Release Notes — v0.6.13¶

Enhanced PPP-RTK SPP seed + RINEX converter frequency unification¶

Release date: 2026-06-07 Type: Positioning improvement (PPP-RTK/VRS-RTK kinematic) + converter refactor Branch: release/v0.6.13

Overview¶

v0.6.13 bundles two independent changes:

Enhanced a-priori SPP seed for PPP-RTK / VRS-RTK (#196) — the per-epoch single-point fix that seeds the CLAS/VRS precise filter now uses the proven v0.6.10 SPP error model (C/N0 weighting + TDCP jump-reject) by default, improving kinematic fix rate and the fixed-epoch error tail. This is a positioning change for PPP-RTK/VRS-RTK; static positioning is unchanged.
RINEX converter unified on obsdef frequency indexing (#71, PR #195) — mrtk convert derives observation-type frequency indices from the same obsdef tables every decoder uses. Converter output change only (some obs-type columns reorder); positioning output is byte-identical.

1. Enhanced a-priori SPP seed for PPP-RTK / VRS-RTK (#196)¶

Mechanism¶

Every epoch, pntpos computes a single-point fix into rtk->sol.rr. On a filter reset (observation-loss gap, persistent-FLOAT, large PPP-RTK↔SPP divergence) the CLAS/VRS filter re-initialises its position from that seed, and resamb_LAMBDA builds the ambiguity-resolution float vector from it. A better seed therefore helps the kinematic, reset-prone case.

The key enabler — no CLAS corruption¶

The seed runs on a private option copy (prcopt_t sppopt = rtk->opt). The C/N0 coefficient slots err[5]/err[6] mean snr_max/snr_error in the SPP varerr, but the iono/trop estimation-error terms in the CLAS varerr. They are therefore set on the copy only, leaving the CLAS measurement model untouched.

Configuration¶

[positioning.clas] enhanced_spp_seed (option key pos1-seedenh):

Value	Seed gets
`off`	nothing — bit-identical to prior behaviour
`cn0+tdcp`	C/N0 weighting + TDCP jump-reject (default)
`cn0+tdcp+robust`	+ IGG-III robust (open-sky opt-in)

Results (PPC-Dataset, 6 urban kinematic runs)¶

cn0+tdcp vs baseline: mean fix rate 23.7 → 24.7 % (+1.0 pp), mean fixed-epoch p95 3.23 → 3.02 m; largest single gain on tokyo_run1 (p95 5.44 → 3.52 m). No large-misfix regression.

Why IGG-III robust is opt-in, not default¶

Robust lifts the mean fix rate a further +1.0 pp and helps open-sky runs, but in a deep urban canyon (tokyo_run3) it shifts the seed enough to trip the filter into fix-and-hold mis-fixes (mis-fixes 5 → 71, worst fixed error 4.4 m → 12 m). The harm is a filter-coupling effect invisible to seed-local diagnostics and could not be auto-gated reliably (self-gate and filter-health gate were both prototyped and rejected), so robust is left to the operator — use it where the environment is open-sky, omit it in dense urban canyons. Full rationale in docs/design/spp-accuracy.md §11.

Safety / scope¶

Static (みなしローバー) positioning unaffected. All six claslib static regression cases run byte-close to the off baseline — deltas sub-cm RMS, inside the existing test tolerances; one VRS case improves (fix 98.9 → 99.9 %). On clean continuously-tracked data the filter rarely falls back to SPP, so the seed is effectively inert.
Plain PPP / PPP-AR are excluded. Investigated and confirmed the lever yields no benefit there: PPP couples to the seed far more weakly (loose VAR_POS prior, no reset-to-SPP snap, AR resolves from the converged float). An empirical probe on madocalib PPP-AR showed identical TTFF and post-convergence accuracy (2D RMS 4.172 → 4.170 cm). See spp-accuracy.md §11.
Real-time too. mrtk run (real-time PPP-RTK/VRS-RTK) drives the same rtkpos() seed path, so the default profile is active in real time as well — arguably where it matters most (stream gaps and slips make real-time the reset-prone regime the seed targets). The benchmark numbers above are post-processing; the real-time effect has not yet been separately measured.

Config cleanup¶

This change ships with a config tidy-up: 31 deprecated commented # constellations / # frequency lines (19 files), superseded by the active systems / signals keys, were removed. Active frequency keys are untouched; no option value changed.

2. RINEX converter unified on obsdef frequency indexing (#71, PR #195)¶

mrtk convert (convrnx) now derives observation-type frequency indices from code2freq_idx() — the obsdef-based mapping already used by every receiver decoder, the RINEX parser, RTCM3 and CLAS — and the legacy fixed per-band code2idx() / band2idx_fixed() are removed.

Output change (converter only, no positioning impact). RINEX obs-type columns now follow obsdef band order, so Galileo lists E5a before E5b and QZSS lists L5 before L2 (BeiDou columns reorder similarly). The -f / --freq N band count keeps its meaning but now selects the same N bands the positioning engines use. Positioning output is byte-identical (the obsdef tables are untouched).

Known limitations (intentional — adding slots would perturb GLONASS/BeiDou positioning):

GLONASS G3 (CDMA, GLONASS-K2 only) — no obsdef_GLO slot.
BeiDou B2a+b (AltBOC) — maps to the 6th obsdef slot, beyond the converter's 5-band --freq mask.

Guarded by a new unit test utest_t_freqidx, which pins the code2freq_idx() (sys, code) → frequency-index contract for GPS/GLONASS/Galileo/QZSS/BeiDou, including the GLONASS G3 regression guard.

Compatibility & migration¶

PPP-RTK / VRS-RTK users: no config change required to gain the enhanced seed; it is on by default. Set [positioning.clas] enhanced_spp_seed = "off" to restore the prior seed exactly. cn0+tdcp+robust is available for open-sky deployments.
RINEX converter users: scripts that key on a fixed obs-type column order must follow the obsdef band order (E5a-before-E5b, L5-before-L2). The signal set and --freq N count are otherwise unchanged.
Everyone else: post-processing positioning output is unchanged by #71, and unchanged by #196 outside PPP-RTK/VRS-RTK.

Tests¶

Full ctest regression: only the two documented pre-existing environment/numerical failures (rtkrcv_rt headless replay, madocalib_pppar_ion_check LAPACK tolerance); no new failures. The enhanced-seed default-off path is byte-identical; all claslib_* / vrs static cases pass. New unit test utest_t_freqidx added for #71. clang-format 21.1.6 / taplo 0.10.0 clean; no test tolerances changed.

Issues / PRs¶

#196 — enhanced a-priori SPP seed for PPP-RTK/VRS (default cn0+tdcp)
#71 / PR #195 — RINEX converter unified on obsdef frequency indexing