A new efficient approach for fast and accurate design of frequency selective surfaces based on geometry estimation networks

作者：Moinzad Sara^*; Abdolali Ali

来源：Neural Computing & Applications, 2017, 28(9): 2727-2734.

DOI：10.1007/s00521-016-2221-z

• Artificial neural network
• Back propagation
• Feed-forward neural networks
• Frequency selective surfaces
• Spatial filters
• Square loop structures

Artificial neural networks are a favorite modern tool for high accuracy designing frequency selective surfaces (FSSs) in RF and microwave engineering field. In this paper, a new fast and precise ANN-based algorithm for designing FSSs is presented. This algorithm, unless the previous works, can develop the structures with due attention to features of incident waves and improve the applicability of developed FSSs. For achieving this algorithm, at first, a new method is presented for the better preparation of training datasets, called frequency sweep method (FSM). The advantage of FSM is to reduce the size of training datasets and prevent from superfluous simulations. So the time needed for the preparation of training datasets and to train the networks is less than before. Following that, FSM is used to train geometry estimation ANN (GEANN) with primary goal of FSSs design in little time and without any optimization algorithm. The proposed design procedure is complete design and analysis unit that consisted of a sequence of GEANN and traditional response calculation ANNs (RCANNs). GEANN is used to estimate geometric dimensions of FSSs with desired incident wave, and RCANNs are used to calculate the frequency response of FSSs under other various incident waves. The results show that required time for designing FSS is less than 30 ms, and errors are <1 %. Both analytical and experimental results confirm the correctness of predicted values.

• 出版日期2017-9