A Unified Framework for Innovation-based Stochastic and Deterministic Event Triggers

Basic Information

• Paper ID: 2503.13043
• Title: A Unified Framework for Innovation-based Stochastic and Deterministic Event Triggers
• Authors: Eva Julia Schmitt, Benjamin Noack (Otto von Guericke University Magdeburg, Germany)
• Classification: eess.SY cs.SY (Systems and Control)
• Publication Date: March 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2503.13043

Abstract

In wireless communication applications, resources such as bandwidth and energy are often limited, particularly when large numbers of sensors and fusion centers need to exchange information frequently. Event-based transmission and estimation represent an effective approach to overcome resource constraints by transmitting only information that significantly contributes to system state reconstruction. The design of efficient triggering strategies and estimators is critical for successful event-driven transmission. While deterministic and stochastic event triggering strategies have previously been addressed separately, this paper unifies both approaches and provides deep insights into designing consistent trigger-matched estimators. Two distinct estimators are proposed, and their performance is evaluated through simulation studies comparing different trigger-estimator pairs.

Research Background and Motivation

Problem Definition

1. Resource Constraint Problem: In modern automated systems such as smart cities, smart manufacturing, and smart agriculture, spatially distributed nodes (sensors, fusion centers, agents) need to propagate large volumes of data. High-frequency wireless data transmission requires substantial bandwidth and energy resources.
2. Information Transmission Optimization Requirement: There is a need to quantify the information content of each potential message packet and transmit only when necessary or useful, thereby reducing the burden on communication systems.
3. Estimator Consistency Challenge: It is necessary to ensure estimator consistency, meaning that the estimated uncertainty matches the filter's estimation error, while simultaneously leveraging implicit information contained when triggering conditions are not satisfied.

Limitations of Existing Methods

• Deterministic Strategies: While ensuring that predefined errors are not exceeded and providing better performance, they increase estimator design complexity because deterministic strategies introduce non-Gaussian noise sources to the system measurement equation at non-transmission moments.
• Stochastic Strategies: Provide simple estimator design but reduce system performance due to the relaxation of stochastic triggering decisions.
• Separate Treatment: The two approaches have been studied independently, lacking a unified framework and in-depth comparative analysis.

Core Contributions

1. Unified Framework: Proposes a generalized framework for stochastic triggers embedding innovation-based deterministic triggers, enabling better understanding of the connections between classical stochastic and deterministic event triggers.
2. Design Guidelines: Based on new insights into the properties of different triggering strategies, investigates the consistency of existing stochastic event-driven Kalman filters (SEBKF) with these triggering strategies.
3. Estimator Extension: Extends event-driven particle filters to stochastic strategies and makes improvements for linear system models and deterministic strategy applications.
4. Performance Evaluation: Evaluates the performance of different trigger strategy and estimator combinations through simulation studies in terms of mean squared error (MSE) and consistency.

Methodology Details

Task Definition

The research objective is to design efficient event-driven transmission and estimation schemes in linear time-invariant systems, which include:

• State Equation: $x_{k+1} = Ax_k + w_k$
• Measurement Equation: $y_k = Cx_k + v_k$
• Noise Model: Process noise $w_l \sim \mathcal{N}(0,Q)$, measurement noise $v_m \sim \mathcal{N}(0,R)$, white noise and mutually uncorrelated

Unified Triggering Framework

Generalized Gaussian-Weighted Function

Proposes a novel shape function based on the generalized normal distribution:

$\phi_\beta(z_k) = \exp\left(-\frac{1}{2}\sqrt{z_k^TZ^{-1}z_k}^\beta\right)$

Where:

• $z_k = y_k - c_k$ is the trigger variable (innovation term)
• $\beta > 0$ is the shape parameter
• $Z \in \mathbb{R}^{n_y \times n_y}$ is a positive definite design variable

Significance of Parameter $\beta$

• $\beta = 2$: Standard Gaussian stochastic trigger
• $\beta \to \infty$: Approaches deterministic trigger
• $\beta \in [2,\infty)$: Provides continuous transition between the two extreme cases

Stochastic Triggering Decision

$\gamma_k = \begin{cases} 1, & \xi_k > \phi_\beta(z_k) \\ 0, & \xi_k \leq \phi_\beta(z_k) \end{cases}$

Where $\xi_k \sim U(0,1)$ is a uniformly distributed random variable.

