We are testing whether a small model aircraft's dynamics can be identified from short flight records, to see if a scale model can stand in for a full-size design. The airspeed channel looked like the weak point: a 1 psi pitot sensor with a 1% error band can have an offset bigger than the whole velocity perturbation.
A constant pressure offset has two effects on the indicated speed: a constant shift and a gain slightly below 1 (0.97 for the 0.25% accuracy figure, 0.89 for 1%). We ran the fit with both, on 20 designs, 10% noise and records of one and two phugoid periods.
- The gain does nothing to the fitted modes. Scaling one channel is a change of state coordinates, which leaves the eigenvalues alone. Median short-period error with a fitted bias stayed at 0.64 to 0.83%, within 0.1 points of the offset-only run (one exception: 8 of 160 fits failed at two periods and the 1% gain, against none before).
- The offset is the problem. A fit that ignores it returned no usable modes in 160 of 160 draws at the 1% size; one extra fitted bias per channel fixed that.
Details, caveats and the 160-draw tables are in the project log (milestone 9ad).