ELECTRICAL SPECIALISATION COURSE

PROJECT REPORT ON

CHANNEL EQUALISATION USING

TRAINING SEQUENCE

GUIDE SYNDICATE

Lt Cdr Kulveer Singh Lt Praveen Kr Duba 51934F

Lt Sumit Joshi 52122-H
Asst Cmdt KK Gupta 5095-D
 CERTIFICATE

This is to certify that the project report titled “channel equalisation using training
sequence” which is being submitted by LT Praveen Kumar Duba, LT Sumit Joshi
and Asst Comdt Kuldeep Kumar Gupta is a record of students own work carried
out by them under my guidance and supervision in fulfillment of the requirements
of ELECTRICAL SPECIALISATION COURSE.

Kulveer Singh
Lt Cdr
Project Guide

2
 ACKNOWLEDGEMENT

We express our gratitude and indebtedness to our guide Lt.Cdr Kulveer Singh for his
encouragement and guidance for this dissertation work. His ever willing attitude and
guidance throughout enabled us to complete this project.
We also thank Lt.Cdr.Saurabh Aggarwal for his motivation and support by providing a
conducive environment and facilities at FTP in LTS.

Lt Praveen Kr Duba Lt Sumit Joshi Asst Cmdt K.K Gupta

51934-F 52122-H 5095-D

3
 INDEX

S No. CHAPTER NAME PAGE No

1 INTRODUCTION 05

2 ADAPTIVE FILTERS 08

3 DECISION FEEDBACK EQUALISER 14

4 MATLAB REALISATION& RESULT 19

5 FUTURE SCOPE 26

6 APPENDICES 27

7 REFERENCES 36

4
 CHAPTER 1

INTRODUCTION

1.1 Wireless communication channels often suffer from the problems of severe inter-
symbol interference (ISI) and multi-pathfading effect. To make up for the channel
distortion caused by these effects, channel equalization is essential to combat the
distortion so that symbols can be correctly determined at the receiving end.

1.1.1 In telecommunication, intersymbol interference (ISI) means a form of distortion

of a signal that causes the previously transmitted symbols to have an effect on the
currently received symbol. This is usually an unwanted phenomenon as the
previous symbols have similar effect as noise, thus making the communication
less reliable. ISI is usually caused by echoes or non-linear frequency response of
the channel. Ways to fight against intersymbol interference include adaptive
equalization or error correcting codes.

1.1.2 MULTIPATH is simply a term used to describe the multiple paths a radio
wave may follow between transmitter and receiver. Such propagation paths
include the ground wave, ionospheric refraction, reradiation by the
ionospheric layers, reflection from the earth’s surface or from more than one
ionospheric layer, and so on. The fig below shows a few of the paths that a signal
can travel between two sites in a typical circuit. One path, XYZ, is the basic
ground wave. Another path, XFZ, refracts the wave at the F layer and passes it on
to the receiver at point Z. At point Z, the received signal is a combination of
the ground wave and the sky wave. These two signals, having traveled
different paths, arrive at point Z at different times. Thus, the arriving waves may
or may not be in phase with each other. A similar situation may result at point A.
Another path, XFZFA, results from a greater angle of incidence and two
refractions from the F layer. A wave traveling that path and one traveling

5
 the XEA path may or may not arrive at point A in phase. Radio waves
that are received in phase reinforce each other and produce a stronger
signal at the receiving site, while those that are received out of phase
produce a weak or fading signal. Small alterations in the transmission
path may change the phase relationship of the two signals, causing periodic
fading. Figure 1-11.—Multipath transmission. Multipath fading may be
minimized by practices called SPACE DIVERSITY and FREQUENCY
DIVERSITY In space diversity, two or more receiving antennas are spaced some
distance apart. Fading does not occur simultaneously at both antennas.
Therefore, enough output is almost always available from one of the antennas
to provide a useful signal.

Fig 1.1

1.2 Channel equalisation is the process of compensating for the effect of the physical
channel between a transmitter and a receiver. It is an important area in
communications as it can greatly improve the quality of transmission which in turn
leads to more efficient communication. Current research focusses on modeling the
transmission channel in wireless communications as a non-linear time-varying
system. The time varying characteristics are approximated by means of a basis and
reduces the non-linear identification problem to a linear regression.

