Hover height thresholds and misclassification
Aliases: hover threshold · near-field false detect · hover hysteresis
What it is
Near-field strength falls continuously with height, so the system must cut it into out-of-range / hover / contact with thresholds. If enter hover and leave hover share one number, the cursor flickers on the boundary. Too high, and a pen passing at a distance—or even the other hand—is tracked. Too low, and the preview never arrives in time. Errors are false positives (a cursor when tracking was not meant) and false negatives (aiming with no preview).
Why it happens
The field–distance curve is a steep near-field decay, not a linear ruler. The same threshold corresponds to different millimetres across pens, films, and screen edges. Contact has its own pressure or capacitive jump; hover is a near-field gate at zero force. To avoid chatter, the enter threshold is usually higher than the leave threshold, forming hysteresis: the cursor unregisters only farther away, and reappears only closer in. Electrode density and metal bezels warp the field, so firmware millimetres at centre and corner disagree. A palm, a second pen, or a charging coil raises the floor, triggering hover early or knocking it out with noise. Using hover height as a precise Z axis (scale the UI with height) amplifies that error. Hover is better treated as a presence state than as millimetre ranging.
Studying it
Lift the pen from contact to leave on a ranging jig, logging the height of state flips and flicker when the same height is crossed repeatedly. Repeat at centre, edges, and with/without film. Then run “sweep past with no intent to write” versus “aim and land”.
Independent variables: enter/leave thresholds, hysteresis width, screen location, film, nearby palm or second pen. Dependent measures: millimetres and variance of trigger height, false positives (cursor on a sweep), false negatives (no cursor while aiming), cursor flicker count.
A single centre average misses corners and film.
Where it stops holding
Resonators and electrode geometry differ by pen; one millimetre spec cannot be a hard product-line rule. Gloves, moisture, or a thick paper-like film change the decay, so factory thresholds bias. Too much hysteresis makes the cursor feel stuck after lift. Using a tall hover as an in-air pointer in a demo or game is a separate product choice and raises the cost of false detection.
Applying it
- Use paired enter/leave thresholds with hysteresis so the cursor does not flicker on the boundary.
- Calibrate centre and four edges separately; do not trust the centre millimetre alone.
- Drive preview only; do not commit, delete, or switch mode before contact.
- How to check: sweep from 3 cm out—the cursor should not appear. Aim from about 1 cm and land—the cursor should be stable before contact and vanish only beyond the leave threshold, with no flicker. Repeat at corners after changing film.