Skip to content

Galileo HAS Support

Status of Galileo High Accuracy Service (HAS) support in libgnss++.

Capability Status
Float PPP with HAS corrections from the HAS Internet Data Distribution (IDD, RTCM 3 SSR) Supported (gnss_ppp --ssr-rtcm <idd.rtc> --ssr-rtcm-profile has-idd)
Static float PPP validated against a known coordinate Validated on the public OBE4 / 2023-08-17 IDD sample, lane gnss reproduce has-idd-ppp
Kinematic float PPP Validated on the same sample (white-noise position, no motion prior): 0.12 m horizontal / +0.06 m up after one hour (hour RMS 0.09 m / 0.10 m), gated by the lane
HAS signal-in-space (Galileo E6-B C/NAV page) decoder Supported (gnss_ppp --has-pages <log>, gnss has-info): u-blox RXM-SFRBX, Septentrio SBF GALRawCNAV and cssrlib page text; CRC-24Q, Reed-Solomon (HPVRS) page recovery, MT1 mask / orbit / clock full-set / clock subset / code bias / phase bias decoding
SIS decoder parity with cssrlib Exact: every decoded orbit, clock and code-bias value of three public recordings (1 x 9 min u-blox X20, 2 x 1 h Septentrio mosaic-X5; 95 k values) equals cssrlib's cssr_has decoder
Static float PPP with SIS corrections Validated, indicative on the Kamakura 2025-02-15 hour (Japan, outside the HAS service area), lane gnss reproduce has-sis-ppp: 0.07 m horizontal / +0.11 m up after one hour
HAS phase biases and PPP-AR Not available: phase biases are decoded (gnss has-info) but not applied; the HAS Service Level 1 streams used here carry none, so ambiguities stay float

Using HAS IDD corrections

The HAS IDD service distributes the HAS corrections as an RTCM 3 SSR stream (GPS 1060 / 1059, Galileo 1243 / 1242) together with the GPS 1019 and Galileo 1046 broadcast ephemerides the corrections refer to.

gnss_ppp --obs OBE42023229c.obs --nav OBE42023229c.nav \
  --ssr-rtcm idd2023229c.rtc --ssr-rtcm-profile has-idd \
  --static --out has_static.pos

--ssr-rtcm-profile has-idd switches the RTCM SSR ingestion to RTCM 10403.3 semantics, which the legacy default (kept byte-identical for existing users) does not apply:

  • IODE-matched broadcast orbits. Every correction sample carries the SSR IODE / IODnav and PPP evaluates exactly that broadcast record. A satellite whose referenced ephemeris is missing is skipped instead of being corrected against a different record.
  • Ephemerides from the stream. GPS 1019 and Galileo 1045 / 1046 messages in the SSR file are merged into the navigation data. A RINEX navigation file converted from the same stream usually keeps only the last record per satellite; on the OBE4 sample, for example, the HAS corrections for G11 and G12 reference IODEs that are not in the RINEX file.
  • Galileo I/NAV only. HAS clocks refer to the I/NAV clock, so F/NAV records with the same IODnav are never selected (NavigationData::setGalileoEphemerisSource(INavOnly)).
  • Updates held until the next one. Each orbit / clock update is applied with its rates and clock polynomial from its epoch until the next update of the same satellite (at most 90 s), and never interpolated across an IODE change.
  • RTCM code biases. Bias signal IDs follow the RTCM SSR tables (GPS 0 = L1 C/A, 10 = L2 P, 8 = L2C(L); Galileo 2 = E1-C, 6 = E5a-Q, 9 = E5b-Q, 16 = E6-C) and are added to the pseudorange. HAS code-bias messages carry a fixed, stale epoch, so the latest bias set per satellite is held by stream order. The HAS biases replace broadcast TGD / BGD.

Orbit corrections are applied with the RTCM convention (x = x_brdc - R_rac->ecef * dRAC, clock dt = dt_brdc + dC / c) and refer to the antenna phase centre of the broadcast orbit, so no satellite antenna offset is applied.

