Neural network implementation for an adaptive maximum-likelihood receiver

An artificial neural network is described for the detection of digital data symbols transmitted over a time-dispersive time-varying channel in the presence of Gaussian noise. The transmitter uses quadrature phase-shift keying modulation. The network computes a maximum-likelihood estimate of the transmitted sequence. Mapping of the maximum-likelihood sequence estimation function onto the artificial neural network structure is described. A neural-network-based receiver structure is presented which can be used for stationary or time-varying channels. Simulation results are presented which show promising error rates over a wide range of intersymbol-interference durations. Unlike the Viterbi algorithm implementation, the neural network does not require a vast amount of memory for storage and the computation time does not increase with increasing channel memory