Abstract
A composite fiber Bragg grating (FBG) structure with several apodized sections is utilized for designing dense wavelength division multiplexing (DWDM) multiband transmission filters. A learning genetic algorithm (LGA) is also developed to determine the optimum design parameters of these filters. By taking advantage of a knowledge base (KB) that stores the FBG parameter sets and the corresponding transmission profile feature sets, our LGA can generate a suitable initial population and perform evolutionary optimization starting from it. This has made the LGA evolve more quickly to more accurate results than the methods without using the KB. The LGA can also store new results into the KB according to its decision procedure and improve its precision of initial prediction as it works through more and more examples.