A Fully Connected Neural Network For Polar Channel Decoding.

One of the most essential components of a communications system is channel coding, which provides protection for the original source message. Polar codes have gained great importance due to the fact that they reach the channel capacity for the Binary Discrete Memoryless Channel. Both encoding and decoding, for this type of codes, are serial processes, which increases latency. In this research, we propose a fully connected neural network architecture that tackles this problem in decoding thanks to the inherent parallelization of this type of structure. Our experimental results demonstrate that a fully connected neural network reduces decoding delay, although further study is still needed to reach the accuracy of the probabilistic decoding methods.