Identification of separable systems using trilinear filtering

Linear filtering methods are well known and have been successfully applied in system identification and equalization problems. However, they become unpractical when the number of parameters to estimate is very large. The recently proposed assumption of system separability allows the development of computationally efficient alternatives to classic adaptive methods in this scenario. In this work, we show that system separability calls for multilinear system representation and filtering. Based on this parallel, the proposed filtering framework consists of a trilinear extension of the classical Wiener-Hopf (WH) solution that exploits the separability property to solve the supervised identification problem. Our numerical results shows the proposed algorithm can provide a better accuracy than the classical WH solution which ignores the multilinear system representation.