Scribd
Upload
38 views

MATLAB Project05

a

Uploaded by

Katamba Rogers
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
38 views

MATLAB Project05

a

Uploaded by

Katamba Rogers
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

EEE4001F: MATLAB project

This document provides an outline for a Matlab project to be


completed by the end of the course. You are expected to investigate, in
detail, methods related to solving the problem. There is a design
element to the project, and a quantitative evaluation of the performance
of the proposed methods must be performed and presented. You are to
write up a comprehensive report (of no more than 8 pages) describing
your method and results. You should work in groups of two, although
you may work alone if you really want to.
If you wish to propose a project of your own, then please come and talk
to me. Project descriptions from previous years are on the course
website, and are also options.
Note that this document is still in preparation, and may be added to
during the course of the project.
The Task: Simple Echo Cancellation
The aim of this project is to investigate methods of restoring signals
that are corrupted by one or more echos. Echos commonly occur over
communications channels such as telephone and ADSL lines. A
reasonable model of the echo process in the time domain is
y[n] = x[n] + a
1
x[n n
d
] + a
2
x[n 2n
d
],
where n
d
is the echo delay, and a
1
and a
2
are reection coecients for
the rst and second echo. The echo process can be thought of in terms
of a system function H(z) = Y (z)/X(z), and the task of echo
cancellation is to nd the inverse function G(z) = 1/H(z). If all the
parameters in the expression above are known, then this inverse can be
found analytically.
1
In reality one doesnt know the parameters of the echo model, which
possibly even change slowly over time. Adaptive lters are therefore
used to do the cancellation. Most formulations are beyond the scope of
this course, but the LMS algorithm is within reach and is relatively easy
to understand. Instead of estimating H(z) and using it to nd the G(z)
that cancels it, we nd G(z) directly.
The echo removal system is modelled as a FIR lter with an unknown
impulse response. We estimate the required impulse response by
transmitting a known signal x[n] over the channel and observing the
corresponding output y[n]. (ADSL modems do this during their
training stage.) We want a system that takes the echo signal y[n] as
input, and outputs the original signal x[n].
The output of the lter at time n is
x[n] = h[n] y[n] =
N

i=0
h
i
y[n i],
where the notation h
i
= h[i] is used for clarity. The desired output is
the known transmitted signal value x[n], so the error in the lter output
at this instant is e[n] = x[n] x[n]. We want to choose the h
i
parameters such that the error is minimised.
The squared error is
e
2
[n] = (x[n] x[n])
2
= (x[n]
N

i=0
h
i
y[n i])
2
,
which we can use to work out how h
i
should be changed to decrease the
error. The derivative of the squared error with regard to the impulse
response value h
j
is
de
2
[n]
dh
j
= 2

x[n]
N

i=0
h
i
y[n i]

y[n j] = 2e[n]y[n j].


2
If this derivative is positive, then it means that e
2
[n] will become bigger
if h
j
is increased. We can then decrease the error by reducing h
j
.
In general we x a step size , and for each sample instant we modify
each h
j
according to
h
j
h
j
+ 2e[n]y[n j].
This is the Least Mean Squares (LMS) algorithm. For suciently small
and enough training data, the values of h
j
will converge to the point
where on average de
2
[n]/dh
j
= 0 for each j (i.e. where the squared error
is minimised).
Some basic code that uses the LMS algorithm to determine a lter for
echo cancellation follows:
% Pseudorandom sequence for training
d = 100; % echo delay
gsig = randn(16000,1);
x = gsig(d+1:end); % desired echo-free signal
y = gsig(d+1:end) + 0.4*gsig(1:end-d); % signal with echo
% Train using LMS
M = 3*d; % FIR filter length
delta = 0.0001; % LMS training step size
h = zeros(M,1); % Initial filter IR values
ev = [];
for i=M+1:length(x)
xn = x(i); % desired sample value
yn = y(i-M+1:i);
xnhat = h*yn; % filter output value
en = xn - xnhat; % current error
h = h + delta*yn*en; % update
3
% Plotting
ev = [ev abs(en)];
if rem(i,100)==0 % too slow to plot every iteration
subplot(2,1,1); plot(h); title(Impulse response);
subplot(2,1,2); plot(ev); title(|error|);
pause(0.001);
end
end
Copy it into a Matlab script, run it, and try to understand it.
Approach
The aim of this project is to explore the problem of echo cancellation.
The implementation of the LMS algorithm in the previous section can
serve as a starting point, but you can look at alternative methods
instead if you prefer.
Some possibilities are:
Do a theoretical analysis of echo cancellation for known delay and
reection parameters. I briey discussed an example in class, which
can be elaborated upon. In the non-adaptive case one can derive
the inverse system analytically.
Explore the adaptation parameter and the eect it has on rate of
convergence.
Explore the eect of dierent echo delays on the cancellation
process. Since the length of the FIR lter used to cancel the echos
must be substantially longer than the echo delay, the number of
lter weights that the LMS algorithm has to estimate increases with
increasing delay.
The LMS formulation does not make use of the knowledge that the
4
signal is corrupted by an echo. It should therefore be possible to use
the algorithm as provided for restoring arbitrary channel distortion,
as long as it is time invariant. You could explore alternative uses for
the cancellation system provided.
The length of the impulse response required to cancel a long delay
is large, so the computational expense becomes high. You could
make the computation more manageable by implementing
overlap-add or overlap-save FFT-based convolution. Alternatively,
Ive seen interesting subband approaches which I know little about.
5

You might also like