Neuro-Fuzzy Network for Adaptive Channel Equalization
Rahib H.Abiyev
1
, Tayseer Al-shanableh
2
1
Near East University, Department of Computer Engineering, P.O. Box 670, Le-
fkosa, TRNC, Mersin-10, Turkey
rahib@neu.edu.tr
2
Near East University, Department of Electrical and Electronic Engineering, P.O.
Box 670, Lefkosa, TRNC, Mersin-10, Turkey
shanableh@neu.edu.tr
Abstract
This paper presents the equalization of channel distortion by using neuro-fuzzy network.
The structure and learning algorithm of neuro-fuzzy network have been described. Using
learning algorithm of neuro-fuzzy network an adaptive equalizer have been developed. The
developed equalizer recovers transmitted signal efficiently. The use of neuro-fuzzy equalizer
in digital signal transmission allows to decrease training time of parameters and the com-
plexity of network. The result obtained from the simulation is compared with the simulation
result of neural equalizer, and it is shown that neuro-fuzzy equalizer has better performance
than other one.
1. Introduction
In today’s communication environment, the channels are affected by both linear and
nonlinear distortion. To equalize channel distortions, such as intersymbol interferences, chan-
nel noises the various equalizers have been applied [1]. Classical equalizers do not perform
well in rapidly fading channels. When channel has time-varying characteristics and channel
model is not precisely known adaptive equalization is applied. Adaptive equalization can be
divided into two types: sequence estimation and symbol detection [2]. Sequence estimation
needs channel estimation, and is computationally complex. In this paper, adaptive channel
equalization that realizes symbol detection technique is considered. This is a classification
problem in which input baseband signal is mapped onto a feature space determined by the
direct interpretation of known training sequence. Here, the aim is to separate the symbols in
the output signal space whose optimal decision regions boundaries are nonlinear.
Recently, different approaches have been proposed for channel equalization. Classical ap-
proaches for adaptive equalizer design are based on knowledge of the parametric channel
model [3]. These are implemented by identifying the dynamic of channel and then construct-
ing an equalizer using the identified channel model. These processes require certain time to
gather statistical data about the channel and are time consuming. Another type of adaptive
equalizer is decision feedback equalizer that can be used to improve the performance of
equalizer. Next approach to equalizer design is based on increasing the number of equalizer
taps and choosing the coefficients from different ranges of values according to the amplitude
of distorted signals [4]. In this approach a large number of coefficients and switching thresh-
olds are required.
Nowadays neural networks are widely used for channel equalization [5-11]. One class of
nonlinear adaptive equalizer is based on multiplayer perceptions (MLP) and radial basis func-
tions (RBF) [3-5,7,9,10]. Different MLP structures have been introduced for channel equali-
zation [8,9]. The MLP equalizers require long training and are sensitive to the initial choice of
Proceedings of the Fifth Mexican International
Conference on Artificial Intelligence (MICAI'06)
0-7695-2722-1/06 $20.00 © 2006