Evolutionary Multi-Objective Optimization Algorithm for Resource Allocation Using Deep Neural Network in Ultra-Dense Networks

It is certain that in the modern era the ultra-dense network (UDN) structure will play a major role for the evolution of 5G and beyond wireless communication system, particularly for blind wireless area and hotspot. In resource constraint environment, obtaining higher energy efficiency (EE), spectrum efficiency (SE), and greater fairness during resource allocation process are conflicting objectives. To obtain the balance among them a multi-objective optimization problem (MOOP) is designed and an enhanced version of non-dominated sorting genetic algorithm II (NSGA-II), which integrates the advantage of evolutionary method and machine learning framework is suggested. Firstly, the chromosome coding scheme is designed which is suitable for spectrum allocation. Afterward, a deep learning framework is designed to enable the self-tuning of crossover and mutation operators to improve the diversity of candidate solutions. Further, an elitist retention strategy is modified by designing variable fraction scheme. This intelligent approach enables micro-cell users to improve their downlink performance of SE, EE, and fairness by assigning resource blocks. The simulated results yield the effectiveness of the proposed scheme in perfect and imperfect channel state information (CSI) environment by analysing the obtained performance gains when compared with other existing allocation methods in terms of EE, SE, and fairness.