Enhanced Centralized Contention Window Optimization with Deep Reinforcement Learning on Highly Dense Wireless Networks

Abstract

With the continuous development and widespread adoption of wireless networks, there is an increasing demand for wireless network performance in highly dense network environments. The contention window (CW) value plays a crucial role in wireless network performance. However, the traditional Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) mechanism using the binary exponential backoff (BEB) algorithm cannot timely set appropriate CW values for highly dense networks. Therefore, we proposed an innovative CW control method called Enhanced Centralized Contention Window Optimization with DRL (ECCOD). This method dynamically controlled the CW values using deep reinforcement learning (DRL) to adapt to different network load levels. What sets ECCOD apart from other related studies is its ability to choose different CW adjustment modes to handle rapidly changing network conditions. This includes linearly increasing or maintaining the CW value, with further fine-tuning after decision-making. Simulation experiments demonstrate that different adjustment modes can achieve more fine-grained control, as expected. Specifically, our method outperforms CCOD (Centralized Contention Window Optimization with DRL) in terms of throughput performance under lightly loaded network conditions. In heavily loaded network conditions, our method achieves an average throughput improvement of 2.34% compared to CCOD, and surpasses the traditional CSMA/CA method by 38.59%.