5G slot antenna design with machine learning

Official URL: https://risc01.sabanciuniv.edu/record=b3077215_(Table of contents)

5G technology is promising to be the future technology due to its higher data output, lower latency, and higher channel capacity. Many transceiver structures are designed for 5G and antennae play a crucial role in these structures with their gain, bandwidth, and directional properties. To satisfy the needs of the system, RF engineers use tools such as HFSS, AWR, and CST to design optimum antennas. These tools can simulate the real behavior of antennae at the cost of time and hardware memory. To solve the computation cost issue, this thesis focused on using machine learning tools to do antenna design. First, a slot antenna topology was chosen based on the 5G antenna and designed using HFSS and traditional optimization methods. The parameters of the chosen topology were swept to create a dataset for machine learning. This dataset was used for predicting realized gain and s-parameters. To find the optimum design, input parameters in the datasets were interchanged with output parameters to locate the best lengths. These lengths were swept with machine learning and deep learning tools to exhaustively search for an improved design. As a result of this multi-step process, a better-performing antenna was designed in a shorter time and with less computation cost.