Performance Analysis with Deep Learning Assay for Cooperative UAV-Borne IRS NOMA Networks under Non-Ideal System Imperfections

Abstract

Internet of Things (IoT) inspired wireless networks foresee intelligent reflecting surfaces (IRSs) and non-orthogonal multiple access (NOMA) as promising techniques to boost the spectral and energy efficiency, while unmanned aerial vehicles (UAVs) as fast and flexible entity for enhancing the wireless connectivity. In this regard, we study a cooperative UAV-borne IRS system employing NOMA transmissions to serve multiple users on the ground. We take into account the impacts of residual hardware impairments (HIs) in devices and imperfect successive interference cancellation (I-SIC) in NOMA, which are inevitable in practical system implementation. We analyze the system performance by deriving the closed-form expressions of outage probability (OP) and ergodic rate of users over the line-of-sight (LoS) Rician fading channels for the aerial IRS links and non-LoS Rayleigh fading for the terrestrial direct links. We further pursue asymptotic analysis for both OP and ergodic rate to reveal useful insights on the high signal-to-noise ratio (SNR) slope and achievable diversity order. Also, we evaluate the system throughput under the delay-limited and delay-tolerant transmission modes. Above all, aiming toward real-time IoT applications of UAV-IRS empowered NOMA system, we present a deep neural network (DNN) framework for OP and ergodic rate predictions with a short execution time under the dynamic stochastic environment. Our results validate the theoretical analyses and accentuate the performance advantages of the proposed UAV-borne IRS NOMA system over the conventional orthogonal multiple access (OMA) equivalent system. IEEE