Reinforcement Learning Assisted Decoding

This paper explores the application of reinforcement learning techniques in the context of the performance improvement of bit-flipping based decoders. We begin with a concise overview of bit-flipping based decoders and reinforcement learning algorithms. We then outline the methodology involved in mapping these iterative decoders into Markov Decision Processes and propose a method to decrease the number of states to make the Q-learning algorithm feasible for low-rate and long-length codes. This enables us to obtain an optimal decision rule and improve the decoding performance through the utilization of reinforcement learning algorithms. Subsequently, we conduct an analysis of the reinforcement aided bit-flipping based decoder and investigate a number of potential optimal solutions achievable through reinforcement learning algorithm. We provide a comparative examination of efficiency and complexity trade-offs between data-driven algorithms and traditional methods across the Binary Symmetric Channel and Additive White Gaussian Noise Channel.