Three Novel Statistical Tests-Inspired Spectrum Sensing Techniques for Cognitive Radio

Abstract

Statistical tests-based approaches have been extensively researched for multiple antenna-based spectrum sensing (SS) in cognitive radio (CR). Nevertheless, their performance is neither satisfactory nor adequate to detect the primary user (PU) particularly in weak signal environments. In this paper, three novel statistical tests-based methods are developed to move forward in this promising domain. The first method is blind autocorrelation and randomness test, which is formulated under the framework of the Durbin Watson test. The second is a non-parametric dispersion test, which is designed in the light of analysis of variance (ANOVA) Bartlett’s test. The last is a non-parametric location test, which is inspired by ANOVA on rank the Kruskal Wallis test. Multipath fading channel with additive Gaussian noise is considered in numerical analysis. Furthermore, Middleton Class A impulsive noise is modeled as a non-Gaussian noise (NGN) to simulate the statistical characterization of realistic noise environments. Several comparison scenarios are performed with state-of-the-art SS techniques at the conditions of low signal-to-noise ratio (SNR), few samples, and few number of receiving elements. Simulation results revealed that our suggested methods outperform the other SS techniques at different levels. Moreover, closed-form expressions of test statistics and asymptotic theoretical thresholds are derived for the three investigated algorithms. Additionally, theoretical analysis is validated by being coincident with simulation findings.