BioGAP-Ultra: Modular Wearable Biosignal Platform

• BioGAP-Ultra is a modular, open-source platform characterized by synchronized multimodal biosignal acquisition and integrated on-device edge-AI processing.
• It employs a dual-SoC architecture with energy-efficient design and flexible sensor configurations for modalities including EEG, EMG, ECG, and PPG.
• Its comprehensive software suite enables real-time visualization, configurable data streaming, and reproducible research in wearable biosensing.

BioGAP-Ultra is a modular, open-source platform specifically engineered for synchronized multimodal biosignal acquisition and on-device edge-AI processing in wearable applications. It advances the state of wearable biosensing by integrating robust hardware, flexible signal acquisition across various electrophysiological and hemodynamic modalities (EEG, EMG, ECG, PPG), high-efficiency embedded computation, real-time wireless connectivity, and a comprehensive software stack for configuration and real-time visualization. Its extensibility, energy efficiency, and reproducible open-source release position BioGAP-Ultra as a reference architecture for next-generation, real-world physiological monitoring and human-machine interface research (Frey et al., 19 Aug 2025).

1. Modular Platform Architecture

BioGAP-Ultra's hardware is centered on a highly modular mainboard supporting sensor, communication, power, and processing subsystems. The mainboard incorporates:

• A multi-rail ultra-low-power PMIC enabling precise battery management and support for multiple voltage domains.
• A dual-SoC configuration, with a Nordic nRF5340 (dedicated to Bluetooth Low Energy communication and experiment control) and a PULP-based GAP9 parallel processor designated for digital signal processing (DSP) and edge-AI workloads.
• High-density connectors for expansion boards, allowing the system to be tailored to specific sensing requirements by addition or removal of dedicated sensor PCBs (e.g., ExG, PPG, IMU).
• Physical form factors are adaptable: the system supports headbands for EEG–PPG, armbands for EMG sensing, and chestbands for ECG–PPG, with typical system footprints of 32.8 mW (EEG–PPG), 26.7 mW (EMG), and 9.3 mW (ECG–PPG).

The mainboard schematic specifies connections among PMIC, SoCs, memory (SRAM, FLASH, PSRAM), and inter-peripheral buses (I²C, SPI, GPIO). Sensor expansion boards include dual 8-channel ADS1298 24-bit analog front-ends, with configuration options for both monopolar and differential acquisition, and compact integrated PPG sensors such as MAX86150.

2. Multimodal Signal Acquisition and Processing

BioGAP-Ultra supports strictly synchronized acquisition from multiple sensor modalities:

• The ExG board employs dual ADS1298 AFEs, facilitating 16 channels of low-noise, high-resolution data acquisition for biosignals including EEG and EMG, with programmable gain, reference configuration, and sample rates from 250 SPS to 32 kSPS.
• PPG acquisition is implemented via dedicated analog front-ends on modular PCBs, enabling concurrent acquisition of cardiac-related photoplethysmograms.
• Multi-sensor expansion enables simultaneous measurement and data fusion for advanced applications such as multimodal seizure monitoring or cardiovascular event detection.

The GAP9 processor performs on-board filtering (IIR, FIR), frequency analysis (FFT), and neural network inference (CNN, RNN) in real time. For instance, the data path for SSVEP EEG processing might involve bandpass filtering: $y[n] = \sum_k a_k y[n-k] + \sum_k b_k x[n-k]$ followed by a 1024-point FFT: $$X[k] = \sum_{n=0}^{N-1} x[n] \cdot \exp(-j2\pi kn/N)$$ Neural inference (classification, regression) is executed via optimized routines leveraging the processor’s on-chip parallel RISC-V cluster and NN accelerator.

The firmware is fully modular, implementing parallel threads for each core task (sensor readout, BLE communication, power management, real-time ML inference). Synchronization approaches guarantee inter-modality temporal alignment and deterministic latency.

