Unified Linearization-based Nonlinear Filtering

Anton Kullberg,Isaac Skog,Gustaf Hendeby

2023-09-14

Abstract:This letter shows that the following three classes of recursive state estimation filters: standard filters, such as the extended Kalman filter; iterated filters, such as the iterated unscented Kalman filter; and dynamically iterated filters, such as the dynamically iterated posterior linearization filters; can be unified in terms of a general algorithm. The general algorithm highlights the strong similarities between specific filtering algorithms in the three filter classes and facilitates an in-depth understanding of the pros and cons of the different filter classes and algorithms. We end with a numerical example showing the estimation accuracy differences between the three classes of filters when applied to a nonlinear localization problem.

Signal Processing

What problem does this paper attempt to address?

The main objective of this paper is to unify nonlinear filtering algorithms and cover various existing filtering techniques through a general algorithmic framework. Specifically, the paper attempts to address the following key issues:

1. **Unifying different types of filters**: The paper proposes a general algorithmic framework that can unify standard filters (such as the Extended Kalman Filter), iterative filters (such as the Iterative Unscented Kalman Filter), and dynamic iterative filters (such as the Dynamic Iterative Posterior Linearization Filter) within a single algorithmic framework. This allows for a better understanding of the similarities and differences between these filters.

2. **Improving linearization methods for nonlinear models**: Traditional linearization methods typically perform local linearization based on the mean of the current state estimate, which can lead to filter divergence in the case of large errors. The paper discusses methods to improve the linearization point through an iterative process, thereby enhancing the performance of the filter.

3. **Demonstrating performance differences among different filter categories**: The paper uses a specific numerical example (the sound source localization problem) to demonstrate the performance differences among three different categories of filters (standard, iterative, and dynamic iterative filters), particularly highlighting the robustness of dynamic iterative filters in handling poor process noise parameters.

Through this general framework, researchers can gain a deeper understanding of the working principles and advantages and disadvantages of various linearization filters, thereby providing a broader reference for practical applications.