1.3 The Channel equalization can be performed either on a symbol basis or on a block
basis. In symbol based equalization, the equalizer coefficients are updated on every
received symbol. This, however, leads to considerable computing overheads for

6
coefficient update. While simple adaptation algorithm such as LMS helps alleviate
the overheads, its slow convergence nature makes it useful only for slowly time
varying channels. In contrast, a block based channel equalization updates the
equalizer coefficients only once in every block. This is because a moderate time
varying channel can be regarded as constant during the transmission of sufficient
small block of data. The consequence is that the coefficient update complexity can be
greatly reduced. The calculation of optimal filter coefficients, however, requires
channel information. Since priori channel information is usually not available,
channel estimation is thus required.

7
 CHAPTER 2

ADAPTIVE FILTERS

2.1 Conventional frequency selective digital filters with fixed coefficients are
designed to have a given frequency response chosen to alter the spectrum of the given
input signal in a desired manner. Their key are as follows :

(a) The filters are linear and time variant.

(b) The design procedure uses the desired passband, transition bands, passband
ripple, and stopband attenuation. We do not need to know the sample values of the
signals to be processed.
(c) Since the filters are frequency selective, they work best when the various
components of the input signal occupy non overlapping frequency bands. For example, it
is easy to separate a signal and additive noise when their spectra do not overlap.
(d) The filter coefficients are chosen during the design phase and are held constant
during the normal operation of the filter.

However there are many practical application problems that cannot be successfully
solved by using fixed digital filters because either we do not have sufficient information
to design a digital filter with fixed coefficients or the design criteria change during the
normal operation of the filter. Most of these applications can be successfully solved by
using special smart filters collectively known as adaptive filters. The distinguishing
feature of these adaptive filters is that they can modify their response to improve
performance during operation without any intervention from the user. These are the
filters that automatically change their characteristics to attain the right response at the
right time. Every adaptive filtering application involves one or more input signals and a
desired response signal that may or may not be accessible to the adaptive filter. We
collectively refer these signals as SOE (signal operating environment) of the adaptive

8
filter. The design of any adaptive filter requires a great deal of a priori information about
the SOE and a deep understanding of the particular application. This information is
needed by the designer to choose the filtering structure and the criterion of performance
and to design the adaptive algorithm. To be more specific, adaptive filters are designed
for a specific type of input signal (speech, binary data, etc), for specific types of
interferences (additive white noise, sinusoidal signals, echoes of the input signals, etc),
and for specific types of signal transmission paths (e.g., linear time invariant or time
varying). After the proper design decisions have been made, the only unknowns, when
the adaptive filter starts its operation, are a set of parameters that are to be determined by
the adaptive algorithm using signal measurements. Clearly unreliable a priori information
and/or incorrect assumptions about the SOE can lead to serious performance degradation
or even unsuccessful adaptive filter applications.
If the characteristics of the relevant signals are constant, the goal of the adaptive filter is
to find the parameters that give the best performance and then to stop the adjustment.
However, when the characteristics of the relevant signals change with time , the adaptive
filter should first find and then continuously readjust its parameters to track those
changes.
A very influential factor in the design of the adaptive filters is the availability of a desired
response signal. We have seen that for certain applications, the desired response may not
be available for use by the adaptive filter. Every adaptive filter consists of the following
modules:

2.1.1 Filtering structure: This module forms the output of the filter using
measurements of the input signal or signals. The filtering structure is linear if the output
is obtained as a linear combination of the input measurements. Otherwise it is said to be
non linear. The structure is fixed by the designer, and its parameters are adjusted by the
adaptive algorithm.

2.1.2 Criterion of performance (COP): The output of the adaptive filter and the
desired response (when available) are processed by the COP module to access its quality
with respect to the requirement of the particular application. The choice of the criterion is

9
a balanced compromise between what is acceptable to the user of the application and
what is mathematically tractable: that is, it can be manipulated to derive an adaptive
algorithm.

2.1.3 Adaptation algorithm: The adaptive algorithm uses the value of the criterion of
performance, or some function of it, and the measurement of the input and desired
response (when available) to decide how to modify the parameters of the filter to improve
its performance.

