Learning-based strategies for millimeter wave radio beamforming and sensing

This doctoral thesis offers learning-based strategies to perform fast and reliable mmWave radio beamforming in drone-base station (BS) environments. A reinforcement learning (RL)-based framework is proposed to maximize the beamforming gains during drone-BS communication following 3rd generation partnership project (3GPP) communication standards. The simulations showed that RL-based beam alignment is generic and converges faster than the state-of-the-art learning-based and conventional schemes in an online manner under real-time conditions. The proposed learning framework is also analyzed for different BS coverage requirements, antenna arrangements, configurations, and varying channel conditions. RL-based framework can also be extended to perform joint optimization with drones involving mmWave beamforming like connectivity-constrained trajectory planning. The simulations showed that RL-based joint path planning and beam tracking method is on par with the learning-based shortest path planning, besides beam tracking comparable to the heuristic beam searching method.

Learning-based strategies are also proposed to perform fast and accurate mmWave beam sensing during communication without additional hardware requirements. The solutions are based on extracting side information during radio communication and simultaneously sensing information like human radio blockages. The simulations showed that learning-based framework using mulit-layered perceptron can sense coarse-grained and fine-grained mmWave blockage directions with reasonable accuracies using received signal measurements.