Deep Learning Aided Power Allocation in An Energy Harvesting Untrusted Relay Network

In an energy harvesting untrusted relay network, power allocation influences the cooperative jamming, the energy harvesting and thus the achievable secrecy rate. In our previous work, theoretical computation of power allocation is derived with high computation. To tackle this issue, in this paper, we propose a deep learning aided power allocation. We here utilize fully-connected deep neural network (FC-DNN) to predict the optimal power allocation factor, where the feature vector and the model structure are carefully designed. Simulation results show the deep learning aided power allocation achieves almost the same power allocation factor and the maximum secrecy rate as the theoretical one, which validates the correctness and accuracy of the proposed scheme. Special case with small optimal power allocation factor is simulated and analyzed in detail. Furthermore, the convergence with different learning rate and batch size is also discussed.