Input signal
Filtering structure

Filter parameters
Performance
Adaptation algorithm evaluation

Fig 2.1

2.2 A very influential factor in the design of adaptive algorithms is the availability of
a desired response signal. For certain applications the desired response may not be
available for use by the adaptive filter. Therefore the adaptation may be performed in one
or two ways:

2.2.1 Supervised adaptation. At each time instant, the adaptive filter knows in
advance the desired response, computes the error (i.e., the difference between the actual
and the desired response), evaluates the criterion of performance, and uses it to adjust its
coefficients.

10
2.2.2 Unsupervised adaptation. When the desired response is unavailable, the
adaptive filter cannot explicitly form and use the error to improve its behavior. In some
applications, the input signal has some measurable property (i.e., constant envelope) that
is lost by the time it reaches the adaptive filter. The adaptive filter adjusts its parameters I
in such a way as to restore the lost property of the input signal.

Fig 2.2

2.3 Channel equalisation To understand the basic principles of channel equalisation

techniques, we consider a binary data communication system that transmits a band
limited analog pulse with amplitudes A (symbol 1) or –A (symbol 0) every Tb s where Tb
is known as the symbol interval and Rb = 1/ Tb as the baud rate. As the signal propagates
through the channel, it is delayed and attenuated in a frequency dependant manner.
Furthermore, it is corrupted by additive noise and other natural or manmade
interferences. The goal of the receiver is to measure the amplitude of each arriving pulse
and to determine which one of the two possible pulses has been sent.

However channels are affected by ISI or inter symbol interference, which can be
compensated by using a linear filter called an equaliser. The goal of the equaliser is to
restore the original pulse as closely as possible to its original shape. The equaliser

11
transforms the channel to a near ideal one if its response resembles the inverse of the
channel
The characteristics of the equaliser are adjusted by some algorithm that attempts to attain
the best possible performance. The most appropriate criterion of performance for data
transmission systems is the possibility of symbol error. However it cannot be used for
two reasons:

(1).The “correct” symbol is unknown to the receiver (otherwise there would be no reason
to communicate), and
(2) The number of decisions (observations) needed to estimate the low probabilities of
error is extremely large.

Thus, practical access their performance by using some function of the difference
between the “correct” symbol and the output. The operation of practical equalisers
involves two modes of operation, dependent on how we substitute for the unavailable
correct symbol sequence
.
(1). A known training sequence is transmitted, and the equaliser attempts to improve its
performance by comparing its output to a synchronized replica of a training sequence
stored at the receiver. Usually, this mode is used when the equalizer starts a transmission
session.
(2). At the end of the training session, when the equalizer starts making reliable
decisions, we can replace the training sequence with the equaliser’s own decisions

There are two modes that adaptive equalizers work;

• Decision Directed Mode: This means that the receiver decisions are used to
generate the error signal. Decision directed equalizer adjustment is effective in
tracking slow variations in the channel response. However, this approach is not
effective during initial acqusition .

12
• Training Mode: To make equalizer suitable in the initial acqusition duration, a
training signal is needed. In this mode of operation, the transmitter generates a data
symbol sequence known to the receiver. The receiver therefore, substitutes this
known training signal in place of the slicer output. Once an agreed time has elapsed,
the slicer output is substituted and the actual data transmission begins.

13
 CHAPTER 3
DECISION FEEDBACK EQUALISER

3.1 In high data rate applications where the transmitted signal bandwidth is larger
than the channel coherent bandwidth, the delay spread due to multipath propagation in
the UW channel results in severe ISI which considerably increases the BER. The decision
feedback equiliser (DFE) is a popular non-linear equiliser used to combat ISI in severe
fading channels. A decision feedback equalizer (DFE) is a nonlinear equalizer that uses
previous detector decision to eliminate the ISI on pulses that are currently being
demodulated. In other words, the distortion on a current pulse that was caused by
previous pulses is subtracted The DFE may be realized in a direct form or as a lattice.
The direct form DFE with symbol spaced taps has been chosen for implementation.
The below Figure shows a simplified block diagram of a DFE where the forward
filter and the feedback filter can each be a linear filter, such as transversal filter. The
nonlinearity of the DFE stems from the nonlinear characteristic of the detector that
provides an input to the feedback filter. The basic idea of a DFE is that if the values of
the symbols previously detected are known, then ISI contributed by these symbols can be
canceled out exactly at the output of the forward filter by subtracting past symbol values
with appropriate weighting. The forward and feedback tap weights can be adjusted
simultaneously to fulfill a criterion such as minimizing the MSE. The advantage of a
DFE implementation is the feedback filter, which is additionally working to remove ISI,
operates on noiseless quantized levels, and thus its output is free of channel noise.

