Introduction
Ensuring the "integrity" of Global Navigation Satellite Systems (GNSS) is a critical safety requirement for autonomous vehicles and urban navigation. While integrity monitoring is well-established in civil aviation, terrestrial applications face unique challenges. In "urban canyons," signals are frequently disrupted by transient interference, multipath reflections, and potential spoofing. These threats are often short-lived, appearing suddenly and lasting for a finite duration.
The Challenge of Transient Change Detection
Traditional Quickest Change Detection (QCD) algorithms, such as the Cumulative Sum (CUSUM) test, are optimized for detecting changes that persist indefinitely. They prioritize minimizing detection delay. However, in safety-critical navigation, the primary concern is the Probability of Missed Detection ($P_{md}$) within a predefined "Time to Alert" window. If a signal anomaly is not detected within a specific timeframe (e.g., 6 seconds), the system fails its integrity requirement. Standard methods like CUSUM can "dilute" evidence of short-lived threats by considering the entire signal history.
Proposed Methodology: Finite Moving Average (FMA)
To address the limitations of traditional QCD in transient scenarios, this research focuses on the Finite Moving Average (FMA) algorithm. Unlike algorithms with infinite memory, FMA utilizes a sliding window of the last $m$ samples, keeping the detection logic focused on recent, relevant data.
Key Contributions
- Tight Statistical Bounds: The authors derive new, tighter theoretical upper bounds for both the probability of false alarms ($P_{fa}$) and missed detections ($P_{md}$) for the FMA algorithm. These bounds are robust and applicable to any statistical distribution, extending beyond standard Gaussian assumptions.
- GNSS Metric Application: The performance of FMA is evaluated across three critical GNSS receiver metrics: C/N0 (Carrier-to-Noise ratio), DLLout (Code Discriminator Output), and SAM (Slope Asymmetry Metric).
Performance Evaluation
The researchers conducted extensive Monte-Carlo simulations ($10^6$ runs) to generate Receiver Operating Characteristic (ROC) curves, comparing FMA against CUSUM, Window-Limited CUSUM (WLC), and the Shewhart test.
As shown in Figure 1, FMA consistently outperforms traditional algorithms for the C/N0 metric. For a fixed false alarm rate, FMA achieves a significantly lower probability of missed detection. This advantage is maintained across more complex metrics as well.
In the complex SAM scenario (Figure 4), which involves simultaneous changes in both mean and variance, FMA maintains its performance advantage. The results confirm that FMA serves as an effective "middle ground" between the Shewhart test (optimal for single-sample changes) and CUSUM (optimal for permanent changes).
Key Insights and Results
- Superior Availability: The tighter theoretical bounds derived in this study are between half to two orders of magnitude more precise than existing literature. In practical navigation, tighter bounds reduce pessimistic "unavailability" declarations, allowing autonomous systems to operate more frequently without compromising safety.
- Transient Optimality: The study highlights that traditional QCD toolkits are ill-suited for transient changes where the duration $m > 1$ but finite. FMA's windowed approach prevents the "dilution" of evidence seen in infinite-memory algorithms.
Conclusion and Future Work
The FMA algorithm, supported by the new performance bounds, offers a superior strategy for real-time safety-critical navigation in challenging environments. Future research will explore formal mathematical proofs of FMA's absolute optimality for general non-Bayesian transient change problems and investigate robust "Generalized Likelihood Ratio" (GLR) versions of the algorithm to handle unknown change parameters in real-world dynamics.
Summary updated