Locally-stationary multivariate AR model analysis of forearm electromyographic signals on handwriting movements

The goal of this study is to construct a mathematical model connecting with motor commands from the brain and handwriting movements on a forearm-hand-pen system. It is assumed that the motor commands can be known indirectly from the electromyographic (EMG) signals on the forearm surface. This is equivalent to be possible to predict written letters from the EMG signals. At first, the EMG signals and the pen-tip movements on writing some letters are measured. The measured EMG signals are analyzed by the locally-stationary multivariate auto-regressive (AR) model. Then, we assume that the locally-stationary EMG signals are the stochastic process based on the AR model driven by the motor commands as Gaussian white noise and we can estimate the electronic signals to the motor commands on each forearm muscle. Moreover, we wish to describe and identify the forearm-hand-pen system as the parametric linear model whose inputs are the estimated motor commands and whose outputs are the pen-tip movements on writing letters. Finally, we would like to reproduce written letters from the measured EMG signals and discuss the results.