Fast Filter Optimization Method Based on Neural Network

keywords:neural network  coupling matrix  microwave filter  fast optimization

Abstract:Currently, the method for designing and optimizing microstrip filters is based on electromagnetic simulation EM software.However, this method has two major drawbacks. First, the initial value of the optimization variable needs to be guessed manually, but due to human experience limitations, the guessed value may deviate greatly from the optimal value, leading to optimization results trapped in local optima. Second, the long simulation time per iteration results in a lengthy optimization process[1] .This paper proposes a proxy model using a parallel neural network and vector fitting to optimize the filter. To verify the credibility of this method, a microstrip filter is designed and tested via simulation and experiments, which showed consistent results. The proposed method overcomes the two major drawbacks of the current filter design and optimization methods, which provides a fast and efficient approach for designing high-performance filters.