3. Energy Efficiency and System Performance

Energy efficiency is a primary design criterion:

• The PMIC employs SIMO buck–boost regulation and dynamic power domain control, ensuring supply rails are active only as required.
• GAP9 delivers up to 32.2 GMACs with highly competitive energy figures under parallel workload, leveraging core gating and DVFS (dynamic voltage and frequency scaling).
• Form-factor dependent power consumption is as low as 9.3 mW (chest-band ECG–PPG configuration), achieving up to 60 hours operation (150 mAh battery) and 17 hours for more intensive applications (EEG–PPG headband).
• Wireless connectivity, via the nRF5340 SoC, supports BLE throughput up to 1.4 Mbit/s—a 4× improvement over BioGAP—enabling high-bandwidth multimodal streaming and rapid experiment reconfiguration.
• The architecture’s in-situ processing model (DSP and inference at the edge, with only relevant results wirelessly transmitted) yields substantial reductions in data bandwidth and overall energy spent on communication.

4. Software Integration and Real-Time Visualization

BioGAP-Ultra is supported by a comprehensive software suite:

• Mainboard firmware runs Zephyr RTOS, dividing the system into modular, maintainable threads: state management, sensor data acquisition, signal processing, BLE advertising, and data streaming.
• The Android-based “BioApp” allows real-time data visualization, configuration, storage, and device management. User interaction includes visualization of raw and filtered streams, modality selection, parameter tuning (gain, filtering, sample rate), battery monitoring, and QC feedback (e.g., low battery, poor electrode connection).
• The software stack is structured for extensibility: additional sensors or processing routines may be easily integrated with firmware and mobile application updates.

5. Research Applications and Case Studies

BioGAP-Ultra is validated across multimodal, real-world scenarios:

• EEG–PPG Headband: 16-channel dry EEG acquisition in a unipolar configuration (with active electrodes) and concurrent PPG at the forehead. Use cases include SSVEP-based BCI, drowsiness detection, and affective computing. SSVEP case studies demonstrated reliable evoked response detection using Normalized Canonical Correlation Analysis (NCCA) on-board.
• EMG Armband/Sleeve: Textile-based, dry-electrode surface EMG for muscle activation monitoring and gesture recognition, with high synchronicity and spatial resolution.
• ECG–PPG Chestband: Partitioned sensing with chest-worn ECG and earlobe PPG for cardiovascular health monitoring, including pulse transit time estimation. This configuration is optimized for ultralow power operation.

Each form factor leverages the modularity of the platform, and data is deliverable immediately to the user or experimenter via the software suite. Flexibility supports both clinical studies (cardiovascular monitoring, epilepsy) and advanced HMI research (neuroergonomics, prosthetic control).

6. Open-Source Resource and Reproducibility

All hardware (PCB schematics, Gerbers), firmware, and software are released under a permissive open-source license and are publicly maintained (https://github.com/pulp-bio/BioGAP). Full transparency in design, implementation, and typical use cases is provided, supporting:

• Rapid replication or adaptation by external groups—including modification of hardware for additional modalities, adaptation of mobile or desktop visualization, and integration of custom signal processing or ML pipelines.
• Standardization of biosignal acquisition and processing paradigms, potentially facilitating common benchmarking and cross-institutional collaboration.
• Democratization of wearable edge-AI biosensing research, with reduced barriers for both methodological innovation and application-driven studies.

7. Conclusion and Impact

BioGAP-Ultra integrates modular hardware, energy-optimal signal acquisition, high-performance parallel edge-AI processing, and robust, configurable software support. It addresses the growing need for synchronized, multimodal physiological monitoring in wearable and mobile scenarios by enabling real-time, on-device data analysis and adaptive application deployment. The platform’s open-source philosophy is intended to accelerate innovation, standardize experimental methodology, and facilitate community-driven advancements for both academic and translational biosignal research (Frey et al., 19 Aug 2025).