Manipulating the Rainbow in Vehicular Relay Networks Using DRL: Beyond Beam Sweeping

Relaying offers promise to improve reliability in millimeter-wave (mmWave) vehicular networks, which are susceptible to link outages caused by blockage. The benefits of relaying, however, may be limited by the time overhead and undesired beam directions resulting from the commonly used exhaustive beam sweeping and arrays based on phased-shifters (PSs). In this paper, we propose a beam training scheme with hybrid arrays using phase shifters (PSs) and true time delay (TTD) elements based on deep reinforcement learning (DRL). The algorithm leverages frequency dependent beam patterns, akin to a rainbow beam, to track relay vehicles with negligible time overhead and point toward the desired direction within the wide bandwidth. Numerical simulations shows that the proposed method outperforms state-of-the-art DRL-based relay selection algorithm using phased arrays, motivating further investigation.