Estimation of time-varying unknown nongaussian noise with DPM

Dirichlet process mixture (DPM) model, which is the state-of-the-art Bayesian nonparametric model, was introduced here to signal processing research field. In present Bayesian statistics it is used to model and inference random nongaussian distributions. We explored its ability to model and estimate nongaussian unknown stationary noise and our work will help dealing with problems in many fields of signal processing. Through some modifications, we also revealed its potential to model and estimate unknown nonstationary nongaussian noise. Sequential Monte Carlo based inference algorithm was developed to estimate time varying unknown nongaussian noise with DPM. Simulation results show the efficiency of our algorithm.