Chapter 9

System Considerations of Pre-filtering & Post-filtering
 Introduction
 Spectra of Sampled Signals
 Prefiltering
 Postfiltering
 Oversampling Approach
 Conventional Switched-Capacitor Approach
 Examples
Prof. Tai-Haur Kuo, EE, NCKU, Tainan City, Taiwan 9-1 郭泰豪, Analog IC Design, 2018
Filters
 Continuous-time filter
 RLC passive
 Active RC
 Sampled-Data filter
 Switched-Capacitor filter
 Digital filter
Continuous-Time Filters
 Example: 1 pole low pass filter
 Passive
R1 L  
Vout RL  1 
Vin RL   
Vout Vin R 1  R L 1 sL 
 R 1  R L 
 Active
RFB
C Vout R FB  1 
Rin   
Vin  Vin R in 1 sR FBC 

Vout
 Equivalence conditions of the above

RR L
Rin  R1 , RFB  1 L , C 
R1  RL R1RL
Switched-Capacitor Filter (SCF)
 Basic concept
V2 V1 1
1 C 2 I eff 2
I eff
V2 V1
R eff
Q V2  V1 T
Q  C(V2  V1 )  I eff    R eff 
T R eff C
SCF (Cont.)
 Example: SC integrator (stray-sensitive)
CI
v out n   v out n  1   v in n  1
CS
CI
v out n   v out n  1  S v in n  1
C vin n 
 v out n 
CI CS
Vout z   z 1Vout z   S z 1Vin z 
C

CI
Vout z  CS z 1
 Hz    where z  e jT
Vin z  C I 1  z 1
If ωT<<1 
  
 jT 
2
      

1 j T 
 
 jT
C e CS  2 
lim H e jT  S  
T 1 C I 1  e  jT CI
jT 
 jT 
2
 
2
CS 1 1 1
  
C I jT T jR eff C I

j C I
 CS 
Digital Filter
 FIR (Finite Impulse Response)
x n  z 1 z 1 z 1 z 1
a1 a2 an1
a0
yn 
 IIR (Infinite Impulse Response)
x n  yn 
z 1
 Operations
a1 b1
 Multiply
 Delay
 Add
z 1
a2 b2
Analog Sampling Circuit
f * t  f n* t 
S1 f t  (Single pulse)
K K
K

f t  …
S2 f t  t 1 t 2
*
t
T 2T nT
 τ
T
T
S1

S2
t1 t 2
 Original continuous-time signal f(t)

 Sampled/Held Signal f*(t)
Analog Sampling Circuit (Cont.)
 Mathematical relationship (assume K=1) f t  f * t 
f n* t 
 Single-pulse signal f n t 
* (Single pulse)
f nT  …
 f n t  
*
 t  nT    t  nT  τ 
τ t
1, t  0 τ T 2T nT
where  t    (13.3)
0, t  0
T
 Sampled signal for all time

 
f nT 
 f t  
*
 f t  
*
n   t  nT    t  nT  τ  (13.4)
n   n   τ
 Laplace transform for fn*(t) and f *(t)
   sτ

 Fn s    f nT e snT
* 1 1 e (13.5)
τ s 
   sτ
  
 F s      f nT e   f nT e snT for τ → 0
* 1 1 e snT
(13.7)
τ  s n -  n - 
Signal Spectra of Zero-Width Samples
 For τ → 0 (zero-width sampling)
f (t) F( j) original signal
f * (t) F*  jω  sampled-data signal
 Spectrum of Sampled Signal
 Method1: replace s  jω in (13.7)
 Method2: Convolution
Fourier
F jω
F  jω  F jω  S jω
1
f t   f t  st 
*
transform *
2π FB  j
 FA  j
2π   2π 
where st    δt  nT  and S jω    δ ω  k  A B

n   T n    T 
F*  jω  
FB  jω 
*

 2π  T
 F*  jω  
1
 F
T k   
jω  jk
T

ω
 Continuous-time signals fAt  and fBt 

FA  jω 
2 *
T T
 Sampled-Data signals fA * t  and fB* t 
Nyquist Theorem
 There is a one-to-one relation between values
FA B  j and
FA B  j

 Replicas forming FB *  j overlap

(This phenomenon is called aliasing or folding.
It is a nonlinear distortion.)
 Low-pass Filter


1 ;
FA  jH j  FA  j H  j   
 T
where 
0 ; 
T
The continuous-time signal f A t  is recovered.


  from B  j

But no such operation can regain B F j F
 Nyquist first observed this phenomenon
 Nyquist Theorem : fsampling > 2fsignal band
S/H Effect
 Due to nonzero-width samples, assume τ = T and from (13.5)
 FSH  jω  HSH  jωF  jω , where F*  jω  is spectra of zero-width sample
*
 Spectrum of S/H response

