MEHRAN UNIVERSITY OF ENGINEERING & TECHNOLOGY, JAMSHORO

INSTITUTE OF INFORMATION & COMMUNICATION TECHNOLOGIES

ADVANCED DIGITAL SIGNAL PROCESSING
Lab 10: Multirate Digital Signal Processing

Objective 2 Decimation
The objective of this lab is to alter the sampling rate of The basic operation required in decimation is the
signal. downsampling of the high-rate signal into a low-
rate signal. MATLAB provides the function [y] =
downsample(x,D) that downsamples input array x into
output array y by keeping every Dth sample starting
1 Introduction with the first sample. An optional third parameter
“phase” specifies the sample offset which must be an
In many practical applications of digital signal pro- integer between 0 and (D-1). For example,
cessing, one is faced with the problem of changing the
sampling rate of a signal, either increasing it or decreas- ≫ x = [1,2,3,4,3,2,1]; y = downsample(x,2)
ing it by some amount. The process of converting a sig-
nal from a given rate to a different rate is called sampling y =
rate conversion. In turn, systems that employ multiple 1 3 3 1
sampling rates in the processing of digital signals are downsamples by a factor of 2 starting with the first
called multirate digital signal processing systems. sample. However,
Let us think of an underlying (or original) analog sig-
nal xa (t) that was sampled using the sampling rate of ≫ x = [1,2,3,4,3,2,1]; y = downsample(x,2,1)
Fs = T1 samples/second to produce a discrete signal
y =
x(n). The resulting digital signal x(n) is subsequently
2 4 2
filtered using a lowpass filter (LPF) with a cutoff fre-
quency of ωc . produces an entirely different sequence by down-
Thus, the output signal y(n) has all its energy in the sampling, starting with the second sample (i.e., offset
band 0 ≤ ω ≤ ωc = 2πfc . According to the sampling by 1).
theorem, such a signal may be represented by the rate
of 2fc/T samples/second instead of its existing rate of 2.1 Example
Fs = T1 . Note that |fc | ≤ 0.5. However, if fc ≪ 0.5, then
2fc/T ≪ Fs . Hence it would seem more advantageous to Let x(n) = cos(0.125πn). Generate a large number of

lower the sampling frequency to a value closer to 2fc/T samples of x(n) and decimate them using D = 2, 4, and
and perform signal processing operations at this lower 8 to show the results of decimation.
rate. MATLAB provides the function y = decimate(x,D)
Other applications include the need for an optimal that resamples the sequence in array x at 1/D times the
interpolation in computer tomography and efficient original sampling rate. The resulting resampled array
multistage designs of narrowband lowpass filters. y is D times shorter i.e., length(y) = length(x)/D.
The MATLAB Signal Processing toolbox may be used Designing an ideal lowpass filter is not possible in the
to carry out a variety of operations associated with MATLAB implementation; however, fairly accurate ap-
sampling rate conversion and multirate processing. The proximations are used. The default lowpass filter used
key functions are listed below, in the function is an 8th-order Chebyshev type-I low-
pass filter with the cutoff frequency of 0.8π/D. Using
additional optional arguments, the filter order can be
1. decimate % resample data at a lower rate after low- changed or an FIR filter of specified order and cutoff
pass filtering frequency can be used.
We will plot the middle segments of the signals to
2. resample % change the sampling rate of a signal avoid end-effects due to the default lowpass filter in
the decimate function. The following MATLAB script
3. interp % resample data at a higher rate using low- shows details of these operations
pass interpolation clc , clear all , close all
n = 0:2048; k1 = 256; k2 = k1 +32; m = 0:( k2 - k1 )
4. upfirdn % upsample, apply a specified FIR filter, ;
Hf1 = figure ( ’ units ’ , ’ inches ’ , ’ position ’
and downsample a signal ,[1 ,1 ,6 ,4] , ...
’ paperunits ’ , ’ inches ’ , ’ p a p e r p o s i t i o n ’
,[0 ,0 ,6 ,4]) ;
Use on line MATLAB help to find out more about % ( a ) Original signal
these functions. Following examples illustrate the use of x = cos (0.125* pi * n ) ;
these functions. subplot (2 ,2 ,1) ;

