Cooperative Multi-Agent Reinforcement Learning-Based Wi-Fi Multiple Access

Carrier sensing multiple access with collision avoidance (CSMA/CA) is a basic medium access control (MAC) protocol specified in the IEEE 802.11 standard. However, the current <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$C$</tex>SMA/<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$C$</tex>A mechanism has limitations in providing high throughput and low latency services in user-dense scenarios. To address this issue, this paper investigates the distributed channel access problem based on communicative multi-agent reinforcement learning (MARL), where each user acts as an agent, deciding whether to transmit data in the current time slot based on its local observation and the messages from other agents. We propose a novel MAC protocol, which employs a centralized training with decentralized execution paradigm. Moreover, we adopt the parameter sharing approach with positional encoding as an identifier to distinguish different agents, which further improves the network adaptability. The simulation results illustrate that the proposed Selective Communication-based Multiple Access (SCMA) algorithm: 1) outperforms the conventional <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$C$</tex>SMA/<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$C$</tex>A in terms of throughput and latency, while maintaining user fairness; 2) improves the model adaptability to accommodate varying numbers of users.