Multibranch Artificial Neural Network Modeling for Inverse Estimation of Antenna Array Directivity

To speed up the design process of antenna arrays, a new multibranch artificial neural network (ANN) modeling technique is proposed and applied to tackle the inverse problem of antenna array directivity estimation. In the inverse modeling of an antenna array, the model inputs are electromagnetic parameters, whereas geometrical or physical parameters are set as the outputs. Due to the nonuniqueness of the inverse problem, it is difficult to directly use ANN for inverse modeling. We are trying to solve this problem utilizing monotonicity. The directional derivatives of the corresponding forward problem are calculated and the derivatives are utilized to determine the monotonicity of training data. Based on the monotonicity, the training data are divided into several groups, which are, respectively, used for training parallel branches of ANN. In this way, valid outputs can be provided without the forward model. Three dipole antenna arrays are utilized as examples to verify the efficiency of the proposed technique.