A Deep Reinforcement Learning-based Contention Window Avoidance Scheme for Distinguishing Network Service Priorities

This study introduces a Deep Reinforcement Learning-based Contention Window Avoidance Scheme aimed at improving service priority differentiation and network throughput under high load conditions within wireless networks that utilize the Network Distributed Coordination Function (DCF) mechanism. This novel approach involves two key strategies: firstly, it enables the differentiation of service priorities within DCF-based channel access by categorizing them in accordance with the varying data types and streams transmitted by different stations; secondly, it leverages Deep Reinforcement Learning to dynamically evaluate the network state through analysis of collision probabilities derived from channel observations. By doing so, the scheme dynamically modulates the ranges of contention windows in alignment with the service priorities, responding to the immediate network circumstances. Empirical evidence indicates that the proposed scheme substantially mitigates collision frequencies and augments data transmission throughput, thus significantly advancing the overall performance of networks that require service priority differentiation.