Deep Learning-Based Millimeter Wave Beam Recommendation via Channel Knowledge Map

Abstract

In this letter, a deep learning based beam recommendation method is proposed for millimeter wave beam alignment, leveraging the concept of Channel Knowledge Map (CKM). To facilitate reliable beam recommendation, an Enhanced-CKM that integrates a beam probability map with the conventional beam index map is proposed to better capture the relationship between user equipment location and optimal beamforming direction. A multi-feature fusion based deep learning model is designed to fuse the input features from Enhanced-CKM and is trained to output the predicted probability of being the optimal for each candidate beam. An adaptive beam recommendation strategy is then proposed to recommend a beam subspace with a variable number of beams, such that the sum of the output probabilities exceeds a pre-determined threshold. Numerical results validate the effectiveness of the proposed method, achieving better tradeoffs between beam subspace size and beam recommendation accuracy compared to existing methods.