4061, Oct. 2021

Deog-Kyoon Jeong, Topics in IC Design - 2.1 Introduction to Phase-Locked Loop [https://ocw.snu.ac.kr/sites/default/files/NOTE/Lec%202%20-%20Charge-Pump%20PLL%2C%20Freuqency%20Synthesizers%2C%20and%20SSCG.pdf]

Limit Cycle Oscillation

limit cycles imply self-sustained oscillators due nonlinear nature

Ouzounov, S., Hegt, H., Van Roermund, A. (2007). SUB-HARMONIC LIMIT-CYCLE SIGMA-DELTA MODULATION, APPLIED TO AD CONVERSION. In: Van Roermund, A.H., Casier, H., Steyaert, M. (eds) Analog Circuit Design. Springer, Dordrecht. [https://sci-hub.se/10.1007/1-4020-5186-7_6]

Digital CDR Category

• DCO part is analogous so that it cannot be perfectly modeled
• Digital-to-phase converter is well-defined phase output, thus, very good to model real situation

Z-domain modeling

The difference equation is \[ \phi[n] = \phi[n-1] + K_{DCO}V_C[n]\cdot T\cdot2\pi \] z-transform is \[ \frac{\Phi(z)}{V_C(z)}=\frac{2\pi K_{DCO}T}{1-z^{-1}} \]

where \(K_{DCO}\) : \(\Delta f\) (Hz/bit)

\(\Delta \Sigma\)-dithering in DCO

Quantization noise

Here, \(\alpha_T\) is data transition density

BBPD quantization noise

DAC quantization noise

M. -J. Park and J. Kim, "Pseudo-Linear Analysis of Bang-Bang Controlled Timing Circuits," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 60, no. 6, pp. 1381-1394, June 2013 [https://sci-hub.st/10.1109/TCSI.2012.2220502]

Time to Digital Converter (TDC)

Digital to Phase Converter (DPC)

IIR low pass filter

simple approximation: \[ z = 1 + sT \] bilinear-z transform \[ z =\frac{}{} \]

FAQ

PLL vs. CDR

PLL or FD(Frequency Detector) for frequency detecting in CDR

reference

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

P. Hanumolu. ISSCC 2015 "Clock and Data Recovery Architectures and Circuits" [slides]

Amir Amirkhany. ISSCC 2019 "Basics of Clock and Data Recovery Circuits"

Fulvio Spagna. INTEL, CICC2018, "Clock and Data Recovery Systems" [slides]

M. Perrott. 6.976 High Speed Communication Circuits and Systems (lecture 21). Spring 2003. Massachusetts Institute of Technology: MIT OpenCourseWare, [lec21.pdf]

Akihide Sai. ISSCC 2023, T5 "All Digital Plls From Fundamental Concepts To Future Trends" [T5.pdf]

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]

—, "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]

Fernando De Bernardinis, eSilicon. "Introduction to DSP Based Serial Links" [https://www.corsi.univr.it/documenti/OccorrenzaIns/matdid/matdid835215.pdf]

Yohan Frans, CICC2019 ES3-3- "ADC-based Wireline Transceivers" [pdf]

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?arnumber=10630516]

Marinaci, Stefano. "Study of a Phase Locked Loop based Clock and Data Recovery Circuit for 2.5 Gbps data-rate" [https://cds.cern.ch/record/2870334/files/CERN-THESIS-2023-147.pdf]

P. Palestri et al., "Analytical Modeling of Jitter in Bang-Bang CDR Circuits Featuring Phase Interpolation," in IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 29, no. 7, pp. 1392-1401, July 2021 [https://sci-hub.se/10.1109/TVLSI.2021.3068450]

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

M.H. Perrott, Y. Huang, R.T. Baird, B.W. Garlepp, D. Pastorello, E.T. King, Q. Yu, D.B. Kasha, P. Steiner, L. Zhang, J. Hein, B. Del Signore, "A 2.5 Gb/s Multi-Rate 0.25μm CMOS Clock and Data Recovery Circuit Utilizing a Hybrid Analog/Digital Loop Filter and All-Digital Referenceless Frequency Acquisition," IEEE J. Solid-State Circuits, vol. 41, Dec. 2006, pp. 2930-2944 [https://cppsim.com/Publications/JNL/perrott_jssc06.pdf]

M.H. Perrott. CICC 2009 "Tutorial on Digital Phase-Locked Loops" [https://www.cppsim.com/PLL_Lectures/digital_pll_cicc_tutorial_perrott.pdf]

—, Short Course On Phase-Locked Loops and Their Applications Day 4, PM Lecture "Examples of Leveraging Digital Techniques in PLLs" [https://www.cppsim.com/PLL_Lectures/day4_pm.pdf]

—, Short Course On Phase-Locked Loops IEEE Circuit and System Society, San Diego, CA "Digital Frequency Synthesizers" [https://www.cppsim.com/PLL_Lectures/digital_pll.pdf]

Gain Kim, "Equalization, Architecture, and Circuit Design for High-Speed Serial Link Receiver" [pdf]

