Real-Time EEG Processing on Embedded Hardware for Brain–Computer Interfaces

Electroencephalography (EEG) is one of the most widely used signal acquisition modalities in Brain–Computer Interface (BCI) systems due to its non-invasive nature, high temporal resolution, and relatively low cost. Among BCI paradigms, Motor Imagery (MI) has attracted considerable attention because it enables users to generate control commands by imagining limb movements without requiring actual muscular activity. However, most EEG-based MI classification systems are typically executed on desktop computers or high-performance computing platforms, limiting their applicability in portable and wearable BCI solutions where local processing, low latency, and reduced power consumption are essential.

This work presents the development and implementation of an embedded EEG classification system for real-time MI recognition using data acquired with the Hybrid Black Unicorn headset. The proposed platform integrates signal acquisition, preprocessing, feature extraction, and machine learning-based classification within a single embedded device. The classification model is deployed locally and optimized to operate under the computational and memory constraints of the target hardware while maintaining real-time performance.

The initial stage of the study focuses on the classification of upper-limb versus lower-limb motor imagery (Arms vs. Legs), exploiting the somatotopic organization of the sensorimotor cortex. In subsequent stages, the system will be extended to discriminate between left-hand and right-hand motor imagery. Particular attention is given to sensorimotor electrodes around C3, Cz, and C4, which cover the primary motor cortex and exhibit characteristic modulations in the μ (8–13 Hz) and β (13–30 Hz) frequency bands during motor imagery. Additional neighboring electrodes may also be incorporated to improve spatial representation and classification performance.

The study evaluates the execution of the complete classification pipeline on embedded hardware through metrics including classification accuracy, inference latency and memory consumption. These measures provide insight into the feasibility of deploying MI-based BCI applications on resource-constrained platforms without relying on external computing resources.