Efficient Microwave Filter Design by a Surrogate-Model-Assisted Decomposition-Based Multi-Objective Evolutionary Algorithm

Abstract

As a crucial frequency selection device in modern communication systems, the microwave filter plays an increasingly prominent role. There is a great demand for the multi-objective design of microwave filters. The filter’s performance affects the quality of the whole communication system directly. However, traditional multi-objective electromagnetic (EM) optimization design demands repetitive EM simulations to adjust the physical parameters of the microwave filters. Accordingly, using electromagnetic simulation directly to design optimization is quite expensive. Given this situation, this paper applies a novel surrogate model based on one-dimensional convolutional autoencoders (1D-CAE) into the multi-objective algorithm evolutionarily based on decomposition (MOEA/D) for the first time. This approach uses MOEA/D as the multi-objective optimizer, and a novel low-complexity surrogate model based on one-dimensional convolutional autoencoders (1D-CAE) is constructed to predict the expensive EM simulation results. The surrogate model based on 1D-CAE is used to generate the results of scalar subproblems of MOEA/D, which greatly improves the design efficiency. Compared with the traditional design methods based on an EM solver, this method not only effectively optimizes multiple design objectives but also completes the design of microwave filters in a shorter time. The proposed method is verified using the design of a sixth-order ceramic filter and a seventh-order metal cavity filter.