Correlation particle filter

Abstract

The nonlinear dynamic sequence Bayesian estimation model consist of 2 parts, the recursive evaluation followed by the filtered and the estimation based on predictive distributions of unmeasured time-varying signal with noise. A new model based on the combination of particle filter (PF) and correlation named correlation particle filter (CPF) is proposed in this paper. On the other hand, the state smoothing is also used for this model. That weights the particles' importance according to the Spearman correlation coefficient between the noisy observations of measured signal and the particles' observations. The sample impoverishment problem is absent because the resampling step is removed in the correlation particle filter. The analysis and results simulated by the proposed model are shown to indicates the versatility and accuracy of the correlation particle filter over those PFs known by us such as the sequential importance resampling (SIR) model, and the Gaussian sum particle filter, the lower time complexity of the correlation particle filter than those PFs such as the SIR model the auxiliary particle filter (APF) and the regularized particle filter, and almost the same time complexity of CPF like Gaussian particle filter (GPF).