Multirate Signal Processing
Multirate : Changing the Sampling
Rate of the Discrete
Time Signal.
Mrs.Aarti Bang,
VIIT, Pune
Need for Multirate
Digital Audio : Three different sampling
rates are employed
Broadcasting : 32 KHz
Digital Compact Disc : 44.1 KHz
Digital Audio Tape : 48 KHz
Mrs.Aarti Bang,
VIIT, Pune
Digital Video :
Luminance signal is sampled at 13.5 MHz
Color difference signal is sampled at
6.75 MHz
Sampling rate of
NTSC Composite signal : 14.31818 MHz
PAL Composite signal : 17.73447 MHz
Mrs.Aarti Bang,
VIIT, Pune
Over sampling A / D Converter :
Sample the signal at much
higher rate than Nyquist rate and
then decimate (lower the sampling
rate) it.
Eliminates the need of very sharp
cut- off frequency antialiasing filter
Mrs.Aarti Bang,
VIIT, Pune
Digital Communication :
Trans multiplexers
Communication Receivers
Biomedical :
Narrow band filter for Fetal ECG & EEG
Speech Processing :
To reduce storage space and transmission
rate
Mrs.Aarti Bang,
VIIT, Pune
The process of converting a signal from
a given rate to a different rate is
called sampling rate conversion.
Systems that employ multiple sampling
rates in the processing of digital
signals are called multirate digital
signal processing systems.
Mrs.Aarti Bang,
VIIT, Pune
Can be accomplished in two ways:
x(n) x1(n)
D/A A/D
converter converter
Adv: New sampling rate can be arbitrarily selected
Disadv: Signal distortion is introduced by
-D/A converter in signal reconstruction
-Quantization effects in A/D converter
Changing the sampling rate in digital domain --Multirate
Mrs.Aarti Bang,
VIIT, Pune
Fundamental Operations in multirate
signal processing are
Decimation ( Down sampling )
Interpolation ( Up Sampling )
Mrs.Aarti Bang,
VIIT, Pune
DECIMATION
Decimation is the process of
decreasing the sampling rate by a
factor of M i.e. from F s to Fs / M
Down sampling : by a factor of M
is achieved by discarding M-1
samples for every M samples.
Mrs.Aarti Bang,
VIIT, Pune
x(n)
M = 3
12 13
0 1 2 3 4 5 6 7 8 9 10 11 n
y(m)
4
0 1 2 3 m
Mrs.Aarti Bang,
VIIT, Pune
|X(F)|
0 5 10 15 F
Fs/2
M = 3
|Y(f)| frequencies
causing aliasing
0 5 10 15
Mrs.Aarti Bang,
VIIT, Pune
This combined operation of filtering and
down sampling is called as DECIMATION
x(n) w(n) y(m)
h(k) M
Fs Fs Fs / M
Block diagram of Decimator
Mrs.Aarti Bang,
VIIT, Pune
The rate compressor reduces the
sampling rate from F s to Fs / M.
To prevent aliasing at lower rate,
digital filter is used to band limit the
i/p signal to less than F s / 2M.
(new folding frequency)
Sampling rate reduction is achieved
by discarding M-1 samples for every
M samples of filtered signal w(n)
Mrs.Aarti Bang,
VIIT, Pune
|X(F)|
0 5 10 15 F
|H(F)| Fs/2
0 5
|W(F)|
0 5
|Y(F)| frequencies
causing aliasing
0 5 10
Spectral interpretation of decimation from 30 to 10 Hz
Mrs.Aarti Bang,
VIIT, Pune
I/P O/P relationship :
y(m) w(mM) h(k)x(mM k)
k
where
w(n) h(k)x(n k)
k
Mrs.Aarti Bang,
VIIT, Pune
Here x(n n0 ) y(n n0 )
Filtering operation is linear & time invariant,
But downsampling is not.
Therefore DECIMATION is a time variant operation
Mrs.Aarti Bang,
VIIT, Pune
Frequency domain characteristics
Let o/p sampling frequency be Fy or ωy
I/p sampling frequency be Fx or ωx
Decimation factor be M
y
M or y xM
x
Thus frequency range 0 <= | ωx | <= Л/M
is stretched into corresponding
Frequency range of 0 <= | ωy | <= Л/M
Mrs.Aarti Bang,
VIIT, Pune
| X(ωx) |
0 ωx
-Л Л
| H(ωx) |
ωx
-Л/M Л/M
| W(ωx) |
ωx
-Л/M Л/M
| y(ωx) |
Mrs.Aarti Bang, ωy
VIIT, Pune -Л -Л
INTERPOLATION
Process of increasing the sampling rate of the
signal by a factor of L i.e. from F s to LFs
Upsampling by a factor L inserting L-1 zeros
between two samples
2
0 1 3 n
Mrs.Aarti Bang, 6
VIIT,0Pune 1 2 3 4 5 7 8
Inserting zeros create image bands.
