ABSTRACT This study presents a reinforcement learning‐based power allocation framework for free‐space optical (FSO) nonorthogonal multiple access (NOMA) systems under moderate atmospheric turbulence and intelligent jamming attacks. Modelling the base station (BS)–jammer interaction as a dynamic zero‐sum game, we employ Q‐learning (QL) to enable the BS to adaptively optimise user power distribution. The novelty of this framework lies in its model‐free QL formulation that jointly optimises BS power allocation against an adaptive jamming agent in a stochastic FSO‐NOMA environment, incorporating geometric losses, atmospheric attenuation and log‐normal turbulence fading without requiring explicit channel state information, thereby enabling emergent robust anti‐jamming policies validated through hyperparameter‐tuned simulations. Extensive simulations, comparing the proposed approach against Fixed Equal Power, Random and MaxPower baselines, validate the strategy. Results demonstrate that the BS develops robust anti‐jamming policies, achieving a 15% increase in valid data rate (from 6.4 to 7.4 bits/s/Hz), outperforming nonadaptive strategies that fail under intelligent interference. The efficacy of this resilient, adaptive strategy confirms its potential for securing high‐throughput FSO‐NOMA deployments in adversarial environments.
Mohammad Ali Amirabadi (Thu,) studied this question.