Performance Analysis for Wireless Localization with Random Sensor Network

Technical Summary: Performance Analysis for Wireless Localization with Random Sensor Network

Problem Statement
This work addresses the challenge of accurately localizing a moving target within $d$-dimensional wireless sensor networks (WSNs) under realistic signal propagation conditions. While satellite-based systems like GPS fail in obstructed or indoor environments, network-based localization using Received Signal Strength (RSS) and Angle-of-Arrival (AOA) measurements offers a viable alternative. However, the spatial configuration of sensor nodes is often random or unknown, and signal measurements are corrupted by propagation effects such as shadowing and multipath fading. The core problem is to characterize the Mean Squared Error (MSE) and Conditional MSE (CMSE) of practical fusion-based estimators in such stochastic environments, specifically when the underlying sensor deployment is not strictly homogeneous but belongs to a broad class of stationary isotropic point processes.

Methodology
The authors adopt a stochastic geometry framework, modeling sensor locations as a stationary and isotropic point process (PP) $\Xi$ on $\mathbb{R}^d$. The target is modeled as a fixed source at the origin, and sensors within a finite observation radius $R$ provide noisy RSS and AOA measurements.

1. Signal Model: The RSS $P_i$ is modeled as a path-loss function $\ell(R_i)$ perturbed by log-normal shadowing $S_i(\sigma)$, where $\sigma$ represents the noise level. The AOA measurements are modeled as wrapped Gaussian distributions centered at the true angles, with variance increasing with distance.
2. Estimators: The study focuses on two fusion-based estimators at a central processor:
   • Simple Average Estimator: Aggregates individual target estimates (derived from RSS and AOA) by taking their arithmetic mean.
   • Weighted Average Estimator: Assigns weights to individual estimates based on estimated distances, typically giving higher weight to closer sensors.
3. Approximation Theorem: A central methodological contribution is the establishment of an approximation theorem. The authors prove that under the Short-Range Dependence (SRD) condition and in the regime of strong propagation noise ($\sigma \to \infty$), the joint distribution of RSS and AOA observations from a broad class of stationary isotropic PPs becomes statistically indistinguishable from that of a Homogeneous Poisson Point Process (HPPP). This allows the complex analysis of general PPs to be reduced to the tractable analysis of HPPPs.
4. Analytical Derivation: Leveraging the HPPP equivalence, the authors derive explicit analytical upper bounds for the CMSE and MSE. They utilize properties of order statistics and the specific distribution of sensor locations within a finite ball to bound the error terms.

Key Contributions

• Approximation Theorem: The paper establishes that for stationary isotropic networks satisfying the SRD condition, the observable process (RSS and AOA) converges in distribution to that of an HPPP as the propagation noise increases. This justifies using HPPP models as proxies for more complex, non-Poisson deployments (e.g., repulsive or clustered networks) in noisy regimes.
• Tractable Performance Bounds: The authors derive closed-form upper bounds for the MSE and CMSE of the simple average estimator in a finite HPPP network. These bounds explicitly characterize the scaling laws with respect to sensor density ($\lambda$), observation radius ($R$), and noise variance.
• Analysis of Weighted Estimators: The paper investigates weighted fusion estimators that utilize estimated distance information. It demonstrates that while weighted estimators may offer improvements for small sample sizes, they are asymptotically inferior to the simple average estimator. The weighted approach fails to effectively utilize information from distant sensors as the network density increases, leading to a slower decay rate of the CMSE ($O(n^{-2/d})$) compared to the simple average ($O(n^{-1})$).
• Validation via Simulation: Extensive simulations are conducted to validate the theoretical bounds across HPPP, Matérn Hardcore (repulsive), and Matérn Cluster (clustered) processes. The results confirm that the HPPP-based bounds serve as reasonable proxies for these diverse deployment models when noise and observation ranges are sufficiently large.

Results

• Scaling Laws: For the simple average estimator, the MSE decays as $O(\lambda^{-1})$ with respect to sensor density and $O(n^{-1})$ with respect to the number of sensors $n$.
• Weighted vs. Simple: Theoretical analysis and simulations show that weighted estimators, which prioritize sensors with smaller estimated distances, do not improve the asymptotic scaling rate for $d \geq 2$. In fact, for large $n$, the simple average estimator outperforms weighted estimators because it effectively averages out errors across all sensors, whereas weighted estimators disproportionately rely on a few "closest" sensors which may be outliers.
• Robustness of HPPP Model: Simulations on repulsive (Matérn Hardcore) and clustered (Matérn Cluster) networks show that their localization performance closely tracks the HPPP theoretical bounds, particularly as the noise parameter $\sigma$ increases. This validates the approximation theorem, suggesting that the specific spatial correlation of sensors has negligible impact on localization accuracy in high-noise environments.

Significance and Claims
The paper claims to provide a robust, cost-aware framework for designing next-generation location-aware wireless networks. By translating deployment parameters (density, region size) and sensing conditions (noise levels) into achievable accuracy targets, the derived bounds offer quantitative guidance for network planners.

The authors emphasize that their work does not analyze specific realizations of a network at fixed coordinates but rather characterizes the average performance across all possible realizations of a spatial point process. This approach allows for general design guidelines rather than case-specific analysis. The significance lies in demonstrating that the mathematically convenient HPPP model is not merely a simplification but a valid and robust proxy for a wide class of realistic, non-Poisson sensor deployments when signal propagation effects are significant. This finding simplifies the design and analysis of localization systems, allowing engineers to rely on tractable HPPP-based bounds even when the underlying sensor distribution is complex or unknown.