International Journal of Research in Engineering and Innovation Vol-3, Issue-3 (2019), 195-199

__________________________________________________________________________________________________________________________________

International Journal of Research in Engineering and Innovation

(IJREI)
journal home page: http://www.ijrei.com
ISSN (Online): 2456-6934

_______________________________________________________________________________________

Channel equalization techniques to mitigate inter-symbol interference in wireless

communication
Nandar Nyein Thu1, Gyamfi Sylvester2, Fahad Ahmed Memon3, Changjiang Bu4
1, 4 Harbin Engineering University, College of Automation, China
2, 3 Harbin Engineering University, College of Information & Communication, China
_______________________________________________________________________________________

Abstract
The major limitation in current wireless communications is the dispersion of time and the interference of the symbols. To counteract
these problems, diverse adaptive equalization techniques are used. The equalization technique is to curb time dispersion is introduced
by the communication channel and counteracts the resulting effect of inter-symbol interference (ISI). Obviously, some form of blind
equalization must be established in the receiver design. Blind equalizers estimates transmitted signals and channel parameters
simultaneously, and may also be time-varying. The aim of this research paper is to study the performance of various adaptive filter
algorithms for both blind channel and Non-blind channel equalization such as Recursive Least means Square (RLS) algorithm
equalizer, the Least Mean Square (LMS) algorithm equalizer and the Constant Modulus Algorithm (CMA) equalizer in terms of the
symbol error rate and convergence speed. The Quadrature Phase Shift Keying (QPSK) and 16-Quadrature Amplitude Modulation
(16-QAM) is used as a modulation scheme. © 2019 ijrei.com. All rights reserved

Keywords: Adaptive Equalizer, Constant modulus algorithm, Recursive Means Least Square
_________________________________________________________________________________________________________

1. Introduction a signal. The implementation of filters with static coefficients

needs proper and prescribed specifications. However, there are
In telecommunication, inter-symbol interference (ISI) is a situations where the specifications are not available, or is time
form of distortion of a signal in which one symbol interferes varying. Therefore there is a solution known as adaptive
with subsequent symbols [1]. This is an undesirable filtering which has coefficients which are time varying and can
phenomenon because the previous symbol has a noise-like adapt to the situation of changing environment [3]. Demands
effect, thus reducing the reliability of communication. ISI is of high data rate transmission with an acceptable error in the
mostly caused as results of the multipath propagation or the presence of available bandwidth are the main targets for each
inherent nonlinear frequency response of the channel, which operator, so we need the proper channel equalization.
results in "fuzziness" of consecutive symbols. ISI in the system
creates errors at the receiver in the decision device. Therefore, 2. Channel Equalization
the purpose of the transmitting and receiving filter is to reduce
the ISI so that the transmission can happen with minimum Channel equalization is an alternative to the techniques of
errors. Ways to fight inter-symbol interference channel identification described previously for decoding of
include adaptive equalization and the raised cosine filter. The transmitted signals across non-ideal communication channels.
receiving filter is designed to reduce the distortion produced In both cases, the transmitted sequences s(n) is known to
by both the channel and the transmitter; it is often stated as transmitter and receiver. However, in adaptive equalization,
Channel equalizing filter or a receiving equalizing filter. the received signal is used as input signal s(n) to an adaptive
Minimization of ISI can be accomplished by applying different filter, which modifies its properties so that its output
types of equalizers [2]. In simple terms an equalizer is the thoroughly matches a delayed versions s(n – Δ) of the known
technique of altering the presence of certain frequencies inside transmitted signal. Translation of the appropriate period,

Corresponding author: Nandar Nyein Thu

Email Address: [email protected] 195
 Nandar Nyein Thu et. al / International journal of research in engineering and innovation (IJREI), vol 3, issue 3 (2019), 195-199

system coefficients are set and used to decode messages as it has a constant modulus and therefore is called CMA, it is
transmitted in the future or adapted using a rough estimate of very effective for achieving channel equalization. The CMA
the operation known as direct adaptation to decisions. The goal attempts to minimize the cost function j(n), which depends on
of channel equalization is to remove the effects of the channel the difference between the received samples of squared
on the transmitted symbol sequence i.e., inter-symbol magnitude and Godard dispersion constants.
interference (ISI), which can be done by inverse filtering,
linear equalizer or decision-feedback equalization or by
applying sequential detection. The following cost functions
below explain how equalizer filter can be optimized.

