Federated Deep Reinforcement Learning-Based Spectrum Sharing and Power Allocation for Mobile Communication System

The escalating demand for heightened performance in global mobile communication systems necessitates novel methods to optimize spectrum and communication resource utilization. Dynamic spectrum access in cognitive radio systems has garnered significant attention, particularly employing deep reinforcement learning (DRL) for resource allocation. In this paper, a federated DRL based algorithm is proposed to deal with both the spectrum access and power allocation quandaries in mobile communication systems. In the proposed algorithm, we model secondary users as learning agents, enabling them to acquire spectrum access and power selection strategies through training with local training and federated model aggregation. By sharing pertinent information and aggregating local models, users enhance the efficiency of the training process. The proposed approach is expected to promote channel access, resource efficiency, and user privacy in the future communication system.