A High SNR, Low-latency Dry EMG Acquisition System for Unobtrusive HMI Devices

Gel-based electrodes are at the core of clinical and HMI applications, given their excellent signal quality. On the other hand, their intrusiveness, preparation times, and non-reusability severely limit their usage in consumer HMIs, most notably when scaling up the number of EMG channels. Dry-electrode EMG systems address the intrusiveness issues of wet electrodes but generally produce noisier signals, resulting in less reliable HMIs. This work introduces a low-cost, zero-preparation, highly dense (16 single-ended channels), unobtrusive dry EMG bracelet system that offers a competitive signal-to-noise ratio and achieves 95% classification accuracy on eight hand gestures. The dry electrodes are coupled with BioWolf16, a small (39x43mm), wireless, low-power, high-sample-rate (up to 4ksps) HMI device for ExG signals that also features a computational engine (Mr Wolf SoC from PULP) for real-time embedded signal processing. The system was validated in the HMI context to control a nano-drone, with the complete processing chain running on the embedded device, demonstrating the system's robustness while achieving very-low latency (~25ms) and long battery life (~12h) thanks to embedded computation.