A mobile embedded platform for high performance neural signal computation and communication

• Published in: 2015 IEEE Biomedical Circuits and Systems Conference (BioCAS) pp. 1 - 4
• Main Authors: Heelan, Christopher, Komar, Jacob, Vargas-Irwin, Carlos E., Simeral, John D., Nurmikko, Arto V.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2015
• Subjects: ALS; Decoding; embedded system; Field programmable gate arrays; FPGA; medical device; mobile; Mobile communication; multichannel; multielectrode; neural decoding; neural prosthetics; paralysis; QNX; real-time; Real-time systems; Signal processing; wireless; Wireless communication; Wireless sensor networks; Xilinx
• DOI: 10.1109/BioCAS.2015.7348356

We have designed and implemented a new compact, high performance neural signal processing system for wearable neurotechnology platforms. The low-power embedded system (referred to hereafter as ESPA) prototype wirelessly receives, records, and processes 200 channels of broadband neural data, demonstrated here in a non-human primate. The subject is implanted with intracortical micro-electrode arrays (MEAs) with a head-mounted wireless neural signal transmitting device. The fully integrated embedded system manages and processes neural data using state-of-the art all-programmable system on a Chip (SoC) technology. The utilized SoC provides substantial digital signal processing hardware in the form of a field programmable gate array (FPGA) subsystem. This programmable logic (PL) subsystem connects through high-bandwidth busses to the SoC's mobile processor system (PS) which provides SoC management and communication capabilities through the use of a real-time operating system. The performance of the system was first benchtop tested using 200 channels of 30ksps simulated neural data which underwent neural signal pre-processing, filtering, and feature extraction implemented on the PL subsystem. The raw and processed data were streamed to the PS where it underwent further processing before being communicated wirelessly to a backend server via Wi-Fi. The SoC only required an estimated 2.01W of power during this test. Next, in an application demonstration, a rhesus macaque performed a dexterous manual grasp task. Wirelessly received broadband neural data was communicated by the prototype embedded system to a backend server for real-time file storage and off-line spike feature extraction and hand-grip classification. Successful grip classification was demonstrated by achieving comparable classification accuracy to a conventional rackmounted commercial neural signal processor.