ML-Driven Feature-Based Spectrum Sensing for NOMA Signal Detection in Spectrum-Agile IoT Networks Under Fading Channels

Efficient spectrum sensing (SS) in cognitive radio (CR) networks (CRNs) is crucial for maximizing spectrum utilization and ensuring reliable communication in nonorthogonal multiple access (NOMA) systems, especially in challenging fading environments. Traditional SS methods rely on predefined detection thresholds, which are difficult to determine in NOMA-based networks and often suffer from poor accuracy in challenging fading environments. This emphasizes the need for advanced machine learning (ML) techniques to overcome detection limitations and eliminate threshold dependence. Despite recent advances, there remains a significant gap in applying ML to enhance SS performance in NOMA-based CRNs, where accurate signal detection is essential for identifying idle spectrum opportunities. This article introduces an ML-driven approach for SS in CR-based NOMA systems within the Internet-of-Things (IoT) environments. The proposed method uses feature-based SS (FBSS) to detect primary user (PR) signals in fading channels without needing a predefined detection threshold by providing adaptive data-driven strategies that improve detection performance. Several ML classifiers, including logistic regression (LR) and random forest (RF), are incorporated into the sensing framework to improve detection performance. The effectiveness of these classifiers is evaluated based on their ability to identify signals accurately. Realistic NOMA signals with cyclic delays and noise are generated using Monte Carlo simulations to simulate real-world communication conditions. The results demonstrate that the ML-based approach significantly improves the detection probability across various signal-to-noise ratio (SNR) values compared to traditional feature-based methods.