System Overview

The sEMG Gesture Wearable is designed to capture muscle activity from the forearm and translate it into digital commands via BLE.

Hardware Architecture

Analog Front-End

The signal chain starts with the AD8232, which provides high-input impedance and excellent common-mode rejection.

• Electrodes: 316L Stainless Steel
• Filtering: Integrated high-pass and low-pass stages
• Gain: Configurable amplification for microvolt signals

Processing & Wireless

The nRF52840 handles both the ADC sampling and the BLE HID profile.

• Sampling Rate: 1kHz per channel
• Connectivity: BLE 5.0
• Power: LiPo 3.7V with LDO regulation

Firmware Logic

The firmware implements a sliding window buffer for the EMG data, which is then passed to the ML inference engine.

ML Implementation

Using Edge Impulse, we trained a neural network to recognize five distinct hand gestures.

• Model: Dense Neural Network (DNN)
• Optimization: Quantized to INT8 for on-device execution
• Accuracy: ~94% on test set