Efficient Neural Network-Based Synthesis of Low Sidelobe Frequency-Invariant Directional Patterns

• Published in: Cross Strait Quad-Regional Radio Science and Wireless Technology Conference pp. 1 - 3
• Main Authors: Su, Xinchang, Cai, WeiQi, Liu, Xinkun, Zhang, QingDong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 04.11.2024
• Subjects: Antenna radiation patterns; frequency invariant directional pattern synthesis; Frequency synthesizers; Iterative Fourier algorithm; Machine learning; Machine learning algorithms; Measurement; neural network; Optimization; Planar arrays; Radial basis function networks; Refining; Wireless communication
• ISSN: 2377-8512
• DOI: 10.1109/CSRSWTC64338.2024.10811524

The synthesis of array antennas is to obtain an antenna array pattern that meticulously fulfills target performance metrics, including the main lobe width, sidelobe height suppression, directivity coefficient enhancement, and the creation of nulls. This intricate design process involves the consideration of factors such as the number of radiating elements, their inter-element spacing, the amplitude excitation ratio, and the excitation phase. This article presents an advancement in the Inverse Fourier Transform (IFT) algorithm, successfully achieving frequency-invariant pattern synthesis for a 32-element, equidistant linear array operating within the X-band, while maintaining sidelobe levels below the stringent threshold of 30dB. Furthermore, to expedite the synthesis process and ensure the generation of low sidelobe, frequency-invariant patterns, an innovative Radial Basis Function neural network (RBFNN) is introduced, harnessing the power of machine learning for rapid and precise pattern optimization.