Wideband Compressive Spectrum Sensing Research Based On Statistical Learning

Cognitive radio is an advanced technique to solve the spectrum scarcity problem and to improve the spectral efficiency. Spectrum sensing is the core technology of cognitive radio, which is the key to find out spectrum hole. A fast and accurate spectrum sensing procedure is the base of cognitive system. However, in wideband communications, the Nyquist sampling rate is too high to afford for hardware equipment. To deal with this problem, the current domestic and international common practice is to introduce compressive sensing. By doing so, we can sample the signal at a lower rate and reduce the requirements for hardware.This article deeply analyzed the theory and merit and demerit of compressive sensing and introduce statistical learning theory to improve compressive spectrum sensing algorithm. The main innovation point as follows:Firstly, when a single cognitive user independently performs spectrum sensing, the proposed algorithm based on bayesian compressive sensing uses Support Vector Machine to set a threshold, which screens valuable elements of original signal to signal reconstruct. By doing so, the proposed algorithm can reduce the computation complexity and weaken the influence of noise, and then improve the accuracy of signal reconstruction and reduce the operating time.Secondly, when many cognitive users cooperatively sensing the spectrum, the proposed algorithm utilizes Boltzman Machine to infer the statistically relationship between cognitive users, on which a cooperatively signal reconstruction procedure performs. By doing so, the proposed algorithm can deal with the channel fading and improve the reconstruction accuracy.In this paper, research has shown that the wideband compressive spectrum sensing based on statistical learning can effectively improve the performance of spectrum detection.