Performance Improvement on Power Allocation Using Deep Neural Network in D2D Multi-cell Underlying Communication

Abstract

Device-to-Device (D2D) underlying communication is communication between two devices without going through the base station by using the resources of Cellular User Equipment (CUE). This communication reduce the workload of the base station and increase network capacity. But the resources used simultaneously by the D2D pair and CUE in the underlying communication systems cause interference. To overcome this problem, power allocation needs to be done using Deep Neural Network (DNN) to overcome non-convex problems in maximizing sum-rate and energy efficiency. DNN can be considered a universal approach that can determine the best scheme in the system because it adapt to different environments and can replace the iterative method such as Convex Approximation (CA) based algorithm. This research aims to provide that power allocation using DNN can improve the performance of CA-based algorithm. An increment in the number of CUEs will be seen in sum-rate and energy efficiency. Simulation results show that an increment of CUEs increase sum-rate and energy efficiency. Besides that, DNN can approach the performance of the CA-based algorithm with accuracy above 98%, and improve 2% performance of the CA-based algorithm. DNN is more suitable to implement because it can improve the performance of the CA-based algorithm and can be implemented in different environment because there is learning process in DNN where the model can predict the output based on the input entered. So that DNN is able to produce the best output in power allocation.