A deep neural network based MU-MIMO beamforming training protocol for IEEE 802.11ay

IEEE 802.11ay supports a multi-user multiple-input-multiple-output (MU-MIMO) beamforming training (BFT); hence, an access point (AP) can perform data transmissions simultaneously with multiple stations (STAs). During MU-MIMO BFT, the AP sends a significant number of action frames to the STAs to determine appropriate directional antenna patterns. However, if transmit sectors (TSs) used in the transmission of action frames are selected inefficiently, the AP could perform redundant transmissions; this could eventually lead to the poor performance of the MU-MIMO BFT in terms of signaling and latency overhead. To select the appropriate TSs, the AP is required to accurately estimate the link qualities measured at the STAs when certain TSs are used simultaneously to transmit an action frame. Therefore, in this study, we propose a deep neural network (DNN)-based scheme that performs efficient MU-MIMO BFT. It achieves this by accurately estimating link qualities measured at the STAs when an action frame is transmitted through multiple TSs. Moreover, it performs this function without collecting any additional channel information that is not specified in the standard. For performance evaluation, we conducted extensive simulations with realistic mmWave channel and antenna array models in four indoor and outdoor propagation scenarios. The simulation results demonstrate that our scheme transmits fewer action frames and achieves a lower BFT time than existing schemes.