CNN-Based Multiantenna Spectrum Sensing for Hybrid Circular and Noncircular Source Signals in Cognitive Radio

Recently, multiantenna spectrum sensing (MSS) using de-ep neural network has received much attention. However, most of them are not explicitly formulated to account for non-circular sources. In practice, the primary users' signals may come from either circular or noncircular sources, or both. We are, therefore, motivated to investigate the MSS problem for hybrid circular and noncircular source signals in cognitive radio. Specifically, we first introduce the augmented sample covariance matrix (ASCM) to exploit the additional statistical information given by the noncircular property. Then, based on the ASCM, a simple convolutional neural network (CNN) structure is designed to extract the features embedded in the ASCM. Simulation results reveal that the proposed method is capable of providing performance improvement over conventional MSS methods, regardless of the spatially white/correlated noise.