Joint Optimization of Channel Bonding and Transmit Power Using Optimized Actor–Critic Deep Reinforcement Learning for Wireless Networks

Abstract

A high-capacity channel access mechanism is desirable for future Wi-Fi networks. This process must address two factors: channel bonding and spatial reuse. Channel bonding increases the transmission capacity of access points (APs), and in spatial reuse, the APs adjust their transmit power and clear channel assessment threshold (CCAT) to allow them to communicate simultaneously with nearby APs. For efficient channel access, simultaneous optimization is required regarding channel bonding and spatial reuse. To resolve this, a novel optimal Actor–Critic Deep Reinforcement Learning (OAC-DRL) algorithm is proposed to select the optimal AP's channel bonding policy, transmit power, and CCAT under random traffic and channel conditions. OAC-DRL incorporates an actor and critic network and a reward-shaping mechanism to regulate the optimal channel bonding policy for a wireless network. The inclusion of reward shaping reduces the learning time to obtain the optimal actions, whereas the optimality of the original optimal policy remains unchanged. The OAC-DRL algorithm is implemented using the Python. The experimental results show that the OAC-DRL algorithm minimizes queue lengths better under realistic traffic loads. In addition, the OAC-DRL algorithm transmits 4.82% more packets per time slot than other learning algorithms.