1  e sT
H SH s  
s
Let s  jω
 T 
 jT jT  sin  sin(πf )
1 e 
or H SH f   T
H SH  j 
fs
 Te  2 
πf 
2
j  T 
  fs
 2  H ( j)
SH

6

4

2 2 4 6

T T T T T T
 3f s  2f s  fs fs 2f s 3f s f
 HSH  j has a linear phase. Its amplitude has sinc response (sinX/X)
S/H Effect (Cont.)
 FSH  jω  HSH  jωF  jω
*
 F*  jω  is formed by replicas of F j

 FSH  jω  is replicated and multiplied by the sinX/X response.
f * t  F jω
f t 
K

… t F  j 
*
τ T 2T nT
T 
f sh t 
f t  HSH( j)
K
… 
t
T 2T nT FSH  j
T

2 2

T T
 S/H before an ADC Reduced high-frequency images
 Allow ADC to have a constant input value during one conversion
 Relax the anti-aliasing requirement
Derivation of S/H Function HSH (s)
 S/H response (refer to p.9-8)
f t  f t  f t  (Single pulse)
* *
n
f t  f sh t  (Single pulse)
… …
t t
τ T 2T nT T 2T nT
T T
Sample Hold

f nT 
 For sampled signal: f t   *
  t  nT    t  nT  τ  (13.4)
n   τ
 Assume τ = T (Nonzero-width sampling)

 f sh t    f nT  t  nT    t  nT  T 
n  
(13.39)
1  e sT 
1  e -sT *
 Laplace transform: FSH s    f nT e F s 
snT
 (13.40)
s n   s
 Sample/Hold transfer function, HSH (s), is equal to
1  e sT
H SH s   (13.41)
s
Sampled-Data System with Continuous-time Input &
Output Signals
 The distortion due to HSH( j) is linear as opposed to the nonlinear
distortion which aliasing introduces.
 F( j) can be recovered from FSH( j) by two steps

 Low-pass filter
 Amplitude equalizer with a transfer function
HEQ j  1
HSH j
Sampled-Data System with Continuous-time Input &
Output Signals (Cont.)
Sampled-data
system
Input Continuous-time Continuous-time

Amplitude
Output
anti-aliasing or smoothing
equalizer
filter filter
A/D DSP D/A
 The anti-aliasing and smoothing filters can be identical lowpass filters,

and should ideally have sharp cutoff frequency (except oversampling
rate signal processing where Decimation and interpolation are used) .
Prefiltering
 Nyquist rate
 Prefilter = Anti-alias filter (AAF)
 Brick wall AAF
 Oversampling rate
 Prefilter = Anti-alias filter + Decimation filter
(AAF) (DF)
AAF : Continuous-time filter
DF : SCF or Digital filter
 
Usually < 60 dB(or 10 bit) Usually > 60dB (or 10bit)
Small size (formerly) Large size (formerly)

Large size (advanced process) Small size (advanced process)
Prefiltering (Cont.)
 Examples: (For Data Acquisition)
 Conventional Nyquist-rate A/D converter
AAF SCF A/D

nfc fc
 Modern high-speed design with Nyquist-rate A/D converter
Auto-tuning
A/D
AAF
fc
 Oversampling A/D converter
AAF A/D DF
mfc fc
Prefilter Strategy for Conventional Data Acquisition
Continuous-time Switched-capacitor A/D
Lowpass filter Lowpass filter converter
Analog Digital
input fSCF fADC output
 Provide anti-aliasing for ADC with SCF sampling at a frequency much

greater than twice the bandwidth of interest.
 Provide anti-aliasing for SCF by continuous-time filter with corner
frequency comfortably between fSCF2 and fBW
fCTLPF
SCF
fBW fADC f S CF fSCF

2
 Since fSCF >> fADC => decimation occurs without aliasing effect since
SCF performs anti-aliasing filter.
 Hence, ADC performs decimation and A/D conversion.
 Sometimes, SCF instead of ADC performs filtering and decimation. For
either way, control clocks had better be synchronized.
Conventional Postfiltering
Digital SC
Smoothing filter Analog
DAC Lowpass
Input ( & equalizer ) Output
filter
fC1 fC2
Smoothing filter
SCF
fBW fC1 fC2
Some discrete-time sinusoidal signals
with different sampling rates
 0 cycles/sample  1/16 cycles/sample
x n  x n 
10 15
1 5 10 15 n 1 5 n
 1/8 cycles/sample  1/4 cycles/sample

x n  x n 
5 15 15
1 10 n 1 5 10 n
Comparison time and frequency of two sampling rates
X S f , X1 
x n 
x c t 
 x s t 
0.25 0.5 t 1 Hz 4 Hz 8 Hz
f
1 2 n  (0.5π) (2π) (4π)  
XS2 f , X2 
x n 
x c t 
 x s t 
0.25 0.5
t 1 Hz 12 Hz 24 Hz
f
3 6 n 
  (2π) (4π)  
 
