Throughput-Aware Beam Optimization for Smart Antennas via Deep Reinforcement Learning and Embedded System Integration

This thesis presents a throughput-aware beam optimization framework for smart antenna systems based on deep reinforcement learning. A PPO agent with recurrent neural networks is trained using real-time throughput measurements collected via an embedded system, enabling online adaptation to dynamic wireless environments. Experimental results demonstrate improved throughput and more robust beam selection compared to conventional heuristic approaches.