1/4
 MEHRAN UNIVERSITY OF ENGINEERING & TECHNOLOGY, JAMSHORO
INSTITUTE OF INFORMATION & COMMUNICATION TECHNOLOGIES
ADVANCED DIGITAL SIGNAL PROCESSING
Lab 10: Multirate Digital Signal Processing

Ha = stem (m , x ( m + k1 +1) , ’g ’ , ’ filled ’ ) ; between 0 and (I-1). For example,

axis ([ -1 ,33 , -1.1 ,1.1]) ; set ( Ha , ’ markersiz e ’ ,2) ;
ylabel ( ’ Amplitude ’) ; title ( ’ Original Sequence x ≫ x = [1,2,3,4]; v = upsample(x,3)
( n ) ’) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’ v =
,[ -1 ,0 ,1]) ; 1 0 0 2 0 0 3 0 0 4 0
% ( b ) Decimatio n by D = 2
D = 2; y = decimate (x , D ) ; 0
subplot (2 ,2 ,2) ;
Hb = stem (m , y ( m + k1 / D +1) , ’ c ’ , ’ filled ’) ; axis upsamples by a factor of 2 starting with the first sample.
([ -1 ,33 , -1.1 ,1.1]) ; However,
set ( Hb , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ;
title ( ’ Decimated by D = 2 ’) ; ≫ v = upsample(x,3,1)
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’
,[ -1 ,0 ,1]) ; v =
% ( c ) Decimatio n by D = 4 0 1 0 0 2 0 0 3 0 0 4
D = 4; y = decimate (x , D ) ; subplot (2 ,2 ,3) ;
Hc = stem (m , y ( m + k1 / D +1) , ’ r ’ , ’ filled ’) ; axis
0
([ -1 ,33 , -1.1 ,1.1]) ;
set ( Hc , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ;
≫ v = upsample(x,3,2)
title ( ’ Decimated by D = 4 ’) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’ v =
,[ -1 ,0 ,1]) ; 0 0 1 0 0 2 0 0 3 0 0
xlabel ( ’ n ’) ; 4
% ( d ) Decimatio n by D = 8
D = 8; y = decimate (x , D ) ; subplot (2 ,2 ,4) ; produces two different signals by upsampling, starting
Hd = stem (m , y ( m + k1 / D +1) , ’ m ’ , ’ filled ’) ; axis
with the second and the third sample (i.e., offset by 1),
([ -1 ,33 , -1.1 ,1.1]) ;
set ( Hd , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ; respectively. Note that the lengths of the upsampled
title ( ’ Decimated by D = 8 ’) ; signals are I times the length of the original signal.
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’
,[ -1 ,0 ,1]) ;
xlabel ( ’ n ’) ; 3.1 Example
We observe that the decimated sequences for D = 2 Let x(n) = cos(πn). Generate samples of x(n) and in-
and D = 4 are correct and represent the original sinus- terpolate them using I = 2, 4, and 8 to show the results
oidal sequence x(n) at lower sampling rates. However, of interpolation.
the sequence for D = 8 is almost zero because the lowpass MATLAB provides the function [y,h] =
filter has attenuated x(n) prior to downsampling. Recall interp(x,I) that resamples the signal in array x
that the cutoff frequency of the lowpass filter is set to at I times the original sampling rate. The resulting
0.8π/D = 0.1π which eliminates x(n). If we had used the resampled array y is I times longer i.e., length(y) =
downsampling operation on x(n) instead of decimation, I*length(x). The ideal lowpass filter is approximated
the resulting sequence would be y(m) = 1, which is an by a symmetric filter impulse response, h, which is
aliased signal. Thus, the lowpass filtering is necessary. designed internally. It allows the original samples to
pass through unchanged and interpolates between so
that the mean square error between them and their
3 Interpolation ideal values is minimized. The third optional parameter
L specifies the length of the symmetric filter as 2*L*I+1,
An increase in the sampling rate can be accomplished and the fourth optional parameter cutoff specifies the
by interpolating new samples between successive values cutoff frequency of the input signal in π units. The
of the signal. This process can be accomplished in two default values are L = 5 and cutoff = 0.5. Thus, if
steps. The first step creates an intermediate signal at I = 2, then the length of the symmetric filter is 21 for
the high rate by interlacing zeros in between nonzero the default L = 5.
samples in an operation called upsampling. In the second We will plot the middle segments of the signals to
step, the intermediate signal is filtered to “fill in” zero- avoid end-effects due to the default lowpass filter in the
interlaced samples to create the interpolated high-rate interp function. The following MATLAB script shows
signal. details of these operations
MATLAB provides the function [v] = clc , clear all , close all
upsample(x,I) that upsamples input array x into n = 0:256; k1 = 64; k2 = k1 +32; m = 0:( k2 - k1 ) ;
Hf1 = figure ( ’ units ’ , ’ inches ’ , ’ position ’
output v by inserting (I-1) zeros between input ,[1 ,1 ,6 ,4] , ...
samples. An optional third parameter, “phase,” ’ paperunits ’ , ’ inches ’ , ’ p a p e r p o s i t i o n ’
specifies the sample offset, which must be an integer ,[0 ,0 ,6 ,4]) ;

