Efficient Beam Selection in mmWave Cellular Systems Using Neural Networks and K-Nearest Neighbors Based on GPS Coordinates

With B5G moving towards 6G, the possibility of having even higher capacity and lower latency is becoming more realistic and expected to be driven more by mmWave frequencies. However, a major issue in these systems remains the downlink beam alignment and training procedure within mmWave cellular networks. Beam selection, as part of the physical layer and the medium access control sublayer, is critical for discovering and pairing superior beams for reliable connections. In this research, machine learning using neural networks (NN) and K-nearest neighbours (KNN) is proposed for selecting the beam based only on the GPS coordinates of the receiver. This method is more efficient than conventional methods that may involve, for instance, protracted or computationally expensive beam searches or hard-to-obtain side information. An improved selection is achieved by proposing a novel selection architecture in the proposed method using NN-KNN while ensuring the best performance out of competing methods by using the average received signal reference power (RSRP) and top-K accuracy metric. This approach has shown that, despite imprecise data of the receiver location, it is a more efficient solution for future wireless communications systems. The results imply potential improvements in beam selection concerning efficiency, which can support the further development of mmWave for future B5G and 6G networks.