Particle Weight Redistribution in Particle Filtering to Improve Nonlinear State Estimation

Particle filters are used to estimate hidden state parameters from the noisy signals. They are based on sequential Monte Carlo theory, where they randomly draw samples from all possible hidden states and then use the previous information of estimation to proceed to the next step. Particle filters are also known for their degeneracy issues where the particles are separated into high probability and low probability regions. Therefore, resampling is incorporated into particle filters. However, resampling brings up another issue of particle impoverishment and consequently reduce particle diversity. This paper proposes an improved resampling process for a standard particle filters to address particle impoverishment in standard sampling importance resampling particle filters (SIR-PF). B y introducing a weight redistribution of particles before resampling, the algorithm expects to mitigate error values across varying signal-to-noise ratios, potentially outperforming conventional SIR-PF. The experiment is validated using a simulated signal, resulting in overall lower RMSE and MAE at various noise levels. The proposed method is a computationally efficient and straight forward process compared to more complex alternatives.