Block-Sparse Modeling for Compressed Sensing of Neural Action Potentials and Local Field Potentials

This paper presents our attempt for the efficient sparse modeling and recovery for Compressed Sensing (CS) of extracellular neural action and local field potentials (APs & LFPs). Both type of neural signals can be modeled as block-sparse in DCT (Discrete-Cosine Transform) domain, where we exploit the spectral information to determine the block boundaries, including bandpass filter pole information used for spike detection and the corner frequency of local filed potentials, respectively. Binary-Weighted ℓ
1
-minimization (BW-ℓ
1
-min) is proposed for neural signal recovery with their respective block boundary information. Experimental results demonstrate that block-sparse modeling and BW-ℓ
1
-min recovery lead to more than 5-dB signal-to-noise ratio improvement for both AP and LFP signals as compared to the standard ℓ
1
-minimization algorithm.