T. Ogunfunmi (USA)
We present a third-order nonlinear Wiener adaptive filter suitable for non-Gaussian inputs. It is applicable even to non-white inputs. In earlier work, we have presented the second- and third-order nonlinear discrete-time Wiener adaptive algorithm and analyzed their performance assuming Gaussian white input. In many practical applications, we do not have the simplicity of Gaussian white input. We apply transform-domain methods in orthogonalizing the nonlinear Wiener adaptive filter input vector. For third-order and higher order models, this method is more efficient than the lattice filter methods that have been proposed in the literature. With our method, we still realize all the advantages of the nonlinear Wiener model over the Volterra model such as faster convergence, smaller number of coefficients, less number of computations and avoidance of the dilemma of multiple local minima. We also present some computer simulations results which verify the suitability of our algorithm.
Important Links:
Go Back
