Cyclostationary Feature Detectors (CFD) have been studied in the past few years as an efficient and feasible candidate for a Primary User (PU) detection method. CFD detector is an interesting alternative to energy detector, as it exploits hidden periodicities present in PU signals, but absent in noise. The CFD use quadratic transformations of the signals to determine the hidden periodicities. However, some knowledge about the signal might be needed at the detector (eg: the cyclic frequency), which leads to a non-blind detection. In this paper, we propose new blind sensing methods based on the investigation of the sparsity of the quadratic transformations in the cyclic frequency domain of man-made input signal. The proposed detectors are based on compressive sensing theory, where a distinctive feature can yield a sparse representation that is defined by only a very small number of so-called atoms. By exploiting the sparse property of the cyclic autocorrelation function in the case of man-made signal, we develop interesting detection algorithms that are not only blind and reliable but also computationally efficient for vacant bands detection. Simulation of one of these proposed methods show promising performance results of the proposed technique in terms of sensing vacant sub-bands in the spectrum.