Using HAS signal-in-space pages

HAS corrections are broadcast on the Galileo E6-B signal as 448-bit HAS pages inside the C/NAV pages. gnss_ppp --has-pages decodes a receiver log of those pages and feeds the corrections to float PPP with the same conventions as the has-idd profile (IODE-matched broadcast orbits, Galileo I/NAV only, held updates, code biases instead of TGD / BGD):

gnss_ppp --obs 046r_rnx.obs --nav BRDC00WRD_S_20250460000_01D_MN.rnx \
  --has-pages 046r_gale6.txt --static --out has_sis_static.pos
Page log --has-pages-format Notes
u-blox UBX (.ubx) ubx RXM-SFRBX with gnssId 2 / sigId 8 (E6-B, 16 words), e.g. X20 / F9 E6 firmware; page time = latest RXM-RAWX epoch
Septentrio SBF (.sbf, *.yy_) sbf GALRawCNAV (block 4024); pages the receiver flags as CRC-failed are dropped
cssrlib page text cssrlib (default for other extensions) wn tow prn type len hex lines as in cssrlib-data *_gale6.txt

gnss has-info --input <log> prints the page / message statistics, the masks (satellites and signals), the flag patterns and validity intervals, and dumps --messages-csv, --corrections-csv (every decoded orbit / clock / bias value with its HAS sign) and --updates-csv (the held per-satellite updates PPP uses).

Decoding follows the HAS SIS ICD Issue 1.0:

  • Pages. CRC-24Q over the 462 reserved + HAS page bits; dummy pages (0xAF3BC3) are skipped. HASS 0 (test) and 1 (operational) pages are used, HASS 3 ("don't use") discards every received message and truncates the held corrections at that time.
  • High Parity Vertical Reed-Solomon. Pages are collected per message ID; any MS distinct page IDs recover the message by inverting the k x k sub-matrix of the RS(255,32) generator matrix over GF(256) (primitive polynomial 0x11D). The generator matrix of ICD Annex B is derived from the generator polynomial (checked against the Annex B file and the Annex C decoding example in the unit tests). Later pages of a decoded message are checked by re-encoding, so a reused message ID starts a new collection.
  • MT1. Mask, orbit, clock full-set, clock subset, code-bias and phase-bias blocks, with Mask ID / IOD Set ID caches: a clock-only message is paired with the orbit block of the same Mask ID and IOD Set ID. DCC "not available" and "do not use" values end the satellite's held correction.
  • Time and validity. TOH is resolved to GST (= GPST seconds of week) from the page reception time (ICD Eq. 28 / 29). Each orbit + clock pair becomes a held update from the later of the two reference times until the next update of the satellite, at most until the end of the shorter validity interval (the recorded streams use 300 s for orbits and biases, 60 s for clocks).
  • Conventions. HAS orbit corrections are added to the broadcast position (ICD Eq. 22), so they are stored negated in the RTCM-convention SSR container; the clock correction is DCC x DCM added to the broadcast clock (Eq. 23); code biases are added to the pseudoranges (Eq. 25) and are mapped from the HAS signal index (ICD Table 20) to the RTCM SSR signal IDs. GPS L2 biases follow the tracked RINEX code: C2W / C2P / C2Y use the HAS L2 P bias, C2L / C2S / C2X the L2 CL bias.
  • Coverage. HAS corrects GPS and Galileo only; with --has-pages any satellite without a valid HAS orbit / clock at the epoch (other constellations, expired validity, missing IODref ephemeris) is excluded instead of falling back to its broadcast orbit.

Validation (HAS SIS, Kamakura, 2025)

Data: the public HAS SIS samples of hirokawa/cssrlib-data (data/doy2025-046: 2025-02-15 17:00-18:00 GPST, data/doy2025-233: 2025-08-21 07:00-08:00 GPST; Septentrio mosaic-X5 with a JAVRINGANT_DM, E6-B pages *_gale6.txt, RINEX observations) and the IGS merged broadcast navigation files of those days (BRDC00WRD_S_2025{046,233}0000_01D_MN.rnx from https://igs.bkg.bund.de/root_ftp/IGS/BRDC/2025/). The reference coordinate is the cssrlib sample value (-3962108.6836, 3381309.5672, 3668678.6720) m. Kamakura is outside the HAS service area (the HAS SDD excludes 60S-60N / 90E-180E), so these numbers are indicative only.

python3 apps/gnss.py reproduce has-sis-ppp --has-sis-data-root /data/cssrlib-data/data --check

Decoder parity with cssrlib (cssr_has, hirokawa/cssrlib main, 2026-09): every message of the three recordings was decoded by both, and all values agree exactly.

