Sparse Bayesian Compressed Spectrum Sensing Under Gaussian Mixture Noise

Improving the performance of spectrum sensing in cognitive radio (CR) systems by exploiting sparse property via compressed sensing has been attracting a lot of research attention. However, existing compressed spectrum sensing approaches mainly focus on the signal model that is impacted by the additive white Gaussian noise. In fact, the noise existed is always with much more complicated form in practical CR systems. So, an underestimate of the noise can degenerate the effectiveness of current methods. To alleviate this problem, this paper attempts to address the signal model with Gaussian mixture noise (GMN) with unknown parameters which is typical in real scenarios. First, an optimization model for compressed spectrum sensing under GMN is constructed. Second, sparsity-inducing penalization term in the Bayesian framework is applied to exploit the sparse property. Third, using the factor graph that describes the hierarchical prior signal models, a variational message passing reconstruction algorithm is proposed. All of the parameters are iteratively calculated in closed form. Simulation results demonstrate the superior performance with respect to conventional and state-of-the-art sparse estimation approaches.