M-QAM demodulation in an OFDM system with RBF neural network

This paper proposes a method which improves the IM-QAM demodulation in the orthogonal frequency division multiplexing (OFDM). The OFDM has several advantages, i.e., the possibility of a simple equalization of the received signal. However, the disadvantages are also considered as the phase ambiguity due to intercarrier-interference (ICI) which causes the significant degradation on the performance of an OFDM system. The proposed method is a radial basis function (RBF) neural network which learns the characteristic of M-QAM signal constellation before reconstructing the correct signal constellation under noisy circumstances. The hybrid learning process is used to train the RBF network. The hidden layer is trained by the hard c means clustering. The supervised learning with given input-output pairs are used to train the output layer. This paper assumes an additive white Gaussian noise (AWGN) channel. The simulation results of the random symbol generations show that the probability of errors closes to ideal with the proposed method.