Reducing False Activations in a Simulated Sip-and-Puff Controller: A Deterministic Test-Bench Study of Smoothing, Hysteresis and Minimum-Duration Filtering
Year: 2026
Type: preprint
Abstract: Sip-and-puff interfaces allow people with limited hand function to operate computers and wheelchairs by breathing into a tube, but coughs, speech and sensor noise can produce commands that the user did not intend. This study quantifies how far three standard signal-processing refinements reduce such false activations in a four-command sip-and-puff detector, and what each costs in command accuracy and response time. A controller for the Arduino Uno was implemented with a software emulation of an MPXV7007-class differential pressure sensor and evaluated on a deterministic synthetic test bench that presented every detector configuration with identical, seeded input: 40 balanced target commands per session under three scenarios of increasing difficulty, with injected cough and speech events. A baseline amplitude-threshold detector was compared with cumulative configurations that added a moving-average filter, hysteresis and a minimum-duration gate, one change at a time, against success criteria fixed in advance. In the main scenario the baseline produced 25 false activations per session at 100% exact-match accuracy. The full configuration (30-sample moving average, release at half the activation threshold, 450 ms minimum duration) reduced this to 5 false activations, an 80% reduction, with no loss of accuracy and a mean release-to-output latency of 284 ms. It met the accuracy and latency criteria but not the criterion of at most three false activations per session. All five remaining errors were coughs whose detected duration had been lengthened by the smoothing stage, revealing an interaction between filters that a one-at-a-time evaluation would otherwise conceal. Because all inputs are synthetic, the results characterise the algorithms on the stated input model rather than clinical performance.
Author: Sahoo, Poonam
Language: English
Subfield: Human-Computer Interaction
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225008902