Fully Cooperative MADRL for Energy-Efficient Resource Allocation in D2D Communications

Device-to-Device (D2D) communication is considered a key technology for improving data transmission rates and reducing energy consumption. To overcome the drawbacks of centralized resource allocation schemes, such as their dependence on global information, this paper investigates the joint spectral subband selection and power control problem for D2D communication in cellular networks, where wireless channels are dynamically changing and a distributed resource allocation scheme is developed. The optimization objective is to maximize the energy efficient (EE) of all D2D pairs while satisfying the minimum transmission rate requirements of both cellular users to the base station (BS) and D2D links. To address the issue of EE resource allocation, this paper proposes a full cooperative multi-agent double deep Q-network distributed algorithm based on low-dimensional fingerprints and soft update framework (FMDL). Through this algorithm, each agent can utilize its locally observed information to cooperatively maximize the system's EE. The analysis of uplink communication cost and computational complexity demonstrates that the proposed algorithm significantly reduces both the uplink transmission overhead and the computational burden on the BS. The simulation results demonstrate that compared to the benchmark algorithm, FMDL significantly improves the EE of all D2D pairs.