P. 1
DSP5

DSP5

|Views: 0|Likes:
Published by Brijesh Singh

More info:

Published by: Brijesh Singh on Jul 14, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

07/14/2012

pdf

text

original

Module 5 - Multirate Signal Processing

Patrick A. Naylor

Digital Signal Processing – p.1/25

Contents
Applications of multirate signal processing Fundamentals decimation interpolation Resampling by rational fractions Multirate identities Polyphase representations Maximally decimated filter banks aliasing amplitude and phase distortion perfect reconstruction conditions

Digital Signal Processing – p.2/25

Introduction
In single-rate DSP systems, all data is sampled at the same rate no change of rate within the system. In multirate DSP systems, sample rates are changed (or are different) within the system Multirate can offer several advantages reduced computational complexity reduced transmission data rate.

Digital Signal Processing – p.3/25

Example: Audio sample rate conversion
recording studios use 192 kHz CD uses 44.1 kHz wideband speech coding using 16 kHz master from studio must be rate-converted by a factor 44.1 192

Digital Signal Processing – p.4/25

5/25 .1 kHz 2 Requires high order analogue filter such as elliptic filters that have very nonlinear phase characteristics hard to design. Bandwidth for audio is 20 Hz < f < 20 kHz Antialiasing filter required has very demanding specification |H(jω)| = 0 dB. f ≥ 44. Digital Signal Processing – p. f < 20 kHz |H(jω)| < 96 dB. expensive and bad for audio quality.Example: Oversampling ADC Consider a Nyquist rate ADC in which the signal is sampled at the desired precision and at a rate such that Nyquist’s sampling criterion is just satisfied.

Digital Signal Processing – p.6/25 .Nyquist Rate Conversion Anti-aliasing Filter.

Digital Signal Processing – p. New (over-sampled) sampling rate is 44.1 × 64 kHz.Consider oversampling the signal at. f ≥ (44. say.7/25 .1 kHz with full precision.1 kHz 2 Could be implemented by simple filter (eg. Then use multirate techniques to convert sample rate back to 44. Requires simple antialiasing filter |H(jω)| = 0 dB. f < 20 kHz |H(jω)| < 96 dB.1 × 64) − 44. RC network) Recover desired sampling rate by downsampling process. 64 times the Nyquist rate but with lower precision.

8/25 .Oversampled Conversion Antialiasing Filter Digital Signal Processing – p.

Overall System This is a simplified version In these lectures we will study blocks like G(z) and ↓ 64 Digital Signal Processing – p.9/25 .

certain frequency bands are less important perceptually because they contain less significant information bands with less information or lower perceptual importance may be quantized with lower precision . This gives an adequate dynamic range. Digital Signal Processing – p.fewer bits. How many bits are required? In general. Divide the spectrum of the signal into several subbands then allocate bits to each band appropriately. 16 bits precision per sample is normally used for audio. In practice.Example: Subband Coding Consider quantizing the samples of a speech signal.10/25 .

5 × 16 + 5 × 8 = 120 kbits/s 4:3 compression Reconstructed signal has no noticeable reduction is signal quality. 10 kHz sampling frequency gives 160 kbits/s Divide into 2 bands: high frequency and low frequency subbands. still satisfying Nyquist criterion.16 bits per sample. Digital Signal Processing – p. Therefore use only 8 bits per sample The sampling frequency can be reduced by a factor of 2 since bandwidth is halved.11/25 . High frequencies of speech are less important to intelligibility.

12/25 .Fundamental Multirate Operations Downsampling by a factor M filter and M-fold decimator Upsampling by a factor L L-fold expander and filter Digital Signal Processing – p.

13/25 . yD (n) = x(M n) Aliasing will occur in yD (n) unless x(n) is sufficiently bandlimited loss of information.M-fold Decimator For an input sequence x(n). Digital Signal Processing – p. select only the samples which occur at integer multiples of M . The other samples are thrown away.

M = 2 Digital Signal Processing – p.Eg.14/25 .

no aliasing. insert L − 1 zeros between each sample.L-fold Expander For an input sequence x(n).15/25 . Digital Signal Processing – p. yE (n) = x(M n) x(n) can always be recovered from yE (n) no loss of information.

16/25 . L = 2 Digital Signal Processing – p.Eg.

17/25 .Frequency Domain View of the Expander From the definition of the z-transform ∞ YE (z) = n=−∞ ∞ yE (n)z −n yE (kL)z −kL k=−∞ ∞ = = k=−∞ x(k)z −kL = X(z L ) For frequency response write z = ejω giving YE (ejω ) = X(ejωL ) YE is a compressed version of X Multiple images of X(ejω ) are created in YE (ejω ) between ω = 0 and ω = 2π Digital Signal Processing – p.

To use the expander for interpolation. Digital Signal Processing – p.18/25 . a lowpass filter is applied after the expander to remove the images (shaded).

Frequency Domain View of the Decimator From the definition of the z-transform ∞ ∞ YD (z) = n=−∞ yD (n)z −n = n=−∞ x(M n)z −n Let x1 (n) = Then ∞ ∞ x(n) 0 if n is an integer multiple of M otherwise YD (z) = n=−∞ x1 (M n)z −n = k=−∞ x1 (k)z −k/M since x1 (k) = 0 unless k is a multiple of M . Digital Signal Processing – p.19/25 .

20/25 .Therefore 1 1/M YD (z) = X1 (z )= M M −1 k X z 1/M WM k=0 as will be shown on the next slide and using k WM = e−j2πk/M For frequency response write z = ejω to give 1 YD (ejω ) = M M −1 X ej(ω−2πk)/M k=0 Digital Signal Processing – p.

We arrive at the previous expression for YD (z) by considering a new sequence cM (n) = and then writing x1 (n) = cM (n)x(n) Further consideration of cM (n) tells us that cM (n) is the inverse Fourier transform of unity and can be written 1 cM (n) = M M −1 −kn WM k=0 1 if n is an integer multiple of M 0 otherwise Digital Signal Processing – p.21/25 .

22/25 .Then 1 X1 (z) = M 1 = M 1 = M M −1 ∞ −kn x(n)WM z −n k=0 n=−∞ M −1 ∞ x(n) k=0 n=−∞ M −1 k X zWM k=0 −n k WM z from the definition of the z-transform. So finally 1 YD (z) = M M −1 k X z 1/M WM k=0 Digital Signal Processing – p.

23/25 .What does YD (z) = 1 M M −1 k=0 k X z 1/M WM represent? stretching of X(ejω ) to X(ejω/M ) creating M − 1 copies of the stretched versions shifting each copy by successive multiples of 2π and superimposing (adding) all the shifted copies dividing the result by M Digital Signal Processing – p.

Digital Signal Processing – p.To use a decimation process we must first bandlimit the signal to π |ω| < M .24/25 .

Summary Downsampling Upsampling Digital Signal Processing – p.25/25 .

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->