Eigenvalue fusion-based machine learning approach for cooperative spectrum sensing in cognitive radio

摘要

Spectrum sensing is the most critical and fundamental function of the cognitive radio (CR) for dynamic spectrum usage. Machine learning (ML) techniques that allow CRs to learn the environment adaptively are most essential in spectrum sensing. This paper proposes a novel machine learning-based cooperative spectrum sensing (CSS) technique using various advanced ML schemes to enhance the probability of detection. ML classifiers used in the fusion centre of the CSS schemes are trained with eigenvalue-based test statistics extracted from the covariance matrix of a received signal. The proposed system recognises the received signal samples as a PU signal or a noise signal and distinguishes the PU signal from noise effectively under low SNR conditions using a threshold that possesses the self-learning ability. The simulation results exhibit the performance analysis of various ML algorithms for CSS under different wireless scenarios and their suitability is compared with conventional approaches.