ECE1371 Advanced Analog Circuits

Lecture 1
Introduction to Delta-Sigma ADCs

Trevor Caldwell
[email protected]
 Course Goals
• Deepen understanding of CMOS analog circuit
design through a top-down study of a modern
analog system – a delta-sigma ADC
• Develop circuit insight through brief peeks at
some nifty little circuits of the day
The circuit world is filled with many little gems that
every competent designer ought to know

2 ECE1371
 Logistics
• Format:
Meet Tuesdays 10:00-12:00 (no class Feb 16)
Twelve 2-hr Online Lectures + Exam + Project
Presentations
• Grading:
30% Homework (5%, 7.5%, 7.5%, 10%)
40% Project
30% Exam
• References
Pavan, Schreier & Temes, “Understanding  …”
Chan Carusone, Johns & Martin, “Analog IC …”
Razavi, “Design of Analog CMOS ICs”

3 ECE1371
 Lecture Plan
Date Lecture (10am-12pm) Reference Homework
2021-01-12 1 MOD1 & MOD2 PST 2, 3, A 1: Matlab MOD1&2
2021-01-19 2 MODN +  Toolbox PST 4, B
2:  Toolbox
2021-01-26 3 SC Circuits R 12, CCJM 14
2021-02-02 4 Comparator & Flash ADC CCJM 10
3: Comparator
2021-02-09 5 Example Design 1 PST 7, CCJM 14
2021-02-16 Reading Week / ISSCC
4: SC MOD2
2021-02-23 6 Example Design 2 CCJM 18
2021-03-02 7 Amplifier Design 1
2021-03-09 8 Amplifier Design 2
2021-03-16 9 Noise in SC Circuits
2021-03-23 10 Nyquist-Rate ADCs CCJM 15, 17 Project
2021-03-30 11 Mismatch & MM-Shaping PST 6
2021-04-06 12 Continuous-Time  PST 8
2021-04-13 Exam
2021-04-27 Project Presentation (Project Report Due at start of class)

4 ECE1371
 What you will learn…
• MOD1: 1st-order  modulator
Structure and theory of operation
• Inherent linearity of binary modulators
• Inherent anti-aliasing of continuous-time
modulators
• MOD2: 2nd-order  modulator
• Good FFT practice

5 ECE1371
 Background
• The Signal-to-Quantization Noise Ratio (SQNR)
of an ideal n-bit ADC with a full-scale sine-wave
input is (6.02n + 1.76) dB
“6 dB = 1 bit”
• The PSD at the output of a linear system is the
product of the input’s PSD and the squared
magnitude of the system’s frequency response

• The power in any frequency band is the integral

of the PSD over that band

6 ECE1371
 What is ?
• Simplified  ADC structure

Loop Coarse
VIN + DOUT
- Filter ADC

Analog In DAC Digital Out

(to digital
filter)

• Key features: course quantization, filtering,

feedback and oversampling
Quantization is often quite course (as low as 1 bit), but
the effective resolution can be as high as 22 bits

7 ECE1371
 What is Oversampling?
• Oversampling is sampling faster than required
by the Nyquist criterion
For a lowpass signal containing energy in the
frequency range (0, fB), the minimum sample rate
required for perfect reconstruction is fS = 2fB

• Oversampling Ratio 𝑺 𝑩

• For a regular ADC, OSR ~ 2-3

Larger than 1 to make the anti-alias filter (AAF) feasible
• For a  ADC, OSR ~ 8 to 200
To get adequate quantization noise suppression
Signals between fB and ~fS are removed digitally

8 ECE1371
 Oversampling Simplifies AAF

9 ECE1371
 How The  ADC Works
• Course quantization  lots of quantization error
So how can a  ADC achieve 22-bit resolution?
• A  ADC spectrally separates the quantization
error from the signal through noise-shaping

10 ECE1371
 A  DAC System

• Mathematically similar to an ADC system

Except that now the modulator is digital and drives a
low-resolution DAC, and the out-of-band noise is
handled by an analog reconstruction filter

