A piezoresistive fingertip sensor does not "see" a texture — it feels a 1-D vibration trace as ridges pass beneath it. As the tip slides at speed v over a surface whose height profile repeats every spacing S, each ridge crossing produces one vibration cycle, so the dominant frequency in the signal is f = v / S. Faster sliding or finer ridges both push the frequency up — exactly like a fingernail scraping a comb.
- Periodic surfaces (fabric weave, ridges) produce a clean spike at one frequency — easy to match.
- Random rough surfaces spread their vibration energy across a band of nearby frequencies — harder to pin to one value, mirroring how real sandpaper-like textures are classified with more uncertainty.
- Smooth surfaces generate almost no vibration at all — the classifier falls back to "no vibration ⇒ smooth" rather than any frequency match.
The classifier here is a tiny matched filter bank: for each candidate texture it predicts the frequency that texture's spacing would produce at the current slide speed, correlates the live signal against that frequency, and normalizes the five scores into a confidence read-out — the same principle behind real tactile texture-recognition systems, which train a classifier on vibration spectra from known material swatches.