Spectrum sensing using non-asymptotic behavior of eigenvalues

The classical random matrix theory is mainly focused on asymptotic spectral properties of random matrices when their dimensions tend to infinity. At the same time, many recent applications, like convex geometry, functional analysis and information theory, operate with random matrices of fixed dimensions. In this paper, we investigate a recently developed non-asymptotic behavior of eigenvalues of random matrices, which is about spectral properties of random sub-Gaussian matrices of fixed dimensions. Then, a new spectrum sensing scheme for cognitive radio is proposed by using the non-asymptotic behavior of eigenvalues. Simulation results show that the proposed scheme has a better detection performance than the classical energy detection technique and the scheme based on asymptotic behavior of eigenvalues of random matrices, even in the case of a small sample of observations.