slide 1 of 57

University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling Converters
David Johns and Ken Martin
University of Toronto
([email protected])
([email protected])
slide 2 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Motivation
Popular approach for medium-to-low speed A/D and
D/A applications requiring high resolution
Easier Analog
reduced matching tolerances
relaxed anti-aliasing specs
relaxed smoothing filters
More Digital Signal Processing
Needs to perform strict anti-aliasing or smoothing
filtering
Also removes shaped quantization noise and
decimation (or interpolation)
slide 3 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Quantization Noise
Above model is exact
approx made when assumptions made about
Often assume is white, uniformily distributed
number between
is difference between two quantization levels
x n ( ) y n ( )
y n ( )
x n ( )
e n ( )
Quantizer Model
e n ( ) y n ( ) x n ( ) =
e n ( )
e n ( )
2

slide 4 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Quantization Noise
White noise assumption reasonable when:
fine quantization levels
signal crosses through many levels between
samples
sampling rate not synchronized to signal
frequency
Sample lands somewhere in quantization interval
leading to random error of

time
T
2
slide 5 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Quantization Noise
Quantization noise power shown to be and is
independent of sampling frequency
If white, then spectral density of noise, , is
constant.

2
12
S
e
f ( )
0
f
s
2
----
S
e
f ( )
f
f
s
2
----
k
x

12
----------


1
f
s
---- =
Height
slide 6 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling Advantage
Oversampling occurs when signal of interest is
bandlimited to but we sample higher than
Define oversampling-rate
(1)
After quantizing input signal, pass it through a
brickwall digital filter with passband up to
f
0
2f
0
OSR f
s
2f
0
( ) =
f
0
N-bit quantizer
u n ( )
y
2
n ( )
y
1
n ( )
H f ( )
0f
0
f
s
2
----
f
f
0

f
s
2
----
H f ( )
1
slide 7 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling Advantage
Output quantization noise after filtering is:
(2)
Doubling OSR reduces quantation noise power by
3dB (i.e. 0.5 bits/octave)
Assuming peak input is a sinusoidal wave with a
peak value of leading to
Can also find peak SNR as:
(3)
P
e
S
e
2
f ( ) H f ( )
2
df
f
s
2
f
s
2

k
x
2
df
f
0

f
0

2
12
------
1
OSR
-----------


= = =
2
N
2 ( )
P
s
2
N
( ) 2 2 ( ) ( )
2
=
SNR
max
10
P
s
P
e
-----


log 10
3
2
---2
2N


log 10 OSR ( ) log + = =
slide 8 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling Advantage
Example
A dc signal with 1V is combined with a noise signal
uniformily distributed between giving 0 dB SNR.
{0.94, 0.52, 0.73, 2.15, 1.91, 1.33, 0.31, 2.33}.
Average of 8 samples results in 0.8875
Signal adds linearly while noise values add in a
square-root fashion noise filtered out.
Example
1-bit A/D gives 6dB SNR.
To obtain 96dB SNR requires 30 octaves of
oversampling ( (96-6)/3 dB/octave )
If ,
3
f
0
25 kHz = f
s
2
30
f
0
54 000 GHz ! , = =
slide 9 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Advantage of 1-bit D/A Converters
Oversampling improves SNR but not linearity
To acheive 16-bit linear converter using a 12-bit
converter, 12-bit converter must be linear to 16 bits
i.e. integral nonlinearity better than LSB
A 1-bit D/A is inherently linear
1-bit D/A has only 2 output points
2 points always lie on a straight line
Can acheive better than 20 bits linearity without
trimming (will likely have gain and offset error)
Second-order effects (such as D/A memory or signal-
dependent reference voltages) will limit linearity.
1 2
4

slide 10 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling with Noise Shaping
Place the quantizer in a feedback loop
H z ( )

x n ( )
u n ( )
y n ( )
H z ( )