6
Use of Oversampling Approach to
Relax Requirements of Prefilter and Postfilter
 Front End
 Use oversampling A/D converter
 Use decimation after A/D conversion
 Back End
 Use interpolation before D/A conversion
 Use oversampling D/A converter
Use of Oversampling Approach to Relax
Requirements of Prefilter and Postfilter (Cont.)
 Example: Block diagram of a signal processing system
Simple
A/D Sampling rate
xc t  xd n
Anti-aliasing
filter xa t  Conversion x̂n reduction by M
Discrete-time
xd n system yd n
yd n Sampling rate ŷn D/A yDA t  Simple yˆ c t 

reconstruction
Increase by L converter
filter
Use of Decimation in A/D Conversion
X c  j 
Signal
High-frequency Simple
1 anti-aliasing filter
noise
 c  N N c

1 X a  j 
Filtered Signal
noise
 c  N N c 
1
T
1
 
X̂ e j Sharp cutoff
Decimation
Aliased noise filter
  T
 2  N  2
N   NT 
M
 
1
T X d e j T   MT
  T 
 2   2
Use of Interpolation in D/A conversion
1
 
Y e j T   MT
T
 2   2   T 
 
Yˆ e j
Compensated T   T   MT
1 L interpolation L L
filter
T
 2 
   2 ω  Ω T 
L L 2 
Simple L
H r  j reconstruction
filter
2
s   N
T
 S
1
Yr  j
S

 N N

Decimation
 Lowpass filtering + downsampling
X n  Low-Pass Down-
Y n 
Filter Sampler
f s1 f S2
 Downsampling (by 3)
 Time-domain
n n
M
0 3 6 0 1 2
L3
 Frequency domain
  

2 2
6 2
Decimation (Cont.)
 Downsampling with aliasing
 Spectrum of original continuous-time signal

 Spectrum of sampled signal
 2  
2  2   T
 Spectrum after decimation

1 ( M  3)
MT
 2  3 3 2   T 
2 2
Decimation (Cont.)
 Downsampling with prefiltering to avoid aliasing
 Low-pass filtering


 2   T '
M 3
 Spectrum after filtering

3
2   T '
 Spectrum after decimation
1
MT
 2   T '
Interpolation
 Upsampling + lowpass filtering
 Time-domain
Y n 
Low Pass
X n  Upsampler
Filter
f S1 f S1
 Upsampling (by 3)
 Time-domain
n n
L
0 1 2 0 3 6
L3
 Frequency-domain

 

2 2
2 6
Interpolation (Cont.)
 Baseband signal
(Analog)
 Original signal
(Digital)
2   T
 Upsampling 1
L  3
T
 Low-pass filtering 2π   T 
2   T 
 Spectrum after filtering
2   T 
Decimate-By-N Filter
 Example: with sinc function
X Decimation Y
fC  filter  fC 
 N
 
Input rate f c
Output rate f N  f d c
Y(n)=(1/N)[x(Nn)+x(Nn-1)+x(Nn-2)+……+x(Nn-N+1)]
x z  N 1 i x z  1  z  N 
Yz    Z   
N 0 N  1  z 1 
Decimate-By-N Filter (Cont.)
 Gain of sinc filter
N
N sin
1 1 z 1 2 
1
 ; z  e j  and
N 1  z 1 N sin  fc
2
N
sinc sinX
 2 ; sincX 
 X
sinc
2
 Example: N=6
Gain
f
- 2f d  f d 0 fd 2 fd 3 fd 4 fd 5 fd 6 fd 7 fd 8 fd 9f d
 1   fc 
  fc 
 6 
Digital Decimator with Sinc Filtering
X 
 
 Y
 
REG REG REG REG
fc fc fc fc
N stages
 
f d   f c 

N
 
 
X 
Y

 
REG REG
fc fd
1 1 Z N
1  z 1
X Y
fc fd
Lowpass Filter Decimator
SC Sampling Stage Without Decimation
 Example: Integrator
2 C
1 C
2
V1
V2
2 1
1
2
KT  T T T
KT  KT KT 
2 2
 T
V2 kT   V2 kT  T   V1  kT  
 2
1

