Cooperative Spectrum Sensing with Small Sample Size in Cognitive Wireless Sensor Networks

Cooperative spectrum sensing based on Dempster–Shafer (D–S) theory has attracted a large amount of interest in cognitive wireless sensor networks. However, most of them employ energy detection (ED) in local sensing, where the classical Gaussian approximation of ED is accurate only with a large number of data samples. In this paper, aiming at drastically reduce the computational cost and the sensing process duration, we consider that a small sample size is collected at each node of the network. In this configuration, to perform the D–S fusion we introduce new basic probability of assignment functions derived from the statistics of the eigenvalues of the samples covariance matrix. To that end, we introduce a relevant approximation of the Tracy–Widom distribution that allows us to cope with the small sample size. Simulation results show that the proposed method allows to improve significantly the detection performance compared to other techniques, even with small number of samples.