Inverse Cubature and Quadrature Kalman filters

Overview

• This paper introduces new inverse stochastic filters for use in cognitive radar applications.
• The proposed filters, including inverse cubature Kalman filter (I-CKF), inverse quadrature Kalman filter (I-QKF), and inverse cubature-quadrature Kalman filter (I-CQKF), are designed to handle highly non-linear system models where the extended Kalman filter (EKF) fails.
• The paper also develops a reproducing kernel Hilbert space (RKHS)-based CKF for unknown system models.
• The stability and consistency of the proposed filters are theoretically analyzed, and their estimation accuracy is demonstrated through numerical experiments.

Plain English Explanation

Cognitive radar is a type of radar system that can adapt and learn from its environment. In some applications, the target being tracked may try to infer what the cognitive radar has learned about them, in order to counter or evade the radar. This is known as the "inverse cognition" problem.

Previous research has addressed this problem by developing inverse Kalman filters (I-KF) and inverse extended Kalman filters (I-EKF) for linear and non-linear Gaussian state-space models, respectively. However, many real-world counter-adversarial settings involve highly non-linear system models, where the EKF's linearization approach often fails to perform well.

To address this, the researchers in this paper have developed several new inverse stochastic filters that use more advanced numerical integration techniques, such as inverse cubature Kalman filter (I-CKF), inverse quadrature Kalman filter (I-QKF), and inverse cubature-quadrature Kalman filter (I-CQKF). These filters are designed to better handle the non-linearities in the system model.

For cases where the system model is unknown, the researchers also developed a reproducing kernel Hilbert space (RKHS)-based CKF approach. The stability and consistency of the proposed filters were theoretically analyzed, and their estimation accuracy was demonstrated through numerical experiments.

Technical Explanation

The paper begins by addressing the inverse cognition problem in cognitive radar applications, where the target being tracked tries to infer the information the radar has learned about them. Previous works have proposed I-KF and I-EKF solutions for linear and non-linear Gaussian state-space models, respectively.

However, the researchers note that many counter-adversarial settings involve highly non-linear system models, where the EKF's linearization approach often fails. To address this, they develop three new inverse stochastic filters: I-CKF, I-QKF, and I-CQKF, which use more advanced numerical integration techniques to better handle the non-linearities.

For the unknown system model case, the researchers develop an RKHS-based CKF approach. They then derive the stochastic stability conditions for the proposed filters in the exponential-mean-squared-boundedness sense and prove the filters' consistency.

The performance of the I-CKF, I-QKF, and I-CQKF filters is evaluated through numerical experiments, with the recursive Cramér-Rao lower bound used as a benchmark. The results demonstrate the estimation accuracy of the proposed filters.

Critical Analysis

The paper presents a valuable contribution to the field of inverse cognition in cognitive radar applications by addressing the limitations of existing approaches in handling highly non-linear system models. The development of the I-CKF, I-QKF, and I-CQKF filters, as well as the RKHS-based CKF for unknown system models, represents a significant advancement in this area.

However, the paper does not discuss any potential limitations or caveats of the proposed methods. For example, it would be interesting to know the computational complexity of the new filters and how they compare to the simpler I-KF and I-EKF approaches in terms of runtime and memory usage.

Additionally, the paper does not explore the real-world applicability of the proposed filters beyond the numerical experiments. It would be valuable to see case studies or simulations of the filters being used in practical cognitive radar scenarios, with a discussion of the challenges and trade-offs involved.

Further research could also investigate the robustness of the filters to model uncertainties, sensor noise, and other practical factors that may arise in operational settings. Leveraging viscous Hamilton-Jacobi PDEs for uncertainty quantification could be a relevant approach to consider in this context.

Conclusion

This paper presents a novel set of inverse stochastic filters, including I-CKF, I-QKF, and I-CQKF, as well as an RKHS-based CKF for unknown system models, to address the inverse cognition problem in cognitive radar applications. The proposed filters are designed to handle highly non-linear system models, where existing approaches such as the EKF often fail.

The theoretical analysis of the filters' stability and consistency, along with the demonstration of their estimation accuracy through numerical experiments, highlights the potential of these methods to improve the counter-adversarial capabilities of cognitive radar systems. The research represents an important step forward in the field of inverse cognition and could have significant implications for the development of more robust and adaptable radar technologies.