Distributed Beamforming by Multi-Agent Active Inference
Tatsuya Otoshi, Masayuki Murata · 2024
Massive MIMO systems have emerged as a promising technology for next-generation wireless communication networks. Beamforming, a key technique in Massive MIMO, significantly enhances the system's performance by exploiting the spatial domain. This paper presents a collaborative beamforming approach that leverages vertical collaboration between base stations to improve adaptation and mitigate fading variations, while also applying the principle of free energy to optimize beamforming. The approach involves sharing estimated channel states and learned models among base stations, enabling them to collectively optimize beamforming based on the principle of free energy. Evaluations in single-user operation scenarios demonstrate that both with and without vertical collaboration, the expected free energy, representing beamforming performance, decreases over time. However, vertical collaboration reduces temporary decreases in signal-to-interference-plus-noise ratio (SINR) caused by independent adaptation to fading variations. Furthermore, in a multi-user switching scenario, the proposed approach ensures stable control by utilizing learned models and state estimation results, leading to improved beamforming performance during switching.