Learning‐Based Reliable Channel Ranking Without CSI for Efficient Spectrum Sensing in Cognitive Radio Networks

The need for effective spectrum usage has increased with the rapid growth of wireless communication systems, which has accelerated the development of cognitive radio (CR) networks. In CR networks, reliable spectrum sensing (SS) and channel selection are major concerns particularly under fading and noise uncertainty where conventional detectors experience considerable performance degradation. This paper proposes a machine learning (ML) assisted reliable pre‐sensing channel ranking framework that is independent of instantaneous channel state information (CSI). Instead of relying upon explicit channel estimation, it uses statistically observable information obtained from the received signal to perform intelligent channel ranking prior to SS. A reliability metric proportional to effective signal‐to‐noise ratio (SNR) is determined by a supervised random forest regression model to enable the selection of the most reliable channel for sensing and access. The proposed method is tested under multiple wireless fading environments, including AWGN, Rayleigh, Rician and Weibull channels. Performance is evaluated in terms of detection probability, ranking reliability, channel selection accuracy, switching rate, and achievable throughput. Simulation outcome indicates that the learning‐based ranking scheme performs close to the oracle approach while providing a relative improvement of up to 8% for the achievable throughput and channel selection accuracy compared to estimation schemes under moderate SNR values in fading channels.