Recording Pages (dummy) MT1 messages Orbit / clock / code-bias values compared Max difference
u-blox X20, Boulder CO, 2025-07-08 19:34-19:43 GPST (rtklibexplorer/GNSS_IMU drive_0708) 3214 (1772) 67 (66 usable; the first clock message precedes any mask) 2288 / 2564 / 1949 0
mosaic-X5, Kamakura, 2025-08-21 07h 34026 (14151) 432 14896 / 18620 / 12721 0
mosaic-X5, Kamakura, 2025-02-15 17h 33451 (17839) 432 13596 / 16995 / 11669 0

The HAS-corrected satellite positions and clocks at 2025-08-21 07:30:00 also agree with cssrlib's satposs() to 0.1 mm in clock and to about 2 cm in position (transmission-time differences). The MRTKLIB mosaic-G5 SBF sample (tests/data/has/has_testdata.tar.gz, Tokyo 2026-06-10, 15 min) decodes to 109 MT1 messages (3 receiver-flagged CRC failures); it carries no RINEX observations, so it is used for decoding only.

Static and kinematic float PPP (--has-pages, IGS BRDC navigation, default 15 degree elevation mask; 2026-10-01, after the solid-earth-tide frame fix described below and the one-measurement-update-per-epoch fix of the static filter), H / U error at the given time after the first epoch; convergence as in the IDD table below:

Run 10 min 20 min 30 min 60 min Converged H < 0.20 m Converged |U| < 0.40 m
2025-02-15 17h, libgnss++ SIS, static 0.473 / +0.257 m 0.102 / +0.075 m 0.031 / +0.126 m 0.184 / -0.123 m 58.7 min 4.0 min
2025-02-15 17h, libgnss++ SIS, kinematic 0.521 / +0.290 m 0.117 / +0.123 m 0.052 / +0.250 m 0.184 / -0.256 m 59.4 min 5.2 min
2025-02-15 17h, cssrlib SIS, static (10 deg, E29 excluded, igs20.atx) 0.531 / +0.450 m 0.220 / -0.033 m 0.077 / +0.193 m 0.153 / -0.065 m 48.4 min 45.1 min
2025-02-15 17h, libgnss++ broadcast only, GPS + Galileo 0.454 / -0.145 m 0.122 / +0.214 m 0.192 / +0.086 m 0.319 / +0.123 m never 9.3 min
2025-08-21 07h, libgnss++ SIS, static 0.849 / -0.317 m 0.641 / +0.014 m 0.668 / -0.066 m 0.520 / -0.400 m never never
2025-08-21 07h, libgnss++ SIS, kinematic 0.404 / +0.314 m 0.368 / +0.504 m 0.385 / +0.322 m 0.372 / -0.540 m never never
2025-08-21 07h, cssrlib SIS, static (10 deg, L1 C/A + L2 CL) 0.303 / -0.663 m 0.112 / -0.501 m 0.079 / -0.186 m 0.079 / +0.212 m 18.1 min 22.2 min
2025-08-21 07h, cssrlib SIS, static (15 deg, L1 C/A + L2 W) 0.381 / -1.214 m 0.117 / -0.629 m 0.139 / -0.607 m 0.084 / +0.563 m 18.4 min never
2025-08-21 07h, libgnss++ JPL GDGPS RTCM SSR, GPS + Galileo 1.050 / +0.886 m 0.719 / -0.007 m 0.640 / +0.092 m 0.524 / -0.113 m never 17.8 min
2025-08-21 07h, libgnss++ broadcast only, GPS + Galileo 0.874 / +0.774 m 0.710 / +0.069 m 0.699 / +0.067 m 0.651 / -0.246 m never 17.8 min
For reference: OBE4 (Germany) HAS IDD, static (table below) 0.056 / -0.147 m 0.067 / -0.170 m 0.092 / +0.049 m 0.014 / +0.050 m 0.8 min 5.8 min

The lane gates the 2025-02-15 static run (H <= 0.20 m and |U| <= 0.40 m at 30 and 60 min) and the decoder (all 432 MT1 messages of each hour, no CRC failure), and reports the other rows. The JPL and broadcast-only rows use the observation file limited to GPS and Galileo.

