Abstract
Estimating unknown signal in Wireless Sensor Networks (WSNs) requires sensor nodes to transmit their observations of the signal over a multiple access channel to a Fusion Center (FC). The FC uses the received observations, which is corrupted by observation noise and both channel fading and noise, to find the minimum Mean Square Error (MSE) estimate of the signal. In this paper, we investigate the effect of the source-node correlation and the inter-node correlation on the performance of the Linear Minimum Mean Square Error (LMMSE) estimator for three correlation models in the presence of channel fading. First, we investigate the asymptotic behavior of the achieved distortion resulting from both the observation and channel noise in a non-fading channel. Then, the effect of channel fading is considered and the corresponding distortion outage probability is found. By representing the distortion as a ratio of indefinite quadratic forms, a closed-form expression is derived for the outage probability that shows its dependency on the correlation. Finally, the new representation of the outage probability allows us to propose an iterative solution for the power allocation problem.
Introduction
In the design of Wireless Sensor Networks (WSNs), a common assumption is that increasing the number of sensors ($N$) will indefinitely improve estimation accuracy. If sensors measure a physical parameter—such as temperature or pressure—conventional models suggest that the estimation error should vanish as $N \to \infty$. However, research by Ahmed, Al-Naffouri, and Alouini challenges this by accounting for spatial correlation. Their work demonstrates that when sensor measurements are spatially correlated, the estimation error does not necessarily go to zero, revealing a fundamental performance floor.
System Model and Methods
The study considers a WSN where a fleet of sensors observes a source signal $s$ and transmits data over a coherent Multiple Access Channel (MAC). In this setup, signals from multiple sensors combine over the air before reaching a Fusion Center (FC).
The FC employs a Linear Minimum Mean Square Error (LMMSE) estimator to reconstruct the source signal. Unlike simplified models that assume "unity correlation" (where every sensor measures the exact same signal plus independent noise), the authors use a distance-based correlation model. Specifically, they employ a power exponential model where the correlation between two points decays as a function of the distance between them.
The Performance Floor
A key contribution of this work is the distinction between "rank-one" models (simplified correlation) and "full-rank" models (realistic spatial correlation). In rank-one or unity correlation models, the distortion (MSE) indeed approaches zero as more sensors are added. However, the full-rank model reveals a persistent distortion floor.
As shown in the figure above, while simplified models suggest error-free estimation is possible with enough sensors, the realistic model hits a hard floor (approximately 0.182 in the simulated scenario). This implies that if measurements are weakly correlated due to distance, simply adding more sensors cannot overcome the inherent uncertainty.
Outage Probability and Analytical Results
Beyond average performance, the paper focuses on Outage Probability—the likelihood that the estimation error exceeds a predefined threshold. This metric is vital for mission-critical applications where reliability is paramount.
By representing the distortion as a ratio of indefinite quadratic forms, the authors derived a closed-form expression for outage probability. This analytical result was verified through Monte Carlo simulations, showing a perfect match. The analysis reveals that outage probability is a monotonically decreasing function of the largest eigenvalue of the correlation matrix; essentially, the "stronger" the primary component of the correlated data, the lower the risk of a high-error outage.
Conclusion
This research provides a more nuanced understanding of WSN performance by incorporating realistic spatial correlation. The discovery of a distortion floor suggests that network density has diminishing returns in spatially correlated environments. Furthermore, the derived closed-form expressions for outage probability offer engineers a robust tool for predicting system reliability and optimizing power allocation without the need for exhaustive simulations.
Summary updated