2/4
 MEHRAN UNIVERSITY OF ENGINEERING & TECHNOLOGY, JAMSHORO
INSTITUTE OF INFORMATION & COMMUNICATION TECHNOLOGIES
ADVANCED DIGITAL SIGNAL PROCESSING
Lab 10: Multirate Digital Signal Processing

% ( a ) Original signal set ( gca , ’ xtick ’ ,[0 ,0.5 ,1]) ; set ( gca , ’ ytick ’
x = cos ( pi * n ) ; ,0:1: I ) ;
subplot (2 ,2 ,1) ; % ( b ) I n t e r p o l a t i o n by I = 4 , L = 5;
Ha = stem (m , x ( m + k1 +1) , ’g ’ , ’ filled ’ ) ; I = 4; [ y2 , h2 ] = interp (x , I ) ; H2 = freqz ( h2 ,1 , w
axis ([ -1 ,33 , -1.1 ,1.1]) ; set ( Ha , ’ markersiz e ’ ,2) ; ) ; H2 = abs ( H2 ) ;
ylabel ( ’ Amplitude ’) ; title ( ’ Original Sequence x subplot (2 ,2 ,2) ; plot ( w / pi , H2 , ’g ’ ) ; axis ([0 ,1 ,0 ,
( n ) ’) ; I +0.2]) ; ylabel ( ’ Magnitude ’) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’ title ( ’I = 4 , L = 5 ’ ) ;
,[ -1 ,0 ,1]) ; set ( gca , ’ xtick ’ ,[0 ,0.25 ,1]) ; set ( gca , ’ ytick ’
% ( b ) I n t e r p o l a t i o n by I = 2 ,0:1: I ) ;
I = 2; y = interp (x , I ) ; % ( c ) I n t e r p o l a t i o n by I = 8 , L = 5;
subplot (2 ,2 ,2) ; I = 8; [ y3 , h3 ] = interp (x , I ) ; H3 = freqz ( h3 ,1 , w
Hb = stem (m , y ( m + k1 * I +1) , ’ c ’ , ’ filled ’) ; axis ) ; H3 = abs ( H3 ) ;
([ -1 ,33 , -1.1 ,1.1]) ; subplot (2 ,2 ,3) ; plot ( w / pi , H3 , ’g ’ ) ; axis ([0 ,1 ,0 ,
set ( Hb , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ; I +0.4]) ; ylabel ( ’ Magnitude ’) ;
title ( ’ I n t e r p o l a t e d by I = 2 ’) ; title ( ’I = 8 , L = 5 ’ ) ; xlabel ( ’\ omega /\ pi ’ , ’
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’ fontsize ’ ,10)
,[ -1 ,0 ,1]) ; set ( gca , ’ xtick ’ ,[0 ,0.125 ,1]) ; set ( gca , ’ ytick ’
% ( c ) I n t e r p o l a t i o n by I = 4 ,0:2: I ) ;
I = 4; y = interp (x , I ) ; subplot (2 ,2 ,3) ; % ( d ) I n t e r p o l a t i o n by I = 8 , L = 7;
Hc = stem (m , y ( m + k1 * I +1) , ’ r ’ , ’ filled ’) ; axis I = 8; L = 7; [ y4 , h4 ] = interp (x ,I , L ) ; H4 =
([ -1 ,33 , -1.1 ,1.1]) ; freqz ( h4 ,1 , w ) ; H4 = abs ( H4 ) ;
set ( Hc , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ; subplot (2 ,2 ,4) ; plot ( w / pi , H4 , ’g ’ ) ; axis ([0 ,1 ,0 ,
title ( ’ I n t e r p o l a t e d by I = 4 ’) ; I +0.4]) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’ ylabel ( ’ Magnitude ’) ;
,[ -1 ,0 ,1]) ; title ( ’I = 8 , L = 7 ’ ) ; xlabel ( ’\ omega /\ pi ’ , ’
xlabel ( ’ n ’) ; fontsize ’ ,10)
% ( d ) I n t e r p o l a t i o n by I = 8 set ( gca , ’ xtick ’ ,[0 ,0.125 ,1]) ; set ( gca , ’ ytick ’
