Scribd
Upload
1K views

MDSP 1

Multi-rate digital signal processing involves employing multiple sampling rates when processing digital signals. There are two main methods for sampling rate conversion: 1) converting the digital signal to analog and resampling at the desired rate, which introduces distortion, or 2) using digital domain methods like decimation and interpolation. Decimation reduces the sampling rate by an integer factor using downsampling after a low-pass filter, while interpolation increases the rate by inserting zeros and filtering. Polyphase filter structures improve efficiency by performing operations at the lower sampling rate. Multi-rate processing has applications like interfacing between systems with different rates and implementing filter banks.

Uploaded by

Soundarya Svs
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
1K views

MDSP 1

Multi-rate digital signal processing involves employing multiple sampling rates when processing digital signals. There are two main methods for sampling rate conversion: 1) converting the digital signal to analog and resampling at the desired rate, which introduces distortion, or 2) using digital domain methods like decimation and interpolation. Decimation reduces the sampling rate by an integer factor using downsampling after a low-pass filter, while interpolation increases the rate by inserting zeros and filtering. Polyphase filter structures improve efficiency by performing operations at the lower sampling rate. Multi-rate processing has applications like interfacing between systems with different rates and implementing filter banks.

Uploaded by

Soundarya Svs
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 48

Multi Rate Digital Signal Processing

The process of employing multiple sampling rates


in the processing of digital signals is called Multi
Rate Digital Signal Processing.
The process of converting a signal from a given
rate or sampling frequency to a different rate or
sampling frequency is called sampling rate
conversion.

Sampling Rate Conversion Methods


There are two general methods are there for accomplishing
sampling rate conversion of a digital signal, one method is,
Pass the digital signal through DAC, filter it if necessary, and
then resample the resulting analog signal at desired rate i.e
pass the analog signal through an ADC.

Disadvantages with this method


Signal distortion is introduced
i. by the DAC during signal reconstruction
ii. By the quantization effects during ADC

Another method is digital domain method.


In digital domain the sampling rate conversion can be
done,
By means of Down sampling - Reducing the sampling rate
by an integer factor D.
By means of Up sampling - Increasing the sampling rate
by an integer factor I.

The process of reducing the sampling rate by an integer


factor D is called Decimation.
The process of increasing the sampling rate by an integer
factor I is called Interpolation.

Decimation by a factor D
Let consider x(n) be a i/p sequence which is passed through
a LPF, characterized by the impulse response denoted by
h(n) and a frequency denoted by HD() for performing
decimation process as shown below:
x(n
)

LPF
h(n)

v(n
)

Down sampler
D

The o/p of the filter is a sequence given as,

Which is then down sampled by the factor D to produce y(m). Thus

The frequency domain characteristics of the o/p sequence


y(m) can be obtained by

Now the Z-Transform of y(m) is

If the filter is properly designed then

c[n ]

Standard Approach
Decimation by a Factor D

Interpolation by a factor I
The process of increasing the sampling rate of a digital
signal by an integer factor I is called Interpolation.
In Interpolation process an I-1 new samples which are
zeros are interpolated between the successive values of
the digital signal.
Let v(m) be a sequence of sampling rate Fy=Ifx is obtained
from x(n) after adding I-1 zeros between successive
values of x(n) and is expressed as,
HI(y)

v(m)

hI(m)

y(m)

The z-transform of v(m) is,

The corresponding frequency spectra is obtained by


substituting

HI(y)

v(m)

hI(m)

y(m)

Sampling rate conversion by a rational


factor I/D

Sampling rate conversion by a rational factor I/D can be


achieved by first performing interpolation by the factor I
and then decimating the interpolator o/p by a factor D.
In this process both the interpolator and decimator are
cascaded as shown in the figure below:
x(n)
Rate Fx

Upsampler
I

LPF
hU(l)

LPF
hd(l)

Interpolator

Downsampler

y(m
)

Decimator
Rate=Fx(I/D)=Fy

Rate=IFx

In this figure the two filters with impulse responses


hu(l),hd(l)are operated at the same frequency=IFx and
hence these two filters are combined in to a single LPF of
impulse response h(l) which is shown in the figure below:
H(v)

x(n)
Rate Fx

Upsampler
I

v(l)

LPF
h(l)

Rate=IFx

w(l)

Downsampler

y(m
)

Rate=Fx(I/D)=Fy

Let v(l) be the o/p of the interpolator and can be


represented as,
Let w(l) be the o/p of the filter and can be obtined as,

Finally the o/p of the sampling rate converter denoted by


y(m) can be obtained as,

The corresponding frequency domain representation is,


=

FIR Filter
In general, a FIR system is described by the difference
equation

or by the system transfer function


According to the equ(1)
y(n)=h(0)x(n)+h(1)x(n-1)+.+h(M-1)x(n-M+1) and
can be realized as

y(n)=h(0)x(n)+h(1)x(n-1)+.+h(M-1)x(n-M+1) and
can be realized as
X(n
)

y(n
)

Filter Design and Implementation for


Sampling Rate Conversion
Here
sampling
rate
conversion
which
is
Decimation
and
Interpolation
is
performed by direct form
FIR filter structures.
The
design
and
implementation of FIR
filter
for
performing
decimation process as
shown in the figure:

This realization is simple but


inefficient because,
1.up sampling process introduces
I-1 zeros between successive
points of the signal.
2.If I is large, most of the signal
components in the FIR filter are
zero.
3.The multiplications and additions
in the FIR filter result in zeros
due to this large I.

Therefore it is necessary
to develop a more
efficient structure.
This can be achieved by
embedding the down
sampling operation
within the filter it self as
shown in the figure.