x n ( )
u n ( )
y n ( )
e n ( )
Quantizer
Delta-Sigma Modulator
Linear model
slide 11 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling with Noise Shaping
Shapes quantization noise away from signal band of
interest
Signal and Noise Transfer-Functions
(4)
(5)
(6)
Choose to be large over 0 to
Resulting quantization noise near 0 where large
Signal transfer-function near 1 where large
S
TF
z ( )
Y z ( )
U z ( )
----------
H z ( )
1 H z ( ) +
-------------------- =
N
TF
z ( )
Y z ( )
E z ( )
----------
1
1 H z ( ) +
-------------------- =
Y z ( ) S
TF
z ( )U z ( ) N
TF
z ( )E z ( ) + =
H z ( ) f
0
H z ( )
H z ( )
slide 12 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Oversampling with Noise Shaping
Input signal is limited to range of quantizer output
when large
For 1-bit quantizers, input often limited to 1/4
quantizer outputs
Out-of-band signals can be larger when small
Stability of modulator can be an issue (particularily
for higher-orders of
Stability defined as when input to quantizer becomes
so large that quantization error greater than
said to overload the quantizer
H z ( )
H z ( )
H z ( )
2
slide 13 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
First-Order Noise Shaping
Choose to be a discrete-time integrator
(7)
If stable, average input of integrator must be zero
Average value of must equal average of
H z ( )
H z ( )
1
z 1
----------- =

x n ( )
u n ( )
y n ( )
Quantizer
z
1
u n ( ) y n ( )
slide 14 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Example
The output sequence and state values when a dc
input, , of is applied to a 1st order modulator
with a two-level quantizer of . Initial state for
is 0.1.
Average of is as expected
Periodic quantization noise in this case
n x(n) x(n + 1) y(n) e(n)
0 0.1000 0.5667 1.0 0.9000
1 0.5667 0.7667 1.0 0.4333
2 0.7667 0.1000 1.0 0.2333
3 0.1000 0.5667 1.0 0.9000
4 0.5667 0.7667 1.0 0.4333
5
u n ( ) 1 3
1.0 x n ( )

y n ( ) 1 3
slide 15 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Transfer-Functions
Signal and Noise Transfer-Functions
(8)
(9)
Noise transfer-function is a discrete-time
differentiator (i.e. a highpass filter)
(10)
S
TF
z ( )
Y z ( )
U z ( )
----------
1 z 1 ( )
1 1 z 1 ( ) +
-------------------------------- z
1
= = =
N
TF
z ( )
Y z ( )
E z ( )
--------- -
1
1 1 z 1 ( ) +
-------------------------------- 1 z
1
( ) = = =
N
TF
f ( ) 1 e
j 2f f
s

e
jf f
s

e
jf f
s

2j
------------------------------------ 2j e
jf f
s

= =
f
f
s
-----


sin 2j e
jf f
s

=
slide 16 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Signal to Noise Ratio
Magnitude of noise transfer-function
(11)
Quantization noise power
(12)
Assuming (i.e., )
(13)
N
TF
f ( ) 2
f
f
s
-----


sin =
P
e
S
e
2
f ( ) N
TF
f ( )
2
df
f
0

f
0

2
12
------


1
f
s
--- 2
f
f
s
-----


sin
2
df
f
0

f
0

= =
f
0
<< f
s
OSR >> 1
P
e

2
12
------



2
3
-----


2f
0
f
s
-------


3

2
36
------------
1
OSR
-----------


3
=
slide 17 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Max SNR
Assuming peak input is a sinusoidal wave with a
peak value of leading to
Can find peak SNR as:
(14)
or, equivalently,
(15)
Doubling OSR gives an SNR improvement or,
2
N
2 ( )
P
s
2
N
( ) 2 2 ( ) ( )
2
=
SNR
max
10
P
s
P
e
-----


log =
10
3
2
---2
2N


log 10
3

2
----- OSR ( )
3
log + =
SNR
max
6.02N 1.76 5.17 30 OSR ( ) log + + =
9 dB
slide 18 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
equivalently, a benefit of 1.5 bits/octave
slide 19 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
SC Implementation
1-bit D/A

u n ( )
y n ( )
Quantizer
z
1
Analog Digital
H z ( )
V
ref
2
V
in
V
out
C
C

2

1
( ) ,

1

2
( ) ,
Latch on
2
falling

1

2
, V
out
high

1
( )
2
( ) , V
out
low
Comparator
C
slide 20 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Second-Order Noise Shaping
(16)
(17)
(18)
Doubling improves SNR by
(i.e., a benefit of 2.5 bits/octave)

