Deep reinforcement learning-based network coding for time-sensitive wireless avionics intra-communications

Wireless avionics intra-communications (WAIC) have emerged as a promising technology to address the challenges of weight, fuel consumption, flexibility and maintenance costs in traditional wired avionics systems. However, when integrating WAIC with backbone networks, such as time-sensitive networking (TSN), interoperability challenges arise due to the heterogeneous characteristics of wired and wireless media at both the physical (PHY) and medium access control (MAC) layers. Furthermore, ensuring high reliability with low deadline-miss probabilities for mixed-criticality avionics traffic remains a crucial issue. In this article, we propose a novel priority-aware random linear network coding (PA-RLNC) strategy optimized by a deep reinforcement learning (DRL)-based framework. Specifically, PA-RLNC classifies packets according to their criticality, applying guaranteed inclusion for high-priority packets and probabilistic activation for low-priority packets, while dynamically adapting coding redundancy based on real-time channel conditions and queue states. To achieve optimal performance under dynamic WAIC conditions, a soft actor-critic (SAC)-based framework is introduced to optimize critical coding parameters. Experimental results demonstrate that the proposed scheme improves the success rate and delay performance of high-priority traffic while providing robust statistical quality-of-service (QoS) improvements for time-sensitive WAIC networks. Critically, the parallel nature of PA-RLNC and SAC algorithms makes the proposed scheme well-suited for high-performance computing acceleration, helping to meet the strict real-time requirements of next-generation avionics.