Multirate Digital Signal Processing

Contents:
1) Introduction
2) Down sampling
3) Decimation
4) Up sampling
5) Interpolation
6) Sampling Rate Conversion
7) Conversion of band pass signals
8) Applications of multi rate signal processing

Introduction:

The process of converting a signal from one sampling rate to another rate is called as sampling
rate conversion. In general, systems that employ multiple sampling rates in the processing are
called as Multirate signal processing systems.

In most of the applications multirate signal processing systems are wont to improve the
performance, or for increased procedure efficiency.

There are two basic operations in a multirate signal processing system are decreasing and
increasing the sampling-rate of a signal. And a few systems conjointly involve each decreasing
and increasing of the rate conversion.

Down sampling:

Used to decrease the sampling rate by an integer factor.

a) Time-Domain Characterization: A down-sampler with a factor M, where M is a positive

integer, which reduces the original sampling rate by factor M.

x[n] M y[n]
The input-output relation can be written as
y[n] = x[nM]
 Where n is time index and M is a positive integer. This relation is obtained by time
frequency relations. To decrease the sampling rate or frequency we want to increase the time
as a result of both are in inverse relation.

Now Down-sampling operation is enforced by keeping each M-th sample of x[n] and
removing intermediate samples to come up with y[n]. or by discarding M-1 samples for each M

samples.

The above figure shows that down sampling by a factor 3. a) original signal x[n] b)
output signal y[n].

Where T is the sampling period and FT is the sampling frequency.

 b) Frequency-Domain Characterization :
Applying the z-transform to the input-output relation of a factor-of-M down-sampler

y[ n ]=x[ Mn]
∞
Y ( z)= ∑ x [ Mn] z−n
n=−∞
The above expression cannot be directly converted in terms of X(Z).
We define a new sequence
x int [ n ]

x int [ n ]= { x [n ], n=0 , ±M , ±2 M , …
0, otherwise
If we substitute this new sequence in to the z- transform formula

We get, ∞ ∞
Y ( z)= ∑ x [ Mn] z = ∑ x int [ Mn ] z −n
−n

n=−∞ n=−∞

Substitute k=Mn, n=k/M.

∞
Y ( z)= ∑ x int [k ] z−k /M = X int ( z 1/M )
k=−∞
Now x int [ n ] can be written interms of x[n]

x int [ n ]=c [ n ]⋅x [ n ]

Where c [ n ]= { 1 , n=0 , ±M , ±2 M , …
0 , otherwise

Here the sequence c[n] is a periodic sequence with the period M, so that this c[n] can be
written in terms of Fourier series equation.
M−1
1
c [ n ]= ∑ W knM
W
kn
− j 2 π kn/ M M
M =e
Where k=0
x int [ n ]=c [ n ]⋅x [ n ]
Apply z transform for by substituting the c[n]

( )
∞ ∞ M−1
1
X int (z )= ∑ c [ n ]x [ n ] z−n = ∑ ∑ W knM x [ n ] z−n
n=−∞ M n=−∞ k =0

by interchanging the order of summation

M−1 ∞ M −1 ∞ M −1

( ) ( )
1 1 −n 1
=
M
∑ ∑ x[n] W knM z−n =
M
∑ ∑ x[ n] ( W −k
M z ) =
M
∑ X ( z W −kM )
k=0 n=−∞ k=0 n=−∞ k=0

∞
−n
Since ∑ x[n] ( W −k
M z ) =X ( z W M)
−k

n=−∞

M −1
1
X int ( z )=
M
∑ X ( z W −k
M )
Now k =0

If we substitute this expression in the Y(Z) we get

∑ X (z )
M −1 1
1
Y ( z )=X int ( z 1/M
)= M
W −k
M
M k =0

jw
Now find the frequency response by substituting z=e
We get

( )
w M−1 w
j 1 j
Y (e jw )=X int ( e M
)=
M
∑ X e M
W −k
M
k=0

Let us consider a down-sampler with a factor of 2 for an input x[n] whose spectrum is as shown
below
 x[n] M y[n]
The DTFTs or frequency response of the output and the input sequences of this down-sampler
are then related as

( )
w M −1 w
j 1 j
Y (e jw
)= X int ( e M
)=
M
∑ X e M
W −k
M
k=0

Here M=2;