11 ECE1371
 Why Do It The  Way?
• Simplified Anti-Alias Filter in ADC
Since the input is oversampled, only very high
frequencies alias to the passband
Simple RC filter is usually sufficient
If a continuous-time loop filter is used, the anti-alias
filter can often be eliminated altogether
DAC: Simplified Reconstruction Filter

 Inherent Linearity
Simple structures can yield very high SNR
 Robust Implementation
 tolerates sizable component errors

12 ECE1371
 MOD1: 1st-Order  Modulator

13 ECE1371
 MOD1 Analysis
• Exact analysis is intractable for all but the
simplest inputs, so treat the quantizer as an
additive noise source:

14 ECE1371
 The Noise Transfer Function (NTF)
• In general, V(z) = STF(z)ꞏU(z) + NTF(z)ꞏE(z)
• For MOD1, NTF(z) = 1 – z-1
 The quantization noise has spectral shape!

• The total noise power increases, but the noise

power at low frequencies is reduced

15 ECE1371
 In-band Quantization Noise Power
• Assume the error is white with power 𝜎
i.e. 𝑺𝒆𝒆 𝝎 𝜎 /𝜋
• The in-band quatization noise power is
𝝎𝑩 𝝎𝑩
𝟐 𝝈𝟐𝒆
𝑰𝑸𝑵𝑷 𝑯 𝒆𝒋𝝎 𝑺𝒆𝒆 𝝎 𝒅𝝎 ≅ 𝝎𝟐 𝒅𝝎
𝟎 𝝅 𝟎

𝝅 𝝅𝟐 𝝈𝟐𝒆
• Since 𝑶𝑺𝑹 , 𝑰𝑸𝑵𝑷 𝑶𝑺𝑹 𝟑
𝝎𝑩 𝟑
• For MOD1, an octave increase in OSR increases
SQNR by 9 dB
1.5-bit/octave SQNR-OSR trade-off

16 ECE1371
 A Simulation of MOD1 (Time)

17 ECE1371
 A Simulation of MOD1 (Freq)

18 ECE1371
 CT Implementation of MOD1
• Ri/Rf sets the full-scale; C is arbitrary
Ri, Rf are typically sized based on noise
Also observe that an input at fS is rejected by the
integrator – inherent anti-aliasing

19 ECE1371
 MOD1-CT Waveforms

• With u=0, v alternates between +1 and -1

• With u>0, y drifts upwards; v contains
consecutive +1s to counteract this drift

20 ECE1371
 MOD1-CT STF
𝟏 𝒛 𝟏
• 𝑺𝑻𝑭 , recall 𝒛 𝒆𝒔
𝒔

21 ECE1371
 MOD1-CT Frequency Responses

22 ECE1371
 Summary
•  works by spectrally separating the
quantization noise from the signal
Requires oversampling 𝑶𝑺𝑹 ≡ 𝒇𝑺 ⁄𝟐𝒇𝑩
• Noise-shaping is achieved by the use of filtering
and feedback
• A binary DAC is inherently linear, and thus a
binary  modulator is too
𝟏
• MOD1 has 𝑵𝑻𝑭 𝒛 𝟏 𝒛
Arbitrary accuracy for DC inputs
1.5 bit/octave SQNR-OSR trade-off
• MOD1-CT has inherent anti-aliasing

23 ECE1371
 MOD2: 2nd-Order  Modulator
• Replace the quantizer in MOD1 with another
copy of MOD1 in a recursive fashion

V(z) = U(z) + (1-z-1)E1(z), E1(z) = (1-z-1)E(z)

 V(z) = U(z) + (1-z-1)2E(z)

24 ECE1371
 Simplified Block Diagrams

25 ECE1371
 NTF Comparison

26 ECE1371
 In-Band Quantization Noise Power
𝒋𝝎 𝟐
• For MOD2, 𝑯 𝒆 𝝎𝟒
𝝎𝑩 𝒋𝝎 𝟐
• As before, 𝑰𝑸𝑵𝑷 𝟎
𝑯 𝒆 𝑺𝒆𝒆 𝝎 𝒅𝝎 and
𝑺𝒆𝒆 𝝎 𝝈𝟐𝒆 ⁄𝝅
𝝅𝟒 𝝈𝟐𝒆
• So now 𝑰𝑸𝑵𝑷 𝑶𝑺𝑹 𝟓
𝟓
With binary quantization to +/- 1,  = 2 and thus
𝝈𝟐𝒆 ∆𝟐 ⁄𝟏𝟐 𝟏/𝟑