u n ( )
y n ( )
Quantizer
z
1

z
1
S
TF
f ( ) z
1
=
N
TF
f ( ) 1 z
1
( )
2
=
SNR
max
6.02N 1.76 12.9 50 OSR ( ) log + + =
OSR 15 dB
slide 21 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Noise Transfer-Function Curves
Out-of-band noise increases for high-order
modulators
Out-of-band noise peak controlled by poles of noise
transfer-function
Can also spread zeros over band-of-interest
0 f
0
f
s
f
s
2
----
N
TF
f ( )
First-order
Second-order
No noise shaping
f
slide 22 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Example
90 dB SNR improvement from A/D with
Oversampling with no noise shaping
From before, straight oversampling requires a
sampling rate of GHz.
First-Order Noise Shaping
Lose 5 dB (see (15)), require 95 dB divided by 9 dB/
octave, or octaves MHz
Second-Order Noise Shaping
Lose 13 dB, required 103 dB divided by 15 dB/
octave, (does not account for reduced
input range needed for stability).
f
0
25 kHz =
54,000
10.56 f
s
2
10.56
2f
0
75 =
f
s
5.8 MHz =
slide 23 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Quantization Noise Power of 1-bit Modulators
If output of 1-bit mod is , total power of output
signal, , is normalized power of 1 watt.
Signal level often limited to well below level in
higher-order modulators to maintain stability
For example, if maximum peak level is , max
signal power is 62.5 mW.
Max signal is approx 12 dB below quantization noise
(but most noise in different frequency region)
Quantization filter must have dynamic range capable
of handling full power of at input.
Easy for A/D digital filter
More difficult for D/A analog filter
1
y n ( )
1
0.25
y n ( )
slide 24 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Zeros of NTF are poles of H(z)
Write as
(19)
NTF is given by:
(20)
If poles of are well-defined then so are zeros of
NTF
H z ( )

x n ( )
u n ( )
y n ( )
Quantizer
H z ( )
H z ( )
N z ( )
D z ( )
---------- =
NTF z ( )
1
1 H z ( ) +
--------------------
D z ( )
D z ( ) N z ( ) +
--------------------------- = =
H z ( )
slide 25 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Error-Feedback Structure
Alternate structure to interpolative
Signal transfer-function equals unity while noise
transfer-function equals
First element of equals 1 for no delay free loops
First-order system
More sensitive to coefficient mismatches
G z ( ) 1
y n ( )
e n ( )
x n ( )
u n ( )

G z ( )
G z ( )
G z ( ) 1 z
1
=
slide 26 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Architecture of Delta-Sigma A/D Converters
Sample-
and-
Digital
filter
hold
low-pass
Mod

x
sh
t ( )
x
lp
n ( )
x
s
n ( ) x
c
t ( )
f
s
f
s
f
s
2f
0
x
dsm
n ( )
OSR
Anti-
aliasing
filter
x
in
t ( )
Analog
Digital
Decimation filter
x
c
t ( )
t f
X
c
f ( )
x
sh
t ( )
f
X
sh
f ( )
f
s
f
0
f
s
f
0
Time
Frequency
slide 27 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Architecture of Delta-Sigma A/D Converters
Time
Frequency
x
lp
n ( )
n

X
lp
( )
x
s
n ( )
n

X
s
( )
2
2f
0
f
s

2 4 6 8 10 12
1
2 3
123

n
x
dsm
n ( ) 1.000000 =
12
3
4

X
dsm
( )
2 2f
0
f
s

slide 28 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Architecture of Delta-Sigma A/D Converters
Relaxes analog anti-aliasing filter
Strict anti-aliasing done in digital domain
Must also remove quantization noise before
downsampling (or aliasing occurs)
Commonly done with a multi-stage system
Linearity of D/A in modulator important results in
overall nonlinearity
Linearity of A/D in modulator unimportant (effects
reduced by high gain in feedback of modulator)
slide 29 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Architecture of Delta-Sigma D/A Converters
Interpolation
filter
(low-pass)
Mod

1-bit
D/A
Analog
filter
low-pass
Analog Digital
x
s
n ( )
x
s2
n ( ) x
lp
n ( ) x
dsm
n ( ) x
da
t ( ) x
c
t ( )
2f
0
f
s
f
s
f
s
OSR
f
s
2f
0
--------
OSR
n

X
s
( )
X
s2
( )
x
s
n ( ) x
s2
n ( )
2 4 6 8 10 12
2 2f
0
( ) f
s

123
1 ( )
2 ( ) 3 ( )

Time
Frequency
slide 30 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Architecture of Delta-Sigma D/A Converters
x
lp
n ( )
x
dsm
n ( ) x
da
t ( )
x
c
t ( )
n
n t ,
t
f
f