14
 Fig 3.1: Block diagram of the adaptive desicion feedbak equalizer

The direct form DFE consists of a feed forward filter (FFF) and a feedback filter (FBF).
The former is driven by the received symbols and its coefficients are adjusted to suppress
the ISI on current symbols from future symbols (precursors). The latter is driven by
decisions from the output of the detector, and its coefficients are adjusted to cancel the
ISI on the current symbol that results from previously detected symbols (postcursors).
The coefficient adjustment may be performed, as in a linear equiliser, by the LMS
algorithm or the faster converging RLS algorithm (conventional RLS or fast-RLS
algorithm).
The DFE was implemented with 30 forward and 15 feedback tap coefficients. The
adaptive algorithms were used to adaptively adjust the DFE tap coefficient vector to track
the dynamics of the defined channel in order to minimize the mean squared estimation
error (MSE). Both the forward and feedback filter coefficients were jointly optimized.

15
The basic idea of a DFE is that if the values of the symbols previously detected are
known (past decisions are assumed to be correct), then the ISI contributed by these
symbols can be canceled out eactly the output of the forward filter by subtracting past
symbols values with appropriate weighting.

3.2 LMS algorithm:

The LMS algorithm uses the criterion of minimization of the MSE between the desired
equiliser output and the actual output. For brevity, the LMS algorithm is described by the
coefficient update relation:-

C(i) = C(i-1) + U(i)e*(i)

Where C(i) is the DFE coefficient vector at time index i;

U(i)* is the complex conjugated DFE data vector;
e(i) = V(i)-U(i)T C(i-1) is the complex equilisation error
is the step size parameter which controls rate of convergence.
(a) Input vector:
U(i)= [r(i)* r*(i-1)…….r*(i-m1) -x*(i-1) -x*(i-2)………..x*(i-m2)]T
T
Where * signifies conjugation and signifies transpose. r(k) and x(k)are the symbols
at time index ‘k’ at the input and output of DFE respectively.

(b) Coefficient vector:

C(i-1)=[c(i-1,0) c(i-1,1)……c(i-1,m1) d(i-1,1) d(i-1,2)…….d(i-1,m2) ]T
Where c(k,n) signifies the nth forward filter estimated coefficient at time index ‘k’
and d(k,m) signifies the mth feedback filter estimated coefficient at time index ‘k’.
(c) From fig , the signal applied to the Decision device is :-
Y(i)= U(i)H C(i-1)
(d) Equalisation error at time index i (time t=1) is defined as follows:-
(i) = V(i)- Y(i) = V(i) - U(i)H C(i-1)
Where V(i) is the desired signal applied to the decision device.

16
The step size parameter controls the rate of adaptation of the equiliser and the stability of
the algorithm. The following conditions ensure the stability of the algorithm:-

0< < 2/ max

Where max is the largest Eigen value of the NxN signal covariance matrix . is a
Hermitian matrix and all the Eigen values are positive. The choice of the step size
parameter close to the maximum value results in rapid convergence, albeit with large
fluctuations in the equiliser coefficients. However, the rate of convergence depends on
the Eigen value spread of the signal covariance matrix. If the ratio max/ mix >> 1 then the
convergence will be a slow even if is chosen to the upper bound of stability.