Deog-Kyoon Jeong Topics in IC(Wireline Transceiver Design) - 3.1. Introduction to All Digital PLL [https://ocw.snu.ac.kr/sites/default/files/NOTE/Lec%203%20-%20ADPLL.pdf]

—, Topics in IC(Wireline Transceiver Design) - 6.1 Introduction to Clock and Data Recovery [https://ocw.snu.ac.kr/sites/default/files/NOTE/Lec%206%20-%20Clock%20and%20Data%20Recovery.pdf]

High-speed Serial Interface Lect. 16 – Clock and Data Recovery 3 [http://tera.yonsei.ac.kr/class/2013_1_2/lecture/Lect16_CDR-3.pdf]

Shiva Kiran. Phd thesis 2018. Modeling and Design of Architectures for High-Speed ADC-Based Serial Links [https://hdl.handle.net/1969.1/192031]

—, et al., "Modeling of ADC-Based Serial Link Receivers With Embedded and Digital Equalization," in IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 9, no. 3, pp. 536-548, March 2019 [https://sci-hub.se/10.1109/TCPMT.2018.2853080]

K. Zheng, "System-Driven Circuit Design for ADC-Based Wireline Data Links", Ph.D. Dissertation, Stanford University, 2018 [https://purl.stanford.edu/hw458fp0168]

S. Cai, A. Shafik, S. Kiran, E. Z. Tabasy, S. Hoyos and S. Palermo, "Statistical modeling of metastability in ADC-based serial I/O receivers," 2014 IEEE 23rd Conference on Electrical Performance of Electronic Packaging and Systems [pdf]

John M. Cioffi. "Decoding Methods" [https://cioffi-group.stanford.edu/doc/book/chap7.pdf]

—. "Equalization" [https://cioffi-group.stanford.edu/doc/book/chap3.pdf]

Iain. [https://youtu.be/rnjy4_gXLAg?si=PC3aowaon-e_mhXX]

—. [https://youtu.be/IJE94FhyygM?si=BMMQ-GmirBWNf4ep]

Noman Hai, Synopsys, Canada CASS Talks 2025 - May 2, 2025: High-speed Wireline Interconnects: Design Challenges and Innovations in 224G SerDes [https://www.youtube.com/live/wHNOlxHFTzY?si=179HZozbtW59H8Lm]

Phase Noise Definition

Eq. (3.25) is widely adopted by industry and academia

using the narrow angle assumption, the two definitions above are equivalent

If the narrow angle condition is not satisfied, however, the two definitions differ

PM vs. FM jitter

Deog-Kyoon Jeong. Topics in IC Design: 1.1 Introduction to Jitter [https://ocw.snu.ac.kr/sites/default/files/NOTE/Lec%201%20-%20Jitter%20and%20Phase%20Noise.pdf]

Phase Noise Profile

Power Spectral Density of Brownian Motion despite non-stationary [https://dsp.stackexchange.com/a/75043/59253]

white noise

\(1/f^2\) Phase Noise Profile

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

flicker noise

\(1/f^3\) Phase Noise Profile

\[ S_{\phi n} = \frac{K}{f}\left(\frac{K_{VCO}}{2\pi f}\right)^2 \propto \frac{1}{f^3} \]

[https://dsp.stackexchange.com/a/75152/59253]

Free-running Oscillator

Note that \(f_{min}\) is related to the observation time. The longer we observe the device under test, the smaller \(f_{min}\) must be

Ali Sheikholeslami ISSCC 2008 T5: Basics of Chip-to-Chip and Backplane Signaling [https://www.nishanchettri.com/isscc-slides/2008%20ISSCC/Tutorials/T10_Pres.pdf]

B. Casper and F. O'Mahony, "Clocking Analysis, Implementation and Measurement Techniques for High-Speed Data Links-A Tutorial," in IEEE Transactions on Circuits and Systems I. [https://people.engr.tamu.edu/spalermo/ecen689/clocking_analysis_hs_links_casper_tcas1_2009.pdf]

Lorentzian spectrum

We typically use the two spectra, \(S_{\phi n}(f)\) and \(S_{out}(f)\), interchangeably, but we must resolve these inconsistencies. voltage spectrum is called Lorentzian spectrum

The periodic signal \(x(t)\) can be expanded in Fourier series as:

Assume that the signal is subject to excess phase noise, which is modeled by adding a time-dependent noise component \(\alpha(t)\). The noisy signal can be written \(x(t+\alpha(t))\), the added excess phase \(\phi(t)= \frac{\alpha(t)}{\omega_0}\)

The autocorrelation of the noisy signal is by definition:

The autocorrelation averaged over time results in:

By taking the Fourier transform of the autocorrelation, the spectrum of the signal \(x(t + \alpha(t))\)​ can be expressed as

It is also interesting to note how the integral in Equation 9.80 around each harmonic is equal to the power of the harmonic itself \(|X_n|^2\)

