A hybrid algorithm for fast optimization-based synthesis of coupling matrices for advanced microwaves filterings specifications

This paper presents a new hybrid algorithm for the optimization-based synthesis of coupling matrices for microwave filter design, and realizing advanced specifications. The proposed algorithm receives as input the coefficients of the polynomials of the filter specification to be realized, and provides as output the corresponding coupling matrix in a few seconds. The algorithm combines a modified binary-coded genetic algorithm (GA) as the global optimizer, and an analytical algorithm based on the adaptive learning coefficient gradient descent method as the local optimizer. Unlike a traditional GAs, the GA presented in this paper uses an initial population selector and a novel mutation and selection operator. In addition, the mutation operator precedes the crossover operator in the flowchart of its GA component, thus avoiding premature homogenization of the population and preventing the global optimizer from getting stuck in a valley of local optimums for too long. This configuration improves the convergence speed of the hybrid algorithm and enables access to advanced topologies. The algorithm has been successfully used to synthesize two coupling matrices that realize complex filtering specifications.