Motor cortical decoding using an autoregressive moving average model.

We develop an Autoregressive Moving Average (ARMA) model for decoding hand motion from neural firing data and provide a simple method for estimating the parameters of the model. Results show that this method produces more accurate reconstructions of hand position than the previous Kalman filter and linear regression methods. The ARMA model combines the best properties of both these methods, producing reconstructed hand trajectories that are smooth and accurate. This simple technique is computationally efficient making it appropriate for real-time prosthetic control tasks.