Optimized D2D Mode Selection in H-CRANs Using Asynchronous Federated Deep Reinforcement Learning with Federated Averaging

In the context of heterogeneous cloud radio access networks (H-CRANs), there's a growing challenge in achieving efficient resource allocation due to the surge in data and complexities related to asynchrony in federated learning frameworks. This research presents an innovative methodology that blends federated averaging with asynchronous federated deep reinforcement learning (AF-DRL). By enabling individual agents to interact with distinct environments and subsequently adjust policy parameters, local updates are consolidated at a central point to generate a global update. Through this iterative procedure, the system attains optimal transmission mode selection and enhanced resource block distribution, thereby significantly boosting the overall performance of H-CRANs while ensuring that latency and reliability benchmarks are met.