An embedded implementation based on adaptive filter bank for brain-computer interface systems

作者：Belwafi Kais^*; Romain Olivier; Gannouni Sofien; Ghaffari Fakhreddine; Djemal Ridha; Ouni Bouraoui

来源：Journal of Neuroscience Methods, 2018, 305: 1-16.

DOI：10.1016/j.jneumeth.2018.04.013

• Embedded brain-computer interface (EBCI)
• Electroencephalography (EEG)
• Motor imagery
• EEG filter optimization
• Weighted overlap-acid (WOLA)
• System on programmable chip (SOPC)
• Embedded Real-time BCI

Background: Brain-computer interface (BCI) is a new communication pathway for users with neurological deficiencies. The implementation of a BCI system requires complex electroencephalography (EEG) signal processing including filtering, feature extraction and classification algorithms. Most of current BCI systems are implemented on personal computers. Therefore, there is a great interest in implementing BCI on embedded platforms to meet system specifications in terms of time response, cost effectiveness, power consumption, and accuracy.
New-method: This article presents an embedded-BCI (EBCI) system based on a Stratix-IV field programmable gate array. The proposed system relays on the weighted overlap-add (WOLA) algorithm to perform dynamic filtering of EEG-signals by analyzing the event-related desynchronization/synchronization (ERD/ERS). The EEG-signals are classified, using the linear discriminant analysis algorithm, based on their spatial features.
Results: The proposed system performs fast classification within a time delay of 0.430 s/trial, achieving an average accuracy of 76.80% according to an offline approach and 80.25% using our own recording. The estimated power consumption of the prototype is approximately 0.7 W.
Comparison-with-existing-method: Results show that the proposed EBCI system reduces the overall classification error rate for the three datasets of the BCI-competition by 5% compared to other similar implementations. Moreover, experiment shows that the proposed system maintains a high accuracy rate with a short processing time, a low power consumption, and a low cost.
Conclusions: Performing dynamic filtering of EEG-signals using WOLA increases the recognition rate of ERD/ERS patterns of motor imagery brain activity. This approach allows to develop a complete prototype of a EBCI system that achieves excellent accuracy rates.

• 出版日期2018-7-15