TL;DR
This paper tackles the challenge of maintaining accurate Global Navigation Satellite System (GNSS) positioning for rotating objects, like missiles or artillery shells, which experience intermittent signal reception. They introduce a novel Kalman filter-based algorithm that dynamically weights satellite data based on its statistical properties, significantly improving positioning accuracy and satellite geometry utilization compared to traditional methods.
What Problem Does This Paper Address?
Accurate positioning is crucial for many applications, including guiding projectiles. However, when these objects rotate, their antennas can only capture GNSS signals intermittently, especially at lower rotation speeds. Traditional methods often discard this "noisy" or "poorly converged" data as outliers. While this simplifies processing, it drastically reduces the available satellite data and degrades the geometric distribution of visible satellites, leading to poor positioning accuracy. This paper aims to overcome this by developing a method that intelligently utilizes this discontinuous data.
Methods
The core of the proposed solution is a Kalman filter algorithm that incorporates a novel weighting scheme. Here's a breakdown:
- Kalman Filter Foundation: The paper starts with a standard Kalman filter framework for state estimation, defining the system's state vector (position, velocity, acceleration, clock, and frequency) and its dynamics.
- Chebyshev Polynomial Extrapolation: To handle the intermittent data, the authors use Chebyshev polynomials to extrapolate pseudorange values. This allows them to estimate expected signal behavior even when direct measurements are unavailable.
- Dynamic Weighting Function: This is the key innovation. Based on the extrapolated pseudorange and the actual observed pseudorange, a weighting function is designed. This function assigns higher weights to data with lower estimated error and lower weights to data with higher estimated error (or anomalous data). This "dynamic weighting" adapts to the quality of the incoming signal in real-time.
- Independent Model Identification: The Kalman filter's observation weights are identified based on independent models, allowing for a more nuanced approach to data utilization.
Datasets
The authors conducted experiments using a rotating carrier test-bed with dual antennas. They collected data from this setup and replayed it using their developed software. Specific datasets mentioned include:
aNavData-New-high-4spr.csv,aNavData-New-high-12spr.csv,aNavData-New-high-18spr.csv: Elevation data at different rotation speeds using the proposed method.aNavData-Tradition-high-4spr.csv,aNavData-Tradition-high-12spr.csv,aNavData-Tradition-high-18spr.csv: Elevation data at different rotation speeds using traditional methods.aReplayData_RC_4spr.dat,aReplayData_RC_12spr.dat,aReplayData_RC_18spr.dat: Positioning data acquired at various rotational speeds.90°_20240717110227_003.dat: Data from robustness validation experiments.
Key Insights and Ideas
- Embrace the Noise: Instead of discarding noisy or discontinuous GNSS data, this paper shows how to intelligently leverage it by assigning it appropriate weights.
- Dynamic Weighting is Key: The adaptive nature of the weighting function, which adjusts based on real-time data quality (error variance), is crucial for improving accuracy in challenging environments.
- Dual Antennas for Coverage: While the core algorithm focuses on weighting, the experimental setup highlights the benefit of dual antennas for ensuring better overall signal reception during rotation.
- Low Rotation Speeds are the Hardest: The paper specifically targets low rotational speeds, where signal interruptions are more pronounced and traditional methods struggle the most. The proposed method shows significant improvements in these scenarios.
Key Results
The paper demonstrates significant improvements in positioning accuracy, particularly at lower rotational speeds:
- Horizontal Positioning Error Reduction: At a rotation speed of 1/18 r/s, the horizontal positioning error was reduced from 37.16 m with the original algorithm to 28.31 m with the proposed method.
- Improved Satellite Geometry: The weighting approach helps retain more satellites in the solution, leading to better Position Dilution of Precision (PDOP) values.
- Robustness: The proposed method outperforms traditional methods and even robust baselines like Huber-M and 30-gate filtering in terms of RMSE, MAE, and percentile errors, especially in scenarios with non-stationary noise and intermittent gross errors. For example, P95 improvements were around 45% for one satellite compared to fixed-weight filtering.
- Stable Performance: The 95% confidence intervals for RMSE indicate comparable stability to the 30-gate baseline and significantly better stability than Huber-M and fixed-weight filtering.
Limitations and Future Work
The authors acknowledge several limitations and suggest future research directions:
- Complex Environments: The current algorithm doesn't fully account for complex environments like urban canyons, mountainous areas, or indoor settings where signals are easily blocked and reflected (multipath effect).
- Signal Obstruction Modeling: Future work should focus on establishing more accurate signal models that incorporate multipath effects and signal obstruction.
- Algorithm Optimization: Further optimization of the weighted positioning algorithm based on Chebyshev polynomials is needed to improve accuracy and reliability in these complex scenarios.
- Practical Applications: The authors suggest that a more robust algorithm would be better suited for practical applications like urban navigation and emergency rescue.
Summary updated