The integral \(S_x(f)\) around harmonic is \[\begin{align} P_{x,n} &= \int_{f=-\infty}^{\infty} |X_n|^2\frac{\omega_0^2n^2c}{\frac{1}{4}\omega_0^4n^4c^2+(\omega +n\omega_0)^2}df \\ &= |X_n|^2\int_{\Delta f=-\infty}^{\infty}\frac{2\beta}{\beta^2+(2\pi\cdot\Delta f)^2}d\Delta f \\ &= |X_n|^2\frac{1}{\pi}\arctan(\frac{2\pi \Delta f}{\beta})|_{-\infty}^{\infty} \\ &= |X_n|^2 \end{align}\]

The phase noise does not affect the total power in the signal, it only affects its distribution

• Without phase noise, \(S_v(f)\) is a series of impulse functions at the harmonics of \(f_o\).
• With phase noise, the impulse functions spread, becoming fatter and shorter but retaining the same total power

[https://community.cadence.com/cadence_technology_forums/f/rf-design/51484/comparing-transient-noise-pnoise-and-pnoise-with-lorentian-approximation-of-a-ring-oscillator/1382911]

Phase perturbed by a stationary noise with Gaussian PDF

If keep \(\phi_{rms}\) in \(R_x(\tau)\), i.e. \[ R_x(\tau)=\frac{A^2}{2}e^{-\phi_{rms}^2}\cos(2\pi f_0 \tau)e^{R_\phi(\tau)}\approx \frac{A^2}{2}e^{-\phi_{rms}^2}\cos(2\pi f_0 \tau)(1+R_\phi(\tau)) \] The PSD of the signal is \[ S_x(f) = \mathcal{F} \{ R_x(\tau) \} = \frac{P_c}{2}e^{-\phi_{rms}^2}\left[S_\phi(f+f_0)+S_\phi(f-f_0)\right] + \frac{P_c}{2}e^{-\phi_{rms}^2}\left[\delta(f+f_0)+\delta(f-f_0)\right] \] ❗❗above Eq isn't consistent with stationary white noise process - the following section

Phase perturbed by a stationary WHITE noise process

Assuming that the delay line is noiseless

Expanding the cosine function we get \[\begin{align} R_y(t,\tau) &= \frac{A^2}{2}\left\{\cos(2\pi f_0\tau)E[\cos(\phi(t)-\phi(t-\tau))] - \sin(2\pi f_0\tau)E[\sin(\phi(t)-\phi(t-\tau))]\right\} \\ &+ \frac{A^2}{2}\left\{\cos(4\pi f_0(t+\tau/2-T_D))E[\cos(\phi(t)+\phi(t-\tau))] - \sin(4\pi f_0(t+\tau/2-T_D))E[\sin(\phi(t)+\phi(t-\tau))] \right\} \end{align}\]

where, both the process \(\phi(t)-\phi(t-\tau)\) and \(\phi(t)+\phi(t-\tau)\) are independent of time \(t\), i.e. \(E[\cos(\phi(t)+\phi(t-\tau))] = m_{\cos+}(\tau)\), \(E[\cos(\phi(t)-\phi(t-\tau))] = m_{\cos-}(\tau)\), \(E[\sin(\phi(t)+\phi(t-\tau))] = m_{\sin+}(\tau)\) and \(E[\sin(\phi(t)-\phi(t-\tau))] = m_{\sin-}(\tau)\)

we obtain \[\begin{align} R_y(t,\tau) &= \frac{A^2}{2}\left\{\cos(2\pi f_0\tau)m_{\cos-}(\tau) - \sin(2\pi f_0\tau)m_{\sin-}(\tau)\right\} \\ &+ \frac{A^2}{2}\left\{\cos(4\pi f_0(t+\tau/2-T_D))m_{\cos+}(\tau) - \sin(4\pi f_0(t+\tau/2-T_D))m_{\sin+}(\tau) \right\} \end{align}\]

The second term in the above expression is periodic in \(t\) and to estimate its PSD, we compute the time-averaged autocorrelation function \[ R_y(\tau) = \frac{A^2}{2}\left\{\cos(2\pi f_0\tau)m_{\cos-}(\tau) - \sin(2\pi f_0\tau)m_{\sin-}(\tau)\right\} \]

After nontrivial derivation

Phase perturbed by a Weiner process

The phase process \(\phi(t)\) is also gaussian but with an increasing variance which grows linearly with time \(t\)

\[\begin{align} R_y(t,\tau) &= \frac{A^2}{2}\left\{\cos(2\pi f_0\tau)E[\cos(\phi(t)-\phi(t-\tau))] - \sin(2\pi f_0\tau)E[\sin(\phi(t)-\phi(t-\tau))]\right\} \\ &+ \frac{A^2}{2}\left\{\cos(4\pi f_0(t+\tau/2-T_D)E[\cos(\phi(t)+\phi(t-\tau))] - \sin(4\pi f_0(t+\tau/2-T_D)E[\sin(\phi(t)+\phi(t-\tau))] \right\} \end{align}\]