Implicit Measurement Selection Strategies

Send-on-Delta (SOD)

$c_k$ is set to the measurement value transmitted at the last event moment $k_e$: $z_k = y_k - y_{k_e}$

Send-on-Delta with Prediction (SODP)

$c_k$ is set to the prediction of the sensor's local state estimate: $z_k = y_k - CA^l\hat{x}^S_{k_e}$

Where $l = k - k_e$ is the number of time steps since the last event.

Estimator Design

Stochastic Event-Driven Kalman Filter (SEBKF)

Optimally designed for linear systems with $\beta = 2$:

Prediction Step: $\hat{x}_{k|k-1} = A\hat{x}_{k-1|k-1}$$P_{k|k-1} = AP_{k-1|k-1}A^T + Q$

Update Step: $\hat{x}_{k|k} = \hat{x}_{k|k-1} + K_k(\gamma_k z_k - \hat{z}_{k|k-1})$$P_{k|k} = (I_{n_x} - K_kC)P_{k|k-1}$

Where the Kalman gain is: $K_k = P_{k|k-1}C^T(CP_{k|k-1}C^T + R + (1-\gamma_k)Z)^{-1}$

Sampling Estimator

To overcome SEBKF limitations when $\beta > 2$, proposes a sampling estimator based on particle filtering:

1. When event is triggered ($\gamma_k = 1$): Execute standard Kalman filtering step
2. When event is not triggered ($\gamma_k = 0$):
   • Sample $N$ particles from the prior distribution
   • Apply system and measurement equations
   • Use triggering condition to reject samples that would cause event triggering
   • Repeat until sufficient accepted samples are obtained

Experimental Setup

System Model

Uses a 2D approximate constant-velocity model:

$A = \begin{bmatrix} 1 & \Delta & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & \Delta \\ 0 & 0 & 0 & 1 \end{bmatrix}, \quad C = \begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 \end{bmatrix}$

Where $\Delta = 0.3$ is the sampling interval.

Evaluation Metrics

• Relative MSE: Mean squared error relative to SEBKF
• ANEES: Average Normalized Estimation Error Squared, used to assess consistency (ideal value is 1)

Experimental Parameters

• Monte Carlo simulation: 500 runs, 150 time steps per run
• Sampling estimator: Uses $N = 1000$ particles ($N = 5000$ for low event rates)
• Test $\beta$ values: 2, 5, 1000

Experimental Results

Main Findings

$\beta = 2$ (Standard Gaussian)

• SEBKF and sampling estimator performance are identical and optimal
• Both estimators maintain consistency across all event rates (ANEES ≈ 1)
• Sampling estimator asymptotically achieves optimal results

$\beta = 5$ and $\beta = 1000$ (Approaching Deterministic)

• SEBKF is slightly conservative at moderate event rates
• When event rate drops below approximately 10%, SEBKF becomes inconsistent (ANEES > 1)
• Sampling estimator maintains consistency across all transmission rates
• In terms of performance, both estimators perform similarly, but the sampling estimator's performance begins to degrade at extremely low event rates

Consistency Boundary Analysis

Through theoretical analysis and simulation verification, the maximum $Z$ value for SEBKF to maintain consistency is determined to be approximately $Z^* \approx 20 \cdot I$, corresponding to an average transmission rate of $\bar{\gamma} \approx 0.1$.