3.3 Steps of LMS algorithm:

Step 1: Initialisation
CO = Zero vector of dimension 45x1
UO =Zero vector of dimension 45x1

Step 2: Compute output of DFE

Y(i) = U(i)T C(i-1)

Step 3: Compute Equilisation error

(i) = V(i) - U(i)H C(i-1)

Step 4: Update tap coefficient vector

C(i) = C(i-1) + K* (i)

Where K = Kalman filter gain

17
3.4 Kalman filter The Kalman filter is a set of mathematical equations that provides
an efficient computational (recursive) means to estimate the state of a process, in a way
that minimises the mean of the squared error. The filter is very powerful in several
aspects: it supports estimations of past, present, and even future states, and it can do so
even when the precise nature of the modeled system is unknown.
(i) Calculate Kalman Gain Vector

K = (P * conj(U))/(lambda + conj(U'
)*P*conj(U))

(ii)Update Inverse of Correlation Matrix

P = (P - K*conj(U'
)*P)/lambda

P= (P+P'
)/2

18
 CHAPTER 4
MATLAB REALISATION

4.1 MATLAB is a high-performance language for technical computing. It integrates

computation, visualization, and programming in an easy-to-use environment where
problems and solutions are expressed in familiar mathematical notation. Typical uses
include Math and computation Algorithm development Data acquisition Modeling,
simulation, and prototyping Data analysis, exploration, and visualization Scientific and
engineering graphics Application development, including graphical user interface
building MATLAB is an interactive system whose basic data element is an array that
does not require dimensioning. This allows us to solve many technical computing
problems, especially those with matrix and vector formulations, in a fraction of the time it
would take to write a program in a scalar non interactive language such as C or Fortran.

4.2 The name MATLAB stands for matrix laboratory. MATLAB was originally
written to provide easy access to matrix software developed by the LINPACK and
EISPACK projects. Today, MATLAB engines incorporate the LAPACK and BLAS
libraries, embedding the state of the art in software for matrix computation. MATLAB
has evolved over a period of years with input from many users. In university
environments, it is the standard instructional tool for introductory and advanced courses
in mathematics, engineering, and science. In industry, MATLAB is the tool of choice for
high-productivity research, development, and analysis. MATLAB features a family of
add-on application-specific solutions called toolboxes. Very important to most users of
MATLAB, toolboxes allow us to learn and apply specialized technology. Toolboxes are
comprehensive collections of MATLAB functions (M-files) that extend the MATLAB
environment to solve particular classes of problems. Areas in which toolboxes are
available include signal processing, control systems, neural networks, fuzzy logic,
wavelets, simulation, and many others

19
4.3 MATLAB Simulation of DFE with LMS algorithm:
The DFE was implemented using the LMS algorithm to update the filter coefficients. The
input to the equaliser was a packet comprising of 1000 training symbols followed by
2000 data symbols. The program was written in a similar manner to the DSP
implementation i.e. for symbol to symbol processing. The packet was filtered through a
45 tap Rayleigh fading channel. Further, different levels of AWGN were added to the
channel corrupted signal for testing the algorithm at various SNRs. The figures below
show the MSE plots which indicate the relative convergence rates for different choice of
step size for 5 db SNR condition. It may be noted that the convergence rates improve ofr
better choices of step size. However the optimum step size depends on eigen values of
signal co variance matrix and hence on the nature of signal itself.

20
4.4 THE RESULTS OBTAINED ARE:

Mean Squared Error

400

350

300

250

200

150

100

50

0
0 500 1000 1500 2000 2500 3000 3500

4000

3500

3000

2500

2000

1500

1000

500

0
0 500 1000 1500 2000 2500 3000 3500

Fig 4.1 Mean square error

21
60

40

20

-20

-40
0 500 1000 1500 2000 2500 3000

Equalisation Error
30

20

10

-10

-20

-30

-40

-50
0 500 1000 1500 2000 2500 3000

Fig 4.2 Equalisation error

22
4.4 DSP Implementation Issues of the LMS algorithm

It must be noted that the algorithm is very sensitive to the choice of the step size. The
upper limit of the step size was chosen after computing the maximum Eigen value of the
signal covariance matrix. The covariance matrix was formed using an ensemble of typical
data vectors of length N (m1+m1-1), which is the size of the filter. For the DFE the data
vector is of the form:-

C(i-1)=[c(i-1,0) c(i-1,1)……c(i-1,m1) d(i-1,1) d(i-1,2)…….d(i-1,m2) ]T

Where r(i)s are the raw input symbols to the DFE and x(i-1)s are the previous hard
decisions outputs of the symbol decision device.

4.5 Symbol Decision Device

The inputs to the symbol decision device are the raw outputs {y(i)s} of the DFE and its
outputs are hard decisions (+1 or -1) made on the inputs.

