Measurement-based Indoor Beam Alignment Utilizing Deep Learning

Millimeter wave (mmWave) band offers an unprecedented opportunity for increasing the wireless throughput. Conversely, mmWave wireless transmissions raise several questions. Multiple-input multiple-output (MIMO) communication systems provide solutions for some of the crucial challenges. The most straightforward mmWave MIMO systems are based on analog beamforming. Therefore, precise beam alignment is essential for the proper operation. This paper presents a deep learning-based beam alignment framework, which relies on measured power levels from few distinct directions. The training, validation, and test set is generated by an open-source ray tracer (RT) based framework. The wireless channel simulation has been validated by measurements. Our analysis reveals the effect of deep neural network (DNN) complexity and the distance between consecutive measurement points on the accuracy of the prediction.