Bayesian Optimisation of a Frequency Selective Surface Using a Regularised Objective Function

This work introduces a penalty-based regularisation to optimise the geometric parameters of a periodic frequency selective surface. The equivalent circuit of a multi-layer bandpass filter is used to illustrate the benefits of the approach for wideband applications. The proposed regularisation is obtained by convolving the frequency response and the desired thresholds before measuring the infraction. We apply bayesian optimisation on the unconstrained optimisation problem and assess the per-formance of the strategy by the number of objective function evaluations. A faster convergence is observed with the proposed regularisation.