Static filter: one measurement update per epoch (2026-10-01). The static rows above used to re-apply each epoch's rows eight times with the geometry frozen at the prior position (see the igs-final-ppp lane). With one update per epoch the static runs follow the kinematic ones more closely: the 2025-02-15 SIS static run ends at 0.184 / -0.123 m instead of 0.010 / +0.041 m (the kinematic run ends at 0.184 / -0.256 m; H at 60 min is still inside the 0.20 m gate), and on 2025-08-21 every static source ends higher (SIS -0.732 -> -0.400 m, JPL -0.809 -> -0.113 m, broadcast -0.655 -> -0.246 m) and 0.06-0.22 m worse horizontally.

2025-08-21 vertical offset: solid-earth tide. Before the fix libgnss++ ended this hour about 1 m low whatever the correction source (HAS SIS static -1.165 m, kinematic -0.959 m, JPL GDGPS -1.235 m, broadcast only -1.081 m at 60 min), while cssrlib ends it at +0.21 / +0.56 m. The main cause was the IERS 2010 (Dehant) solid-earth tide, the default --use-iers-solid-tide model: it was fed the SOFA Sun and Moon in the ICRS, while the routine needs them in the station's Earth-fixed frame, so the Earth's rotation dropped out of the station-body geometry. The modelled radial tide was a nearly constant +0.28 m over that hour instead of -0.10 to -0.14 m (cssrlib tidedisp / tidedispIERS2010, ports of RTKLIB tide_solid(), and the libgnss++ Step-1 model --no-iers-solid-tide agree on the latter), a 0.4 m error in up; on 2025-02-15 17h the true tide was +0.10 to +0.14 m and the error smaller. The Sun and Moon are now rotated to ITRS (icrsToItrs, IAU 2006/2000A, with EOP when --eop-c04 is given). Every 2025-08-21 row moves up by 0.42-0.44 m at 60 min, and the 2025-02-15 static run ends at +0.04 m. The CLAS lane (own tide model) and the per-frequency MADOCA profiles (Step-1 model) do not use this path; the static ionosphere-free MADOCA ppp profile does.

Before the static one-update fix the hour still ended 0.5-0.7 m low; after it the static runs end 0.11-0.40 m low. What was checked (before that fix):

  • Reference coordinate. RTKLIB demo5 b34k PPP-static with IGS final orbits and clocks (GPS only, igs20.atx) ends the hour at -0.21 m (10 degree mask) and -0.28 m (15 degree), the 2025-02-15 hour at -0.02 / -0.20 m.
  • Troposphere. The zenith delay RTKLIB estimates with the position held at the reference is 2.59 m (2025-02-15: 2.41 m). The libgnss++ a priori (UNB3m-style climatology) is 2.395 m on 2025-08-21, 0.2 m short for this humid afternoon; the filter estimate reaches 2.67 m after 45 min. With the zenith delay held at 2.59 m the static run still ended at -0.75 m (before the tide fix), so the troposphere is not the main driver.
  • Model or estimator. A batch least-squares float solution on the libgnss++ corrected observables (clocks per epoch and system, one ambiguity per arc) gives the same up offset as the filter (-1.05 to -1.19 m before the tide fix), so the remaining offset is in the observables of that hour, not in the filter. Code residuals at the reference differ by satellite by up to +/-0.4 m (G28 -0.9 m) with 11-13 satellites in view; GPS-only and Galileo-only runs are both about 1 m low (before the tide fix).
  • Satellite set. The IGS merged BRDC file used by the lane lacks the G12 ephemeris with IODE 8 (uploaded at 07:59:44) that the HAS corrections refer to, so libgnss++ excludes G12 (51-64 degrees elevation) for the whole hour; cssrlib takes it from the receiver's RINEX navigation file. A navigation file that carries it (BRDC00IGS_R) did not help (-1.67 m before the tide fix).
  • Other terms. Receiver ANTEX (--antex igs20.atx, -0.02 m), a 10 degree mask (-1.09 m before and -0.65 m after the fix) and a Galileo clock offset state (horizontal 0.52 -> 0.13 m, up worse) do not explain it; the HAS code biases of each satellite nearly cancel in the ionosphere-free combination (flipping their sign moves the result by 1 cm).

