Spectrum Assignment in 5G and Beyond Ultra-Dense Networks Utilizing Deep Neural Network Framework

Extremely high throughput and data rates can be achieved with ultra-dense networks (UDN). However, the wide-ranging and random deployment of small base stations (SBSs) causes significant interference. A key issue is how to distribute cellular resources among small-cell user equipments (SUEs) in an effective manner to enhance the performance of these mobile services. One of the most significant challenges is figuring out how to distribute cellular resources across network users in a way that maximizes performance for these mobile applications. The generative adversarial networks double deep Q-network (GAN-DDQN) strategy has been proposed as a means of overcoming interference and obtaining effective energy-efficient resource allocation. To mitigate the impact of widely varying utility values, we propose a reward-clipping strategy for GAN-DDQN training. In a simulated setup, the suggested plan is assessed. The suggested strategy considerably increases energy efficiency, SINR, and throughput, as the numerical and simulation findings confirm.