The spectrum of \(y(t)\) is determined by the asymptotic behavior of \(R_y(t,\tau)\) as \(t\to \infty\)

❗❗ \(\lim_{t\to\infty}R_y(t,\tau)\) rather than time-averaged autocorrelation function of cyclostationary process, ref. Demir's paper

We define \(\zeta(t, \tau)=\phi(t)+\phi(t-\tau) = \phi(t)-\phi(t-\tau) + 2\phi(t-\tau)\), the expected value of \(\zeta(t,\tau)\) is 0, the variance is \(\sigma_{\zeta}^2=(k\sigma)^2(\tau + 4(t-\tau))=(k\sigma)^2(4t-3\tau)\) \[ E[\cos(\zeta(t,\tau))]=\frac{1}{\sqrt{2\pi \sigma_{\zeta}^2}}\int_{-\infty}^{\infty}e^{-\zeta^2/2\sigma_{\zeta}^2}\cos(\zeta)d\zeta = e^{-\sigma_{\zeta}^2/2}=e^{-(k\sigma)^2(4t-\tau)} \] i.e., \(\lim _{t\to \infty} E[\cos(\zeta(t,\tau))] = \lim_{t\to \infty}e^{-(k\sigma)^2(4t-\tau)} = 0\)

For \(E[\sin(\zeta(t,\tau))]\), we have \[ E[\sin(\zeta(t,\tau))] = \frac{1}{\sqrt{2\pi \sigma_{\zeta}^2}}\int_{-\infty}^{\infty}e^{-\zeta^2/2\sigma_{\zeta}^2}\sin(\zeta)d\zeta \] i.e., \(E[\sin(\zeta(t,\tau))]\) is odd function, therefore \(E[\sin(\zeta(t,\tau))]=0\)

Finally, we obtain

Amplitude Noise

P.E. Allen - 2003. ECE 6440 - Frequency Synthesizers: Lecture 160 – Phase Noise - II [https://pallen.ece.gatech.edu/Academic/ECE_6440/Summer_2003/L160-PhNoII(2UP).pdf]

reference

A. Hajimiri and T. H. Lee, "A general theory of phase noise in electrical oscillators," in IEEE Journal of Solid-State Circuits, vol. 33, no. 2, pp. 179-194, Feb. 1998 [paper], [slides]

—, "Corrections to "A General Theory of Phase Noise in Electrical Oscillators"" [https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=678662]

—, RFIC2024 "Noise in Oscillators from Understanding to Design"

Carlo Samori, "Phase Noise in LC Oscillators: From Basic Concepts to Advanced Topologies" [https://www.ieeetoronto.ca/wp-content/uploads/2020/06/DL-VCO-short.pdf]

—, "Understanding Phase Noise in LC VCOs: A Key Problem in RF Integrated Circuits," in IEEE Solid-State Circuits Magazine, vol. 8, no. 4, pp. 81-91, Fall 2016 [https://picture.iczhiku.com/resource/eetop/whIgTikLswaaTVBv.pdf]

—, ISSCC2016, "Understanding Phase Noise in LC VCOs"

Antonio Liscidini, ESSCIRC 2019 Tutorials: Phase Noise in Wireless Applications [https://youtu.be/nGmQ0JdoSE4]

A. Demir, A. Mehrotra and J. Roychowdhury, "Phase noise in oscillators: a unifying theory and numerical methods for characterization," in IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications, vol. 47, no. 5, pp. 655-674, May 2000 [https://sci-hub.se/10.1109/81.847872]

Dalt, Nicola Da and Ali Sheikholeslami. "Understanding Jitter and Phase Noise: A Circuits and Systems Perspective." (2018) [https://picture.iczhiku.com/resource/eetop/WykRGJJoHQLaSCMv.pdf]

F. L. Traversa, M. Bonnin and F. Bonani, "The Complex World of Oscillator Noise: Modern Approaches to Oscillator (Phase and Amplitude) Noise Analysis," in IEEE Microwave Magazine, vol. 22, no. 7, pp. 24-32, July 2021 [https://iris.polito.it/retrieve/handle/11583/2903596/e384c433-b8f5-d4b2-e053-9f05fe0a1d67/MM%20noise%20-%20v5.pdf]

Poddar, Ajay & Rohde, Ulrich & Apte, Anisha. (2013). How Low Can They Go?: Oscillator Phase Noise Model, Theoretical, Experimental Validation, and Phase Noise Measurements. Microwave Magazine, IEEE. [http://time.kinali.ch/rohde/noise/how_low_can_they_go-2013-poddar_rohde_apte.pdf]

Pietro Andreani, "RF Harmonic Oscillators Integrated in Silicon Technologies" [https://www.ieeetoronto.ca/wp-content/uploads/2020/06/DL-Toronto.pdf]

Chembiyan T, "Brownian Motion And The Oscillator Phase Noise" [link]

—, "Jitter and Phase Noise in Oscillators" [link]

—, "Jitter and Phase Noise in Phase Locked Loops" [link]