V2 z  z 2
H I z   
V1 z  1  z 1
SC Sampling Stage With Decimation
 Example: Modified Integrator
1
2
2
V1 V2
2 1
1
 fc 
2
 f d  2
N 6
SC Sampling Stage With Decimation (Cont.)
Hz  HI z HD z 
where H D z  is a sinc transfer function
N
sin
HD e  
j 

2 1
;  T
fc
2
 Example : N=6
f
0 fd 2 fd 3 fd 4 fd 5 fd 6 fd
Input
f
Output
f
Cosine Filter
 For N=2
sin 
 
1
HI e j 
 ; 
 fc
sin
2
f
0 fd 2 fd 3 fd 4 fd 5 fd
 f c 
Interpolation Filter
 Example: Linear Interpolation (sinc function)
X Y
 fC  Interpolation  f I   Nf C
Nf C  f I
X   Y
 
 
REG REG
fc Nf c
1 1

fC 1  z 1 
1
1 z N
1
 1 zN 1
Nf C
1  z 1
N
1 z N sin 1 N 6
; H I z  
1
H I z   2 ; 
1  z 1 N sin  fI f
0 fc 2 fc 3 fc 4 fc 5 fc 6 fc 7 fc
2
 fI
SC Sampling Stage With Linear Interpolation
1 C1 Ca
2 C2


V1 V2
3 C3
 3 3
i
1 2 1 2
t
T2 T2 T2
T1  rT2
V1 ,V2
V1
6T2 V2
t
T2 3T2
V2 z  1 1  z  N 1 N 1 i
V2 nT2   V2 nT2  T2   V1 nT2   V1 nT2  NT2 
1 H I z     z
N V1 z  N 1  z 1 N i 0
N
sin
 
H I e j  
1 2
N sin 
; 
1
f
 T2
2
Sinc Filtering Function
 Not ideal lowpass
 FIR (Finite Impulse Response)
ideal lowpass filter
sinc filter
 To implement ideal lowpass function, other approaches can be used.
Changing The Sampling Rate By A Noninteger Factor
Interpolator Decimator
Lowpass Filter Lowpass Filter
xn L ~ Gain  l Gain  l M xd n 

x n  Cutoff 

Cutoff 
 ~
x I ( n)
L M
Sampling
xI n
Period :T T/L T/L T/L TM/L
Two lowpass filters
can be combined
Lowpass Filter
L ~ Gain  L
~x n  M
xn x n    
Cutoff  min , 
L M 
I xd n 
Sampling
: T T/L T/L TM/L
Period
 Noninteger factors can be obtained from properly choosing M and L .
Examples
 Talking Back
Storage
Anti-alias Smoothing
filter ADM D/ A filter
xdb
f BW fS
2
 Speech Reconstruction
Storage
Anti-alias A/D Digital D/A Smoothing

filter converter filter converter filter
Examples (Cont.)
 Approach 1:
Better LPF can be used other than sinc one
continuous-time + oversampling + digital + DAC + SMF
AAF ADC Decimator
and LPF
 Oversampling ADC and digital decimator may be combined.
 Approach 2:
continuous-time + SCF + conventional + DAC + SMF
AAF ADC
(Nyquist rate)

Chapter 9

Uploaded by

Copyright:

Available Formats

Chapter 9

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 9

Uploaded by

Copyright:

Available Formats

System Considerations of Pre-filtering & Post-filtering

 Spectra of Sampled Signals

 Conventional Switched-Capacitor Approach

 Equivalence conditions of the above

 Original continuous-time signal f(t)

 Sampled signal for all time

 Sampled-Data signals fA * t  and fB* t 

 Replicas forming FB *  j overlap

The continuous-time signal f A t  is recovered.

 Spectrum of S/H response

 F*  jω  is formed by replicas of F j

 F( j) can be recovered from FSH( j) by two steps

Input Continuous-time Continuous-time

A/D DSP D/A

 The anti-aliasing and smoothing filters can be identical lowpass filters,

Small size (formerly) Large size (formerly)

AAF SCF A/D

 Provide anti-aliasing for ADC with SCF sampling at a frequency much

fBW fADC f S CF fSCF

fBW fC1 fC2

 1/8 cycles/sample  1/4 cycles/sample

1 2 n  (0.5π) (2π) (4π)  

yd n Sampling rate ŷn D/A yDA t  Simple yˆ c t 

 Spectrum after decimation

ideal lowpass filter

 To implement ideal lowpass function, other approaches can be used.

xn L ~ Gain  l Gain  l M xd n 

Anti-alias A/D Digital D/A Smoothing

You might also like