X
lp
( )
X
dsm
( )
X
da
f ( )
X
c
f ( )
Time
Frequency
f
s
f
s
f
0
f
0
2 2f
0
( ) f
s

2 2f
0
( ) f
s

123
slide 31 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Architecture of Delta-Sigma D/A Converters
Relaxes analog smoothing filter (many multibit D/A
converters are oversampled without noise shaping)
Smoothing filter of first few images done in digital
(then often below quantization noise)
Order of lowpass filter should be at least one order
higher than that of modulator
Results in noise dropping off (rather than flat)
Analog filter must attenuate quantization noise and
should not modulate noise back to low freq strong
motivation to use multibit quantizers
slide 32 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Multi-Stage Digital Decimation
Sinc filter removes much of quantization noise
Following filter(s) anti-aliasing filter and noise
FIR filter Sinc
L 1 + IIR filter
Rate f
s
=
Rate 8f
0
= Rate 2f
0
=

modulator
Lth-order
T
sinc
z ( ) T
IIR
z ( )
FIR filter Sinc
L 1 + Halfband FIR filters
Rate f
s
= Rate 8f
0
= Rate 2f
0
=

modulator
Lth-order
T
sinc
z ( )
Rate 4f
0
=
H
1
z ( ) H
2
z ( ) H
3
z ( )
Sinc compensation
FIR filter
Rate 2f
0
=
slide 33 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Sinc Filter
is a cascade of averaging filters
Averaging filter
(21)
is integer ratio of
It is a linear-phase filter (symmetric coefficients)
If is power of 2, easy division (shift left)
Can not do all decimation filtering here since not
sharp enough cutoff
sinc
L+1
L+1
T
avg
z ( )
Y z ( )
U z ( )
----------
1
M
---- - z
i
i 0 =
M 1

= =
M f
s
8f
0
( )
M
slide 34 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Sinc Filter
Consider applied to
averaging filters in cascade



Converging to sequence of all 1/3 as expected
x
in
n ( ) 1 1 1 1 1 1 , . . . , , , , , { } =
M 4 =
T
avg
z ( ) T
avg
z ( ) T
avg
z ( )
x
in
n ( )
x
1
n ( ) x
2
n ( )
x
3
n ( )
sinc
3
filter
x
1
n ( ) 0.5 0.5 0.0 0.5 0.5 0.0, . . . , , , , , { } =
x
2
n ( ) 0.38 0.38 0.25 0.38 0.38 0.25, . . . , , , , , { } =
x
3
n ( ) 0.34 0.34 0.31 0.34 0.34 0.31, . . . , , , , , { } =
slide 35 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Sinc Filter Response
Can rewrite averaging filter in recursive form as
(22)
and a cascade of averaging filters results in
(23)
Use cascade to roll off quantization noise faster
than it rises in th order modulator
T
avg
z ( )
Y z ( )
U z ( )
----------
1
M
---- -
1 z
M

1 z
1

-----------------



= =
L 1 +
T
sinc
z ( )
1
M
L 1 +
--------------
1 z
M

1 z
1

-----------------



L 1 +
=
L 1 +
L
slide 36 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Sinc Filter Frequency Response
Let
(24)
where
z e
j
=
T
avg
e
j
( )
sinc
M
2
---------


sinc

2
----


------------------------- =
sinc x ( ) x ( ) sin x
2 0

T
avg
e
j
( )
1
slide 37 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Sinc Implementation
(25)
If 2s complement arithmetic used, wrap-around okay
since followed by differentiators
T
sinc
z ( )
1
1 z
1

----------------


L 1 +
1 z
M
( )
L 1 +
1
M
L 1 +
-------------- =
z
1
z
1
z
M
z
M f
s
M
---- -
Out
In
z
1
z
1 f
s
M
---- -
z
1
z
1
(Operate at low clock rate)
Out
In
(Operate at high clock rate)
(Integrators)
(Differentiators)

f
s
M f
s
M
f
s
f
s
slide 38 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Higher-Order Modulators
An Lth order modulator improves SNR by
6L+3 dB/octave
Interpolative Architecture
Can spread zeros over freq of interest using
resonators with and
Need to worry about stability (more later)

