Systems Seminar
Beamspace Signal Processing for Magnetoencephalography and
Electroencephalography
Prof. Barry Van Veen
ECE Department
Abstract
Magnetoencephalographic and electroencephalographic data are typically
characterized by low signal to noise ratio (SNR), spatially and temporally
colored noise with unknown statistics, and relatively few observations of
the phenomena of interest. These constraints create very challenging
problems for signal processing algorithms that image or localize brain
electrical activity. In this talk we show how performing signal processing
in beamspace results in significantly improved performance at low SNR and
with small numbers of data observations. The term "beamspace" refers to
mapping the measured data at the sensors into a lower dimensional space
with a linear transformation prior to applying the signal processing
algorithm of interest. Reducing the dimension significantly improves the
reliability of the estimated noise statistics in algorithms that implicitly
or explicitly estimate the spatial covariance matrix of the data. We show
that relatively small beamspace dimensions are sufficient to accurately
represent anatomically meaningful regions of the cortex. The effectiveness
of beamspace processing is demonstrated by performing source localization
on somatosensory data with three different algorithms. While the three
algorithms involve different tradeoffs between SNR and the number of data
observations, we show that beamspace processing provides dramatic performance
improvement in every case when the number of observations is small.
Time and Place: Wed., Oct. 19, at 3:30 pm in 4610 Engr. Hall.
SYSTEMS SEMINAR WEB PAGE:
http://homepages.cae.wisc.edu/~gubner/seminar/