Application of Sequential Particle Filter in Locating EEG Signal Sources

This paper addresses the problem of EEG source localization using a new method based on particle filter. The proposed method is named sequential particle filter (SPF). SPF estimates each source's state vector parameters separately by considering the effect of other active sources. This method applies some modifications to particle resampling and weighting steps on conventional particle filter algorithms. These modifications, which are kind of applying spatial filtering on particles, lead to less susceptibility to noise for SPF comparing to the conventional particle filters. The SPF method has evaluated using two kinds of simulation scenarios (single-tone and Pseudo-real EEG data) and real EEG data. The SPF results have been compared to conventional particle filter, LCMV beamformer, and sLORETA (schematically comparison). The results have shown that SPF improves localization accuracy in low SNRs compared to the conventional particle filter algorithm and LCMV beamformer methods. In addition, simulation results illustrate that the SPF method is more effective in localizing simultaneous active sources than the others. Also, since the SPF method divides the state vector into several sub-vectors (related to the number of sources), it has higher computational speed than the conventional particle filter.