Compressive Sensing of Neural Action Potentials by Designing Overcomplete Dictionaries

Long-term wireless neural recording systems which are subject to stringent power consumption, are highly desired to reduce the rate of data transmission and computation complexity. In this paper, we propose using a combination of on-chip neural action potentials ('spikes') detection system and compressive sensing (CS) techniques to reduce the power required for data transmission and a circulant and toeplitz matrix to reduce the computation complexity consequently further reduce the power consumption. We use the K-SVD algorithm for adapting dictionaries in order to achieve sparse signal representations of spikes and iterative shrinkage thresholding (IST) algorithm to reconstruct them. Our results show that, using the data from simultaneously recorded from brain cortex of anesthetic SD-rats, the mean compression ratio is 60:1 achieved for 13.7-dB SNDR recovery using this mechanism.