Multi-channel LFP recording data compression scheme using Cooperative PCA and Kalman Filter

Abstract

The power consumption and wireless communication bandwidth restrict the channel count and amount of high-resolution intracortical Brain–Computer Interfaces(iBCIs). So, it is necessary to compress the captured data before transmission. Here, we propose a multi-channel data compression method on the correlation between channels for Local Field Potential(LFPs) recorded through Utah microelectrode arrays. The method improves the compression performance of the Stepwise Expectation-Maximization Principal Component Analysis(SEM-PCA) algorithm by implementing the Kalman filter. The multi-channel correlation is sampled with high resolution while the uniqueness of each channel is retained at a low rate. Then, the two parts of information are fused by the Kalman filter, and finally, the data is reconstructed with high precision and high rate. Our testing results verified the effectiveness of the proposed compression scheme. Compared to the traditional SEM-PCA, this algorithm significantly improved neural LFP reconstruction accuracy from 79.61% to 88%.