Performance Optimization in Multi-Rate IEEE 802.11 WLANs via Traffic-Aware Contention Control Approach with Bidirectional-LSTM Multi-Agent Deep Reinforcement Learning

This study introduces a traffic-aware contention control approach called CIPD-BiDQN, which integrates Contention Interval Probabilistic Deferment (CIPD) with a Bidirectional-LSTM (Bi-LSTM) Dueling Q -Network for performance optimization in multi-rate IEEE 802.11 WLANs. Unlike prior DRL-based Contention Window (CW) tuning methods that adjust CW directly from single-step collision observations that do not capture short-term variations in contention, the proposed CIPD-BiDQN model uses a Bi-LSTM encoder and a dueling DQN model to learn traffic-aware contention parameters that guide how CW is updated through a proportional and probabilistic backoff process. First, each agent collects a short sequence of feature vector that observe recent collision levels and basic CW statistics of high- and low-rate stations (STAs) to give an accurate estimation of short-term contention variation. Second, the Bi-LSTM encoder works as a sequence transduction model that processes the short feature sequence and represents how collision and CW values change across recent observation steps. Third, the dueling DQN agent selects one CIPD parameter set, including three contention control parameters: collision thresholds for load classification, an Exponentially Weighted Moving Average (EWMA) factor for temporal stability, and offset coefficients for raising the backoff lower bound to reduce simultaneous transmissions. Extensive evaluations under static and non-static IEEE 802.11ax conditions show that CIPD-BiDQN delivers higher throughput for overall, both high-rate and low-rate groups, and manages collisions more effectively than the standard protocol, existing backoff schemes, and recent DRL-based CW baselines. 2018 The Korean Institute of Communications and Information Sciences. Publishing Services by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ).