Neural network communication receiver based on the nonlinear filtering

A neural-based network is applied to the data communication receiver to investigate the feasibility of the network over inter-symbol interference (ISI) and additive white Gaussian noise channel environments. With a three-layered perceptron with either backward error propagation or extended Kalman filter training algorithms, it can be shown that it closely approximates the theoretical optimum receiver as the number of network trainings increases. The simulations are made on the network operations, and error rate performance for several important parameters is given. The proposed data receiver is an alternative to the optimum Viterbi channel decoder once the problem on the network training is solved