MRTKLIB v0.4.2 Release Notes¶

Release date: 2026-03-08 Type: Feature (demo5 algorithm improvements ported to PPP-RTK and PPP engines) Branch: feat/ppp-rtk-demo5-improvements

Overview¶

v0.4.2 extends the demo5 kinematic accuracy improvements from v0.4.1 (RTK engine) to the CLAS PPP-RTK (mrtk_ppp_rtk.c) and MADOCA PPP (mrtk_ppp.c) engines. The RTK and VRS-RTK engines are unchanged.

Ported improvements cover four implementation phases:

Highlights¶

• GLONASS clock guard ported to ephpos() — the |taun| > 1 s rejection that already guarded ephclk() (since v0.4.1) now also guards the satellite position computation path used by PPP-RTK and PPP.
• detslp_code() extended to PPP-RTK and PPP — signal-code changes between epochs (receiver re-acquisition events) are detected and used to reset phase-bias states before false integer candidates can accumulate.
• detslp_dop() extended to PPP-RTK and PPP — Doppler-based phase-rate slip detector with clock-jump mean removal, identical to the RTK implementation.
• Full EFACT expansion — varerr() in both PPP-RTK and PPP now handles all seven constellation EFACT factors (GAL/QZS/CMP/IRN) instead of just GPS/GLO/SBS.
• Adaptive outlier threshold (PPP-RTK only) — The residual_test() sigma-normalised threshold is inflated 10× for newly-initialised biases, matching the demo5 RTK behaviour. An equivalent change to undifferenced PPP caused a severe regression (tokyo_run1 MADOCA RMS: 1.8 m → 199 m) and was explicitly reverted.

Disclaimer: The configuration parameters used here have not been tuned or optimised for maximum accuracy. Results are provided for reference only and should not be taken as representative of the best achievable performance.

What Changed¶

Phase A — GLONASS ephemeris rejection in ephpos()¶

ephpos() calls selgeph() followed by geph2pos() to compute the satellite position and clock. Previously, corrupted GLONASS clock data (|taun| > 1 s) was only caught in ephclk(). When SSR corrections are active (PPP-RTK and PPP), satpos_ssr() calls ephpos() directly for the base clock; the dts[0] contribution from geph2pos() could carry a multi-second taun error into the final clock.

Fix: added if (fabs(geph->taun) > 1.0) return 0; immediately after selgeph() in ephpos(), consistent with the guard already present in ephclk().

Files changed: src/data/mrtk_eph.c

Phase B — PPP-RTK PAR enhancements¶

Two AR-quality guards from the RTK engine (v0.4.1) are now also active in the PPP-RTK engine (ppp_rtk_pos()):

B2 — Position variance gate¶

Before invoking resamb_LAMBDA(), the mean diagonal of the 3 × 3 position covariance block is compared to thresar[1]. If posvar > thresar[1] the AR step is skipped entirely (nb = 0; goto ppprtk_ar_done), preventing premature integer fixing before the PPP-RTK filter has converged.

The gate is disabled unless thresar[1] > 0.0 (library default: 0.9999; off = 0.0).

B3 — arfilter¶

When resamb_LAMBDA() returns nb ≤ 1 (AR failed or insufficient satellite pairs), and opt->arfilter is set, any satellite whose phase-bias lock count equals 0 (newly-locked) is backed off by -(minlock + delay) and LAMBDA is retried. The delay is staggered (+2 per satellite) to prevent all newly-locked satellites from becoming eligible simultaneously on the next epoch.

Files changed: src/pos/mrtk_ppp_rtk.c

Phase C — Cycle-slip detectors for PPP-RTK and PPP¶

detslp_dop() — Doppler-based slip detection¶

Added as a static function in both mrtk_ppp_rtk.c and mrtk_ppp.c. Implementation is identical to the RTK version in mrtk_rtkpos.c:

1. For each observation, compute dph = (L[f] − ph[0][f]) / tt (phase rate from carrier phase) and dpt = −D[f] (Doppler-derived phase rate).
2. Accumulate the mean of dph − dpt for all satellites with |diff| < 3 × thresdop.
3. Flag any satellite whose deviation from the mean exceeds thresdop (cycles/s) as a cycle slip.

The mean-removal step is the clock-jump fix from demo5: a large receiver-clock jump shifts every satellite's Doppler by the same amount, so subtracting the mean prevents false slip flags on all satellites simultaneously.

Activated by pos2-thresdop > 0.0 (library default: 0; disabled).

ph[0][f] / pt[0][f] update for PPP: The PPP engine did not previously store phase observations in ssat[].ph[]/ssat[].pt[]. A block was added to update_stat() to record them after each epoch, enabling detslp_dop() to compare consecutive epochs.

detslp_code() — Observation-code-change slip detection¶

Added as a static function in both mrtk_ppp_rtk.c and mrtk_ppp.c. Compares obs[i].code[f] against the stored ssat[sat-1].codeprev[f][0]; any change between non-CODE_NONE values sets the slip flag and logs the transition.

