Sub-Nyquist Sampling for Wideband Spectrum Sensing with Cyclic Feature Detection

In cognitive radio (CR) systems, fast and reliable spectrum sensing is a crucial enabling operation. As the band of interest is very wide in CR systems, the conventional Nyquist sampling is very costly or even impractical, and sub-Nyquist samplers based on compressive sensing (CS) provide an attractive alternative. Among various CS-based sub-Nyquist samplers, we focus on modulated wideband converter (MWC) in this paper, due to its advantages in computational complexity and robustness against model mismatch and hardware non-idealness. The CR systems are required to work in dynamic environments with strong noise and interference, so the spectrum sensing is prone to serious estimation error caused by strong noise and noise uncertainty. To increase the sensing accuracy, we propose to use cyclic feature detection and extract the cyclic statistics of the primary signal directly from the MWC¡¯s output without recovering the signal itself or its frequency spectrum. As the estimation of cyclic statistics is insensitive to noise and noise uncertainty, by avoiding the noise-sensitive input-signal-recovery the proposed technique can work well in dynamic environments with strong noise and interference. The key contribution is the establishment of a transformed linear connection between the output of MWC and the cyclic statistics of the input signal, which does not straightforwardly exist. This technique combines the advantages of both MWC and cyclic feature detection, including low sampling rate, low computational cost, and robustness against noise. Simulation results show that it can achieve relatively good sensing accuracy with an average SNR as low as -10dB and noise uncertainty.