⚡ electrical engineering

A Conditional Timing Protection Level: Holdover-Limited Undetected Time Error Under GNSS Spoofing

This paper addresses the critical vulnerability of GNSS timing receivers to undetected spoofing attacks by demonstrating through field measurements that standard monitors can fail to flag significant time errors, and consequently proposes a conditional "Timing Protection Level" (TPL) that bounds undetected time error based on oscillator coasting and independent cross-satellite consistency checks rather than relying solely on self-referential clock monitoring.

Original authors: Chakshu Baweja

Published 2026-06-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chakshu Baweja

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Problem: The "Silent" Clock Thief

Imagine you have a very expensive, high-precision wristwatch that gets its time from a global satellite network (like GPS). This watch is used to keep critical systems running, like power grids or stock markets.

The paper starts with a scary discovery: Your watch can be lying to you without you knowing it.

In a recent real-world test (called "JammerTest 2024"), researchers watched a high-end receiver get "spoofed." This means an attacker broadcast fake satellite signals that slowly dragged the receiver's clock off course.

The Reality: The clock was actually 1,010,000 nanoseconds (over a millisecond) off.
The Lie: The receiver's own internal "health check" told the user, "Don't worry, I'm accurate to within 51 nanoseconds."

The receiver was wrong by a factor of 20,000. It was like a car speedometer reading "0 mph" while the car was actually speeding down the highway at 100 mph, and the "Check Engine" light was still green.

Why Can't We Just Trust the Receiver?

The authors explain that standard safety checks (called "RAIM") are designed to catch errors in position (where you are), not errors in time when the error is shared by all satellites.

The Analogy:
Imagine a choir where every singer is holding a sheet of music. If one singer starts singing slightly off-key, the conductor (the receiver) can hear the discord and fix it.
But, if a villain sneaks in and quietly changes the sheet music for every single singer at the exact same time, the choir stays perfectly in harmony with each other. The conductor hears no discord, so they think everything is fine. Meanwhile, the entire choir is singing the wrong song.

Because the fake signals are "coherent" (they move together), the receiver sees no conflict and never raises an alarm, even as the time drifts further and further away.

The "Impossible" Truth

The paper proves a hard mathematical fact: You cannot create a perfect, finite safety limit for this kind of attack if you only rely on the receiver's own clock.

The Analogy:
Imagine trying to measure how fast a thief is stealing your money by looking at your bank account balance. If the thief steals $1 every year, and your bank account naturally fluctuates by $100 a day due to normal spending, you will never notice the theft. The thief can just go slow enough that the theft is hidden inside the "noise" of your normal life.

The authors show that if an attacker moves slowly enough, they can corrupt the clock forever without ever triggering an alarm. Therefore, a "guaranteed" safety limit that works all the time is impossible.

The Solution: A "Conditional" Safety Net

Since a perfect guarantee is impossible, the authors propose a Conditional Timing Protection Level (TPL). This is a safety limit that only works if you have a specific type of backup detector.

The Analogy:
Instead of just listening to the choir (the receiver's internal clock), you hire a second, independent observer standing outside the room.

The Detector: This observer doesn't care if the singers are in tune with each other. They compare the singers to a completely different reference (like a recording of the original song). If the singers drift away from the original song, even if they are still in tune with each other, this observer raises a red flag.
The Safety Net (TPL): Once the observer raises the flag, the receiver stops trusting the fake signals and switches to its own internal battery-powered clock (called "holdover").

The TPL answers the question: "If the observer raises the flag, how much time could the clock have drifted in the few seconds it takes for the receiver to switch to its battery?"

The answer is: Very little.

The paper calculates that even if the switch takes 60 seconds, the error would only be about 458 nanoseconds.
This is thousands of times better than the 1,000,000 nanosecond error the receiver would have accepted without the backup detector.

How It Works (The Math in Plain English)

The formula for this safety limit has two parts:

The "Floor": How sensitive is your backup detector? (In the test, it could spot a difference of about 22 nanoseconds).
The "Coast": How much does your internal clock drift while it's running on battery power before it's fixed? (This depends on the quality of the clock's oscillator).

The total safety limit is simply: The Detector's Sensitivity + The Clock's Drift.

The Catch (Limitations)

The authors are very honest about what this solution doesn't do:

It's not a magic shield: It only works if the backup detector actually catches the attack. If the attacker is incredibly sophisticated and perfectly mimics the "noise" of the real world, the detector might miss it.
It's not a certified guarantee: Unlike aviation safety standards which promise a specific risk level (e.g., "1 in a million chance of failure"), this is an engineering calculation based on real data. It's a "best estimate" based on the specific equipment tested, not a universal law.
It needs real data: The safety numbers provided (like 458 nanoseconds) are calculated based on the specific clock used in the test. If you use a cheaper or different clock, the numbers change.

Summary

The paper says: "Don't trust your GPS receiver's 'I'm fine' light when it comes to time. It can be tricked silently. However, if you add a smart, independent checker that compares the signals to each other, you can limit the damage to a tiny, manageable amount, even if the attacker is very slow and sneaky."

The researchers have made their code and data open-source so others can verify these numbers and build better systems.

