Author(s): Yu B, Mak T, Li X, Xia F, Yakovlev A, Sun Y, Poon C
Abstract: Real-time multi-channel neuronal signal recordinghas spawned broad applications in neuro-prostheses and neurorehabilitation.Detecting and discriminating neuronal spikes frommultiple spike trains in real-time require significant computationalefforts and present major challenges for hardwaredesign in terms of hardware area and power consumption. Thispaper presents a Hebbian eigenfilter spike sorting algorithm,in which principal components analysis (PCA) is conductedthrough Hebbian learning. The eigenfilter eliminates the needof computationally expensive covariance analysis and eigenvaluedecomposition in traditional PCA algorithms and, most importantly,is amenable to low cost hardware implementation. Scalableand efficient hardware architectures for real-time multi-channelspike sorting are also presented. In addition, folding techniquesfor hardware sharing are proposed for better utilization ofcomputing resources among multiple channels. The throughput,accuracy and power consumption of our Hebbian eigenfilter arethoroughly evaluated through synthetic and real spike trains.The proposed Hebbian eigenfilter technique enables real-timemulti-channel spike sorting, and leads the way towards the nextgeneration of motor and cognitive neuro-prosthetic devices.
Keywords: Brain-machine interface, Hebbian learning, spike sorting, FPGAs, hardware architecture design.