u n ( )
a
1
a
2
a
3
a
4
c
1
c
2
c
3
f
1
f
2
c
4
c
5
a
5
f
1
f
2
slide 39 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
MASH Architecture
Multi-stAge noise SHaping - MASH
Use multiple lower order modulators and combine
outputs to cancel noise of first stages
z
1
1-bit D/A
z
1
1-bit D/A
z
1
z
1
1
2
e
1
e
2
y n ( )
u n ( )
e
1
n ( )
z
1
u n ( ) 1 z
1
( )e
1
n ( ) +
z
1
e
1
n ( ) 1 z
1
( )e
2
n ( ) +
Analog
Digital
slide 40 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
MASH Architecture
Output found to be:
(26)
Multibit Output
Output is a 4-level signal though only single-bit D/As
if D/A application, then linear 4-level D/A needed
if A/D, slightly more complex decimation
A/D Application
Mismatch between analog and digital can cause first-
order noise, , to leak through to output
Choose first stage as higher-order (say 2nd order)
Y z ( ) z
2
U z ( ) 1 z
1
( )
2
E
2
z ( ) =
e
1
slide 41 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Bandpass Oversampling Converters
Choose to have high gain near freq
NTF shapes quantization noise to be small near
OSR is ratio of sampling-rate to twice bandwidth
not related to center frequency
H z ( ) f
c
f
c
z-plane
Bandpass
f
s
2
f
s
4 MHz =
dc
f
s
4 1 MHz =
zero
f

10 kHz =
z-plane
f
s
4 MHz =
dc
f
s
4 1 MHz =
zero
f
0
10 kHz =
f
s
2
Lowpass
f

2f
o
OSR
f
s
2f
0
-------- 200 = =
OSR
f
s
2f

--------- 200 = =
slide 42 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Bandpass Oversampling Converters
Above has poles at (which are zeros of NTF)
is a resonator with infinite gain at

Note one zero at and one zero at
similar to lowpass first-order modulator
only 9 dB/octave
For 15 dB/octave, need 4th order BP modulator

u n ( )
y n ( )
Quantizer
z
1
z
1
H z ( )

H z ( ) j
H z ( ) f
s
4
H z ( ) z z
2
1 + ( ) =
+j -j
slide 43 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Modulator Stability
Since feedback involved, stability is an issue
Considered stable if quantizer input does not
overload quantizer
Non-trivial to analyze due to quantizer
There are rigorous tests to guarantee stability but
they are too conservative
For a 1-bit quantizer, heuristic test is:
(27)
Peak of NTF should be less than 1.5
Can be made more stable by placing poles of NTF
closer to its zeros
Dynamic range suffers since less noise power
pushed out-of-band
N
TF
e
j
( ) 1.5 for 0
slide 44 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Modulator Stability
Stability Detection
Might look at input to quantizer
Might look for long strings of 1s or 0s at comp output
When instability detected ...
reset integrators
Damp some integrators to force more stable

2
2f
0
f
s

NTF
less stable
more stable
same area under two curves
slide 45 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Linearity of Two-Level Converters
For high-linearity, levels should NOT be a function of
input signal
power supply variation might cause symptom
Also need to be memoryless
switched-capacitor circuits are inherently
memoryless if enough settling-time allowed
Above linearity issues also applicable to multi-level
A nonreturn-to-zero is NOT memoryless
Return-to-zero is memoryless if enough settling time
Important for continuous-time D/A
slide 46 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Linearity of Two-Level Converters
A
1

1
+ A
1
1 1 1 1 1 1 1
A
0

2
+ A
0
V
2
V
1
Binary
Area for
symbol
A
1

1
+ A
1

1
+ A
0

2
+
Ideal
Typical
Ideal
Typical
A
1
A
1
1 1 1 1 1 1 1
A
0
V
2
V
1
Binary
Area for
symbol
A
0
A
1
A
1
A
0
Nonreturn-to-zero (NRZ)
Return-to-zero (RTZ)
slide 47 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Idle Tones
1/3 into 1st order modulator results in output
(28)
Fortunately, tone is out-of-band at
into modulator has tone at
Similar examples can cause tones in band-of-interest
and are not filtered out say
Also true for higher-order modulators
Human hearing can detect tones below noise floor
Tones might not lie at single frequency but be short
term periodic patterns.
could be a tone varying between 900 and 1100 Hz
varying in a random-like pattern
y n ( ) 1 1 1 1 1 1 1 1, . . . , , , , , , , { } =
f
s
3
1 3 1 24 + ( ) 3 8 = f
s
16
f
s
256
slide 48 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Dithering
Add pseudo-random signal into modulator to break
up idle tones (not just mask them)
If added before quantizer, it is noise shaped and
large dither can be added.
A/D: few bit D/A converter needed
D/A: a few bit adder needed
Might affect modulator stability
H z ( )

u n ( )
y n ( )
Quantizer
Dither signal
slide 49 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Opamp Gain
Finite opamp gain, , moves pole at left by
Flattens out noise at low frequency
only 3 dB/octave for high OSR
Typically, require
(29)
A z 1 =
1 A
z-plane
zero
1 A

