Learning-Based Spatial Reuse for WLANs With Early Identification of Interfering Transmitters

In this paper, a reinforcement learning-based spatial reuse scheme for wireless local area networks (WLANs) is proposed and analyzed. In this scheme, when an access point (or a station) overhears an on-going transmission, it decodes the information in the frame header to identify the transmitter and decides whether or not to exploit spatial reuse accordingly. Specifically, it decides whether to stop receiving the remaining part of the frame and start its own transmission or to refrain from channel access until the detected transmission finishes. Through the repeated update Q-learning (RUQL) algorithm, the agent learns the optimal decision in the sense of reducing the media access control layer delay. Moreover, we compare the proposed scheme with the spatial reuse operation in IEEE 802.11ax, which makes the spatial reuse decision only based on a binary identification of the detected interferer, i.e., whether it is in my cell or neighboring cells. The proposed scheme, however, treats different interferers differently for exploiting spatial reuse. From a theoretical perspective, we derive a theoretical bound on the gains in the value function, i.e., the discounted sum of delay, due to making non-binary identifications. Simulation evaluations confirm that the proposed scheme achieves high throughput by reducing the time of freezing backoff counter while not increasing the time of failed transmissions.