I = 8; y = interp (x , I ) ; subplot (2 ,2 ,4) ; ,0:2: I ) ;
Hd = stem (m , y ( m + k1 * I +1) , ’ m ’ , ’ filled ’) ; axis
([ -1 ,33 , -1.1 ,1.1]) ; The first three frequency response plots are for L = 5
set ( Hd , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ; and, as expected, the filters are all lowpass with pass-
title ( ’ I n t e r p o l a t e d by I = 8 ’) ; band edges approximately around π/I frequencies and
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’
,[ -1 ,0 ,1]) ; the gain of I. Also note that the filters do not have sharp
xlabel ( ’ n ’) ; transitions and thus are not good approximations to the
ideal filter. The last plot shows the response for L = 7,
We observe that the interpolated sequences for all which indicates a more sharp transition, which is to be
three values of I are appropriate and represent the ori- expected. Any value beyond L = 7 results in an unstable
ginal sinusoidal signal x(n) at higher sampling rates. In filter design and hence should be avoided.
the case of I = 8, the resulting sequence does not appear
to be perfectly sinusoidal in shape. This may be due the
fact the lowpass filter is not close to an ideal filter. 4 Sampling Rate Conversion
Having discussed the special cases of decimation (down-
3.2 Example
sampling by a factor D) and interpolation (upsampling
Examine the frequency response of the lowpass filter by a factor I), we now consider the general case of
used in the interpolation of the signal in Example 3.1. sampling rate conversion by a rational factor I/D. Basic-
The second optional argument in the interp function ally, we can achieve this sampling rate conversion by first
provides the impulse response from which we can com- performing interpolation by the factor I and then decim-
pute the frequency response, as shown in the following ating the output of the interpolator by the factor D. In
MATLAB script. other words, a sampling rate conversion by the rational
clc , clear all , close all
factor I/D is accomplished by cascading an interpolator
n = 0:256; x = cos ( pi * n ) ; w = [0:100]* pi /100; with a decimator. We emphasize that the importance
Hf1 = figure ( ’ units ’ , ’ inches ’ , ’ position ’ of performing the interpolation first and the decimation
,[1 ,1 ,6 ,4] , ... second is to preserve the desired spectral characteristics
’ paperunits ’ , ’ inches ’ , ’ p a p e r p o s i t i o n ’
,[0 ,0 ,6 ,4]) ; of x(n).

