Experimental Study on a Novel Machine Learning-Assisted Covariance-Based Spectrum Sensing Algorithm in the Presence of Noise

Spectrum sensing (SS) plays a pivotal role in cognitive radio (CR) systems, facilitating efficient spectrum utilization. Covariance-based spectrum sensing (CSS) techniques have demonstrated superior performance compared with traditional methods. Recently, the integration of machine learning (ML) models has further improved CSS algorithms. In this article, we introduce a novel ML-assisted CSS algorithm and evaluate its efficacy using real-world datasets, considering practical hardware imperfections. In addition, we examine its performance in the presence of the
$\alpha $
-stable distributed noise, in its impulsive and white Gaussian forms. In addition, we conduct comparative experiments with various ML-aided CSS techniques. Our experimental results consistently highlight the superiority of the proposed ML-assisted CSS algorithm over alternative methods, regardless of the ML model employed. Furthermore, our findings underscore the impact of realistic noise conditions, attributed to hardware imperfections and correlated noise, on the performance of CSS algorithms. This insight emphasizes the importance of considering practical factors and different types of noise in the design and evaluation of SS techniques.