• An octave increase in OSR increases MOD2’s

SQNR by 15dB (2.5 bits)

27 ECE1371
 Simulation Example
• Input at 75% of Full Scale

28 ECE1371
 Simulated MOD2 PSD
• Input at 50% of Full Scale

29 ECE1371
 SQNR vs. Input Amplitude
• MOD1 & MOD2 @ OSR = 256

30 ECE1371
 SQNR vs. OSR

31 ECE1371
 Audio Demo: MOD1 vs. MOD2
• dsdemo4 in Matlab  Toolbox

32 ECE1371
 MOD1 and MOD2 Summary
•  ADCs rely on filtering and feedback to
achieve high SNR despite coarse quantization
They also rely on digital signal processing
 ADCs need to be followed by a digital decimation
filter and  DACs need to be preceded by a digital
interpolation filter
• Oversampling eases analog filtering
requirements
Anti-alias filter in an ADC; image filter in a DAC
• Binary quantization yields inherent linearity
• MOD2 is better than MOD1
15 dB/octave vs 9 dB/octave SQNR-OSR trade-off
Quantization noise more white
Higher-order modulators are even better
33 ECE1371
 Good FFT Practice
• Use coherent sampling
Have an integer number of cycles in the record
• Use windowing
A Hann window works well
𝒘 𝒏 𝟏 𝒄𝒐𝒔 𝟐𝝅𝒏⁄𝑵 /𝟐

• Use enough points

Recommend 𝑵 𝟔𝟒 · 𝑶𝑺𝑹
• Scale (and smooth) the spectrum
A full-scale sine wave should yield a 0-dBFS peak
• State the noise bandwidth
For a Hann window, 𝑵𝑩𝑾 𝟏. 𝟓/𝑵

34 ECE1371
 Coherent vs Incoherent Sampling

• Coherent sampling: only one non-zero FFT bin

• Incoherent sampling: spectral leakage

35 ECE1371
 Windowing
•  data is usually not periodic
Just because the input repeats does not mean the
output does too!
• Windowing is unavoidable
A finite-length data record is equal to an infinite record
multiplied by a rectangular window 𝒘 𝒏 𝟏, 𝟎 𝒏 𝑵
• Multiplication in time is convolution in frequency

36 ECE1371
 Example Spectral Disaster
• Rectangular window, N=256

37 ECE1371
 Window Comparison (N=16)

38 ECE1371
 Window Properties

39 ECE1371
 Window Length, N
• Need to have enough in-band noise bins to
1. Make the number of signal bins a small fraction of the
total number of in-band bins
<20% signal bins  >15 in-band bins  𝑵 𝟑𝟎 · 𝑶𝑺𝑹
2. Make the SNR repeatable
𝑵 𝟑𝟎 · 𝑶𝑺𝑹 yields std. dev. ~1.4 dB
𝑵 𝟔𝟒 · 𝑶𝑺𝑹 yields std. dev. ~1.0 dB
𝑵 𝟐𝟓𝟔 · 𝑶𝑺𝑹 yields std. dev. ~0.5 dB

• 𝑵 𝟔𝟒 · 𝑶𝑺𝑹 is recommended

40 ECE1371
 FFT Scaling
• The FFT implemented in MATLAB is
𝑵 𝟏
𝟐𝝅𝒌𝒏
𝒋 𝑵
𝑿𝑴 𝒌 𝟏 𝒙𝑴 𝒏 𝟏 𝒆
𝒏 𝟎
• If 𝒙 𝒏 𝑨 sin 𝟐𝝅𝒇𝒏⁄𝑵 , then
𝑨𝑵
𝑿 𝒌 𝒌 𝒇 𝐨𝐫 𝑵 𝒇
𝟐
𝟎 𝐨𝐭𝐡𝐞𝐫𝐰𝐢𝐬𝐞

 Need to divide FFT by 𝑵⁄𝟐 to get A

Note: f is an integer in 𝟎, 𝑵⁄𝟐 . 𝑿 𝒌 ≡ 𝑿𝑴 𝒌 𝟏 ,

𝒙 𝒏 ≡ 𝒙𝑴 𝒏 𝟏 since Matlab indexes from 1 rather than 0.