Smoke test on the u-blox X20 drive (rtklibexplorer/GNSS_IMU drive_0708, Boulder CO, inside the service area, 9 min, RXM-RAWX converted to RINEX, IGS BRDC navigation): kinematic float PPP runs on all 549 1-Hz epochs with the 2850 HAS updates decoded from the same UBX file. The RTK track shipped with the data set is relative to a base of unknown absolute coordinate (constant offset E +6.0 / N +4.7 / U -23.0 m for both HAS and broadcast PPP), so it cannot validate absolute accuracy; after removing that offset the HAS kinematic track scatters 0.46 m (H RMS) / 1.25 m (U RMS) after 5 min against 0.31 m / 1.33 m broadcast only, as expected for an unconverged 9-minute run.

Validation (OBE4, 2023-08-17 02:00-03:00 GPST)

Data: the public HAS IDD sample in hirokawa/cssrlib-data data/doy2023-229/ (OBE42023229c.obs, OBE42023229c.nav, idd2023229c.rtc). Its license is not stated, so the files are not redistributed with libgnss++; download them and point --has-data-root / GNSSPP_HAS_DATA_ROOT at that directory. The reference coordinate is the cssrlib sample value (4186704.2262, 834903.7677, 4723664.9337) m (OBE4, Septentrio AsteRx4 with a SEPCHOKE_B3E6 antenna).

python3 apps/gnss.py reproduce has-idd-ppp --has-data-root /data/cssrlib-data/data/doy2023-229 --check

Lane result (2026-10-01, after the solid-earth-tide frame fix and the static one-update-per-epoch fix, MSVC Release, elapsed time from the first epoch 01:59:12 GPST; the historical rows keep the values of their time):

Run H / U at 10 min 20 min 30 min 60 min Converged H < 0.20 m Converged |U| < 0.40 m
has-idd, static 0.056 / -0.147 m 0.067 / -0.170 m 0.092 / +0.049 m 0.014 / +0.050 m 0.8 min 5.8 min
has-idd, static, before the one-update fix 0.018 / -0.503 m 0.057 / -0.432 m 0.072 / -0.090 m 0.097 / -0.136 m 4.2 min 21.5 min
legacy conversion of the same stream, static 0.517 / -0.120 m 0.207 / +0.072 m 0.273 / +0.551 m 0.105 / +0.284 m 57.8 min 58.7 min
has-idd, kinematic 0.033 / -0.201 m 0.064 / -0.128 m 0.057 / +0.018 m 0.114 / +0.077 m 59.6 min 6.1 min
has-idd, kinematic, before the residual screening 0.035 / -0.195 m 0.064 / -0.127 m 0.136 / +0.058 m 0.018 / -0.390 m 45.0 min never (-0.44 m at the last epoch)
has-idd, kinematic, before the one-update fix 3.03 / -5.95 m 2.47 / -5.13 m 2.20 / -5.71 m 1.87 / -5.61 m never never

The lane gates the static run (H at 60 min <= 0.20 m, |U| at 60 min <= 0.40 m, horizontal convergence <= 10 min, vertical convergence <= 30 min) and the kinematic run (H <= 0.30 m and |U| <= 0.60 m at 30 and 60 min), and reports the other rows. With the kinematic post-fit residual screening (see Broadcast kinematic PPP) the kinematic run stays within 0.21 m horizontally and 0.24 m vertically after the first 10 minutes (RMS 0.086 m / 0.099 m, against 0.156 m / 0.206 m and peaks of 0.27 m / 0.48 m before); it crosses 0.20 m horizontally once more just before the end of the hour, hence the late horizontal convergence time.

