A recurrent neural decoder for convolutional codes

A new neural approach to decode convolutional codes is presented. The method uses a recurrent neural network tailored to the used convolutional code. No supervision is required. As an example, decoders for 1/2 and 1/3 rate convolutional codes with constraint length 9 are studied. Such codes have been proposed, e.g., for the third generation WCDMA cellular system. The new decoders have been tested in a Gaussian channel and it is shown that the performance of the Viterbi algorithm can be approached very closely. The decoder lends itself to pleasing implementations in hardware. Its complexity increases only polynomially with increasing constraint length, which is slower than the exponential increase of the Viterbi algorithm. However, the speed of the current circuits may set limits to the codes used. With increasing speeds of the circuits in the future, the proposed technique may become a tempting choice for decoding convolutional coding with long constraint lengths.