—, "PLLs and reference spurs" [link]

Godone, A. & Micalizio, Salvatore & Levi, Filippo. (2008). RF spectrum of a carrier with a random phase modulation of arbitrary slope. [https://sci-hub.se/10.1088/0026-1394/45/3/008]

Bae, Woorham; Jeong, Deog-Kyoon: 'Analysis and Design of CMOS Clocking Circuits for Low Phase Noise' (Materials, Circuits and Devices, 2020)

• 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

---

The Problem of "Sinusoids Running Around Loops"

The representative of Fourier transform \(\frac{1}{j\omega+j\omega_0}\) back in the time domain \(e^{-j\omega_0 t}\) is infinite extent in time

Running around a loop, chasing one's tail — these are thought pictures that only work in a discretized, time-sequenced conceptual framework that has a beginning and an end

Fix in your mind that oscillations are a type of resonance

rearrangement

The closed loop transfer function of \(Y/X\) and \(Y_1/X_1\) are almost same, except sign

\[\begin{align} \frac{Y}{X} &= +\frac{H_1(s)H_2(s)}{1+H_1(s)H_2(s)} \\ \frac{Y_1}{X_1} &= -\frac{H_1(s)H_2(s)}{1+H_1(s)H_2(s)} \end{align}\]

define \(-Y_1=Y_n\), then \[ \frac{Y_n}{X_1} = \frac{H_1(s)H_2(s)}{1+H_1(s)H_2(s)} \]

Saurabh Saxena, IIT Madras. CICC2022 Clocking for Serial Links - Frequency and Jitter Requirements, Phase-Locked Loops, Clock and Data Recovery

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, Joel L. A Guide to Feedback Theory. Cambridge: Cambridge University Press, 2021.

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

Jens Anders (University of Stuttgart, DE). ESSERC2025 Circuits Insights: The Magic of Feedback in Analog Circuit Design [https://youtu.be/NyXuA6WZ8Hg]

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

voltage doubler

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

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

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:

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

multiphase CP

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

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]

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

• \(x_c(t)\) 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

• \(x[n]\) 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

• 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 \(M = 2\)

Upsampling or Zero Insertion

Rouphael, Tony. (2009). RF and Digital Signal Processing for Software-Defined Radio. [pdf]

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

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

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

sampling identities

downsampling identity

upsampling identity

Polyphase Decomposition

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

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

Polyphase Implementation of Decimation Filters & Interpolation Filters

	Decimation system	Interpolation system
sampling identity

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

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

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

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

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

integral path

integral path gain reduced by \(L\)

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

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

Decimation by Voting

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

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

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

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

Kpd, Kb, Kv

Both decimation factor and factor for voting are 4

• 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]

Rail-to-Rail Op amp

[https://mixsignal.wordpress.com/wp-content/uploads/2013/03/689-604rail2rail.pdf]

[https://toshiba.semicon-storage.com/ap-en/semiconductor/knowledge/faq/linear_opamp/what-does-rail-to-rail-mean.html]

TODO 📅

Noise and Distortion

TODO 📅

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

Slewing in Folded-Cascode Op Amps

In practice, we choose \(I_P \simeq I_{SS}\)

Avoid zero current in cascodes

• left circuit

  \(I_b \gt I_a\)

• right circuit

  \(I_b \gt 2I_a\)

Huijsing, J. H. (2017). Operational Amplifiers: Theory and Design. (3 ed.) Springer

[https://bbs.eetop.cn/forum.php?mod=redirect&goto=findpost&ptid=995502&pid=11548528]

Active Inductor

\[\begin{align} A &= \frac{g_mR_L}{1+(g_{\text{m}_{\text{dio}}}+ g_{\text{ds}_\text{tot}})R_L}\cdot \frac{1+R_pC_Ps}{1+\frac{(1+g_{\text{ds}_{\text{tot}}}R_L)R_PC_P+C_PR_L+R_LC_L}{1+(g_{\text{m}_{\text{dio}}}+g_{\text{ds}_\text{tot}})R_L}s + \frac{R_LC_LR_PC_P}{1+(g_{\text{m}_\text{dio}}+g_{\text{ds}_{\text{tot}}})R_L}s^2} \\ &= \frac{g_mR_L}{1+(g_{\text{m}_{\text{dio}}}+ g_{\text{ds}_{\text{tot}}})R_L}\cdot \frac{R_PC_P}{ \frac{R_LC_LR_PC_P}{1+(g_{\text{m}_{\text{dio}}}+g_{\text{ds}_{\text{tot}}})R_L}}\cdot \frac{1/(R_PC_P)+s}{s^2 + \frac{(1+g_{\text{ds}_{\text{tot}}}R_L)R_PC_P+C_PR_L+R_LC_L}{R_PC_P}s + \frac{1+(g_{\text{m}_{\text{dio}}}+g_{\text{ds}_\text{tot}})R_L}{R_LC_LR_PC_P}} \\ &= A_0 \cdot A(s) \end{align}\]