 Zero forcing criterion: It inverses the channel impulse

response
 MMSE criterion: It minimize the mean-squared-error
 Min. Bit-Error-Rate (BER) criterion.
Figure 2: CMA Adaptive Algorithm Block Diagram
In the following discussion on different equalizers only the
first two criteria are used [4].
𝑗 (𝑛) = 𝐸 [ |𝑦(𝑛)|2 − 𝑅2 ] (1)
In adaptive equalization, the received signal is applied to a
receiver filter. The output of the receiver filter is at a symbol
The phase update equation for carrier recovery loop is;
rate. The sampled signal is then added to the adaptive filter and
these equalizer coefficients apply to minimize output noise and
𝜑̂𝑛+1 = 𝜑𝑛 − 𝜇2 𝐼𝑚 { 𝑎̂𝑛 ∗ 𝑦(𝑛) 𝑒𝑥𝑝(−𝑗𝜑𝑛 )} (2)
ISI. The adaptability of the equalizer is guided by the error
signal. As mentioned in the introduction, interference between
Comparing this to the typical LMS method where a coupling
symbols is a major obstacle to the accuracy required to achieve
exists between the tap update and the carrier tracking loop, one
an increase in digital transmission speed. ISI problem is fixed
can note that the coupling is removed from the CMA. This
by channel equalization in which the purpose is to design an
allows for the equalizer with the CMA to converge. Now the
equalizer so the impulse response of the channel is as close to
cost function of CMA is j(n) = D(p) where p = 2.
z –Δ as possible, where ∆ is a delay. In general, the channel
First, the data symbol constellation is assumed to be
parameters are not known in advance and may change over
symmetric,
time and are important in some application. Therefore,
adaptive equalization provides us means to track the channel
𝐸𝑥 (𝑛)2 = 0 (3)
characteristics. Following is the channel equalization system
diagram depicted.
And data symbols are stationary and uncorrelated i.e.

𝐸𝑥(𝑛) ∗ 𝑥(𝑚) = 𝐸 |𝑥(𝑛)|2 𝛿𝑚 (4)

Also, the noise can be neglected and length of the equalizer is

infinite,

𝐸|𝑦(𝑛)|2 = 𝐸|𝑥(𝑛)|2 ∑|𝑠𝑘 |2

𝑘
Figure 1: Digital transmission system using channel equalization
(5)
𝐸|𝑦 (𝑛)|4 = {𝐸|𝑥(𝑛)|4 −
3. Adaptive Filter 2(𝐸|𝑥(𝑛)|2 )2 ∑𝑘|𝑠𝑘 |4 2𝐸 (|𝑥(𝑛)|2 )2 }

There are three main adaptation algorithms used in this By replacing R2 with 𝐸|𝑥(𝑛)|4 ⁄𝐸|𝑥(𝑛)|2 the above equation
research, one is least mean square (LMS), constant modulus becomes,
algorithm (CMA) and the other is recursive least square (RLS)
filter.
𝐷2 = { 𝐸|𝑥(𝑛)|4 − 2(𝐸|𝑥(𝑛)|2 )2 ∑|𝑠𝑛 |4
3.1 Constant Modulus Algorithm (CMA) 𝑘
+ 2(𝐸|𝑥(𝑛)|2 )2 (∑|𝑠𝑛 |2 )2 }
Godard proposed an algorithm that can be used for this 𝑘
(6)
purpose. This algorithm introduces a different cost function − 2(𝐸 |𝑥(𝑛)|4 ∑|𝑥(𝑛)|2
that exploits the characteristics of the transmitted modulated
𝑘
signal. Godard’s algorithm works for phased-modulated signal + 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡

196
 Nandar Nyein Thu et. al / International journal of research in engineering and innovation (IJREI), vol 3, issue 3 (2019), 195-199

D2 Can be written as the equation below, known as the step size. The convergence of this algorithm is
directly proportional to the step-size parameter μ. When the
𝐷2 = {−|𝑥(𝑛)|4 − 2(𝐸|𝑥(𝑛)|2 )2 ∑|𝑠𝑘 |4 step size is within a range that ensures convergence, the
process leads the estimated weights to the optimal weights.
𝑘
2 Stability is ensured provided that the following condition is
+ 2(𝐸|𝑥(𝑛)|2 )2 (∑|𝑠𝑘 |2 ) (7) met.
𝑘
lim 𝐸{𝑤𝑛 } = 𝑤 = 𝑅𝑥 −1 𝑟𝑑𝑥
− 2𝑅2 𝐸|𝑥(𝑛)|2 ∑|𝑠𝑘 |2 + 𝑅2 𝑥→∞
𝑘
𝐸{𝑤𝑛+1 } = 𝐸{𝑤𝑛 } + 𝜇𝐸 {𝑑(𝑛)𝑥 ∗ (𝑛)}
It was stated that the CM cost reduction is possible, if the − 𝜇𝐸{𝑥 ∗ (𝑛)𝑥 𝑇 (𝑛) 𝑤𝑛 } (11)
following satisfied when |S0|2 is close to unity,
2
0<𝜇<
4(𝐸𝑥(𝑛))2 |𝑠0 |2 4
− 2𝐸|𝑥(𝑛)| ≥ 0 (8) (𝑝 + 1){|𝑥(𝑛)|2 }

