Power Allocation for Multi-user Cooperation: a Multi-Objective and Machine Learning Approach

Energy-efficient optimal power allocation (OPA) has always been an essential optimization for multi-user cooperative diversity systems. However, most of the existing works have mainly focused on optimizing the system’s overall energy efficiency (EE) while neglecting to maximize the EE of each user jointly which is an inherent demand for multi-user systems. To this end, in this letter, we investigate the quality-of-service constrained EE-OPA to maximize the EE for each user simultaneously in a multi-user decode-and-forward cooperative system by a multi-objective optimization approach. This constrained multi-objective optimization problem (MOOP) is solved by jointly leveraging the weighted Tchebycheff method and the Dinkelbach method, however, with a considerably high computational complexity. In order to reduce the computational complexity while still obtaining near-optimal solutions, we further proposed a machine learning approach to solve the MOOP. Specifically, we setup an Elman neural network to model and learn the multi-objective EE-OPA (MO-EE-OPA) for a given multi-user DF cooperative network. Numerical results show that the Elman network can output near-optimal solutions with dramatically low computational complexity.