Entropy-based robust spectrum sensing in cognitive radio

Abstract

Sensitivity to noise uncertainty is a fundamental limitation of current spectrum sensing strategies in detecting the presence/absence of primary users in cognitive radio (CR). Because of noise uncertainty, the performance of traditional detectors such as matched filter, energy detector and even cyclostationary detectors deteriorates rapidly at low signal-to-noise ratio (SNR). Without accurate estimation of noise power, an absolute /SNR wall/ exists in traditional detectors below which robust detection is impossible, no matter how long the observations are. To counteract the effect of noise uncertainty in low SNR, the authors propose a blind frequency-domain entropy-based spectrum sensing scheme. The entropy of the sensed signal is estimated in the frequency domain with probability space partitioned into fixed dimensions. The authors prove that the entropy of noise is a constant and the proposed scheme is thus intrinsically robust against noise uncertainty. Monte Carlo experiments are carried out to verify the robustness and further show that the proposed scheme outperforms classical energy detector and cyclostationary detector in low SNR region with 6 and 4/dB performance improvement, respectively. In addition, the sensing time is reduced to about 75/ by the proposed scheme compared to energy detector under the same detection performance.