Study on Compressed Sensing of Action Potential

TL;DR

This study evaluates various compressed sensing algorithms for neural action potential signals, demonstrating their effectiveness in reducing data size while maintaining signal integrity, which is crucial for efficient neural data processing.

Contribution

It provides a comparative analysis of CS reconstruction algorithms on neural signals, focusing on feasibility, robustness to non-sparsity, and practical implementation considerations.

Findings

Certain algorithms achieved high compression ratios with acceptable SNR.

The MATLAB testing platform facilitated systematic evaluation of algorithms.

Feasibility of dictionary design and thresholding methods was confirmed.

Abstract

Compressive sensing (CS) is a signal processing technique that enables sub-Nyquist sampling and near lossless reconstruction of a sparse signal. The technique is particularly appealing for neural signal processing since it avoids the issues relevant to high sampling rate and large data storage. In this project, different CS reconstruction algorithms were tested on raw action potential signals recorded in our lab. Two numerical criteria were set to evaluate the performance of different CS algorithms: Compression Ratio (CR) and Signal-to-Noise Ratio (SNR). In order to do this, individual CS algorithm testing platforms for the EEG data were constructed within MATLAB scheme. The main considerations for the project were the following. 1) Feasibility of the dictionary 2) Tolerance to non-sparsity 3) Applicability of thresholding or interpolation.