Deep reinforcement learning for initial access beam alignment

• Applicant: Professor Dr.-Ing. Stephan ten Brink
• Subject Area: Communication Technology and Networks, High-Frequency Technology and Photonic Systems, Signal Processing and Machine Learning for Information Technology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 560528105

The challenging propagation environment, combined with the stringent hardware limitations of wireless communication systems operating at mmWave frequencies, creates the need for accurate initial access beam alignment (BA) strategies with low latency and high achievable beamforming gain. Recently, many promising deep learning (DL)-based BA schemes have been proposed and were often shown to greatly outperform their conventional, non-DL based counterparts. However, end-to-end differentiable systems, extensive knowledge of the mmWave channel, and/or a specific hardware architecture are often assumed for learning the neural network (NN) weights of these schemes, limiting their practical relevance. We believe that those shortcomings of current DL-based initial access BA algorithms can be mitigated by the use of the more general deep reinforcement learning (DRL)-framework for NN training. Thus, next to extending and further developing recent advances in DL-based BA, we plan to develop novel methods for applying DRL to BA.

• DFG Programme: Research Grants