That is

\[\begin{align} \omega_z &= \frac{1}{R_PC_P} \tag{1} \\ \omega_n &= \sqrt{\frac{1+(g_{\text{m}_{\text{dio}}}+ g_{\text{ds}_\text{tot}})R_L}{R_LC_LR_PC_P}} = \sqrt{\omega_{p0}\omega_z} \\ \zeta & = \frac{(1+g_{\text{ds}_\text{tot}}R_L)R_PC_P+C_PR_L+R_LC_L}{R_PC_P} \frac{1}{2 \omega_n} \end{align}\]

Where \[\begin{align} \omega_{p0} &= \frac{1}{(R_L||\frac{1}{g_{\text{m}_{\text{dio}}}}||\frac{1}{g_{\text{m}_{\text{tot}}}})C_L} \tag{2} \end{align}\]

Here, relate \(\omega_{p0}\) and \(\omega_z\) by coefficient \(\alpha\) \[ \omega_{p0} = \alpha \cdot \omega_z \tag{3} \] This way \[ \omega_n= \sqrt{\alpha}\cdot \omega_z \]

\[ \zeta = \frac{1}{2}(K\sqrt{\alpha}+\frac{1+C_P/C_L}{\sqrt{\alpha}}) \tag{4} \]

where \[ K = \frac{R_L||\frac{1}{g_{\text{m}_{\text{dio}}}}||\frac{1}{g_{\text{m}_{\text{tot}}}}}{R_L||g_\text{ds\_tot}} \]

And \(A(s)\) can be expressed as \[ A(s) = \frac{\frac{s}{\omega_z}+1}{\frac{s^2}{\omega_n^2}+2\frac{\zeta}{\omega_n}s+1} \] It magnitude in dB \[ A_\text{dB} = 10\log\frac{1+(\omega/\omega_z)^2}{1+(\omega/\omega_n)^4+2\omega^2(2\zeta^2-1)/\omega_n^2} \] Substitute \(\omega_n\) with Eq (2), followed is obtained \[ A_\text{dB} = 10\log{\frac{\alpha^2(\omega_z^4 + \omega_z^2\omega^2)}{\alpha^2\omega_z^4+\omega^4+2\alpha\omega_z^2(2\zeta^2-1)\omega^2}} \] peaking frequency \[ \omega_\text{peak} = \omega_z\cdot \sqrt{\sqrt{(\alpha+1)^2 - 4\alpha \zeta^2}-1} \] If \(\zeta=1\) \[\begin{align} \omega_{A_\text{dB = 0dB} }&= \sqrt{1-2/\alpha}\cdot \omega_{p0} \\ \omega_\text{peak} &= \omega_z\sqrt{\alpha-2} \\ A_\text{dB,peak} &= 10\log\frac{\alpha^2}{4(\alpha-1)} \end{align}\]

Response Speed in Analog Circuits

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

Bandwidth limitation

slew rate limitation

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)

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

\[ I_{NIC} =\frac{V_{im} - V_{ip}}{\frac{2}{g_m}+\frac{1}{sC_c}} = \frac{-2V_{ip}}{\frac{2}{g_m}+\frac{1}{sC_c}} \] 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

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)

Flipped Voltage Follower (FVF)

T&H buffer in ADC

[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 \[ V_o = A_v(\overline{V_\text{ip} - V_\text{im}} - \overline{V_\text{rp} - V_\text{rm}}) \]

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

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

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

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

CM Distortion

Resistive Degeneration

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

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

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]

linearized model

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

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

1st order DDSM (digital accumulator)

assuming \(n_0=2\)

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

2nd order DDSM

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

?? 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

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]

Truncation DAC

accumulator is implicit quantizer

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)

To avoid overflow

\[ \log_2(2^k + 2^{N-m}) \leq N \]

Thus \[ N \ge k - \log_2(1 - \frac{1}{2^m}) = k+m -\log_2(2^m-1) \]

suppose \(m\in [1,+\infty)\) \[ k < k - \log_2(1 - \frac{1}{2^m}) \leq k + 1 \] \(N = k +1\) is sufficient for any \(k\)

In the above Temes's sides, \(N = m_1+m_2\) and \(m=m_1\), we have \[ 2^k \leq 2^{m_1+m_2}\cdot (1-\frac{1}{2^{m_1}}) = 2^{m_1+m_2} - 2^{m_2} \]

Generally speaking, \(N \propto k\) and \(N \propto \frac{1}{m}\), especially \(N_{min} = k+1\) if single-bit quantizer \(m=1\)

Fractional-N PLL

Divider Model

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

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

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

\(\Delta\Sigma\) noise in PLL

\[\begin{align} S_\phi(f) &= \frac{1}{12F_{ref}}|1-z^{-1}|^{2L}\cdot \left|\frac{2\pi z_{-1}}{1-z^{-1}}\right|^2\cdot \frac{1}{T_{ref}^2} T_{ref}^2 \\ &= \frac{1}{12F_{ref}} |1-z^{-1}|^{2L-2} 4\pi^2 \end{align}\]

with \(|1-z^{-1}| = |2\sin\frac{\pi f}{F_{ref}}|\)

\[ S_\phi(f) = \frac{1}{12F_{ref}} \cdot \left|2\sin\frac{\pi f}{F_{ref}}\right|^{2(L-1)}\cdot 4\pi^2 = \frac{\pi^2}{3F_{ref}} \cdot \left|2\sin\frac{\pi f}{F_{ref}}\right|^{2(L-1)} \]