23
 DFE Output
2

1.5

0.5

-0.5

-1

-1.5

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Fig 4.3 DFE output

1
input
output
0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Fig 4.4 Comparison of inputs and outputs

24
If we examine the error signal , we can see that at the beginning of the
training mode there are errors as the equaliser tries to adapt itself to the
channel and achieves this after some period of time. After training mode is
completed decision directed mode begins.in the decision directed mode
actual data transmission begins the reference signal is the output of the
equaliser instead of the training signal.
In spite of this the error level is still low in the decision directed mode. The
reason of this is that the equaliser adapted itself to the channel at the end of
training mode. After this time the channel is not changed the only changes
come from the randomness of the noise.
However if the channel is slowly varying instead of our channel the
equaliser continuous to track channel for some time duration. If the number
of filter coefficients are smaller the adaptation time in the training mode
reduces. However the amount of error in decision directed mode increases.

25
 FUTURE SCOPE

Adaptive decision feedback equalizers can be simulated for different channel responses
and the results can be compared. This work can be expanded by using time varying
channels and by using different signal-to-noise ratio.

26
 APPENDIX (A)

MATLAB CODE:

clear all
close all
clc

%---------Initialisation------------

fftaps=30;

fbtaps=15;

dfelength = fftaps + fbtaps;

C= zeros(dfelength,1);

R = zeros(fftaps,1);

X = zeros(fbtaps,1);
U =[R
X];

Xout=[];

P = 1000 * eye(dfelength);

lambda = 1;

dfeoutput=[];

decision= [];

Eqerror = [];

errsqr =0;

MSE=0;

errorscaling = 1;

errcount=0;

27
H = 1;

%-------Data Generation-----------

rand('
state'
,0);

realtrg= 2*round(rand(1,45))-1;

rand('
state'
,0);

imagtrg= 2*round(rand(1,45))-1;

trgseq= realtrg + i*imagtrg;

for m=1:6

trgseq = [trgseq trgseq];

end

trgseq = trgseq(1:1000);

% V = [complex(ones(1,floor(fftaps/4)), ones(1,floor(fftaps/4))) trgseq];

data = [1+i 1-i -1+i -1-i];

for m=1:10

data = [data data];

end
data = data(1:2000);

datapacket= [trgseq data];

%load channel;

packet = conv(datapacket,H);

packet = AWGN(packet,5,'
measured'
,0);

% noise = (randn(1,length(packet)) + i*randn(1,length(packet)));

% packet = packet + noise;

% packet = packet(1:length(datapacket));

28
% noise = (randn(1,langth(datapacket)) + i*randn(1,length(datapacket)))/2;
% packet = packet + noise;

for n=1:length(datapacket)

nextR = [packet(n)
R];

R = nextR(1:fftaps);

nextX=[-Xout
X];

X = nextX(1:fbtaps);

U = [R
X];

y= conj(U'
)*C;

dfeoutput=[dfeoutput y];

%-------Symbol Decision Device--------

Xout = sign(real(y)) + i* sign(imag(y));

decision=[decision Xout];

% V = [V Xout];

%------Calculate Equalisation Error---------

if n <= length(trgseq)

err = trgseq(n) - y;

else
err = Xout - y;

end

% err = datapacket(n) - y;

Eqerror=[Eqerror err];

29
 errsqr = [errsqr err*conj(err)];

meansqr=sum(errsqr)/n;

MSE = [MSE meansqr];

% if Xout ~= datapacket(n)
% errcount = errcount + 1;
% end

%----------Calculate Kalman Gain Vector----------

K = (P * conj(U))/(lambda + conj(U'
)*P*conj(U));

%----------Update Inverse of Correlation Matrix---------

P = (P - K*conj(U'
)*P)/lambda;

P= (P+P'
)/2;

%---------Update Coefficient Vector----------------

C = C + K*err*errorscaling;

End

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%

if Xout ~= datapacket(n)

errcount = errcount + 1;

end

subplot(2,1,1);

plot(real(Eqerror));

subplot(2,1,2);

plot(imag(Eqerror));

title('
Equalisation Error'
);

