Reinforcement Learning for Dynamic Channel Assignment Using Predicted Mobile Traffic

In mobile networks, determining the allocation of channels is a longstanding challenge. With the surge in mobile device usage and traffic, the availability of wireless channels is increasingly constrained. Though many studies have employed conventional reinforcement learning for channel allocation, they overlooked future mobile traffic predictions. Recent advancements in mobile traffic forecasting have demonstrated notable accuracy. We introduced a dynamic channel assignment methodology using Proximal Policy optimization (PPO) reinforced learning by incorporating mobile traffic predictions. We validated the proposed method with a comprehensive two-month dataset, capturing mobile traffic from 144 base stations in Milano, Italy, each equipped with 1350 channels. Preliminary results suggested that when paired with a reasonably accurate mobile traffic prediction model, the proposed PPO-based technique outperformed both traditional dynamic channel assignment algorithms and alternative reinforcement learning models, as evidenced by minimized blocking probabilities.