41 ECE1371
 How To Do Smoothing
1. Average multiple FFTs
Implemented by MATLAB’s psd() function
2. Take one big FFT and “filter” the spectrum
Implemented by the  Toolbox’s logsmooth()
function
logsmooth() averages an exponentially-increasing
number of bins in order to reduce the density of
points in the high-frequency regime and make a nice
log-frequency plot

42 ECE1371
 Raw and Smoothed Spectra

43 ECE1371
 Simulation vs Theory (MOD2)

44 ECE1371
 What Went Wrong?
• We normalized the spectrum so that a full-scale
sine wave (which has a power of 0.5) comes out
at 0 dB (hence the ‘dBFS’ units)
We do the same for the error signal, use 𝑺𝒆𝒆 𝒇 𝟒⁄𝟑
But this makes the discrepancy 3 dB worse
• We tried to plot a power spectral density
together with something that we want to
interpret as a power spectrum
• Sine-wave components are located in individual
FFT bins, but broadband signals like noise have
their power spread over all FFT bins
The ‘noise floor’ depends on the length of the FFT

45 ECE1371
 Spectrum of a Sine Wave + Noise

46 ECE1371
 Observations
• The power of the sine wave is given by the
height of its spectral peak
• The power of the noise is spread over all bins
The greater the number of bins, the less power there is
in any one bin
• Doubling N reduces the power per bin by a
factor of 2 (i.e. 3 dB)
But the total integrated noise power does not change

47 ECE1371
 How Do We Handle Noise?
• An FFT is like a filter bank
• The longer the FFT, the narrower the bandwidth
of each filter and thus the lower the power at
each output
• We need to know the noise bandwidth (NBW) of
the filters in order to convert the power in each
bin (filter output) to a power density
• For a filter with frequency response H(f),

48 ECE1371
 FFT Noise Bandwidth
• Alternatively, for 𝒉 𝟏 as the L1-norm and 𝒉 𝟐
as the L2-norm
𝟐 𝟐
𝑯 𝒇 𝒅𝒇 𝒉 𝟐
𝑵𝑩𝑾 𝟐 𝟐
, 𝒇𝟎 𝟎
𝑯 𝟎 𝒉 𝟏
• Parseval’s theorem

𝟐 𝟐
𝑯 𝒇 𝒅𝒇 𝒉 𝒏

• If 𝒉 𝒏 𝟎

𝑯 𝟎 𝒉 𝒏

49 ECE1371
 Better Spectral Plot

50 ECE1371
 Homework #1 (Due Jan 26)
A. Create a Matlab function that computes MOD1’s
output sequence given a vector of input
samples and exercise your function in the
following ways:
1. Verify that 𝒗 𝒖 for a few random DC inputs in [-1,1]
2. Plot the output spectrum with a half-scale sine-wave
input. Use good FFT practice. Include the theoretical
quantization noise curve and list the theoretical and
simulated SQNR for OSR = 128.

B. Repeat with MOD2

51 ECE1371
 MOD2 Expanded

52 ECE1371
 Example Matlab Code
function [v] = simulateMOD2(u)
x1 = 0;
x2 = 0;
for i = 1:length(u)
v(i) = quantize(x2);
x1 = x1 + u(i) – v(i);
x2 = x2 + x1 – v(i);
end
return

function v = quantize(y)
if y>=0
v = 1;
else
v = -1;
end
return

53 ECE1371
 What You Learned Today
• MOD1: 1st-order  modulator
Structure and theory of operation
• Inherent linearity of binary modulators
• Inherent anti-aliasing of continuous-time
modulators
• MOD2: 2nd-order  modulator
• Good FFT practice

54 ECE1371