Sei sulla pagina 1di 14

EEE G613: Advanced Digital Signal Processing

Multirate Signal Processing:


M-Channel QMF, Perfect Reconstruction

BITS Pilani Dr. Ananthakrishna Chintanpalli (office no: 2210 - O)


anantha.krishna@pilani.bits-pilani.ac.in
Pilani Campus
M-Channel QMF

^
• Let’s express X (z) in terms of X(z):

• Rewrite the expression:

This expression ignores the coding, decoding and quantization errors


BITS Pilani, Pilani Campus
Analysis Filter Bank (For example)

BITS Pilani, Pilani Campus


M-Channel QMF

For , this represents a shifted version of the original spectrum.


^
Thus, the reconstructed signal X(z) is a linear combination of original spectrum and
(M-1) uniformly shifted versions of the original spectrum.

BITS Pilani, Pilani Campus


Errors created by Filter Bank System
^
• As we know from two-channel QMF that reconstructed signal x(n) differ by x(n)
due to three reasons.
o Aliasing (and Imaging)
o Amplitude Distortion
o Phase Distortion

• The same three errors occur for M-channel QMF.

BITS Pilani, Pilani Campus


M-Channel QMF: Aliasing

Identify the aliasing term:

The presence of the shifted versions are due to decimation and


interpolation.

is the term of aliasing.

is the gain for this aliasing term.

To cancel aliasing:

BITS Pilani, Pilani Campus


M-Channel QMF (M=3)

The aliasing cancellation terms are:

BITS Pilani, Pilani Campus


M-Channel QMF (M=3)

BITS Pilani, Pilani Campus


Amplitude and Phase Distortions

If aliasing is cancelled completely then:

Because

T(z) is a distortion transfer function and is given by

If T(z) is not an all pass filter , we say that


suffers from “amplitude distortion”.

If T(z) is not a linear phase , we say


that suffers from “phase distortion”.
BITS Pilani, Pilani Campus
M-Channel QMF: Perfect Reconstruction
Condition
• If QMF is free from aliasing, amplitude and phase distortions, then it is called
perfect reconstruction.

• If QMF is aliased free, then

• x^[n] should be scaled and delayed version of x[n] to meet perfect


reconstruction condition.

BITS Pilani, Pilani Campus


Synthesis filters computation:

In matrix form:

BITS Pilani, Pilani Campus


Synthesis filters computation:

Condition for aliasing can be written as:

Where,

The M x M matrix H(z) is called the Alias Component (AC) matrix.

BITS Pilani, Pilani Campus


Synthesis Filters Computations:

• Given the analysis filters, we can in principle cancel aliasing by solving for the
synthesis filters.

• This will work as long as determinant {H(z)} is not zero otherwise we have to
compute the pseudo-inverse.

A-1 = AT (AAT)-1

A-1 = (ATA)-1AT

BITS Pilani, Pilani Campus


Reconstructed Signal in Matrix Form

BITS Pilani, Pilani Campus

Potrebbero piacerti anche