Applying Deep Reinforcement Learning for Self-organizing Network Architecture

Inefficient resource allocation and unstable connection quality for mobile devices are the primary challenges of Self-Organizing Networks (SON). Frequent handovers between base stations result in a network burden imbalance. In contrast, unstable connection quality causes disconnection or signal interference between mobile devices and base stations, influencing network performance and reliability. In recent years, wireless communication technology has extensively used Reinforcement Learning (RL) to obtain the optimal strategy through continuous interaction between agents and their environments. Deep Reinforcement Learning (DRL) is based on Deep Neural Networks (DNN) to handle increasingly complex network situations. We proposed a SON architecture based on DRL in response to the aforementioned challenges. We described how the agent learns the optimal parameter settings through training based on various network scenarios to develop handover strategies and enhance overall network performance and resource utilization. The proposed framework can be applied to the present Fifth Generation (5G) network.