Skip to content

GNSS timing and holdover assessment

R9 evaluates receiver-clock estimates without pretending that a position FIX or an offline clock column is an NTP/PTP service. It uses BRUX00BEL because the IGS final CLK product contains an independent station-clock series and the official IGS station metadata identifies an external cesium standard. The released assessment is No-Go: the sealed day exceeded the frozen clock-step gate, and neither day contains a physical PPS outage test.

Commands

Run a bounded 120-epoch smoke entirely from a verified cache:

python3 apps/gnss.py timing-holdover-workflow \
  --phase development --mode smoke \
  --output-dir output/timing_r9/smoke \
  --cache-dir data/timing_r9/cache --offline

The full development replay is:

python3 apps/gnss.py timing-holdover-workflow \
  --phase development --mode full \
  --output-dir output/timing_r9/development-replay \
  --cache-dir data/timing_r9/cache

The 2024-01-02 holdout is closed_no_go. Do not change it back to sealed or rerun it. Select a different hash-pinned day for a future release.

Data and clock contract

The workflow fetches BKG IGS 30-second BRUX observations, broadcast navigation, the IGS final CLK product, the official IGS Network station API snapshot, and the GSI RNXCMP converter. Every compressed and materialized file has byte-count and SHA-256 provenance. The 2024 RINEX headers agree across both days:

  • monument 13101M010;
  • receiver 3057609, SEPT POLARX5TR firmware 5.5.0;
  • antenna 00464, JAVRINGANT_DM SCIS.

The current pinned IGS API snapshot reports the same monument, receiver serial, antenna serial/type, IGS20 membership, and an external cesium frequency standard; its current firmware field is 5.7.0. Historical RINEX, not the later API firmware, is authoritative for the two 2024 runs.

The normal .pos schema intentionally remains unchanged. gnss spp adds an optional --clock-csv artifact containing SPP receiver clock bias in metres, the explicit metres-to-seconds conversion, Doppler clock drift, status, satellites, and PDOP. This is important because the internal SPP bias state is range-equivalent metres even though other solver interfaces may describe clock states differently.

IGS final CLK AR BRUX rows are an independent combined receiver-clock reference aligned to the IGS time scale. They are available every five minutes: 288 points on the development day and 283 on the holdout. The workflow matches by GPST epoch and reports absolute phase error and finite-difference relative frequency error.

Simulated holdover

At each eligible IGS clock epoch, the evaluator fits a line to the preceding 1,800 seconds of GNSS-connected SPP clock bias and extrapolates 300, 600, 900, or 1,800 seconds. It then compares that prediction with the future IGS clock. This emulates an open-loop model but does not disconnect the receiver or measure its PPS output. It therefore cannot reveal real oscillator switching, temperature, power, cable, timestamping, kernel, NIC, or network-servo effects.

The frozen development gates were 95% lock/reference coverage, lock phase p95 at most 39 ns, maximum five-minute lock step at most 12 ns, and holdover p95 at most 40/40/40/42 ns for 5/10/15/30 minutes. Development passed with:

  • 100% reference coverage and lock phase p95 25.71 ns;
  • maximum lock step 7.41 ns;
  • relative-frequency p95 7.94e-12;
  • simulated holdover p95 26.05–27.78 ns.

The sealed day kept 100% coverage, phase p95 21.26 ns, and holdover p95 21.86–22.52 ns, but its maximum five-minute phase-error step was 14.71 ns. That exceeds the unchanged 12 ns gate, so the assessment is degraded and the release is No-Go. Evidence is preserved in the development record and holdout run1 record.

Outputs and service boundary

The bundle contains spp.pos, receiver_clock.csv, SPP summary, workflow log and manifest. The sign-off directory adds clock_comparison.csv, clock_scorecard.png, summary, and manifest.

No NTP/PTP daemon or PPS service is exposed. Before one can be considered, perform a real GNSS disconnect/reconnect test while comparing physical PPS to an independent traceable counter, record oscillator/environment/power state, measure system-clock and network timestamping paths, define UTC/leap-second handling, characterize MTIE/TDEV/Allan deviation, freeze alert/holdover limits, and validate a new untouched multi-day holdout.