Step-relative deadline gating for wireless smart-insole biofeedback

Abstract

A wireless gait-feedback command may remain technically valid after the opportunity to prepare its intended step has passed. Receiver-side deadline gating can limit such delivery, but its cost depends on cadence prediction, clock alignment and the availability of recent contacts. To quantify deadline compliance and lost cue opportunities using independently calibrated comparators, and to determine which timing assumptions most strongly affect performance. An open-loop experiment generated four bilateral contact patterns and three wireless-delay conditions. Calibration used 40 episodes per pattern; held-out evaluation used 120 episodes per pattern and link, each with 500 contacts. Nine policies were evaluated at a 150-ms preparation reserve. Comparator constants were selected on calibration data to maximize on-time coverage subject to at most 1% deadline violations among emitted cues. Paired, stratified episode bootstrap intervals quantified Monte Carlo uncertainty. Clock drift, correlated transport degradation, actuator jitter, command reordering, alternating step asymmetry and missed acquisition contacts were tested separately and jointly. The primary evaluation comprised 207,204 cue-request/link instances. Under congestion, unrestricted delivery provided 69.76% on-time coverage with 23.99% late emitted cues. A 12-interval tenth-quantile gate with a 20-ms margin provided 61.36% coverage with 0.65% late cues. Independently calibrated rolling and fixed gates provided 63.69% and 52.75% coverage, respectively, at similar late fractions of 0.93% and 0.96%. Clock offset and drift raised the reference rolling gate's late fraction to 3.28%; 5% missing acquisition contacts reduced its coverage to 31.75%. Step-relative gating offers a measurable timing advantage over a single calibrated nominal period within the evaluated models. Its benefit is conditional on clock calibration and a sufficiently complete contact history. The results establish computational timing behaviour; human preparation time, cue interpretation and clinical effectiveness remain unmeasured.