( ) ( )
jw 2−1 jw 1 jw
1 1
Y (e jw )=X int ( e )= 2
∑
2 k =0
X e 2
W −k
2 = ∑
2 k=0
X e 2
W −k
2

W
kn
Where − j 2 π kn/ M
M =e
−k
W 2 =e− j 2 π (−k )/ 2=e jπk
Upon substituting these relation in the above expression
 ( )
1 jw
1
Y (e jw )=
2
∑X e 2
e jπk
k =0

¿
1
2
[(
X e
jw
2 ) (
e jπ 0 + X e
jw
2
e jπ 1 )]
jπ 0 jπ 1
Since e =1 and e =−1

1
Y (e )= X e
2
jw
[ ( )+X (−e )]
jw
2
jw
2

Now we can also write

jω/2 j( ω−2 π )/2

X (−e )=X ( e )

Y (e )=
jw 1
2
X e[ ( )+X ( e
jw
2
j( w−2 π )
2 )]
The second term in the above equation will obtain by shifting the first term by 2π positions.

Now the output will become

In the above figure if we observe the plots of 2 terms are overlapping, hence the shape of original signal
jw
is changed or down sampled by 2 and the new shape or Y (e ) will become
This overlap causes aliasing due to under sampling of the signals.

If there is no overlap i.e. no aliasing only with X ( e jω )=0 for |ω|≥π /2

The general form of input-output relations in the down sampler with factor M will be

( )
M−1 j(w−2 πk )
1
Y (e )=
M
jw
∑ X e M

k =0
Decimation:

To avoid aliasing we are band limiting the input signal by

X ( e jω )=0 for |ω|≥π / M

This relation can be obtained by low pass filtering also.

Here H(Z) is the low pass filter which is used to band limit the input signal.
jw
The characteristics of the H(Z) or H (e ) will be

|H ( e jω )|=
{ 1 , |ω|≤ω c / M
0 , π / M≤|ω|≤π

The above system is called as decimator.

The low pass filter is called as decimation filter or anti-aliasing filter.

Up Sampling:

Used to increase the sampling rate by an integer factor.

a) Time-Domain Characterization: An up-sampler with a factor L, where L is a positive

integer, results of increasing the sampling rate of the original signal by a factor of L i.e.
from Fs to LFs
In this operation it inserts L-1 zeros in between each sample.
The block diagram representation shown below.

The input-output relation can be given as

y[ n ]= { x[ n/ L], n=0 , ±L , ±2 L , ⋯
0, otherwise
Where xu[n] is output and x[n] is input to the up sampler.

L is the positive integer

 The above figure shows the up sampling of a signal x[n] with a factor-3 a) shows the
original signal b) shows the up sampled signal y[n].

b) Frequency domain characteristics:

Applying the z-transforms on the input-output relation i.e.

∞
Y ( Z ) = ∑ y [n] z−n
n=−∞

{[ ]
n
x for n=0 , ± L , ±2 L ,± 3 L , ….
where y [ n ] = L
0 for otherwise
If we substitute this y[n] in the above expression we get,

[]
∞
n −n
Y (Z )= ∑ x
L
z
n=0 ,± L, ±2 L ,± 3 L ,…
n
=k then n=kL
let
L
the summation indices n=0 , ± L ,± 2 L, ± 3 L , … . ¿ k=−∞¿ ∞
Now the expression will become,
∞ ∞
−k
Y (Z )= ∑ x [ k ] z −kL
= ∑ x [ k ] ( z L ) =X (Z L )
k=−∞ k=−∞
jw
if we substitute z =e then ,

Y ( e jw )= X (e jwL )
For example, let’s take L=2,
Y ( e jw )= X (e jw 2)
From the figure below, the spectrum of X (e jw ) is given, the spectrum of Y ( e jw ) is X (e jw 2 )
From the above figure it is observed that sampling rate expansion by a factor 2 leads to a
compression of X ( e jw ) of by a factor of 2 and a 2-fold repetition in the baseband [0, 2π]. And
also it inserts a new image in between. This causes the errors in reconstruction of the signal.

