Towards green networking: Efficient dynamic radio resource management in Open-RAN slicing using deep reinforcement learning and transfer learning

Next Generation Wireless Networks (NGWNs) are characterized by agility and flexibility. It introduces new technologies such as network slicing (NS) and Open Radio Access Network (O-RAN). NS supports multiple services with different requirements whereas O-RAN supports different network suppliers and provides Mobile Network Operators (MNOs) more intelligent control. Deep Reinforcement Learning (DRL) techniques have been presented to address resource management and other problems in NGWNs in recent years. However, instability and lateness in convergence are the main obstacles against their adoption in live networks. Moreover, deep learning models consume lots of energy and emit significant amounts of carbon dioxide which badly impacts climate. This paper addresses solving the dynamic radio resource management (RRM) problem in O-RAN slicing with DRL and Transfer Learning (TL), focusing on proposing a green model that minimizes power and energy consumption, decreasing the carbon footprint. A new latency-and-reliability-based reward function is designed. Then, a variable threshold action filtration mechanism is proposed, and a policy TL approach is proposed to accelerate the performance in commercial networks. Compared with the state-of-the-art, this work significantly improved exploration stability, convergence speed, Quality of Service (QoS) satisfaction, power and energy consumption, and emitted carbon footprint.