Abstract
We address the problem of timing-based localization in wireless networks, when an unknown fraction of data is corrupted by nonideal signal conditions. While timing-based techniques enable accurate localization, they are also sensitive to such corrupted data. We develop a robust method that is applicable to a range of localization techniques, including time-of-arrival, time-difference-of-arrival and time-difference in schedule-based transmissions. The method is nonparametric and requires only an upper bound on the fraction of corrupted data, thus obviating distributional assumptions of the corrupting noise distribution. The robustness of the method is demonstrated in numerical experiments.
Introduction
Next-generation wireless networks and IoT devices increasingly rely on timing-based localization to determine device positions in GPS-denied environments. By measuring the time of flight of a signal between a transmitter and receiver, high-precision coordinates can theoretically be obtained. However, real-world environments introduce significant challenges. Signals frequently bounce off obstacles like walls or moving objects, creating non-line-of-sight (NLOS) conditions. These NLOS signals take longer to reach the receiver, appearing as severe outliers in timing data.
Standard localization algorithms, such as nonlinear least-squares, typically assume Gaussian noise. When faced with NLOS outliers, these conventional methods break down, producing wildly inaccurate location estimates. While existing robust methods attempt to mitigate this, they often rely on unrealistic assumptions, such as knowing the exact mathematical distribution of NLOS noise or the precise percentage of corrupted data. This paper proposes a robust, distribution-free localization method that requires only a conservative upper bound on the fraction of corrupted data.
Methodology
The proposed approach frames localization as a problem of robust risk minimization. Instead of minimizing standard squared error across all data points equally, the algorithm simultaneously estimates the node's location and assigns a probability weight to each measurement.
The method is designed to be universal, applying to various timing protocols as illustrated in Figure 2:
- TOA: Two-way ranging between synchronized nodes.
- TDOA: Passive reception from synchronized anchor nodes.
- TDST: Passive reception from asynchronous anchor nodes.
To prevent the algorithm from over-relying on a single data point, an entropy constraint is applied to the weights. This constraint is governed by a single user-defined parameter, $\tilde{\epsilon}$, representing an upper bound on the fraction of corrupted data. The optimization is solved via a blockwise coordinate descent algorithm that alternates between updating the location estimate (weighted nonlinear least-squares) and updating the weights (convex barrier method).
Experimental Evaluation
The method was evaluated using synthetic datasets simulating TOA, TDOA, and TDST configurations. The simulations included standard Gaussian noise for line-of-sight (LOS) signals and an exponential bias to represent corrupted NLOS signals.
A key advantage of this method is its insensitivity to the exact corruption parameter. Experiments demonstrated that even if the actual corruption fraction ($\epsilon$) is small (e.g., 15%), setting a conservative upper bound (e.g., $\tilde{\epsilon} = 0.5$) still yields stable and accurate results. Furthermore, the weights assigned during optimization serve as an effective outlier classification tool, allowing for the identification of LOS vs. NLOS signals with high accuracy.
Results
The robust method significantly outperforms standard least-squares across all tested scenarios. In TOA setups, it matches the performance of specialized tools like the Huber method while maintaining the flexibility to handle the nonlinearities inherent in TDOA and TDST.
As shown in Figure 6, the robust method maintains a low Root Mean Square Error (RMSE) as long as the actual corruption fraction $\epsilon$ remains below the assumed upper bound $\tilde{\epsilon}$. Once the actual corruption exceeds this bound, performance begins to degrade, though it still often remains superior to the standard non-robust approach for a significant range.
Conclusion
This paper presents a versatile and robust framework for wireless localization that effectively handles NLOS signal corruption without requiring detailed prior knowledge of noise distributions. By utilizing an entropy-constrained weighting system, the algorithm dynamically identifies and down-weights outliers. While currently validated on synthetic data, the method's distribution-free nature and computational structure make it a promising candidate for real-world IoT and 5G/6G localization systems. Future work will focus on hardware validation and optimizing the algorithm for low-power edge devices.
Summary updated