A second order adaptive Volterra filter using a two-stage lattice preprocessor

A lattice structure for a second-order adaptive Volterra filter is described. This algorithm includes a two-stage adaptive orthogonalization which greatly reduces nonuniform convergence effects. Significantly improved convergence characteristics compared with direct-form implementations are achieved. Furthermore, the system has a fairly simple structure and a relatively low computational burden. The proposed implementation is especially attractive in those applications in which the error signal needs to be minimized, such as echo canceling. In system identification problems, however, some additional offline processing needs to be performed to transform the lattice reflection coefficients into direct-form filter coefficients.<>