An improved high-accuracy compressed sensing method using a novel constructed dictionary for neural signal detection

This paper constructs a redundant dictionary using neural spike signals and uses a compressed sensing method to compress and reconstruct neural signals, which are cut into several segments of same length. By analyzing neural signals with different signal to noise ratios (SNRs), different types of spikes and different spike widths, we verify the performance of the method. Results show that, when the Compression Ratio (CR) is less than 5, our method can accurately compress and reconstruct high SNR neural signals, which contain several types of spikes. Compared with the spike width used in the redundant dictionary, the width of detected spikes can range from 0.8 to 1.6 times of it. We can also compress and reconstruct low SNR neural signals with the CR less than 2.