Channel Equalization using Self-Organizing Maps

Abstract

In this work, the signals at the output of a wireless channel in the presence of intersymbol interference were visualized using Self-Organizing Maps with the aim of discovering their properties which can then be used to mitigate the effects of inter-symbol interference on signals sent over a wireless channel. The research tried to find out if there are certain properties in an interfering wireless channel output which can be exploited to mitigate the effects of intersymbol interference and then it was demonstrated how the found properties can be used in the mitigation. To achieve these objectives, simulations were carried out. Gray coded 16-QAM symbols were transmitted over a channel which introduces severe inter-symbol interference. The in-phase and quadrature components of the channel output were then used to train a selforganizing map. The fully trained map was used to make observations about the general structure of channel output and how it relates to the transmitted constellation. It was found that the channel output resembles a rotated input constellation. Furthermore, some symbols were found in clusters belonging to other constellation points other than their own. An attempt was then made to classify the channel output using the trained map and an analysis was then done on the misclassified symbols to determine what constellation points that symbols from each of the 16 constellation points is likely to be misclassified to. It was found that a symbol is likely to be misclassified to those symbols whose gray codes differ from its own by one bit. Thus the accuracy of gray codes for the used channel was verified. An examination of the self-organizing map component planes for the in-phase and quadrature components revealed how the classification of the symbols belonging to each of the 16 constellation points is influenced by the value of their in-phase and quadrature components. It was found that the classification of symbols belonging to certain constellation points is strongly affected by the values of their in-phase or quadrature components. The classification of symbols belonging to certain constellation points is found to be weakly dependent on the values of their in-phase or quadrature components while the classification of symbols belonging to certain constellation points is moderately affected by the values of their in-phase or quadrature components. This result can be used for example to minimize the number of symbol classification errors that result when the energy of transmitted signals is reduced. It is demonstrated how this can be done by essentially reducing the magnitude of in-phase or iv quadrature components for the constellation points whose classification is weakly and moderately dependent on the value of their in-phase or quadrature components. The results show that the symbol classification error obtained is less than that obtained when the energy of all constellation points is reduced indiscriminately.