A Planar Coil Design Using Deep Learning

Designing a coil for wireless power transfer involves considering various factors, such as geometry, quality factor, and self-resonance frequency. While there are theoretical approaches, designers face numerous challenges due to the interdependence and trade-offs among these factors. Additionally, the electromagnetic simulation of coils is a time-consuming process, particularly when dealing with intricate details. In this paper, we present an algorithm that addresses these challenges by generating coil images corresponding to desired properties, as represented by the S
11
(reflection coefficient), which encompasses coil self-resonance frequency (SRF), impedances, and quality factors. The proposed algorithm has two deep learning networks for creating electromagnetic coil structures from electronic circuit S
11
diagrams. The first network is a conditional Deep Convolutional Generative Adversarial Network (cDCGAN) that uses coil images and their corresponding S
11
diagrams as its dataset. This network can generate an image of the coil structure from an input S
11
diagram. The second network reconstructs S
11
diagrams for the first network based on the optimized coil model S
11
result. The dataset of the second network is composed of the labels of the first network (coil S
11
diagrams). Upon completion of the training process, the deep learning network generates a novel coil image based on the input S-parameter diagram, which is tailored to fulfill the desired coil properties.