Neural Signal Compression System with Spike Detection Using Compressed Sensing
Ruihan Zheng, Yu Xia, Dongming Li, Liyang Wang, Hung-Chun Li, Peng Un Mak, Mang I Vai, Sio Hang Pun · 2023
This article aims to demonstrate a signal compression method for the wireless invasive neural recording system. A compression system with spike detection for neural signals is proposed. The input signal is firstly detected in the spike detection part and then the intercepted spike segments are sent to the compression part. A compressed sensing technique is applied in the compression part, and the Minimum Euclidean or Manhattan Distance Cluster-based (MDC) matrix is adopted for compressing neural spike segments. During simulation, the compression rate can surpass 99% and the signal-to-noise distortion ratio is around 37 dB. Moreover, the proposed method is also contrasted with the direct compression of input neural signals. When the input neural signal is directly compressed, the compression ratio is 98%, and the signal-to-noise ratio distortion rate is about 28 dB. By employing spike detection and utilizing this MDC matrix, it becomes possible to compress non-sparse spike segments in the time domain, resulting in a higher compression rate and improved reconstruction performance.