High compression rate and efficient spikes detection system using compressed sensing technique for neural signal processing

We design a digital neural signal compression and spikes detection system using compressed sensing technique and root-mean-square method respectively. This system does not only detect spikes from a neural signal but also can compress this neural signal with a high compression rate. In the compression part, due to the fact that neural signals are not sparse in the time domain, we designed a sensing matrix, called Minimum Euclidean or Manhattan Distance Cluster-based (MDC) matrix, to compress neural signals. Using this MDC matrix and a novel reconstruction algorithm, we achieve a compression rate which can be up to 90% with the reconstruction error being around 0.2. Moreover, the proposed system has relatively low power consumption (0.59 mW) and a small chip area (7 μm²).