| X(ωx) |
ωx
-Л 0 -Л
| Y(ωx) |
ωy
-3Л/L -Л/L Л/L 3Л/L
Mrs.Aarti Bang,
VIIT, Pune
Frequency domain characteristics
Let o/p sampling frequency be Fy or ωy
I/p sampling frequency be Fx or ωx
Interpolation factor be L
x
y
L
Thus frequency range 0 <= | ωx | <= Л
is compressed into corresponding
Frequency range of 0 <= | ωy | <= Л/L
and the o/p spectrum is L fold periodic repitition
of i/p spectrum.
Mrs.Aarti Bang,
VIIT, Pune
Spectral components of x(n) in the range
0 <= | ωy | <= Л/L are unique
Images above ωy = Л/L are to be rejected by filtering
x(n) w(n) y(m)
L h(k)
Fs LFs LFs
Mrs.Aarti Bang,
VIIT, Pune
Decimation = Filtering + Downsampling
(antialiasing)
Interpolation = Upsampling + Filtering
(antiimaging)
Mrs.Aarti Bang,
VIIT, Pune
Sampling rate conversion
by
Rational / Non Integer factor
Transferring data from Compact Disc to Audio Tape
44.1 KHz 48 KHz
48 160 L
44.1 147 M
Mrs.Aarti Bang,
VIIT, Pune
y(n)
x(n) L h1(k) h2(k) M
Fs LFs LFs LFs (L/M)Fs
x(n)
L h(k) y(n)
M
Fs LFs LFs
(L/M)Fs
Mrs.Aarti Bang,
VIIT, Pune
Specifications of antiimaging filter
H(e ) L
j
/ L
0 / L ||
Specifications of antialiasing filter
H(e ) 1
j
/ M
0 / M ||
Mrs.Aarti Bang,
VIIT, Pune
Specifications of combined filter
H(e ) L j
(min. / M, / L)
0 otherwise
Mrs.Aarti Bang,
VIIT, Pune
Multistage approach to
sampling rate conversion
High sampling rate conversion
Efficient to change rate in more no. of stages
M = M1 M2 M3…. MI
Reduced computational effort and
storage requirements
Mrs.Aarti Bang,
VIIT, Pune
Estimation of Filter Order ( M ) :
Kaiser’s formula:
10log10 p s 13
M
14.6f
or
10log10 p s 13
M
14.6(st p ) / 2
M is inversely proportional to transition bandwidth ( ωs – ωp)
and not on transition band location.
Mrs.Aarti Bang,
VIIT, Pune
Estimation of Filter Order ( M ) :
• Hermann-Rabiner-Chan’s Formula:
D ( p , s ) F ( p , s )[( s p ) / 2 ]2
M
( s p ) / 2
where
D ( p , s ) [ a1 (log10 p ) 2 a2 (log10 p ) a3 ] log10 s
[ a (log ) a (log ) a ]
4 10 p
2
5 10 p 6
F ( p , s ) b1 b2 [log10 p log10 s ]
with a1 0.005309 , a2 0.07114, a3 0.4761
a4 0.00266, a5 0.5941, a6 0.4278
Mrs.Aarti Bang, b1 11.01217 , b2 0.51244
VIIT, Pune
Filter specifications for each stage
Passband 0 <= F <= Fp
Stopband ( Fi – Fs/2M ) < F < F i-1/2 i= 1,2,….,.,I
Passband ripple δp
Stopband ripple δs
where Fi o/p sampling frequency of ith stage
Fi-1 i/p sampling frequency of ith stage
Fs original sampling frequency
Fp highest frequency of interest
Mrs.Aarti Bang,
VIIT, Pune
Determining the no. of stages and
decimation factors
Least computational effort Optimum no. of stages
I
TSR Ni
I
MPS Ni Fi
i1 i1
where Ni is the no. of filter coefficients per stage
MPS is no. of Multiplications Per Second
TSR is Total Storage Requirement for coefficients
Mrs.Aarti Bang,
VIIT, Pune
Advantages of Multistage design:
•Multistage design yields significant reductions in
computation & storage requirements compared to
single stage.
•Reductions are due to wide transition bands of filters
at early stages (even though sampling rates are high)
leading to small values of N.
•In the last stage though the transition band is small,
the sampling rate is also low, hence filter order (N)
is also small as compared to single stage decimation.
Mrs.Aarti Bang,
VIIT, Pune