An efficient implementation of the PI subspace state-space system identification algorithm

Subspace based methods for linear system identification are robust, require few a priori assumption's, and impose minimal restrictions on the type of input used to probe the system. However, due to their excessive computational and storage requirements, they have seen only limited application within biomedical engineering. In this paper, we develop a fast, low memory variant of an instrumental variable subspace method. Monte-Carlo simulations demonstrate the acceleration achieved by this new implementation.