Browsing by Author "Louw, Ridalise"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Surrogate modelling of performance metrics of a wideband feed for the SKA reflector antenna
    (Stellenbosch : Stellenbosch University, 2018-03) Louw, Ridalise; De Villiers, D. I. L.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: The focus of this thesis is on a design methodology for a pyramidal sinuous antenna for use as a wideband feed in the SKA reflector antenna system. The design objective is to maximise the receiving sensitivity of the feed, while maintaining a reflection coefficient of at most -10 dB over the entire band. Acquiring accurate solutions of the antenna’s performance metrics for each geometric variation in the design space requires a full wave solution for each possibility. This is a time consuming task. Thus, the focus in this thesis is on finding accurate surrogate models that are fast to evaluate, on which the design can be done. Surrogate models require the availability of a coarse model that is less accurate, but faster to evaluate than the fine model. Truncated models of the antenna structure that operate at the band edges of the full fine model are used. These provide a good approximation of the behaviour of the pyramidal sinuous antenna. A simple surrogate model is suggested for the sensitivity of the system, which makes use of an output space mapping technique where a second-order polynomial regression term is applied to the difference between the sensitivity of the coarse and fine models, using only a few fine model evaluations. A rational interpolation model is used to find the input impedance of the antenna from which the reflection coefficient is calculated. Rational interpolation of high-fidelity data, acquired from the fine model, is done first. Low-fidelity data, acquired from the coarse model, are subsequently added to the rational interpolant so as to improve the accuracy of the model without the need for adding additional high-fidelity data points. The constraint which ensures a pole-free rational interpolant restricts the solution of the model. This leads to the introduction of a blended rational interpolation method that locally models the trend of the low-fidelity data and is then blended together into a global interpolant using quadratic B-spline functions. A comparison of this model to other interpolation methods shows this blended rational interpolant to perform well. Brief consideration is given to the application of these surrogate modelling methods on a 5:1 bandwidth pyramidal sinuous antenna. This example illustrates the significant speed-up that is achieved for the design, where a speed-up factor close to 16 is achieved. The design of a 3:1 bandwidth is then considered with two geometric parameters as input to the design. Using very few high-fidelity data points, the blended rational interpolation method leads to a predicted region of where the reflection coefficient is less than -10 dB that has an 11.9% error. From this region, the antenna with the maximum sensitivity is identified, with the surrogate predicting an average sensitivity of 3:651 m2=K. Validation of the results shows the average sensitivity to be equal to 3:7139 m2=K and a reflection coefficient below -10.52 dB over the entire band.