Machine Learning-Assisted Design of Metasurface Radomes

Antennas are indispensable to wireless communications. With high frequency band signals applying in 5G channels, the wireless transmission rate is significantly boosted. While 5G technologies are expected to be the foundation for the next generation’s communication systems, the high frequency band leads 5G antennas be very sensitive to the environment. To set up the base stations efficiently, the demand for a fast and accurate measurement method of antenna radiation patterns is urgently needed. In addition, the development of multi-input-multi-output (MIMO) antenna systems also plays an essential role in 5G technologies, which enables high-speed wireless communications, wide diversities, and multiplexing at the same time. However, the mutual coupling effect significantly affects the performance of MIMO antenna systems which not only degrades the data diversity but also reduce the antenna gain. There are many previous works in mutual coupling reduction. However, most of the design only focus on a specific type of antenna systems and are limited to be widely applied. In this thesis, a series of machine learning (ML)-assisted models are represented for an efficient measurement method of antenna radiation pattern and a design method of metasurface radomes that enable mutual coupling reduction induced by the interference of radiation waves. In general, measuring the antenna's radiation pattern is a time-consuming task and is typically limited to specific planes (e.g., E- and H-planes) or angles. The proposed ML-assisted model based on the generative adversarial network (GAN) can restore the antenna radiation pattern from the sparse measurement data, and hence enables a fast and accurate measurement procedure. In addition, an auto-encoder-decoder (AED)-based inverse design model is presented to design an antenna radomes with broadband, wide-angle unidirectional or bidirectional absorption that can reduce mutual coupling in multi-input-multi-output (MIMO) antenna arrays. The proposed ML algorithm can automatically synthesize the metasurface radome to tailor reflection, transmission, and absorption of EM waves in a unidirectional or bidirectional manner. Three representative metasurface design examples, including an ultrathin, unidirectional and bidirectional metasurface-absorber, and a wide-angle unidirectional metasurface-absorber, are presented to validate the proposed ML-assisted design model. The three applications demonstrated that the ML-assisted model has a great potential in solving EM problems.