Timing Properties of the Starlink Ku-Band Downlink

We develop signal capture and analysis techniques for precisely extracting and characterizing the frame timing of the Starlink constellation’s Ku-band downlink transmissions. The aim of this work is to determine whether Starlink frame timing has sufficient short-term stability to support pseudorange-based opportunistic positioning, navigation, and timing (PNT). A second goal is to determine whether frame timing is disciplined to a common time scale such as GPS time. Our analysis reveals several timing characteristics not previously known that carry strong implications for PNT. On the favorable side, periods of ns-level jitter in frame arrival times across all satellite versions indicate that Starlink hardware is fundamentally capable of the short-term stability required to support GPS-like PNT. But there are several unfavorable characteristics that, if not addressed, will make GPS-like PNT impractical: (1) The v1.0 and v1.5 Starlink satellites exhibit once-per-second abrupt frame timing adjustments whose magnitude (as large as 100s of ns) and sign appear unpredictable. Similar discontinuities are also present in the v2.0-Mini frame timing, though smaller and irregularly spaced. (2) Episodic 15-s periods of high frame jitter routinely punctuate the nominal low-jitter frame arrival timing. (3) Starlink frame timing is disciplined to GPS time, but only loosely: to within a few ms by adjustments occurring every 15 s; otherwise exhibiting drift that can exceed 20 ppm. These unfavorable characteristics are essentially incompatible with accurate PNT. Fortunately, they appear to be a consequence of software design choices, not hardware limitations. Moreover, they could be compensated with third-party-provided corrections.

💡 Research Summary

**
The paper presents a comprehensive study of the timing characteristics of SpaceX’s Starlink Ku‑band downlink, with the explicit goal of assessing whether the constellation can support pseudorange‑based positioning, navigation, and timing (PNT) at a level comparable to traditional GNSS. Using a laboratory‑grade capture system, the authors receive Ku‑band signals with a 40 dBi parabolic dish, down‑convert them to L‑band, and digitize them at up to 62.5 Msps on a radio‑frequency signal analyzer (RFSA). A parallel GPS L1 capture path shares the same GPS‑disciplined 10 MHz oven‑controlled crystal oscillator (OCXO), guaranteeing that both the Starlink and GPS sample streams are synchronized to better than 1 ns when configured with identical sampling rates.

The Starlink downlink employs OFDM frames consisting of 302 symbols (each 4.4 µs) plus a guard interval, yielding a fixed frame period of 1/750 s (≈1.33 ms). Each frame begins with a Primary Synchronization Sequence (PSS) followed by a Secondary Synchronization Sequence (SSS). The authors exploit the known PSS+SSS pattern as a correlation template, allowing them to locate the exact start of every frame with nanosecond precision relative to GPST.

Three major timing behaviours emerge from the analysis. First, “low‑jitter” intervals are observed across all satellite versions (v1.0, v1.5, v2.0‑Mini). During these periods the inter‑frame arrival time varies by only a few nanoseconds, indicating that the onboard crystal oscillators are of high quality and that the capture system’s clock is sufficiently stable.

Second, the authors detect abrupt, once‑per‑second timing adjustments in the v1.0 and v1.5 satellites. The magnitude of these adjustments ranges from 10 ns up to 100 ns, with signs that appear random. Similar, though smaller and irregular, discontinuities are present in the v2.0‑Mini fleet. These adjustments are interpreted as software‑driven clock‑synchronisation events that attempt to align the satellite’s internal time to GPS, but they introduce unpredictable offsets that would corrupt pseudorange calculations.

Third, a 15‑second periodic phenomenon is identified, coinciding with the Fixed Assignment Interval (FAI) during which Starlink re‑assigns downlink beams. At FAI boundaries the frame‑to‑frame jitter spikes to tens of nanoseconds, and the authors note that the timing drift between the 15‑second corrections can exceed 20 ppm (≈20 µs s⁻¹). Consequently, over longer observation windows the accumulated timing error can reach milliseconds, far beyond the sub‑microsecond stability required for GNSS‑grade PNT.

In terms of absolute time reference, the constellation is “loosely” disciplined to GPS time: every 15 seconds a correction of a few milliseconds is applied, but between corrections the satellite clock drifts appreciably. This loose discipline, combined with the per‑second and 15‑second irregularities, renders the raw Starlink timing unsuitable for direct GPS‑like positioning without additional processing.

Crucially, the authors argue that these adverse characteristics stem from software design choices rather than hardware limitations. The onboard oscillator quality is sufficient; the problematic behaviours arise from the timing‑adjustment algorithms and beam‑reassignment schedule. They propose that third‑party services could provide real‑time correction data, or that SpaceX could modify its firmware to smooth the per‑second adjustments and tighten the 15‑second discipline. With such mitigations, the ns‑level short‑term stability demonstrated in the low‑jitter intervals could be leveraged to achieve meter‑level positioning and nanosecond‑level timing, opening a path to low‑cost, opportunistic PNT services that complement or augment existing GNSS.

Overall, the paper fills a critical knowledge gap by quantifying Starlink’s frame‑timing stability and its relationship to GPS time, highlighting both the promise (high‑quality short‑term stability) and the obstacles (software‑induced jitter and drift). The findings suggest that, with modest software revisions or external correction mechanisms, Starlink’s Ku‑band downlink could become a viable platform for high‑precision, opportunistic PNT.