Comprehensive Modeling of Antenna Electrical and Field Responses Using Transformer-Inspired Decoder-Based Transposed Convolutional Neural Networks

Surrogate modeling is a key enabler in antenna engineering, facilitating efficient simulation-based design, including parametric optimization. However, developing reliable behavioral models remains challenging, with most existing methods focusing solely on electrical characteristics (e.g., reflection coefficients) or integral performance metrics (e.g., gain). Modeling full radiation patterns—critical for practical antenna design—has received limited attention in the literature. This paper presents an innovative and comprehensive modeling framework for antennas based on customized deep neural networks. The proposed models capture both electrical responses and radiation patterns across a wide frequency spectrum, various antenna dimensions, and diverse substrate parameters, making them suitable for solving design tasks. Frequency-domain characteristics (e.g., reflection, gain) are accurately modeled using recurrent neural networks (RNNs) incorporating long short-term memory (LSTM) layers, treating frequency as a sequential input. Radiation patterns are learned through decoder-style convolutional neural networks (CNNs), with deconvolutional (or up-sampling) layers inspired by the design principles of Transformer decoders, to provide reliable predictions over wide frequency ranges and azimuth angles as functions of input geometry parameters. Extensive validation involving two planar antennas demonstrates the exquisite accuracy of the presented surrogates compared to several deep learning approaches. Practical design applications, including center frequency tuning, gain maximization, bandwidth widening, and front-to-back ratio improvement, confirm the models’ effectiveness in real-world optimization scenarios at a moderate computational cost. The methodology paves the way for robust, multi-response antenna modeling and expands the utility of data-driven surrogates in modern antenna design.