High-Quality Sampling Inverse Design Scheme with Deep Learning

Artificial intelligence techniques are being used for the inverse design of microwave devices, but challenges such as high computation cost, high-dimensional data, non-uniformity, and low-quality samples in design space can negatively affect modeling performance. To overcome these challenges, a highquality sampling inverse design scheme deep learning method (HQS-DL) is proposed for automated microwave filter design. Particle swarm optimization (PSO) is used for forward simulationbased sampling to tentatively select high-quality samples, and multilabel synthetic minority over-sampling technique (MLSMOTE) is utilized to enlarge the training samples and make them more uniform. A neural network is used to efficiently represent the nonlinear mapping between input S-parameters and required filter structural parameters. The proposed method is verified with a band-pass microstrip hairpin filter and achieves better modeling effectiveness and inverse design efficiency than conventional methods, enabling quick prediction of device parameters for different center frequencies and bandwidths.