Distributed Parameter Estimation under Gaussian Observation Noises

Jiaqi Yan,Hideaki Ishii

2023-03-21

Abstract:In this paper, we consider the problem of distributed parameter estimation in sensor networks. Each sensor makes successive observations of an unknown $d$-dimensional parameter, which might be subject to Gaussian random noises. They aim to infer true value of the unknown parameter by cooperating with each other. To this end, we first generalize the so-called dynamic regressor extension and mixing (DREM) algorithm to stochastic systems, with which the problem of estimating a $d$-dimensional vector parameter is transformed to that of $d$ scalar ones: one for each of the unknown parameters. For each of the scalar problem, an estimation scheme is given, where each sensor fuses the regressors and measurements in its in-neighborhood and updates its local estimate by using least-mean squares. Particularly, a counter is also introduced for each sensor, which prevents any (noisy) measurement from being repeatedly used such that the estimation performance will not be greatly affected by certain extreme values. A novel excitation condition termed as \textit{local persistent excitation} (Local-PE) condition is also proposed, which relaxes the traditional persistent excitation (PE) condition and only requires that the collective signals in each sensor's in-neighborhood are sufficiently excited. With the Local-PE condition and proper step sizes, we show that the proposed estimator guarantee that each sensor infers the true parameter in mean square, even if any individual of them cannot. Numerical examples are finally provided to illustrate the established results.

Systems and Control

What problem does this paper attempt to address?

This paper attempts to address the problem of distributed parameter estimation in sensor networks, particularly in the presence of Gaussian random noise affecting each sensor. Specifically, each sensor continuously observes an unknown d-dimensional parameter and collaborates with other sensors to infer the true value of this unknown parameter. To achieve this goal, the paper proposes a new Local Persistent Excitation (Local-PE) condition and designs a distributed estimation scheme based on the Dynamic Regressor Extension and Mixing (DREM) algorithm.

### Main Issues
1. **Distributed Parameter Estimation**: How to effectively estimate an unknown d-dimensional parameter in a sensor network, even when the observation data of each sensor is affected by Gaussian noise.
2. **Local Persistent Excitation Condition**: How to generate sufficient excitation signals within the local neighborhood of each sensor to ensure the accuracy of parameter estimation.
3. **Estimation Performance Optimization**: How to improve estimation performance by introducing a counter to prevent the reuse of noise measurement values.

### Solutions
1. **Extension of the DREM Algorithm**: Extending the traditional DREM algorithm from deterministic systems to stochastic systems, transforming the d-dimensional parameter estimation problem into d scalar parameter estimation problems.
2. **Local Persistent Excitation Condition**: Proposing a new Local Persistent Excitation (Local-PE) condition, which relaxes the requirements of the traditional persistent excitation condition, requiring only sufficient excitation signals within the local neighborhood of each sensor.
3. **Distributed Estimation Scheme**: Designing a distributed estimation algorithm based on the Least Mean Squares (LMS) method, where each sensor fuses its local neighborhood's regressor and measurement values and updates its local estimate.
4. **Counter Mechanism**: Introducing a counter mechanism to prevent any noise measurement value from being reused, thereby avoiding the impact of extreme values on estimation performance.

### Main Contributions
1. **Theoretical Analysis**: Proving that under the Local Persistent Excitation condition and appropriate step size selection, each sensor can collaboratively complete the parameter estimation task, even if a single sensor cannot complete it independently.
2. **Numerical Experiments**: Verifying the effectiveness of the proposed algorithm and the correctness of the theoretical results through numerical experiments.

### Conclusion
By introducing a new Local Persistent Excitation condition and an improved DREM algorithm, this paper successfully addresses the problem of distributed parameter estimation in sensor networks under Gaussian noise. Experimental results show that the proposed algorithm can effectively estimate the unknown parameter, even if a single sensor cannot complete the task independently.