Multiobjective Reinforcement Learning-Driven Polar Code Construction for B5G IoV

Internet of Vehicles communications need high-reliability and low-latency channel codes. This paper aims to improve the error probability performance of the rate-compatible polar codes (RCPC) by optimizing the common coding construction. To this end, the novel construction method is proposed for the RCPC cluster with multiple lengths and patterns, particularly where a multiobjective reinforcement learning (MORL) approach is utilized. Both the reconstruction-aided fast construction algorithm and the successive cancellation list decoding-oriented integrated construction algorithm are designed, by combining the code properties and MORL. Meanwhile the construction algorithm complexity is reduced by optimizing the action space and reward calculation. Simulation results and analysis verify the overall reliability performance advantages of the proposed method for the RCPC cluster, compared to existing construction schemes.