ADAPTIVE THRESHOLDING IN CYCLOSTATIONARY SPECTRUM SENSING USING ARTIFICIAL NEURAL NETWORK UNDER NON-STATIONARY NOISE CONDITIONS

Abstract

An efficient spectrum sensing is a primary requirement in cognitive radio networks to ensure optimal utilization of underused spectral resources without interfering with licensed users. The cyclostationary feature detection shows periodic statistical properties of modulated signals to distinguish them from noise, offering superior detection reliability at low signal-to-noise ratios. However, the performance of cyclostationary feature detection can degrade when the noise characteristics or signal-to-noise vary dynamically. This work proposes an artificial neural network assisted adaptive thresholding framework for cyclostationary feature detection based spectrum sensing to address these challenges. The artificial neural network is trained to predict optimal detection thresholds under different SNR and noise variance conditions, thereby improving detection accuracy while maintaining optimal computational complexity. Simulation results show that the proposed method achieves high detection probability, P_d under severe noise fluctuations and low SNR environments. Specifically, M-ary phase shift keying exhibits reliable performance down to –30 dB, while M-ary quadrature amplitude modulation maintains accuracy up to –20 dB. The findings indicate that artificial neural network driven adaptive thresholding can significantly enhance the adaptability and robustness of cognitive radio systems in spectrum sensing process.