Release Notes — v0.7.13¶

MADOCALIB 2.2 sync: the BeiDou PRN reassignment¶

Release date: 2026-10-31 Type: Upstream sync + bug-fix release. Configuration action needed if you enable BeiDou. ABI change in prcopt_t. Branch: release/v0.7.13

Overview¶

Between 2026-04-10 and 2026-04-21 BeiDou retired most BDS-2 satellites and moved BDS-3 satellites onto the PRNs they freed:

• C01–C08 and C11–C14 are now BDS-3.
• BDS-2 remains only on C09 and C10.
• The BDS-3 IGSO satellites moved from C38–C40 to C06–C08, and C38–C40 are now MEO.

MRTKLIB, like MADOCALIB up to 2.1, treated every BeiDou PRN below 19 as BDS-2. On current data this gave those satellites the wrong signal and frequency model, put them in the BDS-2 receiver-clock group, and placed them in the wrong PPP-AR groups.

MADOCALIB 2.2 (2026-10-30, "Support IS-QZSS-MDC-005") makes the classification configurable. v0.7.13 ports that change, its MDC-005 decoder change, and fixes found while porting.

The PRN table and the rules for choosing a value are in the new BeiDou PRN Reassignment guide.

BeiDou BDS-2 / BDS-3 classification (#401)¶

• Defaults. They are the same as MADOCALIB 2.2. Legacy .conf files use pos2-bds2prn / pos2-bds3igso.
• List syntax. Entries are satellite IDs or ranges, e.g. ["C09", "C10"] or ["C01-C18"].
• Invalid lists. A malformed entry or a list longer than 255 characters is rejected with a warning. It is not partly applied. For mrtk bias -bds2prn a list over 1023 characters is an error.
• The IGSO list replaces a hard-coded rule. PPP-AR used C38–C40 as IGSO. The IGSO satellites are never taken as the PPP-AR reference satellite and are dropped first in partial AR.
• When the setting is applied. mrtk post applies it before reading the input files, and mrtk run applies it at startup.
• Warning. Both print a warning when BeiDou is enabled and bds2_sats is absent. bds2_sats = [] states "no BDS-2" and silences the warning.
• mrtk bias reads no configuration file. It gains -bds2prn LIST (--bds2-sats), the same name as the MADOCALIB 2.2 cssr2ssr option (#403).

Bundled configurations are set for the data they process:

MADOCA-PPP decoder (IS-QZSS-MDC-005)¶

• BDS-3 satellites in GNSS ID 3. MDC-005 may send a BDS-3 satellite on PRN 1–18 in GNSS ID 3. Its code and phase biases are now read with the BDS-3 (GNSS ID 7) signal table.
• Out-of-bounds write for an undefined signal. A signal-mask bit that is undefined in the table in use used to index cbias[-1] / pbias[-1], an out-of-bounds stack write. It is now skipped.
• Sparse cell masks (also in MADOCALIB 2.2). A skipped cell left the cell-mask cursor where it was. With a sparse cell mask, the decoder then read a bias for an absent cell and shifted every later bias in the message.
• Unsupported satellites (also in MADOCALIB 2.1 / 2.2, #402). A masked satellite that MRTKLIB cannot number had its biases written to ssr[-1], over the rtcm_t members before ssr[].

The current L6E broadcast and the bundled data do not trigger these faults: all MADOCA regression outputs are byte-identical before and after.

Validation¶

Unchanged outputs. On an internal-LU build, before vs after:

• Bundled data: all 13 MADOCA-PPP and IGS regression outputs are byte-identical.
• PPC benchmark: the SPP and RTK outputs of all six runs are byte-identical.

These datasets predate the reassignment, and their configurations are set to the old rule.

Live data (2026-10-13, 1 h). The before/after numbers come from one 1-hour session with a mosaic-G5. The reference is a GPS + Galileo static PPP over the same session.

The CODE MGEX biases cover BeiDou only for the BDS-3 B1C / B2a signals. They list C06–C08 and C11–C14 among the BDS-3 satellites.

Tests¶

• MDC-005 regression on the MADOCALIB 2.2 sample (MIZU, 2026-03-28, L6E PRN 207; dataset madocalib_mdc005-v1): PPP and PPP-AR.
• Upstream parity. On the same platform, MRTKLIB is bit-identical to upstream 2.2.
• Parity tolerances. PPP 0.16 m, PPP-AR 0.090 m. The references come from upstream built on macOS. Upstream built on Linux (gcc / glibc) differs from them by up to 3D RMS 12.07 cm, because the PPP float solution has not yet converged. The tolerances cover that platform difference.
• Absolute accuracy against the IGS week-2411 SINEX gates accuracy: 2D 95 % ppp 6.0–7.1 cm, pppar 1.4 cm.
• utest_mcssr_bd3. A synthetic ST1 / ST4 / ST5 message covering:
• BDS-3 in GNSS ID 3
• an undefined signal
• a sparse cell mask
• an unsupported satellite
• the BeiDou generation API

Each decoder fault fails this test on the old code. - utest_t_toml. Covers the new keys and the absent / [] / invalid / over-long cases. - madocalib_pppar_ion_l6d_template. It no longer compares two runs byte for byte, which failed intermittently on Accelerate (#404). It checks that no .l6d input is ignored, that both runs produce the same solution epochs, and that the solutions agree within the pppar_ion tolerance.

Upgrade notes¶

• If your configuration enables BeiDou, set bds2_sats:
• ["C09", "C10"] for data after 2026-04-21;
• ["C01-C18"] for data before 2026-04-10, plus bds3_igso_sats = ["C38-C40"] for PPP-AR;
• for 2026-04-10 to 04-21 see the guide (no single list is right).

Without the key, every PRN 1–18 satellite is BDS-3, which differs from v0.7.12 for pre-reassignment data. - mrtk bias: pass -bds2prn in the same way. - External drivers that embed prcopt_t must rebuild: the struct gains bds2prn, bds3igso and bds2prn_set (appended). - New API: set_bds2prn(), is_bds2prn(), set_bds3igso(), is_bds3igso(), bds_gen(), set_bdsopt().

Also in this release¶

The repository was repositioned as a library for open PPP / PPP-RTK correction services (PR #399):

• new GitHub description;
• a "Supported Correction Services" table in the README;
• an updated .zenodo.json description, which reaches Zenodo with this tagged release.

Test status¶

Full suite: 180 / 180 (macOS, Accelerate). RT replay tests are timing-sensitive and can false-fail under CPU contention; re-run them in isolation.

Next¶

• v0.7.14: BeiDou PPP-B2b.
• v0.7.15: SouthPAN PVS (under evaluation).
• v0.8.x: CLAS CMA authentication.
• Still open: full TOML save for mrtk run (#342) and several PRN files per CLAS channel in post-processing (#373).