Performance Analysis and Deep Learning Assessment of Full-Duplex Overlay Cognitive Radio NOMA Networks Under Non-Ideal System Imperfections

• Published in: IEEE transactions on cognitive communications and networking 2023-06, Vol.9 (3), p.664-682
• Main Authors: Singh, Chandan Kumar, Upadhyay, Prabhat Kumar, Lehtomaki, Janne J.
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Closed form solutions; Cognitive radio; Cognitive radio (CR); deep learning (DL); deep neural network (DNN); Defects; Ergodic processes; Exact solutions; full-duplex (FD); Hardware; hardware impairments (HIs); Interference; Interference cancellation; Machine learning; NOMA; non-orthogonal multiple access (NOMA); Nonorthogonal multiple access; overlay system; Receivers; Relays; Throughput; Wireless networks
• ISSN: 2332-7731; 2332-7731
• DOI: 10.1109/TCCN.2023.3246532

In this paper, we investigate the effectiveness of an overlay cognitive radio (OCR) coupled with non-orthogonal multiple access (NOMA) system using a full-duplex (FD) cooperative spectrum access with a maximal ratio combining (MRC) scheme under the various non-ideal system imperfections. In view of practical realization, we ponder the impact of loop self-interference, transceiver hardware impairments, imperfect successive interference cancellation, and channel estimation errors on the system performance. We investigate the performance of the proposed system by obtaining closed-form expressions for outage probability and ergodic rate for primary as well as secondary users using Nakagami- {m} fading channels. As a result, we reveal some notable ceiling effects and present efficacious power allocation strategy for cooperative spectrum access. We further evaluate the system throughput and ergodic sum-rate (ESR) to assess the system's overall performance. Our findings manifest that the FD-based OCR-NOMA can comply with the non-ideal system imperfections and outperform the competing half-duplex (HD) and orthogonal multiple access (OMA) counterparts. Due to the massive complexity of the suggested system model, direct derivation of the closed-form formula for the ESR becomes cumbersome. To address this problem, we develop a deep neural network (DNN) framework for ESR prediction in real-time situations.