Channel-Aware Satisfactory Learning for Robust QoS Self-Provisioning in 5G and Beyond

Today’s services require a certain level of Quality of Service (QoS) to run smoothly and provide a satisfactory experience. Incongruously, most of the research efforts seek to maximize the perceived QoS whilst meeting a certain threshold would be way enough. In this paper, we build a satisfactory and distributed power control for fifth-generation (5G) and future sixth-generation (6G) wireless networks, allowing users to meet the target QoS level. We formulate the distributed power control as a satisfaction game, where each device seeks to achieve its desired QoS level. Next, we propose two strategic learning schemes to be implemented at the device level, and which allow exploring the game’s Nash equilibria under uncertainty. We adapt the Banach-Picard (BP) iterates to learn the satisfactory transmit power equilibrium, in a fully distributed manner. To boost the speed of convergence and improve the accuracy of our power control, we include a prediction phase before the BP iterates. We refer to it as Exponentially Moving Average Banach-Picard (EMA-BP). Extensive simulations were conducted to evaluate the behavior of the proposed schemes under different channel variability and different QoS demands. Result demonstrates that using the EMA-BP distributed learning scheme provides faster convergence toward satisfactory equilibrium transmit power. With the ability to track QoS levels, EMA-BP delivers less fluctuation over the service and maintains satisfaction under noisy channels.