On the Effect of Correlated Measurements on the Performance of Distributed Estimation

Mohammed F. A. Ahmed, Tareq Y. Al-Naffouri, Mohamed-Slim Alouini

Abstract

We address the distributed estimation of an unknown scalar parameter in Wireless Sensor Networks (WSNs). Sensor nodes transmit their noisy observations over multiple access channel to a Fusion Center (FC) that reconstructs the source parameter. The received signal is corrupted by noise and channel fading, so that the FC objective is to minimize the Mean-Square Error (MSE) of the estimate. In this paper, we assume sensor node observations to be correlated with the source signal and correlated with each other as well. The correlation coefficient between two observations is exponentially decaying with the distance separation. The effect of the distance-based correlation on the estimation quality is demonstrated and compared with the case of unity correlated observations. Moreover, a closed-form expression for the outage probability is derived and its dependency on the correlation coefficients is investigated. Numerical simulations are provided to verify our analytic results.

Introduction

In Wireless Sensor Networks (WSNs), distributed estimation is a fundamental task where multiple spatially distributed sensors measure a physical phenomenon and transmit their data to a central Fusion Center (FC). Traditional research often simplifies these systems by assuming sensor observations are either independent or perfectly correlated with the source (unity correlation). However, physical phenomena exhibit spatial correlation: sensors closer to each other or the source typically yield more similar readings. This paper investigates how a realistic, distance-based correlation model affects estimation accuracy and outage probability over noisy, fading wireless channels.

System Model

The authors consider a network of $N$ sensors observing a scalar parameter $s$. Each sensor $i$ captures a noisy observation $x_i$, which is then amplified and transmitted simultaneously over a coherent Multiple Access Channel (MAC) to the FC. The FC employs a Linear Minimum Mean Square Error (LMMSE) estimator to reconstruct the original signal $\hat{s}$ from the received signal $z$.

Distributed estimation system model over a coherent Multiple Access Channel.

The key innovation lies in the correlation model. Instead of a rank-one matrix (idealized), a full-rank, distance-based correlation matrix is used. Here, the correlation between any two nodes decays exponentially with their physical distance, providing a more faithful representation of real-world environments.

The Impact of Spatial Correlation

A critical finding of this study is the "Infinite Sensor Floor Effect." In idealized models, increasing the number of sensors $N$ allows the estimation error to vanish. However, under realistic distance-based correlation, the Mean-Square Error (MSE) hits a non-zero floor even as $N \to \infty$.

Distortion floor in realistic distance-based correlation models vs. idealized models.

As shown in the results, the full-rank model exhibits a distortion floor (approximately 0.182 in the simulated scenario), whereas the unity and rank-one models incorrectly predict that error will continue to decrease toward zero. This suggests that spatial correlation fundamentally limits the information gain from additional sensors.

Outage Probability Analysis

Beyond average distortion, the paper derives a closed-form expression for the "outage probability" — the chance that the estimation error exceeds a predefined threshold. By utilizing indefinite quadratic forms and analyzing the eigenvalues of the correlation matrices, the authors provide a mathematical tool that matches Monte Carlo simulations with high precision. The analysis reveals that the physical distance of the source from the sensor cluster significantly impacts the outage probability; as the source moves further away, the system's largest eigenvalue decreases, leading to a higher likelihood of estimation failure.

Conclusion

This research demonstrates that neglecting spatial correlation leads to overly optimistic performance estimates in WSNs. The existence of a distortion floor implies hard physical limits on sensing accuracy that cannot be overcome simply by increasing sensor density. Future work may explore these effects in non-coherent MAC schemes or under optimal power allocation strategies to mitigate the impact of correlation-induced performance ceilings.

Summary updated

Abstract

We address the distributed estimation of an unknown scalar parameter in Wireless Sensor Networks (WSNs). Sensor nodes transmit their noisy observations over multiple access channel to a Fusion Center (FC) that reconstructs the source parameter. The received signal is corrupted by noise and channel fading, so that the FC objective is to minimize the Mean-Square Error (MSE) of the estimate. In this paper, we assume sensor node observations to be correlated with the source signal and correlated with each other as well. The correlation coefficient between two observations is exponentially decaying with the distance separation. The effect of the distance-based correlation on the estimation quality is demonstrated and compared with the case of unity correlated observations. Moreover, a closed-form expression for the outage probability is derived and its dependency on the correlation coefficients is investigated. Numerical simulations are provided to verify our analytic results.
BibTeX
@article{ahmed2013effect,
  author = {Mohammed F. A. Ahmed and Tareq Y. Al-Naffouri and Mohamed-Slim Alouini},
  title = {{On the Effect of Correlated Measurements on the Performance of Distributed Estimation}},
  journal = {IEEE Trans. Signal Process.},
  year = {2013},
  url = {https://openreview.net/forum?id=cU_4SIu5_5L},
}

Loading related content...

Figure