Blind System Identification of Autoregressive Model Using Independent Component Analysis
Masuhiro Nitta, Kenji Sugimoto, Atsushi Satoh · Transactions of the Society of Instrument and Control Engineers · 2005
This paper proposes a new method for identifying a multi-input multi-output autoregressive model without observation of input signals. This is achieved by making use of an independent component analysis technique, which rebuilds signals from their linear mixture under the assumption that the source signals are independent. The method firstly introduces an augmented state-space expression of the observed signal, and then adopts a numerical search using the so-called natural gradient. Finally, numerical simulation has been carried out to illustrate the effectiveness of the proposed method.