Threshold Estimation for Multi-Channel Electrotactile Stimulation Using Discharge Voltage Waveforms

Electrotactile stimulation provides compact and energy-efficient tactile feedback for wearable and XR applications, yet maintaining consistent perception across users and electrode sites remains challenging due to variations in skin impedance. This study investigates the feasibility of using voltage waveforms recorded during the discharge phase for threshold estimation in multi-channel electrotactile stimulation. Data were collected from eleven fingertip sites under both cathodic and anodic stimulation. A sub-threshold fixed-height pulse was inserted periodically to monitor skin properties without using prepulses, enabling near-simultaneous multi-channel operation. Long Short-Term Memory (LSTM) networks with attention were trained to estimate stimulation current from discharge waveforms of different lengths (1/3, 2/3, full of 75 samples). Results show that using only the first third of the discharge (25 samples$\approx 225 \mu \mathrm{s}$) achieves accuracy comparable to full sequences. Joint training across channels improved generalization, and combining polarities slightly enhanced performance (MAE$\approx$0.36-0.41 mA), suggesting that a unified model can estimate thresholds for diverse electrode sites and polarities.