Release Notes — v0.6.5¶

First official release of mrtk cssr2rtcm3 — real-time CSSR→RTCM3 converter¶

Release date: 2026-05-10 Type: Feature — first official release of a new subcommand and a new hardware integration path Branch: release/v0.6.5

Overview¶

v0.6.5 marks the first official release of the mrtk cssr2rtcm3 subcommand, previously available only on the feat/integrate-cssr2rtcm3 development branch. It converts QZSS CLAS L6D corrections into standard RTCM3 MSM7 messages on the fly, so any RTCM3-capable receiver (including those that cannot decode CLAS L6D natively) can take advantage of CLAS as a Virtual Reference Station (VRS) source.

The release also ships a Septentrio mosaic-G5 P3 hardware integration guide, a 24-hour static endurance test, and a smaller utility, mrtk l6extract, for pulling L6D / L6E frames out of SBF and UBX logs.

The 62-test CTest suite passes unchanged; this release does not modify any existing positioning-engine numerics.

Major additions¶

mrtk cssr2rtcm3 — CSSR-to-RTCM3 converter (public release)¶

Real-time converter from QZSS CLAS L6D (Compact SSR) to RTCM3 MSM7, designed to be inserted between a CLAS-capable receiver (e.g. Septentrio mosaic-G5 P3) and an arbitrary RTK rover. Functional highlights:

• MSM7 output for GPS / Galileo / QZSS (configurable via TOML).
• 1005 / 1006 RTCM3 base-station messages with auto-detected GNSS indicators and configurable mosaic-CLAS-compatible station ID.
• Septentrio SBF single-stream input (multiplexed L6D + decoded NAV + PVT) plus UBX and BINEX paths.
• Elevation-based L6D PRN auto-selection with configurable threshold (l6d_elmin, default 10°), 30-second timeout, and 5° hysteresis; l6d_prn_fixed available to lock the source PRN.
• Configurable signal remap (e.g. G2X→2W, E1X→1C, E5X→5Q, J2X→2L) for receivers that prefer specific signal codes.
• Same-frequency fallback in clas_osr_corrmeas() so CLAS bias matching does not strand satellites that broadcast a different code on the same frequency.
• PVT-time-based RTCM3 pacing (1 s grid) for stable downstream RTK consumption.
• Selectable SNR model (fixed 50 dBHz or elevation-dependent).
• Default config in conf/cssr2rtcm3.toml.

mrtk l6extract — L6D / L6E frame extractor¶

Small companion utility that reads an SBF or UBX log and writes per-PRN L6D / L6E frame files. Useful for recovering raw CLAS / MADOCA streams from receiver-side recordings.

Hardware integration: Septentrio mosaic-G5 P3¶

New documentation page, docs/hardware/cssr2rtcm3-mosaic-g5.md, walks through the full end-to-end setup of cssr2rtcm3 with a mosaic-G5 P3 evaluation kit:

• RxTools configuration (USB ports, SBF Output Support block set, QZSS L6 tracking, RTCM3 input port, Solution Sensitivity).
• Wiring of mrtk relay + mrtk cssr2rtcm3 for VRS-RTK operation.
• Alternative path: native MRTKLIB engine via mrtk run.
• Real-world setup gotchas captured from field testing (macOS cu.* vs tty.*, explicit RTCMv3 input type required).

24-hour endurance test¶

The hardware guide ships with the first published 24-hour static endurance result of the cssr2rtcm3 → mosaic-G5 chain. Headline numbers (2026-05-08 → 2026-05-09 GPST, 87,839 epochs):

A reference 24-hour run of a mosaic-CLAS receiver at a separate Kantō site on the same day reached 97.96 % Fix, indicating roughly 20 % of remaining headroom — see Known Issues below.

Robustness fixes that landed during the v0.6.5 development cycle¶

These fixes shipped progressively on feat/integrate-cssr2rtcm3 and are all bundled into this release.

