Independent component analysis of multichannel neural recordings at the micro-scale

The problem of blind source separation from an observed array mixture is encountered in many biomedical engineering applications. In this work, we compare two methods aimed at separating multiple signal sources from multichannel extracellular neural recordings in the microenvironment of the brain. The framework is part of our ongoing effort to enhance the communication and signal processing technology of microimplanted devices used for recording and stimulating neural cells at the micro-scale. Simulation results show that Independent Component Analysis (ICA) and the Minimum Variance (MV) methods have equal degrees of success in separating signal components consisting of mixtures of spike trains and local field potentials. The MV method exhibits less noisy estimates but prior knowledge of the mixing matrix and noise parameters is required. On the other hand, the ICA method doesn't require prior knowledge of the mixing matrix but exhibits higher variance. Two examples are given to illustrate the advantage of applying both methods to multichannel neural recordings.