Comparison with cssrlib (main, samples/test_ppprtcm.py case 1, which processes the same files from 02:00:00 GPST with igs20.atx). libgnss++ was run on the observation file cut to start at 02:00:00 as well (2026-09-29, before the solid-earth-tide frame fix, which moves the lane rows above by at most 2 cm):

Solver (static float PPP, from 02:00:00) 10 min 20 min 30 min 60 min Converged H < 0.20 m Converged |U| < 0.40 m
libgnss++ has-idd 0.036 / -0.460 m 0.076 / -0.407 m 0.081 / -0.061 m 0.105 / -0.137 m 3.8 min 20.3 min
libgnss++ has-idd + --antex igs20.atx 0.037 / -0.568 m 0.077 / -0.515 m 0.081 / -0.169 m 0.105 / -0.244 m 3.8 min 27.9 min
cssrlib (static) 0.214 / -0.487 m 0.162 / -0.215 m 0.140 / -0.100 m 0.152 / -0.300 m 11.0 min 12.9 min
libgnss++ legacy (develop behaviour) 0.306 / -0.606 m 0.225 / -0.248 m 0.262 / +0.253 m 0.148 / +0.060 m 55.1 min 15.4 min

Both solvers end the hour at 0.10-0.15 m horizontal and 0.14-0.30 m vertical error. cssrlib's kinematic mode in that sample uses a 0.01 m/sqrt(s) position random walk and gives the same numbers as its static mode. The libgnss++ kinematic filter re-seeds the position from SPP every epoch (white-noise position, no motion prior); from 02:00:00 it gives 0.051 / -0.183, 0.072 / -0.143, 0.064 / +0.009 and 0.099 / +0.114 m at 10 / 20 / 30 / 60 min (with --antex igs20.atx: 0.052 / -0.290, 0.072 / -0.251, 0.063 / -0.098 and 0.099 / +0.007 m).

H is the horizontal error and U the up error at the given time after the first epoch; convergence is the first time after which H stays below 0.20 m (resp. |U| below 0.40 m) until the end of the hour. No receiver ANTEX is applied; with --antex igs20.atx the up error shifts by about -0.1 m.

Known limitations

  • SIS: service area. Both public SIS hours are from Japan, outside the HAS service area; the in-area X20 recording is 9 minutes of driving with an RTK track of unknown absolute datum. There is no in-area static SIS validation yet.
  • SIS: navigation data. The RINEX 4 navigation file shipped with the 2025-02-15 cssrlib-data hour is not usable by the libgnss++ RINEX reader (every satellite is rejected with ~100 km residuals, also without HAS), and the RINEX navigation files of both hours hold only the ephemerides received during the hour, so earlier IODrefs cannot be matched. The lane uses the IGS merged BRDC files instead.
  • SIS: phase biases are decoded and dumped but not applied (float PPP).
  • SIS: SBF observations. SBF input is used for the HAS pages only; the observations must be converted to RINEX separately.

  • IDD profile unchanged. The has-idd profile keeps its stage-A behaviour: GPS L2 code biases are chosen by the coarse signal (L2C) even for C2W observations, and satellites without corrections keep their broadcast orbit. --has-pages does both the tracked-code selection and the exclusion.

  • Kinematic PPP (fixed). Until the one-update fix, --kinematic settled about 2 m horizontal / 5.6 m vertical off on this hour with HAS, legacy SSR and broadcast-only input alike. The PPP filter re-applied the same epoch's measurement update up to eight times while the observation geometry stayed at the prior (SPP-seeded) position, so every extra pass pushed the position again by the innovation it had already absorbed and the troposphere and float ambiguities soaked up the difference. Kinematic PPP now commits one update per epoch, as RTKLIB / MADOCALIB do, and since 2026-10-01 so do static and --low-dynamics runs (only the coherent MADOCA static ionosphere-free profile keeps the historical passes); at start-up the repeated push had moved static solutions tens to hundreds of metres. See the igs-final-ppp lane. Kinematic PPP also screens its post-fit residuals (w-test, 4 sigma); see Broadcast kinematic PPP.

  • Galileo inter-system bias. Galileo shares the GPS receiver clock unless GNSS_PPP_ESTIMATE_ISB=gal is set (as on the other non-MADOCA PPP paths). On this sample Galileo code residuals sit about 2 m below GPS; estimating the ISB did not improve the one-hour static result.
  • Satellite antenna frequency dependency. Corrections are applied at the broadcast antenna phase centre without the per-signal satellite PCO / PCV differences cssrlib applies (centimetre level).
  • One-hour, one-station validation. The public IDD sample is a single hour at one station; there is no multi-day validation yet.