A novel decoder structure for convolutional codes based on a multilayer perceptron

A novel decoder structure for convolutional codes based on a multilayer perceptron (MLP) is presented. It is shown, that by an appropriate preprocessing based on the feedback of previous decisions, the complexity of the MLP decoder and of the training algorithm leading to the optimal coefficients can be reduced substantially. Simulation results are given, comparing the performance-expressed in terms of the bit error rate as a function of the signal to noise ratio (E_b/N₀)-of the MLP decoder with an optimal maximum a posteriori probability (MAP) decoder based on a Bayes classifier. Furthermore, a preprocessing scheme is presented, which involves the feedback of soft decisions. For the MAP decoder we compare the performance of preprocessing with feedback of correct information symbols, hard decision feedback, and soft decision feedback