• Long-running daemon stability — actualdist() now caps satellite enumeration at MAXOBS and skips non-CLAS systems (d028b0c, 7fb497f), eliminating the MAXSAT overflow / GLONASS RK4 hangs that would silently stall the converter on multi-day runs.
• No more 10-minute DGPS dips — broadcast-ephemeris IODE rollover used to leave CLAS SSR pointing at a vanished IODE for ~55 s, causing RTK to fall back to DGPS. A new nav_t.eph_prev slot keeps the prior IODE alive long enough to bridge the gap (3a8cc51).
• L5 / E5a corrections — the OSR engine now applies CLAS L5/E5a corrections instead of leaving them invalid (361b6df).
• Galileo / QZSS in MSM — Galileo and QZSS satellites are now emitted in the RTCM3 stream (21f7b1f, c2ba3b6, f5f2748, 985fc34).
• VRS base position latched on first SPP fix (4fa4d65) — eliminates base-coordinate drift that was the dominant cause of Float-only RTK on the converter side.
• osr[] per-epoch clearing in clas_ssr2osr() (0145c30) — prevents previous-epoch P / Φ values from being carried forward when no new correction arrived for a satellite.
• δBIAS discontinuity compensation (IS-QZSS-L6-005 §5.5.3.2, 2b7580e).
• Ephemeris lookup via seleph() with a 30 s rebroadcast cycle (d64893b, c2ba3b6).
• -prn deprecated in favour of l6d_elmin auto-selection (2c6e251, 64617b5, ef369a3, 16db68f).
• Same-frequency code fallback so CLAS bias matching does not strand satellites that broadcast a different code on the same frequency (d8d2f4b).
• VRS PP improvements — base position init enabled for VRS-RTK in post-processing, and an selfreqpair fallback in zdres_sat for E5a observations (b034d48, ea296db). Post-processing VRS now reaches 98.5 % Fix on the regression dataset.
• PVT-based RTCM3 pacing at the 1 s grid (e729983, 95dn58ca, 03a798d) — RTCM3 delivery rate up to ~99.5 %.
• mosaic-G5 RTCMv3 input gotcha documented after a 24h SPP-stuck diagnosis: auto does not reliably detect RTCMv3 on firmware 20250611b; the receive-side port must be set to RTCMv3 explicitly (0ea0147).

Other changes worth highlighting¶

• l6extract subcommand for offline L6D / L6E frame extraction.
• module:cssr2rtcm3 GitHub label added now that the tool is publicly released (PR #99).
• Plotting utilities under scripts/plotting/:
• parse_pvt.py — NMEA/SBF parser with rigorous ECEF→ENU.
• plot_pos.py — ENU comparison plotter with CSV save mode.
• sbf_plot.py — real-time SBF position monitor.
• Positioning-engines reference (docs/guide/positioning-engines.md): RTK vs VRS-RTK vs PPP-RTK overview.

Known limitations¶

Vertical-component dispersion (~30 s sawtooth) — #97¶

The current cssr2rtcm3 SSR extrapolation holds rate terms at zero, so each ~30 s SSR snapshot introduces a small step in the corrected satellite position/clock that propagates into the vertical component as a few-cm peak-to-peak sawtooth. Horizontal accuracy is unaffected. Estimating SSR rate by numerical differentiation between consecutive snapshots is being investigated.

Fix-rate gap vs. mosaic-CLAS reference (~20 %) — #98¶

In the 24-hour endurance test summarised above, the cssr2rtcm3 → mosaic-G5 chain achieved 72 % Fix, versus 98 % on a mosaic-CLAS reference at a different Kantō site on the same day. The gap is concentrated in two windows (h ≈ 04–05 and h ≈ 14–18 GPST), in which the tracked-satellite count and the differential-correction age are both healthy — so the cause is neither observation starvation nor a correction-link outage. Investigation continues; candidates include PRN-specific cycle-slip cascades, ionospheric activity, and CLAS-network-region transitions (#66).

Files Changed¶

Upgrade notes¶

• No positioning-engine numerics changed. Existing mrtk run, mrtk post, rnx2rtkp invocations behave bit-identically. The 62-test CTest suite passes unchanged.
• No TOML-schema break for existing modes. The new [cssr2rtcm3] options (l6d_elmin, l6d_prn_fixed, signal remap, MSM systems) are scoped to the new subcommand.
• The legacy -prn flag on cssr2rtcm3 is deprecated but still accepted; new usage should rely on l6d_elmin / l6d_prn_fixed.

Test Results¶

62/62 tests pass (no regressions).