Containing errors in computations for neural sensing: does a hierarchical-referencing strategy lead to energy savings?

Recent experimental evidence and theoretical results challenge the belief that ultra high-density EEG sensing will not yield higher spatial resolutions. It raises the exciting possibility that source-localization accuracy can be improved substantially with ultra high-density systems; however, these systems are hindered by implementation constraints in circuit volume and energy consumption. Recently, an information-theoretic hierarchical referencing mechanism has been proposed to exploit the decay of high-spatial frequencies during volume conduction from source to scalp - and the induced high local spatial correlations - to reduce the required circuit power and volume. In this paper, human EEG data is used to experimentally test and validate theoretical inter-electrode correlations and bit savings when employing a hierarchical referencing strategy. Extrapolating from electrodes that are at least 2 cm apart, we observe that on average savings can exceed 3 bits per electrode at inter-electrode distances of 3 mm.