Identification of Widely Linear Systems Using Data-Dependent Superimposed Training

Abstract

In this correspondence, we address the identification of widely linear (WL) systems using data-dependent superimposed training (DDST). The analysis shows that the nonlinear nature of WL systems can be exploited to decouple the finite impulse responses of the filters that constitute the system under identification. Unlike the DDST scheme considered for strictly linear systems, for the WL scenario two data-dependent sequences are required to distort the transmitted data and avoid interference with the training sequence during the estimation process. Closed form expressions for sequences with a complete second order characterization (good periodic autocorrelation and zero complementary periodic autocorrelation) are also provided. The performance of the method is compared with other techniques using numerical simulations.