Low Complexity Multichannel Neural Data Compression by Exploiting Spatial Signal Correlation

Abstract

Wireless data transmission accounts for a significant fraction of total power consumed by an implanted high-density microelectrode array neural recording system. It is well known that the problem can be mitigated by using an on-chip data compressor. The suitable data compressor should offer high compression and low signal distortions. In addition it should feature very low power consumption, low memory footprint and low latency. In this paper we evaluate the feasibility of exploiting spatial correlation between neural signals being recorded with closely spaced electrode array for compression purpose. We show that optimal signal decorrelation and compression is possible by linear signal transformation with the matrix obtained with the Principal Component Analysis (PCA). Furthermore, we demonstrate that the optimal PCA based matrix has a similar structure to the Discrete Cosine Transform (DCT) matrix, which is widely used in image compression. Finally, the performance of data compressor exploiting the PCA and the DCT is compared. Along with compression ratio the power consumption and the area of decorrelation processors are reported.