Exponential Linear Unit and Swish with Deep Neural Network Based Resource Allocation in Wireless Communication Systems

The Device-to-Device (D2D) communications in wireless communication system ensures the energy and spectral efficiency, data rates and total system capacity. However, the dynamic and complex nature of a 5G system, various technological scenarios, provide significant challenges in Resource Allocation (RA) of D2D communication. To overcome these limitations, an Exponential linear unit and Swish based Deep Neural Network (ES-DNN) for energy efficient RA in D2D wireless communication system. The combination of ELU and swish activation function that helps to optimize resources such as power, and user allocation in without excessive computational overhead, improve system performance under high traffic. The ELU activation function introduces a smoother gradient when values are negative, that allowing the DNN model to avoid dead neurons and maintain a steady flow of information in dynamic environment. The experimental results of the proposed method achieved energy consumption for 200, 400, and 500 signals are 86.72%, 92.49%, and 93.68% which is lesser than the existing approaches like Deep Deterministic Policy Gradient (DDPG).