In this structure all multiplications and additions are


performed at the lower sampling rate Fx/D.
Thus desired efficiency can be achieved.

consider
the
Next
interpolation
process
which can be performed
by means direct form FIR
filter structures as shown
in the figure.
This structure is realized
by first inserting I-1 zeros
between the samples of
x(n) and then filtering the
sequence.

The major problem in this


realization is that the filter
computations
are
performed
at
high
sampling rate Ifx.
This problem is solved by
using transposed form of
FIR filter and embedding
the up sampler within the
filter as shown in the
figure.
So all multiplications are
performed at the lower
rate Fx.

Design and Implementation of Poly Phase Filter


Structures for Sampling Rate Conversion
The sampling rate conversion which is interpolation
(decimation) is also performed by means of poly phase
filter structures as shown in the figure below which results in
better computational efficiency than FIR systems.

Here each sub filter is


defined with unit impulse
responses
pk(n)=h(k+nI) where

k=0,1,I-1,
n=0,1,K-1
This structure is achieved
by reducing the large FIR
filter of length M in to a
set of smaller filters of
length K=M/I where M is
selected to be a multiple of
I.

Here all sub filters are


basically al pass filters of
different
phase
characteristics and are
arranged in a parallel form.
The o/p of each filter can
be
selected
by
a
commutator.
The rotation of the
commutator is in the
counter
clockwise
direction.
filter
structure
This
performs computations at
low sampling rate Fx.

Next is the decimator which can be realized by


transposing the Interpolator structure as shown below:
Here each sub filter is defined with unit impulse responses
pk(n)=h(k+nD) where k=0,1,D-1, n=0,1,K-1

This structure is achieved by reducing the large FIR filter


of length M in to a set of smaller filters of length K=M/D
is an integer and M is selected to be a multiple of D.

Applications of Multi-rate Signal Processing


Design of Phase Shifters
Interfacing of Digital Systems with different sampling
rates
Implementation of narrow band LPFs
Implementation of Digital filter banks
Sub band coding of speech signals
Quadrature mirror filters
Transmultiplexers
Oversampling A/D and D/A conversion

Design of Phase Shifters


Here a network is designed that delays the signal x(n) by a
rational fraction of a sampling interval Tx i.e d=(K/I)Tx,
where d is the delay.
In the frequency domain this delay corresponds to a linear
phase shift of the form ()=-(K/I) .
Let consider the system which performs both
interpolation and decimation as shown below:

The interpolator increases the sampling rate by a factor I.


The LPF eliminates the images or frequency duplications
in the spectrum of the interpolated signal.
Next the o/p of the filter is delayed by k samples at the
sampling rate IFx.
Then the delayed signal is decimated by factor D=I.
Thus the desired delay of (K/I)Tx will be achieved.

Interfacing of Digital Systems with different


sampling rates
Let consider the interfacing of two digital systems A and
B through a digital sample and hold block as shown below:

The sapling rate of system A is Fx and the sampling rate


of system B is Fy .

The o/p of system is fed to an interpolator which increases


the sampling rate by a factor I.
The o/p of the interpolator which is sampling rate IFx is fed
to a digital sample-and-hold system.
The signals from the digital sample-and-hold system are
read out in to system B at a rate IFy of system B.
Thus the desired interfacing is achieved and the o/p rate of
digital sample-and-hold system is not synchronized with
the i/p rate.

Implementation of Digital filter banks


Digital filter banks are categorized as two types based on
decimation and interpolation
1. Analysis filter banks
2. Synthesis filter banks
An analysis filter bank consists of a set of filters with system
function {Hk(z)} are arranged in parallel with i/p x(n) as
shown below:

The frequency response characteristics of this filter bank


splits the signal in to a corresponding number of sub bands.
Next, a synthesis filter bank consists of a set of filters with
system function {Gk(z)} are arranged in parallel with i/p
yk(n) as shown below:

The o/p this filter bank are summed to form a synthesized


signal x(n).

These filter banks are often used for


performing spectrum analysis and signal
synthesis.
When a filter bank is used for computing
DFT of a sequence {x(n)} then the filter bank is
called DFT filter bank.
The analysis filter bank for computing DFT consists
of N filters of system function {Hk(z)} where
{k=0,1,2,.N-1}.
If {Hk(z)} where {k=1,2,.N-1} then the analysis
filter bank is called uniform DFT filter bank and the
corresponding frequency domain representation is

The frequency response characteristics of the filters {Hk(z),


k=1,2,.N-1} are obtained by uniformly shifting the
frequency response of the filter having system function
{H0(z)} by multiples of 2/N .
In the time domain the impulse responses are expressed as,

The uniform DFT analysis filter bank can


be realized as shown below:

In this structure the frequency components in


the sequence {x(n)} are translated in frequency
to low pass by multiplying x(n) with
and the resultant signals are passed through
LPFs with impulse responses denoted by h 0(n).
The resulting decimated signal can be
expressed as

Where Xk(m) are samples of DFT at


frequencies
k=2k/N .

The corresponding synthesis filter for each element in the


filter bank can be viewed as shown below:

The i/p signal sequences [Yk(m),k=0,1,N-1] are upsampled by a factor of I=D, filtered to remove the images
or image frequency components and translated in
frequency by, multiplication by the complex exponentials
{exp(j2nk/N), k=0,1,N-1}.

The resulting frequency translated signals from the N filters


are then summed for obtaining v(n) as shown below:

Where the factor 1/N is a normalizing factor.


{yn(m)}represent samples of the inverse DFT sequence
corresponding to {yk(m)}.
{g0(n)} is the impulse response of the interpolation filter.

Sub band coding of speech signals

Quadrature mirror filters

Two channel QMF Bank

You might also like