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]

\(\Delta\Sigma\) DAC

LPF (RC Filter)

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

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

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

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

ZOH (Zero-Order Hold Models)

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

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

Sinc Corrector

Neil Robertson, Design a DAC sinx/x Corrector [https://www.dsprelated.com/showarticle/1191.php]

Dan Boschen. how to make CIC compensation filter [https://dsp.stackexchange.com/a/31596/59253]

—. Core Building Blocks for Software Defined Radio (SDR): DDC, DUC, NCO) [https://lnkd.in/p/e2MtC9QK]

Equalizing Techniques Flatten DAC Frequency Response [https://www.analog.com/en/resources/technical-articles/equalizing-techniques-flatten-dac-frequency-response.html]

aka. Inverse Sinc Compensation

TODO 📅

OSR & NS

maximum output signal 22kHz

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

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));

Y. Liu, J. Gao and X. Yang, "24-bit low-power low-cost digital audio sigma-delta DAC," in Tsinghua Science and Technology, vol. 16, no. 1, pp. 74-82, Feb. 2011 [https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6077939]

1
2
3
4
5
6
7
8
9
10
11
12
13
14

snr_if = 144;
snr_ds = 135;
snr_ana = 95;

n_if = 1/10^(snr_if/10);
n_ds = 1/10^(snr_ds/10);
n_ana = 1/10^(snr_ana/10);

snr_tot = 10*log10(1/(n_if + n_ds + n_ana))

% 
% snr_tot =
% 
%    94.9995

MASH 1-1-1 implementaion

J. W. M. Rogers, F. F. Dai, M. S. Cavin and D. G. Rahn, "A multiband /spl Delta//spl Sigma/ fractional-N frequency synthesizer for a MIMO WLAN transceiver RFIC," in IEEE Journal of Solid-State Circuits, vol. 40, no. 3, pp. 678-689, March 2005 [https://sci-hub.se/10.1109/JSSC.2005.843604]

(a) a fractional accumulator, and (b) a triple-loop \(\Delta\Sigma\) accumulator for \(N(z) = 100 + 1/32\)

\(n=5\)

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

import numpy as np
import matplotlib.pyplot as plt
import scipy.fft as fft
import scipy.signal.windows as windows

def acc_nbit(din, nbit=5, ncycles=2**10):
    mod = 2**nbit
    acc = 0
    eq = 0
    colist = []
    eqlist = []
    for i in range(ncycles):
        acc = din[i] + eq
        eq = acc % mod
        #print(acc, eq)
        co_tmp = int(acc >= mod)
        colist.append(co_tmp)
        eqlist.append(eq)
    return colist, eqlist

Ncyl = 2**16
cin1 = [1]*Ncyl

c1,eq1 = acc_nbit(cin1, ncycles=Ncyl)
cin2 = [0,*eq1[:-1]]
#cin2 = eq1

c2,eq2 = acc_nbit(cin2, ncycles=Ncyl)
cin3 = [0,*eq2[:-1]]
#cin3 =  eq2

c3,eq3 = acc_nbit(cin3, ncycles=Ncyl)

ctot = []

for i in range(2, Ncyl):
    # C1(z) + (1 − z^−1)C2(z) + (1 − z^−1)^2 C3(z)
    ctot_cur = c1[i] + c2[i]-c2[i-1] + c3[i]-2*c3[i-1] + c3[i-2]
    ctot.append(ctot_cur)

print(sum(c1)/len(c1))      # 0.03125
print(sum(ctot)/len(ctot))  # 0.03128147221289712

plt.figure(figsize=(20,8))
plt.subplot(1,2,1)
plt.plot(c1[:200], 'o-', label='c1')
plt.legend(loc='upper left'); plt.grid(True)
plt.subplot(1,2,2)
plt.plot(ctot[:200], 'o-', label='ctot')
plt.legend(loc='upper left'); plt.grid(True)


Ntot = int(2**np.floor(np.log2(len(ctot))))
c1_4fft = np.array(c1[:Ntot])
ctot_4fft = np.array(ctot[:Ntot])
whann = windows.hann(Ntot)
Y_c1 = abs(fft.rfft(c1_4fft*whann))/sum(whann)
Y_ctot = abs(fft.rfft(ctot_4fft*whann))/sum(whann)

print(Y_c1[0])      # 0.031250000002717625
print(Y_ctot[0])    # 0.031249999999526525

plt.figure(figsize=(20,8))
plt.subplot(2, 1, 1)
_, stemlines, _ = plt.stem(Y_c1, markerfmt=" ", label='c1')
plt.setp(stemlines, 'linewidth', 4)
plt.ylim([0,0.04]); plt.grid(True); plt.legend(loc='upper left')
plt.subplot(2, 1, 2)
_, stemlines, _ = plt.stem(Y_ctot, markerfmt=" ", label='ctot')
plt.setp(stemlines, 'linewidth', 4)
plt.ylim([0,0.04]); plt.grid(True); plt.legend(loc='upper left')

plt.show()

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

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

Rick Lyons. How Discrete Signal Interpolation Improves D/A Conversion [https://www.dsprelated.com/showarticle/167.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.

