Joint Optimization of Spectral Efficiency and Energy Harvesting in D2D Networks Using Deep Neural Network

In this work, we study the joint optimization of energy harvesting and spectrum efficiency in wireless device-to-device (D2D) networks where multiple D2D pairs adopt simultaneous wireless information and power transfer (SWIPT) functionality with a power-splitting policy. To observe the trade-off relationship between spectrum efficiency and energy harvesting via SWIPT, we construct an objective function using the weighted sum method, which scalarizes the dominant with spectrum efficiency and energy harvesting, and attempt to find the optimal transmit power and power-splitting ratio to maximize the objective function. Typical iterative search algorithms such as exhaustive search (ES) or gradient search (GS) with a log barrier function are employed to find the global optimum and sub-optimum, respectively. Furthermore, we apply a deep neural network (DNN) learning algorithm to deal with the non-convexity of the objective function with an effective loss function. The simulation results verify the trade-off relationship between spectrum efficiency and energy harvesting, and show that the DNN-based algorithm can achieve a near-global optimal solution with computational complexity much lower than that of the optimization-based iterative algorithms.