Optimal classifier based spectrum sensing in cognitive radio wireless systems

In this work, we present and investigate the performance of novel classification schemes for spectrum sensing in cooperative multiple-input multiple-output (MIMO) wireless cognitive radio (CR) networks. In this context, we consider several optimal classification schemes such as support vector classifiers (SVC), logistic regression (LR) and quadratic discrimination (QD) for primary user detection. It is demonstrated that these classification techniques have a significantly reduced complexity of implementation in practical CR applications compared to conventional likelihood based detection schemes as they do not require knowledge of the channel state information and noise power. Further, in the presence of disruptive malicious users, the proposed classifiers have a significantly lower detection error compared to conventional detection schemes. Also, we propose a novel QD classifier for blind MIMO spectrum sensing scenarios. The detection performance of the proposed classifiers is compared with existing schemes in co-operative CR scenarios. It is demonstrated through simulation of several scenarios including the presence of malicious users, Doppler shift, and carrier frequency offset that the proposed classifiers offer a robust and significantly superior alternative to existing schemes for co-operative MIMO CR spectrum sensing.