Date:2008/DEC/18(thu)15:00- Place: Speaker:Dr.Danilo P. Mandic(Communication and Signal Processing Research Group Department of Electrical and Electronic Engineering Imperial College) TitleFComplex Adaptive Filters: Noncircularity, Widely Linear and Neural Models AbstractFI will present a unified framework for adaptive filtering, both linear and nonlinear, of the generality of complex valued signals. The models introduced will be both feedforwad and feedback, and I will illustrate how the correspnding learning algorithms can be derived based on CR calculus. The main part of the talk will be dedicated to dealing with complex circularity, a unique property of complex probability distributions. To deal with second order non-circularity, a class of widely linear models will be introduced and the augmented complex least mean square (ACLMS) algorithm will be outlined. This allows for a unified treatment of both circular (symbols in communications) and noncircular (directional processes, array signals, vector fields) processes. Examples will be provided on the forecaseting of wind and on radar data.