A New Statistical Model of Electroencephalogram Noise Spectra for Real-time Brain-Computer Interfaces

摘要

$Objective$: A characteristic of neurological signal processing is highlevels of noise from sub-cellular ion channels up to whole-brain processes. Inthis paper, we propose a new model of electroencephalogram (EEG) backgroundperiodograms, based on a family of functions which we call generalized van derZiel--McWhorter (GVZM) power spectral densities (PSDs). To the best of ourknowledge, the GVZM PSD function is the only EEG noise model which hasrelatively few parameters, matches recorded EEG PSD's with high accuracy from 0Hz to over 30 Hz, and has approximately $1/f^\theta$ behavior in themid-frequencies without infinities. $Methods$: We validate this model usingthree approaches. First, we show how GVZM PSDs can arise in population of ionchannels in maximum entropy equilibrium. Second, we present a class of mixedautoregressive models, which simulate brain background noise and whoseperiodograms are asymptotic to the GVZM PSD. Third, we present two real-timeestimation algorithms for steady-state visual evoked potential (SSVEP)frequencies, and analyze their performance statistically. $Results$: Inpairwise comparisons, the GVZM-based algorithms showed statisticallysignificant accuracy improvement over two well-known and widely-used SSVEPestimators. $Conclusion$: The GVZM noise model can be a useful and reliabletechnique for EEG signal processing. $Significance$: Understanding EEG noise isessential for EEG-based neurology and applications such as real-timebrain-computer interfaces (BCIs), which must make accurate control decisionsfrom very short data epochs. The GVZM approach represents a successful newparadigm for understanding and managing this neurological noise.