GLRT-Based Spectrum Sensing with Blindly Learned Feature under Rank-1 Assumption

Using signal feature as the prior knowledge can improve spectrum sensing performance. In this paper, we consider signal feature as the leading eigenvector (rank-1 information) extracted from received signal's sample covariance matrix. Via real-world data and hardware experiments, we are able to demonstrate that such a feature can be learned blindly and it can be used to improve spectrum sensing performance. We derive several generalized likelihood ratio test (GLRT) based algorithms considering signal feature as the prior knowledge under rank-1 assumption. The performances of the new algorithms are compared with other state-of-the-art covariance matrix based spectrum sensing algorithms via Monte Carlo simulations. Both synthesized rank-1 signal and real-world digital TV (DTV) data are used in the simulations. In general, our GLRT-based algorithms have better detection performances, and the algorithms using signal feature as the prior knowledge have better performances than the algorithms without any prior knowledge.