A computationally efficient second order adaptive lattice filter

A nonlinear second order Volterra filter can be reformulated as a linear multichannel lattice structure. Such a formulation reduces eigenvalue disparity for gradient descent adaptive implementations. For a filter of memory span L, this approach requires O( L³) coefficients, with corresponding computational burden and memory requirements. A more efficient two-stage lattice structure is described, where a similar result is achieved with computational complexity proportional to O(L² )