Fundamental concepts about equalizers, blind channel 3.3 Recursive Least Square Algorithm (RLS)
equalization and along with four different versions of constant
modulus algorithm (CMA) have been presented that are This algorithm recursively finds the filter coefficients which
derived from the same cost function introduced by Godard in reduces a weighted linear least squares cost function
[5]. concerning to the input signals [7]. The purpose of this
algorithm is to reduce the mean squares error. In RLS, the input
3.2 Least Mean Square (LMS) signals are assumed deterministic. The RLS exhibits extremely
fast convergence as compared to other conventional
As described earlier the LMS algorithm is built around a algorithms. Nevertheless, this advantage comes at the price of
transversal filter that performs a filtering process. The high computational processing complexity, and possibly not
weighting factor mechanism is accountable for execution the very good tracking performance when the filter to be estimated
adaptive control method on the tape weight of the transversal changes. RLS and LMS algorithms are similar as shown in
filter. The LMS algorithm is consists of two processes: Figure 3 but RLS algorithm gives sufficient tracking ability for
Filtering process, which involves calculating the output (d(n – fast fading channel [8]. Additionally, RLS algorithm have
d)) of a linear filter with respect to the input signal and stability problems because of the covariance update formula
resulting an estimation error by subtracting this output with a p(n), which is used for electronic adjustment in accordance
desired response as shown in equation bellow: with the estimation error as follows the figure below illustrate
the RLS algorithm block diagram:
𝑒(𝑛) = 𝑑(𝑛) − 𝑦(𝑛) (9)

d(n) is the desired response and y(n) is filter output at time n.

Adaptive process, involves the automatic adjustments of the
parameter of the filter with respect to the estimation error.

𝑤 ̂𝑛 + 𝜇(𝑛)𝑒 ∗ (𝑛)
̂(𝑛+1) = 𝑤 (10)

µ is the step size, (n +1) = estimate of tape weight vector at

time (n +1) and if preceding knowledge of the coefficients of
vector (n) is not available, set (n) = 0; additionally, an LMS
adaptive algorithm having p+1 coefficients require Figure 3. Block diagram for RLS adaptive equalizer
multiplication and p + 1 additions to upgrade the filter
coefficients. Hence, single addition is required to calculate the 𝑝(0) = 𝛿 −1 𝐼 (12)
error e(n) = d(n) – y(n) and single multiplication is required to
perform product pe(n). Finally, p+1 multiplication and p Where; p is inverse correlation matrix and δ-1 is regularization
additions are required to calculate the output y(n), of the parameter, positive constant for high SNR and negative
adaptive filter. Thus, a total of 2p + 3 additions per output point constant for low SNR. (n = 1, 2, 3...)
are required to perform the adaptive filter. The LMS algorithm
[6] was by Widrow. In LMS, the weights are upgraded at every 𝜋(𝑛) = 𝑝(𝑛 − 1)𝑢(𝑛) (13)
repetition by approximating the gradient of the quadratic mean
square error (MSE) surface, and then stirring the weights in the 𝜋(𝑛)
𝑘(𝑛) = (14)
opposite direction of the gradient through a small amount, 𝜆𝜇 𝐻 (𝑛) + 𝜋(𝑛)

197
 Nandar Nyein Thu et. al / International journal of research in engineering and innovation (IJREI), vol 3, issue 3 (2019), 195-199

Time varying gain vector

̂ 𝐻 (𝑛 − 1) 𝑢(𝑛)
𝜉(𝑛) = 𝑑(𝑛) − 𝑤 (15)

𝑤
̂(𝑛) = 𝑤
̂(𝑛 − 1) + 𝑘(𝑛) 𝜉(𝑛) (16)

4. General System Model

System model is shown in the following figure as block

diagram which is generalized for communication channel
utilizing any of the CMA, RLS or LMS equalizer to overcome
ISI.

Figure 4: System model diagram

The model used and shown in Fig.4, consists of a binary data Figure 5: Results of RLS, LMS and CMA Equalization with QPSK
b(n) and a modulated using QPSK and 16-QAM technique to
produce the transmitted signal s(n). s(n) goes through the
channel that has the transfer function. In this research work one
Gaussian communication channel with its z-transform or
transfer function is considered in a form as;