% ( a ) I n t e r p o l a t i o n by I = 2 , L = 5;
I = 2; [ y1 , h1 ] = interp (x , I ) ; H1 = freqz ( h1 ,1 , w
4.1 Example
) ; H1 = abs ( H1 ) ;
subplot (2 ,2 ,1) ; plot ( w / pi , H1 , ’g ’) ; axis ([0 ,1 ,0 ,
Consider the sequence x(n) = cos(0.125πn) discussed in
I +0.1]) ; ylabel ( ’ Magnitude ’) ; Example 2.1. Change its sampling rate by 3/2, 3/4, and
title ( ’I = 2 , L = 5 ’ ) ; 5/8.

3/4
 MEHRAN UNIVERSITY OF ENGINEERING & TECHNOLOGY, JAMSHORO
INSTITUTE OF INFORMATION & COMMUNICATION TECHNOLOGIES
ADVANCED DIGITAL SIGNAL PROCESSING
Lab 10: Multirate Digital Signal Processing

MATLAB provides the function [y,h] = xlabel ( ’ n ’) ;

resample(x,I,D) that resamples the signal in ar-
The original x(n) signal has 16 samples in one period
ray x at I/D times the original sampling rate. The
of the cosine waveform. Since the first sampling rate
resulting resampled array y is I/D times longer (or the
conversion by 3/2 is greater than one, the overall effect
ceiling of it if the ratio is not an integer) i.e., length(y)
is to interpolate x(n). The resulting signal has 16×3/2 =
= ceil(I/D)*length(x). The function approximates
24 samples in one period. The other two sampling rate
the anti-aliasing (lowpass) filter by an FIR filter, h,
conversion factors are less than one; thus, overall effect is
designed (internally) using the Kaiser window. It also
to decimate x(n). The resulting signals have 16 × 3/4 =
compensates for the filter’s delay.
12 and 16 × 5/8 = 10 samples per period, respectively.
The length of the FIR filter h that resample uses is
proportional to the fourth (optional) parameter L that
has the default value of 10. For L = 0, resample per- 5 Exercise
forms a nearest-neighbor interpolation. The fifth op-
tional parameter beta (default value 5) can be used to 1. A continuous time signal is characterized by the fol-
specify the Kaiser window stopband attenuation para- lowing function,
meter β. The filter characteristics can be studied using
the impulse response h. x(t) = A cos(2πf1 t) + B cos(2πf2 t)
The following MATLAB script shows the details.
(a) Using MATLAB, generate a discrete time equi-
clc , clear all , close all
valent of the signal. Assume a sampling fre-
n = 0:2048; k1 = 256; k2 = k1 +32; m = 0:( k2 - k1 )
; quency of 1 kHz, f1 = 50 Hz, f2 = 100 Hz,
Hf1 = figure ( ’ units ’ , ’ inches ’ , ’ position ’ A = 1.5, B = 1.
,[1 ,1 ,6 ,4] ,...
’ paperunits ’ , ’ inches ’ , ’ p a p e r p o s i t i o n ’ (b) Interpolate the discrete time signal by a factor
,[0 ,0 ,6 ,4]) ; of 4 using the interp command.
% ( a ) Original signal
x = cos (0.125* pi * n ) ; subplot (2 ,2 ,1) ; (c) Decimate the output of the interpolator in step
Ha = stem (m , x ( m + k1 +1) , ’g ’ , ’ filled ’ ) ; axis (b) by a factor of 4 using the decimate func-
([ -1 ,33 , -1.1 ,1.1]) ; tion.
set ( Ha , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ;
title ( ’ Original Sequence x ( n ) ’) ; (d) Plot the original, interpolated and decimated
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’ discrete time signal.
,[ -1 ,0 ,1]) ;
% ( b ) Sample rate Conversio n by 3/2: I = 3 , D =
2
I = 3; D = 2; y = resample (x ,I , D ) ; subplot
(2 ,2 ,2) ;
Hb = stem (m , y ( m + k1 * I / D +1) , ’c ’ , ’ filled ’) ; axis
([ -1 ,33 , -1.1 ,1.1]) ;
set ( Hb , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ;
title ( ’ Sample Rate I / D : I = 3 , D = 2 ’ ) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’
,[ -1 ,0 ,1]) ;
% ( c ) Sample rate Conversio n by 3/4: I = 3 , D =
4
I = 3; D = 4; y = resample (x ,I , D ) ; subplot
(2 ,2 ,3) ;
Hc = stem (m , y ( m + k1 * I / D +1) , ’r ’ , ’ filled ’) ; axis
([ -1 ,33 , -1.1 ,1.1]) ;
set ( Hc , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ;
title ( ’ Sample Rate I / D : I = 3 , D = 4 ’ ) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’
,[ -1 ,0 ,1]) ;
xlabel ( ’ n ’) ;
% ( d ) Sample rate Conversio n by 5/8: I = 5 , D =
8
I = 5; D = 8; y = resample (x ,I , D ) ; subplot
(2 ,2 ,4) ;
Hd = stem (m , y ( m + k1 * I / D +1) , ’m ’ , ’ filled ’) ; axis
([ -1 ,33 , -1.1 ,1.1]) ;
set ( Hd , ’ markersize ’ ,2) ; ylabel ( ’ Amplitude ’) ;
title ( ’ Sample Rate I / D : I = 5 , D = 8 ’ ) ;
set ( gca , ’ xtick ’ ,[0 ,16 ,32]) ; set ( gca , ’ ytick ’
,[ -1 ,0 ,1]) ;

4/4