Maximum Correntropy Unscented Kalman Filter for Spacecraft Relative State Estimation

Xi Liu, Hua Qu, Jihong Zhao, Pengcheng Yue, Meng WangSensors 2016 · DOI

TL;DR

This paper introduces the Maximum Correntropy Unscented Kalman Filter (MCUKF) to improve the robustness of spacecraft relative state estimation in the presence of non-Gaussian, particularly impulsive, noise. By incorporating the Maximum Correntropy Criterion (MCC) into the Unscented Kalman Filter (UKF) framework, the MCUKF demonstrates superior performance in challenging noise environments compared to standard filters.

What Problem Does This Paper Address?

Accurate relative state estimation is crucial for various space missions, including satellite formation flying, rendezvous, and docking. Traditional Kalman filters (KF) and their non-linear extensions like the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) are widely used. However, these filters are typically designed for Gaussian noise. In real-world scenarios, especially in space communications, measurements can be corrupted by heavy-tailed, impulsive noises, which significantly degrade the performance of EKF and UKF, potentially leading to filter divergence. This paper aims to address this limitation by developing a more robust filtering approach.

Methods

The core contribution of this paper is the development of the Maximum Correntropy Unscented Kalman Filter (MCUKF). The MCUKF builds upon the Unscented Kalman Filter (UKF) by integrating the Maximum Correntropy Criterion (MCC).

Here's a breakdown of the approach:

  • Unscented Kalman Filter (UKF) Foundation: The paper starts by outlining the standard UKF algorithm. The UKF uses a deterministic sampling approach (sigma points) to approximate the probability distribution of the state, propagating it through non-linear functions without requiring explicit Jacobian calculations, which is an advantage over the EKF.
  • Maximum Correntropy Criterion (MCC): The MCC is introduced as a robust criterion for state estimation. It's based on the concept of correntropy, a generalized similarity measure that captures higher-order statistics of data, making it less sensitive to outliers than traditional minimum mean-square error (MMSE) methods (which are equivalent to L2-norm minimization). The paper defines correntropy using a Gaussian kernel and provides its sample mean estimator.
  • MCUKF Formulation: The MCUKF combines these two elements. The Unscented Transform (UT) is used to predict the state and covariance. Crucially, the measurement update step is reformulated using a non-linear regression model under the MCC. This involves using the MCC to process the measurement information, effectively making the filter more resilient to impulsive noise. The paper details how the MCC is applied to a transformed measurement model to derive the modified filter equations.

Datasets

The paper does not introduce new datasets. Instead, it uses two illustrative examples for evaluation:

  1. Univariate Nonstationary Growth Model (UNGM): A standard benchmark for nonlinear filtering, used to test the filters under both Gaussian and non-Gaussian (heavy-tailed) noise conditions.
  2. Two-Spacecraft Relative Motion: A practical example simulating the relative motion of two spacecraft in orbit, using a realistic dynamic model and measurement equations derived from radar observations. This example also evaluates performance under Gaussian and non-Gaussian noise.

Key Insights and Ideas

  • Robustness through Correntropy: The fundamental insight is that the MCC, by considering higher-order moments and being less sensitive to outliers than L2-norm minimization, provides a powerful mechanism to enhance the robustness of Kalman filters against impulsive noise.
  • UKF as a Robust Base: Leveraging the UKF's ability to handle non-linearities without Jacobians makes it a suitable foundation for developing a robust filter. The MCUKF retains the UKF's structure while improving its noise resilience.
  • Kernel Bandwidth is Key: The performance of the MCUKF is significantly influenced by the choice of the kernel bandwidth (σ) in the correntropy function.
  • A smaller bandwidth generally leads to greater robustness against outliers.
  • A larger bandwidth makes the MCUKF behave more like the traditional UKF.
  • There's a trade-off: too small a bandwidth can lead to filter divergence or accuracy issues, while too large a bandwidth diminishes the robustness advantage.
  • MCUKF Approaches UKF with Large Bandwidth: A theoretical result (Theorem 1) is presented, showing that as the kernel bandwidth approaches infinity, the MCUKF converges to the standard UKF. This provides theoretical backing for the observed behavior.

Key Results

The paper demonstrates the MCUKF's effectiveness through simulations:

  • Gaussian Noise: In purely Gaussian noise scenarios, the standard UKF generally performs best. The MCUKF's performance approaches that of the UKF as the kernel bandwidth increases.
  • Non-Gaussian (Impulsive) Noise: This is where the MCUKF shines.
  • In the UNGM example with heavy-tailed measurement noise, the MCUKF with a properly chosen kernel bandwidth (e.g., σ = 2.0) achieved the smallest Mean Squared Error (MSE), outperforming the UKF and other robust filters like HEKF and NRUKF.
  • In the spacecraft relative motion example with mixed-Gaussian noise, the MCUKF with a proper kernel bandwidth (σ = 2.0) again exhibited the smallest TAMSDp and TAMSDÿ, indicating superior estimation accuracy.
  • Computational Cost: The MCUKF has a moderate computational cost, being slightly higher than the UKF but significantly lower than other robust filters like HEKF and NRUKF.

Limitations and Future Work

  • Kernel Bandwidth Selection: The paper acknowledges that selecting the optimal kernel bandwidth is crucial and often requires trial-and-error or manual tuning. This is a practical challenge for real-world deployment.
  • Convergence Speed: The paper notes that correntropy is a local similarity measure, and the MCUKF might converge slowly, especially if initial estimates deviate significantly from true values. In their second example, they addressed this by using the UKF for an initial period before switching to MCUKF.
  • Future Work: While not explicitly detailed, the findings suggest that further research into adaptive methods for kernel bandwidth selection or exploring other robust criteria within the UKF framework could be valuable. Investigating the MCUKF's performance in even more complex and dynamic non-Gaussian noise environments would also be a logical next step.

Summary updated

Abstract

A new algorithm called maximum correntropy unscented Kalman filter (MCUKF) is proposed and applied to relative state estimation in space communication networks. As is well known, the unscented Kalman filter (UKF) provides an efficient tool to solve the non-linear state estimate problem. However, the UKF usually plays well in Gaussian noises. Its performance may deteriorate substantially in the presence of non-Gaussian noises, especially when the measurements are disturbed by some heavy-tailed impulsive noises. By making use of the maximum correntropy criterion (MCC), the proposed algorithm can enhance the robustness of UKF against impulsive noises. In the MCUKF, the unscented transformation (UT) is applied to obtain a predicted state estimation and covariance matrix, and a nonlinear regression method with the MCC cost is then used to reformulate the measurement information. Finally, the UT is adopted to the measurement equation to obtain the filter state and covariance matrix. Illustrative examples demonstrate the superior performance of the new algorithm.
BibTeX
@inproceedings{liu2016maximum,
  author = {Xi Liu and Hua Qu and Jihong Zhao and Pengcheng Yue and Meng Wang},
  title = {{Maximum Correntropy Unscented Kalman Filter for Spacecraft Relative State Estimation}},
  booktitle = {Sensors},
  year = {2016},
  doi = {10.3390/s16091530},
}
Figure