Abstract:
The paper investigates new algorithm for decoding convolutional codes based on neural networks. The novelty of the algorithm is in its capability to generate soft output estimates of the message bits encoded. The log likelihood function is derived, related to the noise energy function and then used as a criterion to decide which message bits are transmitted. The algorithm is demonstrated on a systematic 1/2-rate convolutional code for the assumed input message bits and the presence of the white Gaussian noise in the channel.