John Rogers, Calvin Plett, and Foster Dai. 2006. Integrated Circuit Design for High-Speed Frequency Synthesis (Artech House Microwave Library). Artech House, Inc., USA. [pdf]

K. Hosseini and M. P. Kennedy, Minimizing Spurious Tones in Digital Delta-Sigma Modulators (Analog Circuits and Signal Processing). New York, NY, USA: Springer, 2011.

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

Lacaita, Andrea Leonardo, Salvatore Levantino, and Carlo Samori. Integrated frequency synthesizers for wireless systems. Cambridge University Press, 2007.

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)

"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

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

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]

Noise Shaping

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')

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

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

first order delta-sigma modulator

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

second order delta-sigma modulator

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

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]

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

Quantizer Bits & ENOB

without the delta-sigma loop

High-Order \(\Delta\Sigma\) Modulators

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

Quantizer overload

\(\Delta \Sigma\) 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

DAC

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

[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

\(M \gt N\)

[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]

ADC

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

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

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]

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

[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

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

Both integrator and quantizer are delay free

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

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

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

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

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

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 \[\begin{align} v[n] &= nu - \sum_{k=1}^{n-1}v[k] + e[n] \\ &= u + \left((n-1)u - \sum_{k=1}^{n-2}v[k]\right) - v[n-1] + e[n] \\ &= u + v[n-1] - e[n-1] -v[n-1] + e[n] \\ &= u + e[n] - e[n-1] \end{align}\]

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]

MOD1 with DC Excitation

TODO 📅

Mismatch Shaping

Data-Weighted Averaging (DWA)

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

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:

[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]

John Rogers, Calvin Plett, and Foster Dai. 2006. Integrated Circuit Design for High-Speed Frequency Synthesis (Artech House Microwave Library). Artech House, Inc., USA. [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

32 to 56 Gbps Serial Link Analysis and Optimization Methods for Pathological Channels [https://docs.keysight.com/eesofapps/files/678068240/678068273/1/1629077956000/tutorial-32-to-56-gbps-serial-link-analysis-optimization-methods-pathological-channels.pdf]

Feedforward Equalizer

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]

Vivek Telang, Equalization for High-Speed Serdes: System-level Comparison of Analog and Digital Techniques 2012 [https://ewh.ieee.org/r5/denver/sscs/Presentations/2012_08_Telang.pdf]

Kevin Zheng, Boris Murmann, Hongtao Zhang, and Geoff Zhang. Feedforward Equalizer Location Study for High-Speed Serial Systems [https://www.signalintegrityjournal.com/articles/1228-feedforward-equalizer-location-study-for-high-speed-serial-systems]

—, “System-Driven Circuit Design for ADC-Based Wireline Data Links”, Ph.D. Dissertation, Stanford University, 2018 [https://purl.stanford.edu/hw458fp0168]

Hanumolu, P. K., Wei, G. Y., & Moon, Y. K. (2005). Equalizers for high-speed serial links. International Journal of High Speed Electronics and Systems [https://people.engr.tamu.edu/spalermo/ecen689/hslink_eq_overview_hanumolu_jhses05.pdf]

Gain Kim. Equalization, Architecture, and Circuit Design for High-Speed Serial Link Receiver [pdf]

TX-FFE

TODO 📅

RX-FFE

TODO 📅

overlapped tuning range

TODO 📅

Mueller-Muller PD

Mueller-Muller type A timing function

Mueller-Muller type B timing function

MMSE-based algorithms

minimum mean squared error (MMSE)

There are three major MMSE-based algorithms:

• least mean square (LMS),
• normalized least mean square (NLMS)
• recursive least square (RLS)

LMS (Least-Mean-Square)

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)

[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]

Mueller-Muller CDR

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

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]

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

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

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/]

Hongtao Zhang, DesignCon 2016. PAM4 Signaling for 56G Serial Link Applications − A Tutorial [https://www.xilinx.com/publications/events/designcon/2016/slides-pam4signalingfor56gserial-zhang-designcon.pdf]

Noise Analysis

sampling (amplification) phase

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

Transient Noise Method

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

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

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

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

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

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

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

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 📅

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

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

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"

Pre-amp (preamplifier)

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

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 \[ \Phi(x) = \frac{1}{\sqrt{2\pi}}\int_{-\infty}^x e^{-z^2/2} \mathrm{d}z. \]

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

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

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]