Deep Reinforcement Learning-Based Beam Training with Energy and Spectral Efficiency Maximisation for Millimetre-Wave Channels

The millimetre-wave (mmWave) spectrum has been investigated for the fifth generation wireless system to provide greater bandwidths and faster data rates. The use of mmWave signals allows large-scale antenna arrays to concentrate the radiated power into narrow beams for directional transmission. The beam alignment at mmWave frequency bands requires periodic training because mmWave channels are sensitive to user mobility and environmental changes. To benefit from machine learning technologies that will be used to build the sixth
generation (6G) communication systems, we propose a new beam training algorithm via deep reinforcement learning. The proposed algorithm can switch between different beam training techniques according to the changes in the wireless channel such that the overall beam training overhead is minimised while achieving good performance of energy efficiency or spectral efficiency. Further, we develop a beam training strategy which can maximise either energy efficiency or spectral efficiency by controlling the number of activated radio frequency
chains based on the current channel conditions. Simulation results show that compared to baseline algorithms, the proposed approach can achieve higher energy efficiency or spectral efficiency with lower training overhead.