HydratEar: Non-Invasive Hydration Monitoring using In-Ear Acoustic Reflectometry

Monitoring hydration is critical for maintaining physical, cognitive, and physiological performance, yet current monitoring still largely relies on invasive laboratory assays or cumbersome non-pervasive methods such as bioimpedance straps, sweat patches, or intake logs. The ear canal offers a unique opportunity for unobtrusive monitoring through acoustic reflectometry, leveraging its stable anatomy and acoustic properties. In this work, we propose HydratEar, an earable-based system that transmits FMCW chirps through in-ear speakers and records echoes with embedded microphones to infer hydration status between two states (hydrated vs mildly dehydrated). By extracting acoustic biomarkers such as echo decay and resonance characteristics, and modeling them with temporal convolutional networks (TCN), the system classifies hydration state in 8 controlled but diverse lab and outdoor environments (quiet/noisy/hot/cold/low or high humidity/windy/post-exercise). A 21-participant study showed that HydratEar attains a macro-F1 of 0.88 in within-subject evaluation and 0.83 under strict leave-one-subject-out (LOSO) generalization. Under the same protocol, HydratEar improves F1 by ≈ 10% over the vision-based IR Noir imaging systems, while remaining hands-free and lighting-invariant, and runs in real time on our prototype with an end-to-end latency of ≈68 ms per probe. This study highlights the feasibility of earable-based acoustic probing as the first in-situ hydration monitoring system, paving the way for continuous and user-friendly hydration tracking in daily life.