Computational Complexity Trade-off

• SEBKF: High computational efficiency but consistency issues when $\beta > 2$
• Sampling estimator: Higher computational burden but provides better generality and consistency guarantees

Related Work

Deterministic Triggering Strategies

• Send-on-Delta (SOD) schemes
• SOD with prediction (SODP)
• Matched sampling and variance-based sampling
• Set membership methods and event-driven particle filters

Stochastic Triggering Strategies

• Stochastic versions of SOD and SODP
• Modified versions using Gaussian error-weighted functions
• Linear Kalman filter extensions for standard cases and correlated input data

Theoretical Foundations

Yu et al. proved that for any stochastic trigger, there exists a better deterministic trigger in the scalar linear case, providing theoretical support for the performance comparisons in this paper.

Conclusions and Discussion

Main Conclusions

1. Unified Framework Effectiveness: The generalized Gaussian-weighted function successfully unifies deterministic and stochastic triggering strategies, providing a continuous design space.
2. Estimator Applicability: SEBKF is suitable for high transmission rate scenarios, while the sampling estimator provides better generality, particularly at low transmission rates.
3. Performance Trade-off: Deterministic strategies (high $\beta$ values) provide lower estimation error at the same transmission rate but require more complex estimator design.

Limitations

1. Computational Complexity: The sampling estimator's computational burden is significantly higher than SEBKF
2. Parameter Tuning: Joint optimization of $\beta$ and $Z$ requires further research
3. Nonlinear Extension: The current framework is primarily for linear systems; extension to nonlinear systems remains to be explored

Future Directions

1. Theoretical Analysis: Establish functional relationships between event rate, design parameter $Z$, and triggering error
2. Consistency Conditions: Establish rigorous consistency conditions for SEBKF
3. Extended Applications: Investigate other weighting functions and nonlinear system models
4. Practical Applications: Consider practical communication issues such as packet loss and sensor unavailability

In-Depth Evaluation

Strengths

1. Theoretical Innovation: Provides the first unified mathematical framework for deterministic and stochastic event triggering, with significant theoretical value
2. Strong Practicality: The proposed method is directly applicable to practical scenarios such as wireless sensor networks
3. Comprehensive Analysis: Provides complete theoretical and experimental analysis from trigger design to estimator consistency
4. Method Generality: The sampling estimator is extensible to other triggering schemes with good generality

Weaknesses

1. Computational Efficiency: The sampling estimator's computational complexity is high, potentially limiting real-time applications
2. Parameter Selection: Lacks adaptive selection mechanisms and optimization criteria for the $\beta$ parameter
3. Experimental Scope: Verification only on 2D linear systems; lacks validation on more complex systems
4. Theoretical Completeness: Sufficient and necessary conditions for SEBKF consistency have not been fully established

Impact

1. Academic Contribution: Provides an important theoretical unification framework for the event-driven estimation field
2. Practical Value: Has direct application value in IoT, intelligent sensor networks, and other domains
3. Reproducibility: Clear algorithm descriptions and detailed experimental settings ensure good reproducibility
4. Extensibility: Provides a solid theoretical foundation and multiple research directions for subsequent studies

Applicable Scenarios

1. Wireless Sensor Networks: Large-scale sensor deployment with limited bandwidth
2. Internet of Things Systems: State monitoring of energy-constrained smart devices
3. Industrial Control: Industrial systems with high real-time requirements but limited communication resources
4. Intelligent Transportation: Cooperative sensing and state estimation in vehicular networks

References

This paper cites 28 relevant references covering important works in event-driven control, state estimation, particle filtering, and related fields, providing a solid theoretical foundation for the research. Key references include:

• Foundational work on event-driven triggering (Miskowicz, 2006)
• Pioneering research on stochastic event triggering (Han et al., 2015)
• Theoretical analysis comparing deterministic and stochastic triggering (Yu et al., 2021)
• Applications of particle filtering in event-driven estimation (Gasmi et al., 2022-2024)

Overall Evaluation: This is a high-quality paper in the systems and control field that successfully unifies deterministic and stochastic event triggering methods, providing important theoretical contributions and practical value. While there is room for improvement in computational efficiency and theoretical completeness, its innovation and practicality make it an important advance in the field.