So we need to remove these extra images in the output spectrum. This image needs to be
filtered out with a low pass filter (anti-imaging filter) of band-width π/2. With L = 2 this is the
only image in (0, 2π).

Suppose L=3, then the low pass filter (anti-imaging filter) of band-width will be π/3
So, in general for up sampling factor L the low pass filter (anti-imaging filter) of band-width will
be π/L

Interpolation:
After up sampling we are using a filter to remove the extra images that filter is called as anti
imaging filter. This complete system is called as Interpolator. And the process is called as
interpolation. The interpolation system is shown below,
Where

|H ( e jω )|=
{
1 , |ω|≤π / L
0 , π / L≤|ω|≤π

Decimation = Filtering (anti aliasing) + Down sampling

Interpolation = Up sampling + Filtering (anti imaging)

Sampling Rate Conversion:

Sampling rate conversion by a factor L/M, i.e. both up sampling and down sampling are taken
place in one operation only.

Here either we can do up sampling (Interpolation) first then down sampling (Decimation) next
or first down sampling (Decimation) then up sampling (Interpolation).

Form the two methods in the first one we need to use 2 low pass filters one for up sampling
and another for down sampling, where as in second method instead of using two low pass
filters we use only one for both the purposes.
Hence, the cut off frequency for the filter will be

ω c=min ( π π
,
L M )
Decimation and Interpolation are not commutative.

For example,
Transferring data from Compact Disc to Audio Tape 44.1 KHz to 48 KHz
48 160 L
= =
44.1 147 M
Sampling rate conversion of band pass signals:
In many practical systems it is important to deal with band pass signals.
Here in this operation the signal must be band pass in nature. This means that the signal
occupies a narrow band and at other frequencies it is having sufficiently less energy. The figure
shown below is band pass signal.

In most of the cases the signal will be created using band pass filter before decimation is
performed. Here the band width BW must be less than the original sample rate f s divided by
two times the decimation factor D.
BW< fs/2D

In conversion of band pass signals the most mechanism is aliasing. Aliasing is that the method
wherever signals are translated from one location to a different location on the frequency axis.
In most of the applications this frequency translation destructs the signals, as a result of
multiple signals are often translated to the new same frequency, therefore destroying one
another signals and therefore the information they contain. Once this method done properly
the frequency translation are often controlled, and therefore the result's a lot of helpful
In the above figure the decimation by factor two is shown. The original sample rate is fs and the
new sample rate will be fs/2. A band pass signal with bandwidth BW is centered at frequency
(fs/4 + Fp). After decimation the signal is centered at (fs/4 - Fp). In this case the spectrum has
been flipped.
If the signal isn't band pass in general and there is significant energy present at (fs/4 - Fp), then
the signal of interest at (fs/4 + Fp) will alias into the signal at the lower frequency and can be
corrupted. If the new sample rate isn't a minimum of twice the bandwidth of the signal, then
the signal won't fit into the new frequency area without aliasing.
In the above figure decimation by factor 3 is shown. the initial sample rate is fs and therefore
the new sample rate are fs/3. The figure shows the abstract idea of the signal of interest
‘flipping’ over every multiple of fs/3. If the number of flips is odd, as in Figure 2, the spectral are
inverted. If the number of flips is even, as in Figure 3, the spectrum will not be inverted. This
rule will be extended to any price of D. To avoid a spectral inversion, select D such that the
number of ‘flips’ is even.

Applications of multi rate signal processing:

1. Oversampling A/D and D/A converters: The idea of oversampling A/D converters is to
increase the rate of the signal to the point where a low-resolution quantizer will be
adequate. Therefore, the dynamic range of the signal values between serial samples are
often reduced. Associate oversampling A/D converter is enforced by a cascade of
associate analog sigma-delta modulator (SDM) followed by a digital technique
devastation filter and a digital HPF.

An analog SDM produces a 1-bit per sample output at a awfully high rate. This 1-bit
output is passed through a LPF producing a high exactitude (multiple-bit) output that is
decimated to a lower rate. The decimated output is passed through a HPF that
attenuates the division noise at the lower frequencies.

