Optimum design of a novel Ku-band rasorber for RADAR warfare systems using ML neural network

Abstract

The design, fabrication, and testing of a conformal Frequency Selective Rasorber (FSR) operating at A-T-A mode is designed for RADAR warfare systems. In the design of a miniaturized single-layer FSR element, multiple parameters influencing the absorption and transmission characteristics need to be optimized. This simulation using conventional methods consumes more simulation time. Thus, the geometrical parameters are optimized and predicted using supervised machine learning (ML) techniques to expedite the process. The multiple output regression neural network (MORNN) is used to generate multiple input and output features from the dataset. The ML algorithm is trained using the datasets generated from the electromagnetic solver using which a scalable FSR is synthesized. The reflection coefficient, (|S₁₁|), and transmission coefficient (|S₂₁|) are used as input data, and the dimension of the FSR unit cell for the user input frequency requirements are derived as output data. The extracted dimensions of the FSR offered a small mean square error (MSE) of 0.02 between the desired and observed results. The designed FSR offers absorption at 11.5 GHz and 18.9 GHz while the transmission window extends from 15.02 GHz to 16.09 GHz. The neural network results are endorsed using the EM simulation tool and validated by experimental measurements.