Multi-source Neural Activity Estimation and Sensor Scheduling: Algorithms and Hardware Implementation

Electroencephalography (EEG) and magnetoencephalography (MEG) measurements are used to localize neural activity by solving the electromagnetic inverse problem. In this paper, we propose a new approach based on the particle filter implementation of the probability hypothesis density filter (PF-PHDF) to automatically estimate the unknown number of time-varying neural dipole sources and their parameters using EEG/MEG measurements. We also propose an efficient sensor scheduling algorithm to adaptively configure EEG/MEG sensors at each time step to reduce total power consumption. We demonstrate the improved performance of the proposed algorithms using simulated neural activity data. We map the algorithms onto a Xilinx Virtex-5 field-programmable gate array (FPGA) platform and show that it only takes 10 ms to process 100 data samples using 6,400 particles. Thus, the proposed system can support real-time processing of an EEG/MEG neural activity system with a sampling rate of up to 10 kHz.