𝐴
𝐻(𝑧) =
𝐵 + 𝐶𝑧 −1

Where H(z) is the channel used, and Additive White Gaussian

Noise (AWGN) with Signal to Noise Ratio (SNR) with a range
from 0dB and 30 dB has been added, and then the receiver gets
the signal which is now referred to as x(n) or the received
signal. The objective of equalizer is to cancel the channel effect
or minimize the effect of ISI on the transmitted signal and to
also obtain an estimate for it, which is given as the output of
the equalizer y(n). The equalizer models are shown in Fig .4
for a general model. Implementing the number of the adaptive
FIR filter coefficients is 3 with 300000 transmitted samples is
carried out. The step size for LMS is 0.002 and 0.001 for QPSK
and 16-QAM respectively and 0.0001 for CMA. The Figure 6: Results of RLS, LMS and CMA Equalization with 16-QAM
simulation results were performed in Matlab. Figures below
present the transmitted signal which are QPSK and 16-QAM 6. Results Discussions
modulated signals; also figure below shows the received signal
which is the transmitted signal passed through the distortion From the simulation results, for both QPSK and 16-QAM we
channel. can notice that the proposed adaptive algorithms used to
combat ISI give good channel equalization (CMA, LMS and
5. Simulation Results RLS) among the three algorithms, thus the RLS equalizer is
better, followed by the LMS and CMA equalizers and also
In order to estimate the performance of these adaptive RLS has high convergence and accuracy, trailed by LMS and
algorithms, analytical simulations are carried out in MATLAB. CMA. Cost, convergence and accuracy determine equalizer
choice. Fig. 5 and Fig. 6 show a SER versus SNR for (QPSK)
and 16-QAM using AWGN channel by Adaptive Equalizers
respectively. In both results from the figures, it can be seen that
the CMA works well after SNR is 5db. This is due to the fact
that Recursive Least Square and Least Means Square using
training symbols in every frame to approximate channel effect

198
 Nandar Nyein Thu et. al / International journal of research in engineering and innovation (IJREI), vol 3, issue 3 (2019), 195-199

then equalize it; which Constant Modulus Algorithm uses References

blind algorithm for its estimate. Decreasing in noise power the
Recursive Least Square and Least Means Square are more [1] S. S. Haykin, Adaptive Filter Theory, Prentice Hall, Englewood Cliffs,
efficient and better than Constant Modulus Algorithms. NJ, 4th edition, . 1996.
[2] J. Sahoo, L. P. Mishra, S. Panda, and M. N. Mohanty, "Channel
Adaptive filters using Constant Modulus Algorithm, Least
Equalization Using Adaptive Zero Forcing Technique in Rayleigh
Means Square and Recursive Least Square algorithm have Fading Channel," in 2015 International Conference on Information
been done using MATLAB environment and their responses Technology (ICIT), 2015, pp. 60-64.
have been studied. The comparison diagrams of the (LMS, [3] P. S. R. Diniz, Adaptive Filtering Algorithms and Practical
Implementation. Springer Science+Business Media, LLC, 233 Spring
CMA and RLS) has been carried out based on their SER. It was
Street, New York, NY 10013, USA),: Springer Science+Business Media,
shown that the SER of RLS was less followed by LMS and LLC, 233 Spring Street, New York, NY 10013, USA),, 2008.
CMA equalizer. The RLS, LMS and CMA in both (QPSK) and [4] C. Sooyong and L. Te-Won, "A negentropy minimization approach to
16-QAM meet channel equalization by reducing channel adaptive equalization for digital communication systems," IEEE
Transactions on Neural Networks, vol. 15, no. 4, pp. 928-936, 2004.
effects. From the simulation results, for both QPSK and 16-
[5] N. S. Randhawa, "An Overview of Adaptive Channel Equalization
QAM, it can be noticed that 4-QAM or (QPSK) performed Techniques and Algorithms," International Journal of Science and
better than 16-QAM. Also CMA performed better for 4-QAM Research (IJSR).
than 16-QAM based on the symbol error rate against the signal [6] A. Beasley and A. Cole-Rhodes, "Performance of an adaptive blind
equalizer for QAM signals," in MILCOM 2005 - 2005 IEEE Military
to noise ratio which proved that for increased value of M the
Communications Conference, 2005, pp. 2373-2377 Vol. 4.
data rate will be increased but with an increasing value of SER, [7] A. H. I. Makki, A. K. Dey, and M. A. Khan, "Comparative study on LMS
however QPSK will perform better than 16- QAM. and CMA channel equalization," in 2010 International Conference on
Information Society, 2010, pp. 487-489.
[8] C. V. Sinn and J. Gotze, "Comparative study of techniques to compute
FIR filter weights in adaptive channel equalization," in Acoustics,
Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03). 2003
IEEE International Conference on, 2003, vol. 6, pp. VI-217-20 vol.6.

Cite this article as: Nandar Nyein Thu, Gyamfi Sylvester, Fahad Ahmed Memon, Changjiang Bu, Channel equalization
techniques to mitigate inter-symbol interference in wireless communication, International Journal of Research in Engineering and
Innovation Vol-3, Issue-3 (2019), 195-199.

199