Reproducible testing for embedded BCIs: a demultiplexing PCB and acquisition system for EEG signal emulation

Abstract

Validating machine learning models for Brain-Computer Interfaces (BCIs) on resource-constrained edge devices is challenging, as traditional methods rely on costly EEG equipment or simulations that fail to capture real-world electronic characteristics. To bridge this gap, we introduce the DEEGMUX, a low-cost, open-source hardware system for high-fidelity, hardware-in-the-loop (HIL) testing of EEG classification algorithms. The system comprises an EEG Demultiplexer Board that converts a multiplexed EEG signal into 8 parallel channels, and an EEG Acquisition and Processing Board featuring an ADS1299 24-bit ADC interfaced with an Arduino Nano 33 BLE. This setup enables the use of real EEG datasets, such as the PhysioNet Motor Imagery dataset, to generate precisely timed electronic signals. Characterization demonstrated high signal fidelity, with a Mean Squared Error of 1.7·10⁻¹⁰ V² and a Signal-to-Noise Ratio of 16 dB relative to the original digital data. Furthermore, an EEGNet motor imagery classifier evaluated on hardware-acquired signals showed a negligible accuracy difference of (−0.3 ± 5)% compared to evaluation on the original data, confirming that the emulation chain preserves classification-relevant features. The DEEGMUX provides a scalable, reproducible, and affordable platform for rigorously testing edge-deployed CNN models against realistic electronic inputs, accelerating the transition from simulation to robust real-world BCI deployment.