Automated FSS Design and Optimization with Time Series Forecasting Process Through Combined CNN-RNN Model

In this work, an automated-oriented methodology in combination with intelligent technique is presented, leading to design and optimize a frequency selective surfaces (FSS) geometry. The FSS configurations include periodic unit cells of which designs iterative cycles of simulations are usually required. For this case, firstly we present a procedure for automatically configure the FSS structure and afterward, a neural network that is the combination of convolutional neural network (CNN) and recurrent neural network (RNN) is employed for its optimization. The proposed methodology leads to automatically configure the FSS structure, and to predict the performances of the generated design in at specific frequencies within the initial frequency band. The modeling process is executed considering the combination of electronic design automation (EDA) tool as CST studio suite, and numerical analyzer as MATLAB. The effectiveness of the proposed method is validated by designing an FSS operating as multi-band device in the 6.2-6.4 GHz, 7.9-8.4 GHz, and 10.7- 11.4 GHz frequency bands. Finally, a prediction of the input scatting parameter of the optimized structure obtained through CNN-RNN model is performed.