Call order in udbias_ppp():

detslp_ll()    (LLI flag)
detslp_gf()    (geometry-free combination)
detslp_dop()   ← new
detslp_code()  ← new

Files changed: src/pos/mrtk_ppp_rtk.c, src/pos/mrtk_ppp.c

Phase D — Measurement noise model improvements¶

D1 — Full per-constellation EFACT in varerr()¶

Both mrtk_ppp_rtk.c and mrtk_ppp.c previously used a three-way conditional for the system error factor:

/* Before */
fact *= sys == SYS_GLO ? EFACT_GLO : (sys == SYS_SBS ? EFACT_SBS : EFACT_GPS);

This silently applied EFACT_GPS = 1.0 to Galileo, QZSS, BeiDou, and IRNSS. The corrected expansion uses all seven EFACT constants from mrtk_const.h:

/* After */
fact *= sys==SYS_GLO?EFACT_GLO:(sys==SYS_GAL?EFACT_GAL:
        (sys==SYS_QZS?EFACT_QZS:(sys==SYS_CMP?EFACT_CMP:
        (sys==SYS_IRN?EFACT_IRN:(sys==SYS_SBS?EFACT_SBS:EFACT_GPS)))));

EFACT values (mrtk_const.h): | Constellation | EFACT | |---------------|------:| | GPS | 1.0 | | GLONASS | 1.5 | | Galileo | 1.0 | | QZSS | 1.0 | | BeiDou | 1.0 | | IRNSS | 1.5 | | SBAS | 3.0 |

The only effective change is for IRNSS (1.0 → 1.5). GPS, GLO, GAL, QZS, CMP, SBS are unchanged by value.

D2 — Adaptive outlier threshold (PPP-RTK only)¶

In residual_test() (PPP-RTK's outlier rejection loop), the per-residual threshold inno0 is inflated 10× when the corresponding phase-bias state was just initialised:

double inno0_eff = inno0;
if (type == 0 && rtk->opt.std[0] > 0.0) {
    int ib = IB_RTK(sat2, freq, &rtk->opt);
    if (rtk->P[ib + ib * rtk->nx] == SQR(rtk->opt.std[0])) inno0_eff *= 10.0;
}

PPP-RTK uses a sigma-normalised test (v² < Q × inno0²) where Q already incorporates the large initial bias variance via the Kalman innovation covariance. When bias variance is large, inno0 × √Q is already a generous threshold; the 10× inflation is a belt-and-suspenders guard that has no adverse effect.

Why D2 was not applied to PPP: PPP's outlier check compares |v| > maxinno in metres (not sigma-normalised). A newly-initialised PPP phase bias has undifferenced residuals that can legitimately reach tens of metres (ionosphere, troposphere, orbit errors not yet estimated). Inflating maxinno by 10× (default 30 m → 300 m) admitted extreme outliers into the filter, corrupting the bias state. Benchmark confirmed: tokyo_run1 MADOCA RMS degraded from 1.8 m to 199 m. The change was reverted; PPP retains the original fixed-threshold check.

Files changed: src/pos/mrtk_ppp_rtk.c

Benchmark Results¶

All benchmarks use --skip-epochs 60, PPC-Dataset urban driving runs (6 × nagoya/tokyo). See docs/benchmark.md for full three-tier tables.

CLAS (PPP-RTK) — FIX tier (Q = 4)¶

• 2/6 Tokyo runs: FIX RMS_2D improved (tokyo_run1: −14%, tokyo_run2: −13%)
• 4/6 runs: Fix% and accuracy unchanged
• nagoya_run2: −3.5 pp fix rate, but 1σ accuracy improved (0.717 m → 0.461 m) and FF rate improved (+4 pp: 63.9% → 67.9%); fewer but higher-quality fixes

MADOCA (PPP) — PPP tier (Q = 3)¶

All six MADOCA runs are identical to the v0.3.3 baseline. The Phase A/C/D1 improvements did not produce a measurable change on this dataset; Phase D2 was explicitly excluded.

Test Suite (57 tests, unchanged)¶

All 57 regression tests from v0.4.1 continue to pass. No new test entries were added.

Acknowledgements¶

RTK algorithm improvements in this release (Phases A–D) are adapted from demo5 RTKLIB by rtklibexplorer.

Benchmark results use the PPC-Dataset from the Precise Positioning Challenge 2024 (高精度測位チャレンジ2024), organised by the Institute of Navigation Japan (測位航法学会):

Taro Suzuki, Chiba Institute of Technology. PPC-Dataset — GNSS/IMU driving data for precise positioning research. https://github.com/taroz/PPC-Dataset