Jamming Source Localization Using Augmented Physics-Based Model

Andrea Nardin, Tales Imbiriba, Pau Closas

TL;DR

This paper introduces an Augmented Physics-Based Model (APBM) to localize GNSS jamming sources in complex urban environments. By hybridizing a traditional pathloss physics model with a neural network that learns environmental corrections, the authors achieve significantly higher localization accuracy than either pure physics or pure data-driven approaches, even when the jammer's transmit power is unknown.

What Problem Does This Paper Address?

Global Navigation Satellite Systems (GNSS) like GPS are the backbone of modern positioning and timing, but they are incredibly fragile. Because GNSS signals arrive at Earth with very low power, they are easily drowned out by "jammers"—malicious or accidental transmitters operating on the same frequency.

To stop a jammer, you first have to find it. A common approach is crowdsourcing: using signal strength (RSS) measurements from many agents (like smartphones or connected cars) in an area. Traditionally, researchers use a pathloss model (a physics equation describing how signal strength drops over distance) to estimate the source location. However, in cities, buildings cause reflections (multipath) and blockages (shadowing) that the simple physics equations can’t account for. Conversely, pure "black-box" neural networks require massive amounts of data to learn these complexities from scratch. This paper seeks a "best of both worlds" solution.

Methods

The authors propose the Augmented Physics-Based Model (APBM). The core idea is to model the received signal power $y_n$ as: $$y_n = f(x_n; \theta) + g(x_n; \phi) + \xi_n$$

  • $f(x_n; \theta)$ (The Physics): A modified pathloss model that provides the general structure of signal decay based on the jammer's location $\theta$.
  • $g(x_n; \phi)$ (The Data): A feed-forward neural network (NN) with two hidden layers that acts as a nonlinear correction term to account for urban complexities like reflections.
  • $\xi_n$: Measurement noise.

Key Innovations: 1. Joint Optimization: Unlike methods that train a model and then use it, this approach jointly estimates the jammer's location ($\theta$) and the neural network's weights ($\phi$) simultaneously by minimizing a regularized cost function. 2. Singularity Avoidance: Standard pathloss models go to infinity as you get closer to the source. The authors introduce a "far-field distance" ($d_F$) constraint to smooth the optimization landscape, making gradient-based search much more stable. 3. Regularization: They use $\ell_2$ regularization on the NN parameters to ensure the physics model remains the primary driver and the NN only provides necessary corrections.

Datasets

The study relies on high-fidelity simulations to validate the model:

  • Ideal Scenario: An "open-sky" environment where signal propagation perfectly follows the nominal pathloss model.
  • Urban Scenario: A complex environment generated using ray tracing techniques. This includes 3D geometry and electromagnetic analysis, accounting for free-space loss and up to four signal reflections.
  • Crowdsourced Setup: 10,000 observations were simulated over a 1 $km^2$ area. To mimic a realistic, bandwidth-constrained system, only the 15 most powerful observations were used for the actual localization.

Key Insights and Ideas

  • Physics as a Prior: The physics model provides a "global" hint of where the jammer is, while the NN handles "local" environmental quirks. This allows the system to work with far less data than a pure NN approach.
  • Handling the Unknown: One of the most practical insights is the "$P_0$-blind" estimation. In the real world, we don't know how powerful a jammer is. The APBM's neural network component is effectively able to learn a constant offset that compensates for this missing information, allowing for accurate localization without knowing the jammer's transmit power.
  • Seamless Adaptation: The same APBM architecture performed optimally in both the ideal scenario (where it defaulted to the physics model) and the urban scenario (where it leaned on the NN corrections), demonstrating high flexibility without manual tuning.

Key Results

  • In Ideal Conditions: The APBM matched the performance of the Maximum Likelihood Estimator (MLE) and closely approached the Cramér-Rao Bound (CRB), which represents the theoretical limit of estimation accuracy.
  • In Urban Conditions: The APBM significantly outperformed the standard pathloss model. While the pure physics model's error (RMSE) remained high (often >40m) due to model mismatch, the APBM achieved much lower errors (approaching 10-20m) as the Interference-to-Noise Ratio (INR) increased.
  • Vs. Pure Neural Networks: A stand-alone NN struggled to learn the propagation rule from the limited data (15 points), whereas the APBM's hybrid nature allowed it to converge to a correct estimate quickly.

Limitations and Future Work

  • Static Jammers: The current model assumes the jammer is stationary. The authors acknowledge that moving or intermittent jammers are a more difficult, real-world challenge.
  • Continual Learning: Urban environments change (new buildings, moving trucks). Future work will look into "continual learning" so the model can adapt to changes in the data distribution over time.
  • Privacy Concerns: Since the data is crowdsourced, the authors suggest integrating Federated Learning in future iterations to process data locally on devices rather than sending raw measurements to a central server.

Summary updated

Abstract

Monitoring interferences to satellite-based navigation systems is of paramount importance in order to reliably operate critical infrastructures, navigation systems, and a variety of applications relying on satellite-based positioning. This paper investigates the use of crowdsourced data to achieve such detection and monitoring at a central node that receives the data from an arbitrary number of agents in an area of interest. Under ideal conditions, the pathloss model is used to compute the Cramér-Rao Bound of accuracy as well as the corresponding maximum likelihood estimator. However, in real scenarios where obstructions and reflections are common, the signal propagation is far more complex than the pathloss model can explain. We propose to augment the pathloss model with a data-driven component, able to explain the complexities of the propagation channel. The paper shows a general methodology to jointly estimate the interference location and the parameters of the augmented model, showing superior performances in complex scenarios such as those encountered in urban environments.
BibTeX
@inproceedings{nardin2023jamming,
  author = {Andrea Nardin and Tales Imbiriba and Pau Closas},
  title = {{Jamming Source Localization Using Augmented Physics-Based Model}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  doi = {10.1109/icassp49357.2023.10095731},
}

Loading related content...

Figure