CLAS DD PPP-RTK Filter A2

A2 adds gated DD-native LAMBDA conditioning to the A1 double-differenced CLAS scaffold and removes the A1 static position anchor pseudo-measurement. The path remains behind GNSS_PPP_CLAS_DD_FILTER=1; the default CLAS and MADOCA paths do not call this code.

Implemented

• DdFilterScaffold::processFloatUpdate() now attempts integer fixing after a successful DD float update.
• The ambiguity vector is built from the same phase rows selected by buildDdMeasurementSystem(), using StateLayout::ambiguityIndex() for the reference and target satellites in each DD group.
• DD ambiguity covariance and ambiguity-vs-head covariance are extracted through rtk_measurement::buildAmbiguityTransform() on the DD scaffold state and covariance. The old PPP ambiguity maps in ppp_ar.cpp and ppp_wlnl.cpp are not used.
• The full DD candidate vector is attempted first. If the full vector does not pass the ratio gate, A2 attempts the highest available per-frequency DD subset from the same row builder groups. This is a minimal PAR bridge for CLASLIB's posopt7=on oracle configuration; post-fit fixed validation and hold policy remain A3.
• On acceptance, the DD head state is conditioned with xa = x - Qab * Qb^-1 * (b_float - b_fixed), the head covariance is downdated with Pa = Paa - Qab * Qb^-1 * Qab', and compatible single-satellite ambiguity values are restored into the fixed snapshot.
• Accepted fixed snapshots are fed back into the DD scaffold so subsequent epochs carry the accepted ambiguity datum. No hold pseudo-measurements are added in A2.
• The A1 anchor row was removed: no appendStaticPositionAnchor(), no anchor variance constant, and no static distance fallback guard remain. The only stabilization retained is the initial DD position covariance (0.003 m^2) used when seeding the DD state from the first native float.

Anchor Removal Measurement

2019-08-27 CLAS, same-epoch /tmp/s31_ref/claslib_oracle_xyz_full.pos oracle, --ar-ratio-threshold 2.0, 3599 matched epochs.

Path	Fixed epochs	Fixed 3D mean / p95	Fixed Up mean / p95	All 3D mean / p95	LAMBDA accepted / rejected	Mean attempted ratio
A1 anchor, no A2 LAMBDA	0	n/a	n/a	1.721532 / 3.335225 m	0 / 0	n/a
A2 with A1 anchor still present	169	2.138678 / 3.870272 m	1.785294 / 2.749376 m	1.721540 / 3.335829 m	169 / 2420	1.450616
A2 final, anchor removed	1073	1.342069 / 2.868478 m	0.709820 / 1.918809 m	1.161354 / 2.914994 m	1073 / 1576	84.636654

The final no-anchor path fixes more epochs than the native baseline fixed count (301) and improves the A1 float oracle mean/p95 3D and Up errors, but it does not yet reach CLASLIB/native fixed accuracy. That remaining gap is the A3 hold/validation target.

A3 Entry Points

• DdFilterScaffold::conditionFixedSnapshot(): add CLASLIB-style post-fix residual validation around the accepted fixed_snapshot_ before feedback.
• DdFilterScaffold::processFloatUpdate(): add holdamb-style pseudo-row feedback only after min-fix, ratio, PDOP, and postfit gates pass.
• buildDdMeasurementSystem(): preserve reference-satellite continuity and expose reference changes so A3 can reset or withhold ambiguity feedback on unstable groups.
• DdUpdateDiagnostics: surface accepted/rejected reasons and subset type if A3 needs per-epoch validation telemetry.