LEO IoT Reliability: The 6-Hour TLE Ageing Cliff
The 20-Second Summary
Many LEO IoT devices conserve power by waking up only for a tightly scheduled satellite pass, but that schedule depends on Two-Line Element (TLE) sets whose accuracy degrades with age. This paper quantifies how “TLE ageing” translates into missed passes for Starlink-scale systems, and it identifies a sharp reliability cliff: keeping TLEs fresher than ~6 hours is a practical requirement for >99% pass reliability under the modeled duty-cycled schedule.
The Problem
TLE sets are the ubiquitous ephemerides used to propagate LEO orbits via SGP4. They are “mean elements” fit from observations, and their predictive accuracy decays over time due to unmodeled forces (drag variability, radiation pressure, perturbations). For many systems that’s tolerable, but it becomes a reliability problem for energy-constrained IoT terminals.
A typical satellite IoT terminal is duty-cycled: it wakes up shortly before the predicted Acquisition of Signal (AOS), listens briefly, and goes back to sleep. If the prediction error is tens of seconds, the device can miss the true visibility window entirely, turning what looks like “timing drift” into a hard communication failure for that pass.
Our Approach
We build a simulation pipeline that compares predicted pass windows computed from aged TLEs to “truth” pass windows derived from the most recent available TLE, then applies a concrete device schedule to determine whether the pass would be missed.
In the TeX, the modeled gateway uses:
- $\tau_{\text{wake}} = 15\,\text{s}$ (wake/initialization lead time)
- $\tau_{\text{miss}} = 30\,\text{s}$ (fixed listening window)
- Mask angles $\theta_{\text{mask}} \in {5,10,15,20}^\circ$
- Aged TLE offsets $\Delta t_{\text{age}} \in {3,6,12,24,48,96}$ hours
What We Measured
The study analyzes over 2.8 million pass events for 150 Starlink satellites, observed from four geographically diverse ground stations (San Francisco, Austin, Toulouse, Munich). The primary operational metric is missed-pass probability $P_{\text{miss}}$.
Key Results
The main finding is a non-linear reliability transition:
- For TLE age ≤ 6 hours, $P_{\text{miss}}$ stays below 1% across scenarios.
- At 12 hours, average $P_{\text{miss}}$ jumps to about 3.1%.
- At 24 hours, it deteriorates to > 6.8%.
The mechanism is AOS timing error tails. In the reported CDF:
- For fresh TLEs (3–6 hours), >95% of passes have absolute AOS error < 20 s, which fits inside the modeled 30 s listening window.
- For 24-hour-old TLEs, the median absolute error approaches 30 s.
- For 48-hour-old TLEs, the 95th percentile exceeds 100 s.
The practical implication is straightforward: if you want >99% pass reliability without inflating the device’s listening window, the TeX supports refreshing TLEs at ≤ 6-hour intervals.
The paper also notes that the data overhead of frequent refresh is small. For $N{=}150$ satellites and 4 stations, a 6-hour cadence is reported as ~96 kB/day; scaling to $N{=}3000$ is ~1.9 MB/day.
Limitations and Next Steps
These results are grounded in a specific duty-cycled schedule and Starlink-era TLE cadence. Space weather and catalog differences can shift timing-error tails, and different device timing budgets change the miss criterion. A useful next step is integrating the paper’s “CDF mapping” into an operations playbook: choose a target reliability, pick a device window budget, and derive the required ephemeris refresh interval (including assistance-data distribution constraints).