A Neural Network-Aided Viterbi Receiver for Joint Equalization and Decoding

In recent years, many works have been focusing on applying machine learning techniques to assist with communication system design. Instead of replacing the functional blocks of communication systems with neural networks, a hybrid manner of ViterbiNet symbol detection was proposed to combine the advantages of Viterbi algorithm and neural networks, which achieves guaranteed performance with reasonable complexity. However, this block-based design not only degrades the system performance but also increases hardware complexity. In this work, we propose a ViterbiNet receiver for joint equalization and channel decoding, which simultaneously considers both the code structure and channel effects, thus achieving global optimum with 3 dB gain. Furthermore, a dedicated neural network model is proposed to avoid the need for perfect channel state information (CSI). It is shown to be more robust under CSI uncertainty with 1.7 dB gain.