TL;DR
This paper investigates the performance of Kalman Filter (KF) based carrier tracking loops for Global Navigation Satellite Systems (GNSS) in challenging, high-dynamic environments with weak signals. By introducing self-adaptive algorithms for both process and measurement covariance matrices, the authors demonstrate improved tracking accuracy and sensitivity compared to standard KF/EKF approaches, particularly in low signal-to-noise conditions.
What Problem Does This Paper Address?
Modern applications like autonomous vehicles and drones rely heavily on GNSS for positioning, navigation, and timing (PNT). However, GNSS signals are inherently weak due to the long distances from satellites to receivers. This weakness, combined with high-dynamics environments (e.g., fast-moving aircraft), makes traditional phase-locked loops (PLLs) struggle to maintain a lock on the signal. This leads to reduced availability and accuracy of GNSS services. The core challenge is to effectively track weak GNSS signals in dynamic scenarios, balancing the trade-offs between integration time, signal strength, and receiver dynamics.
Methods
The paper proposes and evaluates two main Kalman Filter (KF) based carrier tracking loop architectures: a standard KF and an Extended Kalman Filter (EKF). The novelty lies in the introduction of self-adaptive algorithms for the covariance matrices:
- Adaptive Process Covariance Matrix (Q): An algorithm is introduced to adjust the process noise covariance matrix based on the "innovation sequence" (the difference between predicted and actual measurements). This helps the filter better adapt to changing dynamics.
- Adaptive Measurement Covariance Matrix (R): A dedicated noise channel is used to automatically adjust the measurement noise covariance matrix. This adjustment is based on the estimated carrier-to-noise density ratio (C/No), which is crucial for handling varying signal strengths.
The tracking loop models are based on a 4th-order state vector that includes carrier phase error, angular velocity, angular acceleration, and angular jerk. The measurement is derived from a four-quadrant arctangent carrier discriminator.
Datasets
The paper uses simulated GNSS signals generated using a high-dynamic signal model defined by the Jet Propulsion Laboratory (JPL). This model simulates challenging motion profiles including rapid changes in acceleration and jerk, as described in Table 1 of the paper. The simulations are conducted across a range of carrier-to-noise density ratios (C/No) from 25 dB-Hz to 45 dB-Hz. No real-world datasets are used for evaluation.
Key Insights and Ideas
- Adaptive KF is Crucial for Weak Signals: The self-adaptive KF algorithms (A-SKF and A-EKF) significantly outperform their non-adaptive counterparts (SKF and EKF) when GNSS signals are weak (low C/No). This is attributed to better noise reduction and improved tracking sensitivity.
- EKF Offers Slight Edge in Weak Scenarios: While both adaptive approaches show improvement, the EKF-based adaptive methods (A-EKF) tend to slightly outperform the standard KF-based adaptive methods (A-SKF) in challenging, weak signal conditions.
- Diminishing Returns in Strong Signals: As the C/No increases (above approximately 39 dB-Hz), the performance advantage of the adaptive algorithms diminishes. In fact, in strong signal conditions, the standard KF might even show slightly better performance in some metrics.
- Trade-off in Adaptivity: While adaptive algorithms improve tracking in dynamic situations, they might require more time for the tracking loop to "sense" and adapt to sudden changes in dynamics, potentially leading to larger initial tracking errors during these transitions compared to non-adaptive methods.
Key Results
- Improved Tracking Accuracy: The adaptive KF algorithms (A-SKF and A-EKF) demonstrated reduced Root Mean Square Error (RMSE) for carrier tracking errors, especially at lower C/No values (e.g., 25 dB-Hz).
- Enhanced Tracking Sensitivity: The Phase Lock Indicator (PLI) results showed that the adaptive algorithms could maintain lock on the signal in a 25-dB-Hz situation where non-adaptive methods failed.
- EKF vs. SKF Nuances: While A-EKF generally performed slightly better in weak signals, the PLI RMSEs were very close. The mean PLI favored adaptive EKF, while PLI RMSE favored adaptive SKF.
- Performance Convergence: At higher C/No (e.g., 45 dB-Hz), the performance differences between all four tested methods (SKF, EKF, A-SKF, A-EKF) became very small.
Limitations and Future Work
- Simulation-Based: The study relies entirely on simulated data. While the JPL high-dynamic model is realistic, real-world multipath, interference, and atmospheric effects are not explicitly modeled.
- Specific Model: The performance is evaluated against a specific high-dynamic model. Its effectiveness across a broader range of dynamic scenarios might need further investigation.
- Future Work: The paper implicitly suggests that further refinement of adaptive algorithms could lead to even better performance across all signal conditions. Exploring hybrid approaches or more advanced filtering techniques could also be avenues for future research. The authors also mention that the adaptive algorithms can be adjusted depending on real-time estimation results, hinting at potential for online tuning.
Summary updated