Latency-Bounded Reliability-Oriented Degree Distribution of Rateless Codes: A Knowledge-Assisted Reinforcement Learning Method

Existing Luby transform (LT) codes struggle to maintain reliability in latency-bounded industrial systems. To address this, we propose an imitation learning and reinforcement learning based degree distribution optimization (ILRL-DO) method. Initially, imitation learning (IL) pretrains the policy to alleviate sparse reward issues during reinforcement learning (RL). Subsequently, RL minimizes overhead while ensuring latency constraints, guiding the policy toward the optimal degree distribution. Simulations show that ILRL-DO improves reliability under bounded latency while maintaining competitive overhead performance.