System considerations for the compressive sampling of EEG and ECoG bio-signals

Analog domain Compressed Sensing (CS) has enabled dramatic levels of sub-Nyquist sampling of sparse signals in applications such as electrocardiogram (ECG) and electromyogram (EMG) bio-signals. This work describes extensions of CS to electroencephalogram (EEG) and electrocorticogram (ECoG) brain signals. Specifically, the time-, frequency- and wavelet-domain sparsity of these signals is investigated. For EEG/ECoG signals it is shown that the time-and frequency-domain capture essential spike features even at high threshold levels and are useful for coarse EEG/ECoG processing. Wavelet domain can be used for fine processing with signal de-noising properties without using thresholding. CS signal-reconstruction performance for time-, frequency- and wavelet domain (Daubechies, Symlets, Coiflets and Meyer wavelets) are presented along with system design considerations.