Reinforcement Learning in 5G Heterogenous Network: A Survey

The advent of 5G heterogenous network (HetNets) with densely deployed femtocells, offer enhanced capacity, handoff, throughput, and spectrum utilization. Yet, the intricacies of radio resource allocation pose significant challenges, exacerbated by interference, penetration losses, reduced capacity and throughput. Consequently, mitigating interference within femtocell networks is imperative for optimal network performance. By surveying existing methodologies and exploring machine learning (ML) driven solutions, the reinforcement learning (RL) algorithm has emerged as a promising solution for optimize network performance. The RL Algorithm has been surveyed as a capable solution to the various problems of 5G HetNet focusing on double deep neural network (DDNN). We reviewed existing RL approaches including Q-learning and deep q-networks (DQN), highlighting their limitation. Our analysis revealed that using DDQN approach which can dynamically adjust power control and frequency allocation strategies to optimize network performance, DDNN superiority stem from ability learn complex pattens, adapt to dynamic environments and generalize well. DDQN ability to handle non-stationarity and reduce overestimation bias makes it a promising approach for improving spectral efficiency and reducing interference in complex HetNet environments. this research contributes to the evolving landscape of 5G network optimization and interference mitigation strategies.