MRTKLIB v0.5.1 Release Notes¶

Release date: 2026-03-10 Type: Bug fix (dual-channel CLAS real-time performance) Branch: 35-bug-rtkrcv-notable-performance-degradation-in-dual-channel-clas-realtime-positioning

Overview¶

v0.5.1 resolves a significant performance degradation in dual-channel CLAS real-time PPP-RTK positioning. The fix rate improves from 67% to 93% (RT) and from 87% to 99% (PP).

Root Cause¶

The rtkrcv_2ch.toml configuration used frequency = "l1+2+3" (nf=3), but CLAS does not provide Galileo E5a (L5) phase bias corrections. This caused:

1. pbias[2] (L5) remained CLAS_CSSRINVALID for all Galileo satellites
2. The L1-L5 geometry-free cycle slip detector (detslp_gf() in mrtk_ppp_rtk.c) fired false slips at every epoch
3. Galileo ambiguities were destroyed repeatedly, preventing AR convergence

Fix¶

Set frequency = "l1+2" (nf=2) to match the bias corrections actually available from CLAS. With nf=2, the L1-L5 GF slip detector is skipped entirely (if (rtk->opt.nf <= 2) return), and Galileo ambiguities converge normally.

Performance Improvement¶

Dual-channel CLAS (2025/157 dataset)¶

Single-channel CLAS (2019/239 dataset) — unchanged¶

Changes¶

Fixed¶

• conf/claslib/rtkrcv_2ch.toml — Changed frequency from "l1+2+3" to "l1+2". CLAS does not provide Galileo E5a bias corrections; using nf=3 caused false L1-L5 geometry-free cycle slip detection, destroying ambiguities and degrading AR fix rate by ~25 percentage points (RT) and ~12 pp (PP).

Improved¶

• scripts/tools/gen_l6_tag.py — Two fixes:
• tick_scale: now always matches the master tag's timing scale, not only when the master is detected as "compressed".
• UTC→GPST time basis: L6 tag time_time now includes GPS-UTC leap seconds (+18 s) to match the GPST basis used by gen_bnx_tag.py and RTKLIB's strsync(). Previously the ~18 s mismatch caused L6 data to replay early relative to observations.

Updated¶

• tests/data/claslib/claslib_testdata.tar.gz — Updated ref_L6_2CH_RT.nmea reference to reflect the improved fix rate with nf=2 configuration.

Test Results¶

All 59 tests pass (56 non-RT + 3 RT), unchanged from v0.5.0.

Technical Details¶

Why nf=3 worked for single-channel but not dual-channel¶

The 2019/239 single-channel dataset has fewer Galileo E5a observations, so the false GF slip detection had minimal impact. The 2025/157 dual-channel dataset has full Galileo E5a tracking, making every Galileo satellite vulnerable to the false slip issue at every epoch.

Code path analysis¶

nf=3 + CLAS (no E5a bias):
  clas_osr_corrmeas() → pbias[2] = CLAS_CSSRINVALID (no E5a bias available)
  detslp_gf() → gfmeas_L1L5() returns non-zero (L5 obs exists)
                → |g1 - g0| > thresslip → slip flagged
  ppp_rtk_pos() → ambiguity reset for Galileo → AR fails

nf=2 + CLAS:
  detslp_gf() → if (rtk->opt.nf <= 2) return → no L1-L5 GF check
  → Galileo ambiguities converge normally → AR succeeds

Files changed¶