Energy Efficient Design of Low-Profile Wideband Microstrip Patch Antennas Using Deep Learning

A deep-learning (DL) method is proposed for the energy-efficient design of low-profile wideband microstrip patch antennas with maximized gain-bandwidth product. The two-step method includes data preparation through feature extraction and network training for prediction. The trained DL network enables a fast and accurate bi-directional prediction of the geometry and antenna performances in seconds. For proof of concept, three microstrip antennas are designed by the DL network in three predefined frequency bands, including the 2.4-GHz Wi-Fi bands, the 2.6-GHz 5G bands and the 3.3-GHz 5G bands respectively, all with identical overall size of 0.98λ 0 × 0.98λ 0 × 0.05λ 0 , where λ 0 is the free-space wavelength at 2.45 GHz. The achieved bandwidths are all above 760 MHz for 10-dB return loss and gain above 8 dBi. One of the three antennas is experimentally verified in good agreement with full-wave simulations. Without wasting the energy consumed by simulations, the proposed method makes full use of all the simulated data at once, thereby providing a new paradigm toward energy-efficient design of all types of antennas.