Antenna Modelling Based on Image-Model-Oriented CNN Exploiting LSTM

Abstract

To address the time-consuming and computationally intensive challenges associated with antenna performance analysis using full-wave electromagnetic simulation software combined with global optimisation methods, this study proposes an efficient strategy based on deep learning, applied to high-precision antenna modelling. Considering the excellent performance of Convolutional Neural Networks (CNN) in pattern recognition and the high efficiency of Long Short-Term Memory (LSTM) structures of Recurrent Neural Networks (RNN) in handling sequential data, this paper combines CNN and LSTM structures to form a hybrid CNN-LSTM network. Furthermore, to enhance network performance and leverage the characteristic of CNNs in extracting image features within the receptive field by mimicking the biological visual cortex, the antenna to be modelled is constructed as a two-dimensional image. Thus, an Image-Model-CNN-LSTM hybrid network is proposed. This study employs two different antenna models to validate the generalisation capability of the proposed approach. Experimental results demonstrate that the proposed network exhibits significant advantages in terms of prediction accuracy and model fitting. Compared to the CNN-LSTM network, the proposed Image-Model-CNN-LSTM network applied to different antenna configurations achieves a reduction in Mean Squared Error (MSE) by 51.5% and 40.9%, respectively, while improving model fitting R² by 5.6% and 4.0%.