Technical Summary: A Conditional Timing Protection Level for GNSS Under Spoofing

Problem Statement
Critical infrastructure increasingly relies on Global Navigation Satellite System (GNSS) for precise timing. However, a coordinated, coherent spoofing attack presents a unique threat to timing receivers: it can slowly pull the served clock time (a common-mode ramp) while maintaining a clean position solution and keeping the receiver's internal integrity flags "green." Unlike position-domain errors, which create geometric residuals detectable by Receiver Autonomous Integrity Monitoring (RAIM), a coherent time-synchronization spoofer drives all satellites coherently. This allows the receiver to absorb the error entirely into its clock estimate, leaving no range residuals to bound. Consequently, a receiver may report high time accuracy (e.g., 51 ns) while actually serving a time error orders of magnitude larger (e.g., ~1 ms). The paper addresses the lack of a quantitative bound on the undetected time error under such attacks.

Methodology
The authors employ a three-pronged approach combining field data analysis, theoretical impossibility proofs, and a conditional bounding framework:

Field Measurement: Using a public dataset from the JammerTest 2024 campaign, the authors reconstructed the clock-solution trajectory of a survey-grade u-blox ZED-F9P receiver. They solved for the clock offset using raw L1 pseudoranges, a fixed clean dual-band position, and broadcast ephemeris, referencing the result against a holdover extrapolation of the pre-attack segment.
Theoretical Analysis: The paper establishes an impossibility result regarding unconditional bounds. It demonstrates that against an adversary free to choose the ramp rate, a self-referential clock-aided monitor (one that compares the current solution to a disciplined reference) cannot bound the error. A sufficiently slow ramp keeps the disciplined reference "in lock-step" with the attack, preventing any alarm while the integrated error grows without limit.
Conditional Bound Construction: To overcome the impossibility, the authors propose a Timing Protection Level (TPL). This is a conditional bound that holds given detection by an independent, model-free cross-satellite consistency monitor. The TPL is defined as the sum of two terms:
- Detectability Floor: The static $k$ -sigma offset floor of the cross-satellite monitor (the maximum error the monitor fails to flag).
- Holdover Uncertainty: The oscillator's coast uncertainty (phase error variance) accumulated over the detection latency ( $\tau_{det}$ ), calculated using van Loan's matrix-exponential integration of the oscillator's power spectral densities (derived from measured Allan deviation).

Key Contributions

Empirical Gap Quantification: The study documents a real-world attack where a receiver served a time error of approximately 1.01 ms while reporting a claimed accuracy of 51 ns. This represents a claim-versus-reality gap of roughly $2 \times 10^4$ , demonstrating that standard receiver self-assessment is insufficient for timing integrity under coherent spoofing.
Impossibility Observation: The paper formally argues that no finite unconditional bound exists for undetected time error under a single self-referential monitor against an unconstrained-rate ramp. This necessitates a conditional approach.
The Timing Protection Level (TPL): The authors define and calibrate a conditional bound. The TPL is a "holdover budget" representing the maximum error that can accrue after a model-free monitor triggers an alarm but before the system recovers.
- Calibration: On the recorded attack, the TPL was calculated to be 114 ns at a 1-second recovery latency and 458 ns at a 60-second latency (with a swept band up to 1205 ns to account for long-tau red noise uncertainty).
- Comparison: These values are 2,200 to 8,800 times smaller than the 1.01 ms error the receiver silently accepted.
Monitor Performance Analysis: The study contrasts two monitoring strategies:
- Clock-Aided Sequential Test (CUSUM): On the recorded slow ramp, this test provided essentially no protection (protection factor of 1x), only alarming after ~993 µs of error had already accrued.
- Model-Free Cross-Satellite Consistency: This monitor detected the attack during the ramp (minutes before capture) by identifying differential inconsistencies that a coherent spoofer could not perfectly suppress.

Results and Significance
The paper concludes that while a receiver's internal integrity flag is not a safety bound, a conditional bound is achievable. The TPL provides a reproducible, closed-form estimate of the maximum undetected time error, provided a model-free cross-satellite monitor is active and the oscillator's holdover characteristics are known.

The authors are explicit about the limitations and boundaries of their claims:

Conditional Nature: The bound is strictly conditional on the model-free monitor detecting the attack. The paper does not characterize the missed-detection probability of this monitor against an optimally coherent adversary.
No Integrity-Risk Budget: Unlike aviation protection levels, the TPL does not carry a certified integrity-risk-per-hour budget (e.g., $10^{-7}$ /hr). It is an engineering bound based on calibrated primitives, not a certified guarantee.
Calibration vs. Validation: The bound is calibrated on a single recorded attack and a specific receiver/oscillator. It has not been field-validated against ground-truth time errors across a population of receivers and attacks.
Long-Coast Uncertainty: For long detection latencies, the bound is governed by the oscillator's long-tau red-noise floor, which is reported as a swept band rather than a single scalar value.

The significance of the work lies in shifting the paradigm from seeking an impossible unconditional bound to providing a rigorous, conditional "holdover budget" that quantifies the risk during the recovery phase of a spoofing event. The construction, including the Kshana simulator and calibration examples, is open-source to facilitate reproducibility and further research into monitor detectability and empirical coverage.