An oversampling D/A converter performs the reverse operations

A digital signal passes through a HPF, whose output is interpolated (up sampled and
filtered by anti-imaging filter). This high rate high exactness digital signal is input to the
digital SDM that provides a high rate 1-bit per sample output. This 1-bit output is
regenerate to AN analog signal by low pass filtering and more smoothing with an analog
filter.
 2. Interfacing systems with different sampling rates: Frequently, 2 digital systems
controlled by independently operating clocks should be interfaced. the only approach is
to use basic sample-rate conversion strategies

The system A output is at a rate Fx is upsampled by a factor of I, sent at the rate Ifx to a
digital sample-and-hold system that serves as an interface to system B, and fed into
system B at the clock rate DFy.
Therefore, once decimation by issue D, 2 systems are synchronal.
In a special case once I = D, 2 clock rates are comparable but not identical.

Problems:
1) Consider the signal x(n) = anu(n) , a < 1.

a) Determine the spectrum X(ω)

b) If x(n) is applied to a decimator that reduces the sampling rate by a factor of 2
determine the output spectrum
c) Show that the spectrum in part (b) is simply the Fourier transform of x(2n)
d) Plot the spectra of x(n) and x(2n) for a = 0.905
Answer
The sequence will be

Then y(n) = x(2n), with M = 2, is its 2-fold down-sampled version and is obtained by
keeping every other sample of x(n) and dropping the samples in between:
a) The spectrum of x(n) is given by its DTFT

This spectrum is not band-limited but we may pretend it is. This may also be
obtained as X(ω) = X(Z)|z=ejw
b) The spectrum of y(n) = x(2n) is given by

which, with M = 2, becomes

c) The Fourier transform of y(n) = x(2n) = a2n u(2n)=a2n u(n)

 d) The spectrum,

2) if x(n) = a u(n) n , a < 1, up sample the sequence by a factor 2

Answer
The sequence is

Then y(n) = x(n/2), with L = 2, is its 2-fold up-sampled version and is obtained by
inserting a 0 between each pair of consecutive values in x(n)
Multiple choice Questions:
1. decimation is:
a) increasing sampling rate
b) decreasing sampling rate
c) no change in sampling rate
d) random change in sampling rate
Answer: b
2. interpolation is:
a) increasing sampling rate
b) decreasing sampling rate
c) no change in sampling rate
d) random change in sampling rate
Answer: a
3. anti aliasing filter is kept at
a) before up sampler
b) after up sampler
c) before down sampler
d) after down sampler
Answer: c
4. anti imaging filter is kept at
a) before up sampler
b) after up sampler
c) before down sampler
d) after down sampler
Answer: b
5. down sampling by D introduces how many additional images
a) D images
b) D-1 images
c) No images
d) D/2 images
Answer: c
6. Up sampling by I introduces how many additional images
a) I images
b) I-1 images
c) No images
d) I/2 images
Answer: a
7. If x[n]={1,2,3,4,5,6,7} then x[n/2]=
 a) {1,0,2,0,3,0,4,0,5,0,6,0,7,0}
b) {1/2,2/2,3/2,4/2,5/2,6/2,7/2}
c) {1,3,5,7}
d) {2,4,6}
Answer: a
8. If x[n]={1,2,3,4,5,6,7} then x[2n]=
a) {1,0,2,0,3,0,4,0,5,0,6,0,7,0}
b) {1/2,2/2,3/2,4/2,5/2,6/2,7/2}
c) {1,3,5,7}
d) {2,4,6}
Answer: c
9. Up sampler and down sampler are
a) Time variant
b) Time in-variant
c) Both a& b
d) None of the above
Answer: a
10. The low pass filter which is used in up sampling is called as
a) Anti aliasing filter
b) Anti imaging filter
c) Both a& b
d) None
Answer: b
11. The process of decimation is
a) Up sampling+ anti aliasing filter
b) Up sampling+ anti imaging filter
c) Down sampling+ anti aliasing filter
d) Down sampling+ anti imaging filter
Answer: c
12. The process of Interpolation is
a) Up sampling+ anti aliasing filter
b) Up sampling+ anti imaging filter
c) Down sampling+ anti aliasing filter
d) Down sampling+ anti imaging filter
Answer: b
13. In sampling rate conversion first which operation to be performed
a) Interpolation
b) Decimation
c) Either a or b
d) None
Answer: a