Physical decoding of the Applanix reference

The 14 April 2009 ATM .013 sample supports six signed, little-endian 32-bit increments: velocity X/Y/Z followed by angle X/Y/Z. Independent numerical fits support the expected scales, axis order and signs. This establishes an empirical decoder for the tested AV-510 VER5, firmware04.60, ICD15.00, IMU8, rate-code2 configuration. It does not establish every onboard correction or an independent Earth-model measurement.

The historical sample corpus remains unchanged apart from its living explanatory notes. Its original inspector treated the 24-byte payload as opaque. The new streaming parser reproduces 143,471 valid outer frames, 132,444 IMU packets, 662 fused navigation records, 5,934 primary receiver packets, and 660 checksum-valid sentences of each GGA, VTG and ZDA type. No resynchronization, checksum repair or science interpolation is used. Source SHA-256: 18c48772bcd7e83566f2301ae3689d691c1675dfc2e60c0650b6e0017b764096.

Independent scale and axis checks

All 662 navigation timestamps match an IMU timestamp exactly. Rates use the actual adjacent IMU interval. Group-1 gyro rates are converted from degrees/second to radians/second. Each gyro scale and intercept is fitted freely before comparison with 2^-18 rad/count. Accelerometer validation fits three scales and offsets with one shared gravity magnitude, using roll/pitch to express gravity in the aircraft frame. Comparing specific force directly with translational acceleration would be physically wrong. The fitted shared gravity is about 9.80553 m/s². It is a decoder cross-check, not a gravity or Earth-shape measurement.

The engineering criteria were correlation at least 0.9999 and scale agreement within 0.1%, also required in both chronological half-record fits. These are conservative interpretation checks, not statistical confidence bounds or manufacturer accuracy specifications. All 48 axis/sign mappings are tested with positive, physically plausible scales; allowing negative fitted scales would make a sign check meaningless. The expected identity mapping wins. Full reports retain intercepts, maximum and 99th-percentile residuals, fixed-binary-scale residuals, alternative mapping scores and chronological holdout results.

The timing sensitivity report also compares preceding navigation-interval averages with the endpoint fused reference. That is a diagnostic of averaging during maneuvers, rather than an alternative acceptance rule or an interpolated scientific observation.

Gyro reference offsets are nonzero: approximately 1.60, 6.12 and 14.75 degrees/hour. Their origin could involve frame conventions, navigation corrections, calibration, transport or timing. Agreement does not prove that either stream includes or removes Earth's rotation. These offsets are never subtracted from the exported IMU observations.

Evidence and remaining uncertainties

The public Applanix ICD describes Group-1 aircraft-frame dynamics and its user-entered lever-arm reference. Its Group-4 table gives the outer layout and 24 payload bytes, rather than the physical scale table for this older instrument. No proprietary manufacturer scale table has been found.

The geometry-only acquisition screen finds no qualifying one-minute level-flight window in the reference. Its newly generated bulk products were discarded after recording that result; the original source and historical corpus were preserved. This folder keeps compact reports, 662 matched validation samples and the first 1,000 raw/physical IMU rows as decoder regression evidence, selected by timestamp order rather than Earth-model fit quality.

The catalog tracks all 326 .013 URLs and rounded metadata sizes. The six-file compatibility trial used authenticated archive access.

This supports broader acquisition and screening, rather than applying IMU8 scales to every sensor. Full-catalog screening has completed with two numerical threads and per-file checkpoints: 18 retained, 58 confirmed no-level rejections, 250 unresolved. Retained files contain 22 windows totaling 7329.62 seconds across instrument streams; these are not necessarily independent flights. Two retained ATM files have supported physical decoding and 692.01 seconds of qualifying geometry. All unresolved originals remain preserved. The completion report records final counts and provenance hashes. No model fit, simulation campaign, bootstrap, calibration or production promotion has run for ILVIS0.

See the acquisition guide for commands, retention policy and recovery.

The 250 unresolved files divide into 114 with GPS-time-tagged IMU logs, requiring an explicitly validated mapping to receiver UTC, and 136 with UTC tags but insufficient valid navigation/GPS context. Sixteen of the latter have some valid context. The next engineering work is timing support and a navigation-status/context audit, followed by independent physical checks for each IMU configuration. Originals are preserved. The diagnosis records the per-file grouping without changing the completed screen or eligibility rules.