30
figure;

subplot(2,1,1);

stem(MSE, '
x');

title('
Mean Squared Error'
);

subplot(2,1,2);

stem(errsqr, '
o');

figure

plot(dfeoutput,'
b*'
);

title('
dfeoutput'
);

AXIS([-2 +2 -2 +2]);

figure;

plot(datapacket, '
go'
);

hold on;

plot(decision, '
r*'
);

legend('
input'
,'
output'
);

31
 APPENDIX B

Applications of adaptive filters:

The best way to introduce the concept of adaptive filtering is by describing some
typical application problems that can be effectively solved by using an adaptive filter.
The applications of the adaptive filters can be sorted for convenience into four classes:

(a) System identification

(b) System inversion
(c) Signal prediction
(d) Multisensor interference cancelation

System identification This class of applications is also known also as system

modeling. The system to be modeled can be either real as in control system applications,
or some hypothetical signal transmission path (e.g., the echo path). The distinguishing
characteristic of the system identification application is that the input of the adaptive
filter is noise free and the desired response is corrupted by additive noise that is
uncorrelated with the input signal. Applications in this class include echo cancelation,
channel modeling, and identification of systems for control applications. In control
applications, the purpose of the adaptive filter is to estimate the parameters or the state of
the system and then to use this information to design a controller. In signal processing
applications, the goal is to obtain a good estimate of the desired response according to the
adopted criterion of performance.

Acoustic echo cancelation we have a typical audio teleconferencing system that helps
two groups, located at two different places, to communicate effectively. However, the
performance of this system is degraded by the following effects:

32
(1) The reverberations of the room result from the fact that the microphone picks up not
only the speech coming from the talker but also the reflections coming from the walls of
the room and the furniture.

(2) Echoes are created by the acoustic coupling between the microphone and the
loudspeaker located in the same room. Speech from room B is not only heard by the
listener in room A but also is picked up by microphone in room A, and unless it is
prevented, will return as an echo to the speaker in room B.

TRANSMISSI TRANSMISS
ON ION
EQUIPMENT TRANSMISSION EQUIPMENT
CHANNEL

LOCATION A LOCATION B

Fig : Typical teleconferencing system without echo control

Several methods to deal with acoustic echoes have been developed. However, the most
effective technique to prevent or control echoes is adaptive echo cancelation. The basic
idea is very simple: To cancel the echo, we generate a pseudo echo or a replica and
subtract it from from the real echo. To synthesize the echo replica, we pass the signal at
the loudspeaker through through a device designed to duplicate the reverberation and
echo properties of the room (echo path), as is illustrated in fig.

33
 Fig: Principle of acoustic echo cancelation using an adaptive echo canceller

In practice, there are two obstacles to this approach:

(1) The echo path is usually unknown before actual transmission begins and is quite
complex to model.

(2) The echo path is changing with time, since evn the move of a talker alters the
acoustic properties of the room. Therefore, we cannot design and use a fixed echo
canceller with satisfactory performance for all possible connections. There are two
possible ways around this problem:

(a) Design a compromise fixed echo canceller based on some “average” echo path,
assuming that we have sufficient information about the connections to be seen by the
canceller.
(b) Design an adaptive echo canceller that can “learn” the echo path when it is first
turned on and afterwards “tracks” its variations without any intervention from the

34
designer. Since an adaptive canceller matches the echo patch for any given connection, it
performs better than a fixed compromise canceller.

However it should be noted that the main task of the canceller is to estimate the echo
signal with sufficient accuracy; the estimation of the echo path is simply the means for
achieving this goal. The performance of the canceller is measured by the attenuation of
the echo. The adaptive echo canceller achieves this goal, by modifying its response, using
the residual echo signal.

System inversion This class of applications, which is illustrated in fig, is also known
as inverse system modeling. The goal of the adaptive filter is to estimate and apply he
inverse of the system. Dependent on the application, the input of the adaptive filter may
be corrupted by additive noise, and the desired response may not be available. The
existence of the inverse system and its properties creates additional complications.
Typical applications include adaptive equalization, etc.

35
 REFERENCES

(i) “Modern digital and analog communication systems” by BP Lathi

(ii) “Statistical and Adaptive signal processing” by D.G. Manolakis and
Stephen Kogon
(iii) “DSP Application LAB workbook” by Edutech Systems

36