Zero-Input Compensation Mechanism-Based Linear Estimation for Systems with Multiple Packet Dropouts and Multiplicative Noises

Herui Zha, Tong Wang, Hao Jin, Jing MaInternational Journal of Signal Processing Image Processing and Pattern Recognition 2016 · DOI

TL;DR

This paper introduces a new linear estimation method for systems that suffer from both lost data packets and multiplicative noise. By employing a "zero-input" compensation strategy, the proposed estimators (filter, predictor, and smoother) offer reduced computational complexity compared to existing methods, while achieving comparable performance.

What Problem Does This Paper Address?

Networked control systems (NCS) are increasingly prevalent in applications like target tracking and environmental monitoring. A significant challenge in these systems is the potential for data loss during transmission due to network issues like congestion or interference, leading to "packet dropouts." When a measurement is lost, the system needs a way to compensate for the missing information to maintain accurate state estimation. Furthermore, many real-world systems are subject to uncertainties, which can be modeled as additive or multiplicative noise. This paper specifically tackles the problem of estimating the state of a system that experiences both multiple packet dropouts and multiplicative noise in its state and measurement equations. Existing approaches often rely on "hold-input" compensation, where the last known value is used, or more complex augmented estimators. This work aims to provide an efficient and effective alternative using a "zero-input" compensation mechanism.

Methods

The core of this paper lies in developing optimal linear estimators (filter, predictor, and smoother) based on the zero-input compensation mechanism. When a measurement is lost (i.e., a packet dropout occurs), this mechanism simply assumes the measurement value is zero.

The authors utilize the innovation analysis approach to derive these estimators. This approach focuses on the difference between the actual measurement and the predicted measurement (the innovation) to update the state estimate.

Key steps include:

  • Deriving a linear one-step predictor: This forms the foundation for the subsequent filter and smoother.
  • Developing optimal linear filter and smoother: These are built upon the one-step predictor, extending the estimation capabilities to the current time step (filter) and past time steps (smoother).
  • Comparing computational cost: A significant claim is that the proposed zero-input based estimators have a computational complexity of O(n³) compared to O((n+m)³) for augmented estimators using the hold-input mechanism, where 'n' and 'm' relate to system dimensions.

Datasets

Not discussed in the paper. The paper uses simulated examples for evaluation.

Key Insights and Ideas

  • Zero-Input Compensation as a Viable Alternative: The paper demonstrates that a simple zero-input compensation strategy can be effectively integrated into linear estimation frameworks, offering a computationally lighter alternative to more complex methods.
  • Reduced Computational Burden: The proposed estimators offer a significant reduction in computational cost, making them potentially more suitable for real-time applications or systems with limited processing power.
  • No Universal Superiority: Crucially, the simulation results indicate that neither the zero-input nor the hold-input compensation mechanism is universally superior. The performance depends on the specific system parameters and noise characteristics. This suggests that the choice of compensation mechanism should be carefully considered based on the application.
  • Smoother Outperforms Predictor: As expected in most estimation problems, the smoother consistently provides the best performance (lowest error variance), followed by the filter, with the predictor being the least accurate.

Key Results

The paper presents theoretical derivations for the optimal linear filter, predictor, and smoother under the zero-input compensation mechanism. Simulation results are provided to compare the performance of these estimators and to contrast them with estimators based on the hold-input mechanism.

  • Tracking Performance: Figures 1-3 show that the smoother provides the best tracking of the true state, followed by the filter, and then the predictor.
  • Estimation Error Variances: Figure 4 illustrates that the smoother has the lowest estimation error variance, followed by the filter, and then the predictor.
  • Impact of Packet Dropout Rate ($\eta$): Figure 5 shows that for the proposed filter, higher packet dropout rates (larger $\eta$) generally lead to higher estimation error variances.
  • Impact of Noise Covariances ($Q_\epsilon, Q_v$): Figure 6 indicates that the filter performs better (lower error variances) with larger packet dropout rates ($\eta$) and smaller noise covariances ($Q_\epsilon, Q_v$).
  • Zero-Input vs. Hold-Input: Figure 7 and Figure 8 present comparisons between the zero-input and hold-input mechanisms. In some scenarios (Figure 7), the hold-input mechanism performs better, while in others (Figure 8), the zero-input mechanism is superior. This reinforces the conclusion that neither is universally better.

Limitations and Future Work

  • No Universal Best Compensator: The paper explicitly concludes that neither the zero-input nor the hold-input compensation mechanism is always superior. This highlights the need for further research into designing an "optimal compensator" that can adapt to system conditions.
  • Focus on Linear Estimation: The methods presented are for linear systems. Extending these concepts to nonlinear systems would be a valuable future direction.
  • Specific Noise Models: While multiplicative noise is considered, the specific forms and distributions of these noises might be further explored.

This work provides a valuable contribution by offering a computationally efficient linear estimation method for systems with packet dropouts and multiplicative noise, while also shedding light on the trade-offs between different compensation strategies.

Summary updated

BibTeX
@article{zha2016zero,
  author = {Herui Zha and Tong Wang and Hao Jin and Jing Ma},
  title = {{Zero-Input Compensation Mechanism-Based Linear Estimation for Systems with Multiple Packet Dropouts and Multiplicative Noises}},
  journal = {International Journal of Signal Processing Image Processing and Pattern Recognition},
  year = {2016},
  doi = {10.14257/ijsip.2016.9.11.30},
}
Figure