As the first sage to develop the design method for a GenLOT (generalized linear-phase lapped orthogonal transforms) in multiobjective eases by nonlinear optimization methods, I selected promising optimization methods and compared their performance in a singleobjeceive ease (coding gain) by experiments. I selected the following six optimization methods: the pattern search method, the simplex method, the implicit filtering method, the DIRECT (Dividing RECTangles) algorithm, the adoptive simulated annealing algorithm (ASA) and the determinstic genetic algorithm (DCDGA). Main results are as follows. The pattern search and simplex methods attained relatively better results than the three global search methods with extremely fewer function evaluations. The best results were obtained by the pattern search method. Concerning the global search capability, the best one is DIRECT-A and ASA and DCDGA follow it. On the contrary, the local search capabilities of these three methods are relatively weak.
展开▼