Globally Optimal Multisensor Distributed Random Parameter Matrices Kalman Filtering Fusion with Applications

Yingting Luo, Yunmin Zhu, Dandan Luo, Jie Zhou, Enbin Song, Donghua WangSensors 2008 · DOI

TL;DR

This paper introduces a novel distributed Kalman filtering fusion method designed for systems with uncertain parameters (random matrices). It proves that this distributed approach achieves the same optimal performance as a centralized system, even when sensor noises are correlated. This work extends Kalman filtering to handle complex real-world scenarios like uncertain observations and false alarms.

What Problem Does This Paper Address?

In many modern applications, such as robotics, aerospace, and defense, systems rely on multiple sensors to estimate the state of an object or environment. While centralized Kalman filtering, where all sensor data is processed at a single location, offers optimal performance, it's often impractical due to communication bandwidth limitations or the need for system resilience. Distributed Kalman filtering, where each sensor processes its own data and then fuses estimates, is a more practical alternative. However, a significant challenge arises when the system's parameters (like state transition and measurement matrices) are not fixed but vary randomly. This randomness, especially when sensor noises or parameter matrices are correlated across sensors, can violate the assumptions of standard distributed Kalman filtering, leading to suboptimal performance. This paper tackles the problem of achieving globally optimal state estimation in multisensor systems with random parameter matrices, even when sensor correlations exist.

Methods

The core contribution of this paper is a new distributed Kalman filtering fusion algorithm specifically designed for systems with random parameter matrices. The authors first convert the original system with random matrices into an equivalent system with deterministic matrices but state-dependent process and observation noises. They then rigorously prove that this converted system still satisfies the conditions for standard Kalman filtering.

The key innovation lies in demonstrating that their proposed distributed fusion scheme, even when dealing with correlated sensor noises or random measurement matrices, achieves a fused state estimate that is equivalent to that of a centralized Kalman filter using all sensor measurements. This equivalence ensures global optimality. The paper also shows how this framework can be extended to handle more complex scenarios like uncertain observations and systems with multiple dynamic models.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and uses simulated data for numerical examples.

Key Insights and Ideas

  • Equivalence of Distributed and Centralized Fusion: The most significant insight is proving that a distributed Kalman filtering fusion with random parameter matrices can achieve the same optimal performance as a centralized approach, even under mild conditions and with correlated sensor noises. This is a crucial theoretical result for practical distributed systems.
  • Handling Randomness: The paper provides a robust framework for dealing with systems where parameters are not fixed but vary randomly. This is a more realistic representation of many real-world dynamic systems.
  • Unified Approach for Uncertainties: The proposed method offers a unified way to handle various sources of uncertainty, including random parameter matrices, uncertain observations, and false alarms, by framing them within the context of random parameter matrices.
  • Performance Loss of Ignoring Randomness: Numerical examples clearly demonstrate that ignoring the randomness of parameter matrices leads to a significant performance degradation in state estimation.

Key Results

The paper presents a theoretical framework and demonstrates its effectiveness through numerical examples. The key results include:

  • Proof of Global Optimality: The proposed distributed Kalman filtering fusion is proven to be equivalent to centralized Kalman filtering, thus achieving the best possible performance in terms of Mean Square Error (MSE).
  • Handling Correlated Sensor Noises: The method is shown to be effective even when sensor noises or random measurement matrices are correlated across sensors, a scenario that typically breaks standard distributed filtering assumptions.
  • Application to Uncertain Observations and False Alarms: The framework is successfully applied to Kalman filtering with uncertain observations, including the special case of measurements with false alarm probabilities. Numerical examples show that the proposed random Kalman filtering fusion performs significantly better than standard Kalman filtering when these uncertainties are present.
  • Application to Multi-Model Systems: The method can also be applied to randomly variant dynamic systems with multiple models, offering an optimal MSE solution where previous methods might not have.

Limitations and Future Work

  • Synchronous Observations: The paper assumes synchronous sensor observations. While the authors note that extensions to asynchronous cases are possible by treating missing observations as a special case, this remains a practical challenge.
  • Fixed Model Probabilities: In the application to multi-model systems, the paper assumes fixed occurrence probabilities for each model. The authors acknowledge that in practice, these probabilities are often state or observation-dependent, and their method might not offer an advantage in such cases.
  • Future Research Directions: The authors suggest that further research could explore decentralized algorithms that are more resilient to the loss of sensing nodes, as well as investigate the random Kalman filtering framework for such scenarios. They also point out that the random Kalman filtering in the framework of [16] may be another future research direction.

Summary updated

Abstract

This paper proposes a new distributed Kalman filtering fusion with random state transition and measurement matrices, i.e., random parameter matrices Kalman filtering. It is proved that under a mild condition the fused state estimate is equivalent to the centralized Kalman filtering using all sensor measurements; therefore, it achieves the best performance. More importantly, this result can be applied to Kalman filtering with uncertain observations including the measurement with a false alarm probability as a special case, as well as, randomly variant dynamic systems with multiple models. Numerical examples are given which support our analysis and show significant performance loss of ignoring the randomness of the parameter matrices.
BibTeX
@inproceedings{luo2008globally,
  author = {Yingting Luo and Yunmin Zhu and Dandan Luo and Jie Zhou and Enbin Song and Donghua Wang},
  title = {{Globally Optimal Multisensor Distributed Random Parameter Matrices Kalman Filtering Fusion with Applications}},
  booktitle = {Sensors},
  year = {2008},
  doi = {10.3390/s8128086},
}
Figure