Optimal Input Selection for MISO Systems Identification: Applications to BMIs

Abstract

We have developed an algorithm for selecting an optimal set of inputs for use in linear multiple-input, single-output system identification processes. The algorithm provides a decomposition of the system output such that each component is uniquely attributable to a specific input This reduces the complexity of the estimation problem by optimally selecting inputs according to the uniqueness of their output contribution and is useful in when subsets of the inputs are highly correlated or do not contribute significantly to the system output. The algorithm was evaluated on experimental data consisting of up to 40 simultaneously recorded motor cortical signals and peripheral electromyograms (EMGs) from four upper limb muscles in a freely moving primate. It was used to select the optimal motor cortical signals for predicting each of the EMGs and significantly reduced the number of inputs needed to generate accurate EMG predictions. For example, although physiological recordings from up to 40 different neuronal signals were available, the input selection algorithm reduced this 10 neuronal signals that made significant contributions to the recorded EMGs