Blockage-Aware Power Allocation Algorithm for Millimeter-Wave Communication with Dynamic Reward Q-Learning

Area of Science:

• Wireless Communication
• Machine Learning
• Network Engineering

Background:

• Millimeter-wave (mmWave) systems face challenges like signal attenuation and interference.
• Maintaining Quality of Service (QoS) under dynamic conditions is crucial.

Purpose of the Study:

• To develop a lightweight, distributed power allocation framework for mmWave base stations.
• To address challenges posed by signal blockage and interference.

Main Methods:

• Implemented a tabular Q-learning policy for independent base station updates.
• Utilized locally observable metrics: blockage ratio, serving distance, and aggregate interference.
• Designed a state-dependent dynamic reward function adapting to real-time network conditions.

Main Results:

• The proposed framework achieved performance comparable to fixed Q-learning.
• Maintained a transparent blockage-aware state with low-complexity distributed implementation.
• Outperformed DQN, greedy, and uniform power in specific small-scale network scenarios.

Conclusions:

• The developed method offers an effective solution for power control in mmWave networks.
• Demonstrates empirical benefits and clarifies limitations for deployment.
• Focuses on power control post-beam establishment, excluding joint beam tracking.