Robust spectrum sensing based on statistical tests

Spectrum sensing, in particular, detecting the presence of licensed or incumbent users in licensed spectrum, is one of the pivotal tasks in cognitive radio network. In this study, the authors tackle the spectrum sensing problem by using statistical test theory and derive novel spectrum sensing approaches. The authors apply the classical Kolmogorov-Smirnov (KS) test under the assumption that the noise probability distribution is known. However, as in practice, the exact noise distribution is unknown, a sensing method for Gaussian noise with unknown noise power is proposed in this article and referred to t-sensing. The proposed sensing scheme is asymptotically robust and can be applied to non-Gaussian noise distributions. A closed form equation determining the miss-detection probability for the t-sensing is derived. The authors compare with the performance of our sensing algorithms with the energy detector and Anderson-Darling (AD) sensing methods proposed in literature. Simulation results show that the proposed sensing methods outperform both ED- and AD-based sensing, especially for the case when the received signal-to-noise ratio is low.