QUEIROZ, W. J. L.; http://lattes.cnpq.br/7663004390139625; QUEIROZ, Wamberto José Lira de.
Resumo:
This work presents results of the application of Bayesian detection, along with neural networks and radial basis functions (NRBF) to the channel equalization problem. Equalization is used here in the sense of recovering degraded symbols from communication channels. The purpose of this study is therefore, to model the main effects of the channel and propose a solution that relies on NRBF. As a result of this work, two detection structures were implemented in order to evaluate the performance of the detection schemes. The Bayesian hypothesis testing is a statistical method used in pattern classification and assumes that the decision problem is based on known probabilities. An
introduction to the theory is initially presented, using binary symbols for simplicity, to ensure that the concepts can be extended to the equalization problem. The detection structures, with and without feedback, are based on this method. The adopted modulation scheme was quadrature phase shift keying (QPSK), which is used by several mobile cellular air interfaces, including the code division multiple access (CDMA) system. But, the obtained structures can be used with other modulation schemes. The equalizer structure is similar to an NRBF. This, certainly, influenced the used of this type of neural network in the implementation of the Bayesian method. Other factors, including implementation facility, fast training and small complexity also were
taken into consideration. The detection structures were compared and, for a channel with Doppler effect, the Bayesian feedback detector showed a superior performance.