Sparse Representation for Blind Spectrum Sensing in Cognitive Radio: A Compressed Sensing Approach

Abstract

Cognitive radios enable opportunistic transmission for secondary users (SUs) without interfering the primary user (PU). Cyclo-stationary-based spectrum sensing methods are better than the energy detection methods in negative signal-to-noise (SNR) decibel (dB) regime, in which case the noise variance cannot be exactly estimated. However, blind cyclo-stationary methods require a large number of symbols (and hence measurements). This paper aims to reduce the number of measurements in a blind sensing method (using a combination of linear prediction and QR decomposition), by employing compressed sensing at the receiver front-end, so as to reduce the A/D requirements needed with a large number of measurements, along with oversampling the received signal. Till now, compressed sensing has not been investigated at very low negative SNR (dB), e.g., $$-$$-12 dB, which is very crucial in spectrum sensing. The novel algorithm, in this paper, overcomes this shortcoming, and its simulation results show that the SU is able to detect the PU signal, using much less measurements, even at very low negative SNR (dB). The proposed method also investigates the effect of joint and individual measurement matrices at multiple oversampled branches.