DQRL-Based Dynamic Resource Allocation for Energy-Spectral Efficient Next-Generation Clustered Radio Access Networks

In next-generation Radio Access Networks (RAN), dynamically and intelligently selecting Radio Units (RUs) with varying power levels is essential to maximize throughput, ensure energy efficiency, and maintain reliable transmission quality. The 6Gparadigm relies heavily on real-time adaptive decision-making, requiring an optimal RUs assignment for each User Equipment (UE) based on critical parameters such as Signal-to-Noise Ratio (SNR), transmit power, Spectral Efficiency (SE) and traffic load. To achieve this, we propose a Deep Q-Reinforcement Learning (DQRL)-based framework that evaluates the SNR between all possible UE-RU pairs under different power configurations. The framework optimizes its selection policy by balancing energy ef f iciency and transmission quality, encoded in a carefully designed reward function. Initially, the agent explores the environment through random RU selection (exploration phase), but over time, it transitions to an epsilon-greedy policy, gradually favoring high reward actions to minimize power consumption while sustaining Quality of Service (QoS) requirements. By continuously refining its policy through interactions with the network environment, the DQRL agent converges to an optimal strategy that dynamically allocates UEs in each time interval, achieves a strong trade off between Energy Efficiency (EE) and SE compared to static or heuristic approaches. This method ensures scalable and intelligent resource management in 6G networks, adapting to fluctuating channel conditions and traffic demands.