Bayesian Compressive Sensing Using Adaptive Threshold for Block Sparse Wideband Signal Recovery

By using Relevance Vector Machine (RVM) to solve the problem of sparse signal recovery, Bayesian Compressive Sensing (BCS) can obtain good performance in spectral discrete spike signal detection. However, in cognitive radio (CR) system, the spectrum of primary user's signal, which is continuous in narrowband and is block sparse in wideband, cannot be exactly recovered by BCS. In this paper, a Bayesian Compressive Sensing Using Adaptive Threshold (AT-BCS) is proposed, in which we improve the iterative algorithm of RVM for enhancing the accuracy of block sparse wide band signal recovery. Specifically, before every iteration, we set a threshold to select potentially valuable weights for signal recovery so that we can reduce the total number of iterations and develop the accuracy of signal recovery simultaneously. Furthermore, we take the change of noise into account, exploiting Support Vector Machine (SVM) to fit a function to describe the relationship between SNR and threshold. Thus, we can offer different thresholds for each signal recovery procedure with different SNR to guarantee the robustness of our algorithm.