Papers
Topics
Authors
Recent
Search
2000 character limit reached

Robust Gaussian Process Regression for Real-Time High Precision GPS Signal Enhancement

Published 3 Jun 2019 in stat.ML, cs.LG, and eess.SP | (1906.01095v1)

Abstract: Satellite-based positioning system such as GPS often suffers from large amount of noise that degrades the positioning accuracy dramatically especially in real-time applications. In this work, we consider a data-mining approach to enhance the GPS signal. We build a large-scale high precision GPS receiver grid system to collect real-time GPS signals for training. The Gaussian Process (GP) regression is chosen to model the vertical Total Electron Content (vTEC) distribution of the ionosphere of the Earth. Our experiments show that the noise in the real-time GPS signals often exceeds the breakdown point of the conventional robust regression methods resulting in sub-optimal system performance. We propose a three-step approach to address this challenge. In the first step we perform a set of signal validity tests to separate the signals into clean and dirty groups. In the second step, we train an initial model on the clean signals and then reweigting the dirty signals based on the residual error. A final model is retrained on both the clean signals and the reweighted dirty signals. In the theoretical analysis, we prove that the proposed three-step approach is able to tolerate much higher noise level than the vanilla robust regression methods if two reweighting rules are followed. We validate the superiority of the proposed method in our real-time high precision positioning system against several popular state-of-the-art robust regression methods. Our method achieves centimeter positioning accuracy in the benchmark region with probability $78.4\%$ , outperforming the second best baseline method by a margin of $8.3\%$. The benchmark takes 6 hours on 20,000 CPU cores or 14 years on a single CPU.

Citations (15)

Summary

  • The paper introduces a novel Filter-Reweight-Retrain (FRR) framework that integrates clean and reweighted noisy signals to achieve centimeter-level positioning accuracy.
  • The paper demonstrates that the robust Gaussian Process regression method outperforms conventional approaches, achieving 78.4% precision improvement in high-noise settings.
  • The paper validates the scalability and practicality of the FRR methodology for real-time applications, addressing the challenges of high computational costs in autonomous systems.

Robust Gaussian Process Regression for GPS Signal Enhancement

The paper "Robust Gaussian Process Regression for Real-Time High Precision GPS Signal Enhancement" (1906.01095) explores the use of Gaussian Process (GP) regression models to enhance the precision of GPS signals, especially for real-time applications where positioning accuracy is crucial. It introduces a novel approach to handle noisy GPS data using a three-step framework known as Filter-Reweight-Retrain (FRR) to achieve high precision in positioning systems.

GPS Signal Enhancement Using Gaussian Process Regression

Background

GPS signals are affected by various sources of noise, significantly degrading the accuracy of positioning systems needed for applications like self-driving cars and drones. Conventional single-frequency GPS receivers provide only meter-level accuracy, which is insufficient for these applications. The paper addresses this issue by focusing on the noise caused by the ionosphere, specifically the vertical Total Electron Content (vTEC).

Proposed Method: Filter-Reweight-Retrain (FRR)

The authors propose a three-step approach for GPS signal enhancement using FRR methodology:

  1. Signal Screening: Initially, incoming GPS signals are validated to categorize them as 'clean' or 'dirty' (noisy). This segregation is pivotal since the subsequent steps are built upon the cleanliness of data.
  2. Initial Model Training: A Gaussian Process regression model is trained using clean signals exclusively, ensuring the model starts with high-quality data.
  3. Signal Retraining and Weight Adjustment: Dirty signals are reweighted based on their residual errors after initial model application. The GP model is then retrained using both clean signals and reweighted dirty signals.

Theoretical Analysis

The paper extensively discusses the theoretical underpinnings proving that the FRR approach holds robustness against high levels of noise. Critical aspects of their analysis include:

  • Tolerance to Noise: The proposed model is shown to handle noise levels, surpassing conventional regression methods.
  • Consistency: Through their theoretical proofs, the authors establish that the FRR method remains consistent and effective even when the data corruption rates are high.

Experimental Validation

Experiments were conducted using a large-scale GPS receiver grid system to evaluate the approach against several robust regression methods. The FRR methodology achieved centimeter-level positioning accuracy with a probability of 78.4%, outperforming the second-best method by 8.3%. This was validated through extensive computational validation, taking six hours on 20,000 CPU cores, or equivalently, 14 years on a single CPU.

Practical Considerations

The implementation of FRR is scalable and practical for deployment in real-time GPS systems where high precision is non-negotiable. Despite the computational cost, the benefits in accuracy for critical applications like autonomous navigation justify the resource use.

Conclusion

The paper explores a substantive solution for enhancing GPS signal precision through robust Gaussian Process regression. The Filter-Reweight-Retrain methodology offers a viable pathway to achieving real-time centimeter-level accuracy by efficiently handling noisy data. Theoretical insights and strong experimental results showcase the method's superiority over existing approaches, particularly in high-noise scenarios. Future research may explore optimization of computational resources and expansion to broader datasets and environments. This research is a significant step towards precision GPS applications in AI-driven autonomous systems.

Paper to Video (Beta)

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Collections

Sign up for free to add this paper to one or more collections.