Skip to content

Release Notes — v0.7.4

CLAS PPP-RTK claslib parity — storm-day fix-rate gap closed

Release date: 2026-07-04 Type: Accuracy (CLAS PPP-RTK claslib parity) + independent fixes Branch: release/v0.7.4

Overview

v0.7.4 is a five-part sync series that aligns the CLAS PPP-RTK engine with upstream claslib across three layers — the correction-stream timing, the measurement model, and the SPP seed — closing the storm-day fix-rate gap tracked in #225.

On the deterministic-BLAS storm benchmark the kinematic fix rate rises 89.44 % → 93.54 %, now exceeding upstream claslib's 92.88 % on the same data. The single largest contributor is the BackupCSSR / BackupGrid port that bridges short correction gaps (+118 fixed epochs, confirmed by a 5-cell ablation). A new opt-in MRTK_DETERMINISTIC_BLAS build links single-thread OpenBLAS so these ±0.1 pp comparisons are bit-reproducible run-to-run — with no positioning-code change and the default build unaffected.

The release also carries several independent bug fixes: a ~0.10 m Galileo E5a antenna-PCV error, an mrtk l6extract stack overflow, all-GNSS solution-status output, and traceopen path handling.

All CLAS-scoped changes are applied for CLAS only; the MADOCA-PPP references are unaffected. All five #225 PRs (#244–#248) passed CI and Copilot review.

1. Storm-day fix-rate gap — what closed it

The #225 investigation traced the CLAS PPP-RTK deficit to three layers rather than a single bug. Each layer was brought to claslib parity independently:

Layer Change PR
Correction-stream timing mask-time-aligned CSSR selection + per-epoch bank refresh #246
Gap bridging BackupCSSR / BackupGrid (≤ 180 s) #246
Measurement model claslib slot layout + OSR/adaptive-iono + PCV #247
Ionosphere decay est-adaptive AR(1) time constant wired into udion() #245
SPP seed Galileo E5a rows restored + large-residual exclusion #248

BackupCSSR / BackupGrid gap bridging

Ported from claslib: CLAS now bridges correction gaps (≤ 180 s) with the last usable CSSR / grid snapshot instead of dropping the affected epochs to a single solution. On the #225 storm benchmark this is the largest single contributor (+118 fixed epochs, confirmed by a 5-cell ablation).

CLAS PPP-RTK correction-stream timing

CSSR is read through the correction set whose mask time equals the observation epoch, stopping at the first future mask — this fixes one-epoch-stale correction selection and a startup epoch lost to an empty grid. Grid status is bootstrapped by a network scan on the first epoch. mrtk ssr2osr now refreshes the bank snapshot every epoch instead of only on a network change (it previously served the first correction set until it aged out at ~60 s and then stopped) (PR #246).

CLAS PPP-RTK measurement model

Brought to the claslib observation slot layout — GPS {L1, L2, L5} / GAL {E1, –, E5a, E6, E5b, E5ab} / QZS {L1, L2, L5, L6} — with the OSR ambiguity index for est-adaptive ionosphere, a facility-change filter-reset guard, residual wavelength-ratio parity, and CPC freq-major persistence across zdres passes. All of this is applied for CLAS only.

QZSS L1 signal priority (C > S > L > X > Z) is a CLAS-scoped runtime override: claslib and MADOCALIB genuinely disagree on the order, so it is enabled only from mrtk post while correction = CLAS, leaving the MADOCA-PPP references unaffected (PR #247).

Ionosphere Gauss-Markov decay

The est-adaptive ionosphere time constant was parsed but never wired into udion() state propagation; the AR(1) decay now matches claslib (PR #245).

CLAS SPP velocity seed

The GAL / SBS iono-free second frequency is now taken from slot 2 (E5a), restoring the Galileo rows to the seed like upstream (PR #248).

2. Deterministic-BLAS builds

MRTK_DETERMINISTIC_BLAS is a new opt-in CMake option that links a deterministic LP64 single-thread OpenBLAS instead of Apple Accelerate. Accelerate's non-deterministic floating-point reductions scatter storm-day fix rates run-to-run because the AR ratio test amplifies ULP noise into fix/float flips; OpenBLAS makes those comparisons bit-reproducible so the ±0.1 pp deltas above are meaningful.

It is intended for benchmarking / CI reproducibility only. There is no positioning-code change and the default build is unaffected. A configure-time check verifies OpenBLAS is the resolved LAPACK provider (PR #244).

3. New configuration

  • pos2-rejethres / pos2-rejeminsat — upstream SPP large-residual exclusion. Default 0 keeps standalone SPP bit-compatible; the CLAS PPP-RTK / VRS private seed maps 0 to upstream's 20 m / 7-sat defaults (PR #248).
  • pos1-frequency = l1+l2+l5 — config alias for the CLAS three-frequency observation slot layout (PR #247).

4. Bug fixes

  • Galileo E5a antenna PCV error (~0.10 m) — receiver PCV slot mapping now uses claslib-compatible ANTEX compaction. The stale offset had been cascading into ambiguity resets (the E34 case in the #225 ledger) (PR #247).
  • mrtk l6extract 2-byte stack overflow in sbf_extract_l6_payload (PR #237).
  • traceopen filename handling — guard path length before reppath and expand time keywords in the trace filename (#83, PR #239).
  • Solution-status (.stat) output now emits all GNSS, not just GPS (PR #240).
  • Test harness — BSD mktemp suffix collision in run_rtkrcv_test.sh (#194, PR #241).

5. Documentation

  • ICD & academic-paper reference catalogs — docs/reference/icd/ and docs/reference/papers/ index pages added to the MkDocs Reference nav (link-only rows; collected PDFs held locally) (PR #238).

6. Validation

  • Storm-day CLAS kinematic PPP-RTK (deterministic OpenBLAS): fix rate 89.44 % → 93.54 %, above upstream claslib's 92.88 % on the same data. The BackupCSSR contribution (+118 epochs) is confirmed by a 5-cell ablation.
  • claslib label 34/34 (including RT replay and F5 absolute-accuracy checks); the MADOCA-PPP references are unaffected (the QZSS L1 priority change is CLAS-scoped).
  • Full ctest suite green apart from the two perennial environment failures on the maintainer's machine (rtkrcv_rt headless RT replay, madocalib_pppar_ion_check LAPACK-vs-reference tolerance), both of which reproduce identically on a clean develop.
  • All five #225 PRs (#244–#248) passed CI and Copilot review.

Compatibility & migration

No configuration changes are required. Existing CLAS PPP-RTK and MADOCA-PPP workflows are unaffected; the CLAS parity work is scoped to correction = CLAS and the MADOCA-PPP references are untouched. The new pos2-rejethres / pos2-rejeminsat and pos1-frequency = l1+l2+l5 options default to backward-compatible behavior, and MRTK_DETERMINISTIC_BLAS is off by default with no effect on the standard build.

Issues / PRs

References

  • IS-QZSS-L6 — CLAS (L6D) CSSR message format.
  • claslib — the upstream reference implementation for CLAS PPP-RTK.