Software interpolation cannot fully replace a hardware report rate

What it is

Report rate is set by the touch IC’s scan, filtering, and bus. Process-level interpolation, Bézier smoothing, or prediction cannot turn unobserved in-between positions into measurements. Software can make a polyline look rounder and can guess a point between frames, but the guess has no matching capacitive reading. Treating interpolation as a delivered higher report rate breaks on hard stops, reversals, and gesture classification.

Why it happens

Interpolation assumes smooth motion between samples—usually constant velocity or acceleration. A real finger can brake, corner, or plant a second finger in that gap. The interpolator still fills from the old model, inventing a trail in empty time. Smoothing filters also delay observable turns, dragging an already sparse signal by extra milliseconds. If a recognizer computes speed or curvature from those invented points, thresholds fire or miss on fake motion. More fundamentally, interpolation adds no information: in the Shannon sense the high-frequency content between two 60 Hz samples is already gone, and a denser curve is only a redraw of the baseband.

Studying it

On one panel compare three paths: the raw sparse points, a cubic-spline interpolation, and raw points from a higher hardware report rate. Tasks are a stop-and-reverse, a fast signature, and a second finger landing abruptly. Independent variables are interpolation order, prediction horizon, and whether interpolated points are labeled as “real events” into the recognizer. Dependent measures are stop overshoot, turn delay, false velocity on the first pinch frame, and recognizer false fires. If interpolation is display-only, report “looks rounder” and “classifies better” separately; they often diverge.

Where it stops holding

For decorative strokes, replay, and screenshots, interpolation is entirely appropriate; nobody needs those points to be measurements. In a slow target drag the gap is nearly linear and interpolation error is negligible. Hardware filtering already running on the IC plus another application layer double-lags the signal. In cloud gaming or screen mirroring the network is the sampling bottleneck, and local interpolation fills a transport hole, not a gap on glass—that is a different patch. An accessibility magnifier that hit-tests interpolated coordinates will activate a control the user never touched at a corner.

Applying it

• Feed gesture classification and hit-testing only real reports; use interpolated points for stroking, and paint them a different debug color.
• Write a minimum report rate into the hardware spec as a constraint software cannot substitute, especially for signing, games, and fling scrolling.
• Accept on stops and reversals: if a smoothed trail keeps traveling after the finger has halted, interpolation or prediction is being treated as input fact.

Related

• Same group: C2.17.1 Touch report rate is the sensor's position samples per second · C2.17.2 Super-refresh report rates feed predictive rendering with earlier samples · C2.17.3 Undersampled fast swipes look lagged and polygonal
• Nearby: C2.10 Touch latency and directness · C3.29 Gesture velocity thresholds
• Search terms: interpolation · report rate · Nyquist