The Spectrum Detection And Location Based On Variational Sparse Bayesian Learning

The sparse Bayesian learning method is based on a Bayesian framework. Thesparse Bayesian learning can fully mining and make use of a priori information, theassumed a priori information of probability distribution, reasonable modeling of theproblem to be solved to implement low dimensional model of learning. The sparseBayesian learning was still difficulty used for idle spectrum sensing. The reasons areoptimal hypothesis computational cost is larger, and the sensing accuracy is imprecise.This research paper revolves around these issues.A novel spectrum detection algorithm based on variational sparse Bayesianlearning algorithm is proposed for reducing the complexity of the compressedspectrum sensing for wideband. The algorithm directly uses the compressedmeasurement to estimate the location and number of the primary users and adoptsvariational sparse learning to obtain the sparse weights although the priori knowledgeis unknown. And it reduces the computation difficulty of marginal likelihood functionby adopting the approximation of simple factorial function. And the variational sparseBayesian approach can provide a consistent framework of the derivation for derivingthe model of graphic.. The experimental results show that the algorithm can obviouslyimprove the sensing speed and accuracy..In order to overcome the effects of deep fading, shadowing effect and hiddenterminal on cognitive sensing accuracy, each cognitive users can obtain globaloptimal solution, under in the case of without fusion center, the algorithm ofdistributed variational sparse Bayesian Spectrum detection algorithm based on factorgraph is proposed. The method can combines the spectrum estimation and locationtogether, and mapping the problem to the model of factor graph. The algorithm.canpassing message by using sum-product algorithm to achieve the sharing ofinformation. In order to reduce the complexity of the algorithm, it adopts the methodof variational to approximation the probability of posterior. And the algorithm prunsthe divergence of hyperparameters and the corresponding basis functions for reducingthe amount information of kl,ukland the load of communication. The experimentalsimulation results show that the distributed spectrum detection algorithm can achievesensitive detection for authorized user information under the low SNR.