DESIGN OF WIDEBAND REFLECTARRAYS BY USING NEURAL NETWORKS

Date

2023-11-7

Author

Karateke, Gürsu

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

126
views

193
downloads

Cite This

During the design of reflectarray antennas, the phase design curve of a unit cell with specified geometry is utilized. These curves are obtained by simulating the unit cell at the center frequency of the array and plotting the phase of the reflection coefficient with respect to the variable of the unit cell used to alter the phase. However, this approach is not sufficient to design reflectarrays operating at widebands. As the operating frequency deviates from the design frequency, the reflection phase changes, that results in increased sidelobes and wider beamwidths. For the elements at different distances from the center of the array, the variation of phase with respect to frequency is desired to be different to achieve wideband operation. Therefore, a single variable of the unit cell will not be enough to control the variation of phase with respect to frequency in the operational bandwidth in addition to the reflection phase at the center frequency. Hence a comprehensive design approach is required to optimize multiple parameters of the unit cell. In the literature, approaches based on filter design that utilizes equivalent circuit model of the unit cell are proposed. Although these approaches are effective, they tend to be infeasible when dealing with complicated unit cell geometries such that the relation between the inductance and capacitance values in the equivalent circuit and the parameters of the unit cell can not be expressed explicitly. In this thesis, a unit cell that is characterized by three independent parameters is considered and a neural network model is constructed by using full-wave simulation results of the unit cell both at the center and also at the lower and upper frequencies at the edges of the desired band for different combinations of the three independent parameters of the unit cell. The inputs to the neural network model are the required reflection phase values at three frequencies and outputs are the three parameters of the unit cell that results in the desired phase values. The radiation patterns of the array designed with this approach are compared to the results of the array designed by using a phase design curve that is obtained by eliminating two of the independent parameters by expressing them in terms of the third parameter. It is observed that the neural network model based design with three optimized parameters provides a stable main beam within the operational bandwidth and sidelobe levels below -19 dB even at the edges of the band as opposed to the design with a single optimized parameter that provides similar performance at the center frequency but results in wider beamwidth and sidelobe levels as high as -12dB at the edge frequencies of the band.

Subject Keywords

reflectarray, neural network, wideband, machine learning

URI

https://hdl.handle.net/11511/106422

Collections

Graduate School of Natural and Applied Sciences, Thesis