The Research And Application Of Reference Signal In EEG Signal

EEG signal includes three types that are scalp EEG signal, macrowireintracranial EEG and microwire intracranial EEG signal. The source signals fromthese three EEG signals, in turn, are more sophisticated, more close to the neurons.Currently most of researchers focus on scalp Electroencephalograms (EEG) that arerecordings of the spontaneous electrical potentials produced by the brain，reflects thepotential difference between the scalp recorded potential and the reference potential.Preprocessing plays a very important role in the analysis and study of the EEG, forexample, identification and removal of the reference signal which has been an opentechnical problem for several decades. Average reference signal method is currentlythe most commonly used method, the defects of the method have been described inthe literature.In this paper, under the key assumption that the reference signal is independentfrom EEG sources, based on the theory of independent component analysis, wepropose four reference signal removal methods, namely R1, R2, R3and R4and alsogive a detailed theoretical derivation. In order to satisfy the independence assumption,we use the two ways: the first one is that we select the earlobe of one subject as thereference site, propose one special experimental method for the subject by taping thereference electrode about once every second and record scalp EEG for some timeperiod; for the second one, we select five reference electrode locations of one subject:left thumb, left arm, left shoulder, left chest and left back. The experiment resultsshow that in first method, taping the reference electrode of one subject regularly canintroduce EMG component to all scalp EEG channel sources, this EMG component isindependent from all scalp EEG sources; in the second method, the referenceelectrode is placed on one of the above five locations, due to the long distancebetween the reference electrode and the scalp electrodes, so the real reference signalfrom these locations should be independent from all the scalp sources.Once the assumption is satisfied, we can use the above four methods to calculatethe reference signal of the original EEG signal. In this paper, the corrected EEG afterremoving the R1or R2involves much more brain neural activities than the corrected EEG after removing the average reference, and the all common signals (EMG、ECGetc) in original EEG are completely removed out, this fact actually suggests that R1and R2are superior to the average reference signal method. We also confirm that thereal reference from earlobe or mastoid is correlated with some scalp sources, in ourexperimental data, i.e., the correlation coefficients is0.69.Finally under the assumption that the reference signal is independent from EEGsources, we prove theoretically the equivalence of R2and R4methods. And we alsogive a detailed theoretical proof to show that1）R2is superior to the average referencesignal method；2）when the number of channels tends to infinity, the obtained R2isvery close to the real reference, and the average reference is never close to the realreference.