2
NTF
low opamp gain
A
2f
0
f
s

A OSR >
slide 50 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Multi-bit Oversampled Converters
A multi-bit DAC has many advantages
more stable - higher peak
higher input range
less quantization noise introduced
less idle tones (perhaps no dithering needed)
Need highly linear multi-bit D/A converters
Example
A 4-bit DAC has 18 dB less quantization noise, up to
12 dB higher input range perhaps 30 dB improved
SNR over 1-bit
Large Advantage in DAC Application
Less quantization noise easier analog lowpass
filter
NTF
slide 51 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Multi-bit Oversampled Converters
Randomize thermometer code
Can also shape nonlinearities
C
C
C
C
C
C
C
C
E
i
g
h
t
-
l
i
n
e

r
a
n
d
o
m
i
z
e
r
T
h
e
r
m
o
m
e
t
e
r
-
t
y
p
e

d
e
c
o
d
e
r
b
1
b
2
b
3
Analog
output
slide 52 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Third-Order A/D Design Example
All NTF zeros at
(30)
Find such that
Use Matlab to find a Butterworth highpass filter with
peak gain near 1.4
If passband edge at then peak gain = 1.37
(31)
z 1 =
NTF z ( )
z 1 ( )
3
D z ( )
------------------ =
D z ( ) NTF e
j
( ) 1.4 <
f
s
20
NTF z ( )
z 1 ( )
3
z
3
2.3741z
2
1.9294z 0.5321 +
-------------------------------------------------------------------------------- =
slide 53 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Third-Order A/D Design Example
Find as
(32)
(33)
z-plane
1
1
j
j
0 =
2 =
=
Three zeros at
Butterworth poles

H z ( )
H z ( )
1 NTF z ( )
NTF z ( )
---------------------------- =
H z ( )
0.6259z
2
1.0706z 0.4679 +
z 1 ( )
3
--------------------------------------------------------------------- =
slide 54 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Third-Order A/D Design Example
Choosing a cascade of integrator structure
coefficients included for dynamic-range scaling
initially
last term, , initially set to so input is stable for
a reasonable input range
Initial found by deriving transfer function from 1-bit
D/A output to and equating to
1-bit D/A

u n ( )
y n ( )
Quantizer
z
1
z
1
z
1

1

2

3

1

2

3
V
3
V
2
V
1
Analog
Digital

2

3
1 = =

1

1

i
V
3
H z ( )
slide 55 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Third-Order A/D Design Example
(34)
Equating (33) and (34) results in
(35)
H z ( )
z
2

1

2

3
+ + ( ) z
2
2
3
+ ( )
3
+
z 1 ( )
3
--------------------------------------------------------------------------------------- =

1
0.0232 = ,
2
1.0 = ,
3
1.0 =

1
0.0232 = ,
2
0.1348 = ,
3
0.4679 =
slide 56 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Third-Order A/D Design Example
Dynamic Range Scaling
Apply sinusoidal input signal with peak value of 0.7
and frequency
Simulation shows max values at nodes of
0.1256, 0.5108, and 1.004
Can scale node by by multiplying and by
and dividing by
Can scale node by by multiplying and
by and dividing by
(36)
256 rad/sample
V
1
V
2
V
3
, ,
V
1
k
1

1

1
k
1

2
k
1
V
2
k
2

2
k
1

2
k
2

3
k
2

1
0.1847 = ,
2
0.2459 = ,
3
0.5108 =

1
0.1847 = ,
2
0.2639 = ,
3
0.4679 =
slide 57 of 57
University of Toronto
D.A. Johns, K. Martin, 1997
Third-Order A/D Design Example

2a

1a

1a

1a
+V
DAC
+V
DAC V
DAC

V
DAC

2a

2a

1a

2a

2a

1
1

3
C
A
1 =
C
A
1 =
C
A
1 =
C
A
1 =
C
A
1 =
C
A
1 =

1
+
V
i n

+
V
3

+
V
3

1
+
V
ref

+V
DAC
V
DAC