Kernel canonical correlation analysis for robust cooperative spectrum sensing in cognitive radio networks

Spectrum sensing is a key operation in Cognitive Radio (CR) systems, where secondary users (SUs) are able to exploit spectrum opportunities by first detecting the presence of primary users (PUs). In a CR network composed of several SUs, the detection accuracy can be much improved by cooperative spectrum sensing (CSS) strategies, which exploit the spatial diversity among SUs. However, cooperative detection strategies, which are typically based on energy sensing, do not perform satisfactorily under impairments such as non-Gaussian noise or interferences. In this paper, we propose a scheme based on kernel canonical correlation analysis (KCCA), which is able to operate in non-ideal scenarios and in a totally blind fashion. This technique is performed at the fusion center (FC) by exploiting the non-linear correlation among the received signals of each SU. In this manner, statistical tests are extracted, allowing the SUs to make decisions either autonomously at each SU or cooperatively at the FC. The performance of the KCCA-based detector is evaluated by means of simulations and over-the-air experiments using a CR testbed composed of several Universal Radio Peripheral (USRP) nodes. Both the simulations and the measurements show that the KCCA-based detector is able to obtain a significant gain over a conventional energy detector, whose sensing performance is severely degraded by the presence of external interferers. Copyright c © 2014 John Wiley & Sons, Ltd.