Optimum pole postion for digital Laguerre network with least square error criterion (Q2735086)

The paper proposes a global optimization method to solving the optimum pole selection problem for a digital Laguerre network structure with a frequency domain Least Square Error (LSE) criterion. The method is based on a modification of the bridging technique developed in \textit{Y. Liu} (1998), by incorporating a local search relying on the Golden Section method (Minoux, 1986). The bridging method finds a local minimum point of the bridged function every time a local solution is obtained until the global minimum of the original cost function is found. Since the method requires the continuous differentiability properties of the bridged function, a smoothing procedure is introduced in every iteration. By incorporating a local search method (such as the Golden Section), the bridging technique can be improved, since no smoothing procedure is needed.NEWLINENEWLINENEWLINEThe proposed optimization method yields an effective algorithm for locating the optimum pole position for the Laguerre network with high accuracy. Simulation results on practical numerical examples are presented, and a comparison between the Laguerre structure and the FIR (Finite Impulse Response) filter structure shows that the former requires a smaller number of coefficients than the latter, especially for the specifications that involve over-sampling.NEWLINENEWLINEFor the entire collection see [Zbl 0962.00004].