Multiple sensor sequential tracking of neural activity: Algorithm and FPGA implementation

Abstract

We investigate the use of the particle filtering sequential Bayesian estimation technique and its hardware implementation for tracking neural activity. We propose using the multiple particle filter (MPF) approach in order to reduce the computational intensity incurred due to the large number of sensors required to observe the noninvasive magnetoencephalography (MEG) measurements needed to estimate the brain current dipole source locations and moments when tracking neural activity. The improved tracking performance of the MPF is demonstrated using numerical simulations on synthetic and real data. We also investigate the parallel implementation of the MPF algorithm on a Xilinx Virtex-5 field-programmable gate array (FPGA) platform. Our results of significant reduction in timing requirements demonstrate that the implementation is suitable for real-time tracking.