A stream-based Hebbian eigenfilter for real-time neurophysiological signal processing

Abstract

Rapid advances in multi-channel microelectrode neural recording technologies in recent years have spawned broad applications in implantable neuroprosthetic and rehabilitation systems. The dramatic increases in data bandwidth and data volume associated with multichannel recording also come with a large computational load which presents major design challenges for implantable systems in terms of power dissipation and hardware area. In this paper, we present a new design methodology that utilizes Hebbian learning for real-time neural signal processing. A stream-based technique is proposed that can effectively approximate the hardware learning kernel while significantly reducing hardware area and power. The proposed method is validated using benchmark problems including spike sorting and population decoding. Experimental results show that the stream-based approach can achieve up to 98% and 43.4% reduction in equivalent slice look-up table and power of Xilinx Spartan6 Low Power FPGA.