Sense-Bandits: AI-based Adaptation of Sensing Thresholds for Heterogeneous-technology Coexistence Over Unlicensed Bands

Abstract

In this paper, we present Sense-Bandits, an AI-based framework for distributed adaptation of the sensing thresholds (STs) over shared spectrum. This framework specifically targets the coexistence of heterogenous technologies, e.g., Wi-Fi, 4G Licensed-Assisted Access (LAA), and 5G New Radio Unlicensed (NR-U), over unlicensed channels. To access the channel, a device compares the measured power with a predefined ST value and accordingly decides if the channel is idle or not. Improper setting of the ST values creates asymmetric sensing floors, resulting in collisions due to hidden terminals and/or reduction in the spatial reuse due to exposed terminals. Optimal ST setting is challenging because it requires global knowledge of mobility, traffic loads, and channel access behavior of all contending devices. Sense- Bandits tackles this problem by employing a clustering-based multi-armed bandit (MAB) algorithm, which adapts its learning behavior based on network dynamics. Clustering allows the algorithm to track network changes in real-time, ensuring fast learning of the best ST values by classifying the state and dynamics of coexisting networks. We develop a C++-based network simulator that implements Sense-Bandits and we apply it to evaluate the coexistence of Wi-Fi and 5G NR-U systems over the unlicensed 5 GHz U NII bands. Our simulation results indicate that ST-adaptive devices employing Sense-Bandits do not harm neighboring devices that adopt a fixed ST value.