Deep Reinforcement Learning for MU-MIMO Beamforming Training in mmWave WLAN

In IEEE 802.11ay wireless local area network (WLAN), a single access point (AP) performs multi-user multiple-input-multiple-output (MU-MIMO) beamforming training (BFT) to enable simultaneous directional communications with multiple stations (STAs). During MU-MIMO BFT, the AP transmits a significant number of action frames to multiple STAs, and the performance of MU-MIMO BFT depends on how the AP configures its transmit antenna arrays for the transmission of these action frames. In this paper, we develop an algorithm that utilizes a deep reinforcement learning model to learn from the configuration of transmit antenna arrays and the BFT feedback of the STAs in previous MU-MIMO BFT processes, enabling the accurate configuration of transmit antenna arrays for the transmission of action frames in the current MU-MIMO BFT process. Through performance evaluation, our proposed deep reinforcement learning scheme demonstrated improved performance in terms of latency and failure probability of MU-MIMO BFT compared to existing studies that require estimating the signal-to-noise ratios (SNRs) measured at the STAs during the transmission of action frames.