Title:Distributed Spectrum and Power Allocation for D2D-U Networks: A Scheme based on NN and Federated Learning

Abstract:In this paper, a Device-to-Device communication on unlicensed bands (D2D-U) enabled network is studied. To improve the spectrum efficiency (SE) on the unlicensed bands and fit its distributed structure while ensuring the fairness among D2D-U links and the harmonious coexistence with WiFi networks, a distributed joint power and spectrum scheme is proposed. In particular, a parameter, named as price, is defined, which is updated at each D2D-U pair by a online trained Neural network (NN) according to the channel state and traffic load. In addition, the parameters used in the NN are updated by two ways, unsupervised self-iteration and federated learning, to guarantee the fairness and harmonious coexistence. Then, a non-convex optimization problem with respect to the spectrum and power is formulated and solved on each D2D-U link to maximize its own data rate. Numerical simulation results are demonstrated to verify the effectiveness of the proposed scheme.