An Automatic Decomposition Approach to Optimization of Large Microwave Systems

We present a novel and general technique applicable to the optimization of large microwave systems. Using sensitivity information obtained from a suitable Monte Carlo analysis, we extract possible decomposition properties which could otherwise he deduced only through a detailed physical and topological investigation. The overall problem is automatically separated into a sequence of subproblems, each being characterized by the optimization of a subset of circuit functions w.r.t. variables which are sensitive to the selected responses. A heuristic algorithm for automatic decomposition is developed. The decomposition patterns are dynamically updated until a satisfactory solution is reached. The partitioning approach proposed by Kondoh for FET modeling problems is verified. The technique was successfully tested on Iarge-scale optimization of microwave multiplexer involving 16 channels, 399 nonlinear functions, and 240 variables.