Neural net based continuous phase modulation receivers

Feed-forward neural nets (NN) are proposed as receivers for partial-response CPM (continuous phase modulation) systems. The approach involves replacing the entire receiver structure, excluding timing recovery, with a neural-net unit whose inputs are time samples of the incoming baseband signals and whose outputs are the decoded symbols. The authors present simulation results for coherent and incoherent NN-based receivers, and compare their performance with the optimum maximum-likelihood (ML) receiver. A performance analysis of NN-based receivers at high signal-to-noise ratio (SNR) is presented. There is an excellent match between the predicted and actual noise in the network at high SNR. It is found that at high SNR the ratio between the output and input noise variance is a constant depending on the number of saturated nodes, the product of the input and output weights, and the temperature parameter of the nonlinearity