The aerolog-scaling programme asks what a small Froude-scaled model of a cargo UAV can tell us about the full-size aircraft. To read the model's phugoid and short-period modes from a flight record you need airspeed, and the airspeed sensor is the weak channel.
Sized from a 1 psi differential pressure part (MS4525DO datasheet figures, span assumed): at the model's trim dynamic pressure of 272 Pa the sensor's offset error is 0.37 (0.25% accuracy) to 1.48 (1% total error band) times the rms of the whole velocity perturbation. In simulation over 20 designs and 8 draws each:
- plain output-error fitting rises from 0.78% to 2.76% median short-period error at offset 0.37, and fails in 160 of 160 draws at 1.48;
- the same fit with one fitted bias per channel stays near 0.7 to 0.8%;
- a speed-proportional error (a gain plus an offset) changes nothing a bias does not remove. One cost: 8 of 160 fits failed at the 1% gain with two periods of data, which I read as lower signal-to-noise and did not check further.
So: fit a bias on the airspeed channel. Limits are in the paper draft, Section 4.10c: constant offset, one scale, 8 draws, no repeat seed.
Source: research/aerolog-scaling/ (LOG.md, milestones 9aa to 9ad).