A highly accurate and robust source reconstruction method of printed circuit boards based on complex-valued neural network

Printed circuit boards play an increasingly important role in modern electronic systems. Equivalent dipole moments are widely used to reconstruct the radiated fields of printed circuit boards in view of the extremely high complexity. In this paper, an improved source reconstruction method based on complex-valued neural network is proposed to investigate the radiation properties of printed circuit boards working in microwave frequency band. Firstly, the magnetic field data obtained from the near-field scanning are utilized to construct a matrix equation based on the source reconstruction theory. Secondly, the kernel of the complex-valued neural network is constructed through the algorithm of complex numbers, which eventually end up with the novel neural network with the help of the principle of regression optimization. Thirdly, the data obtained from the near-field scanning are fed into the neural network for effective training to get the equivalent dipole model. Finally, several numerical examples and available measurement results are given to validate the method. Compared with the commonly used Tikhonov regularization method, the proposed method possesses higher accuracy and better robustness in noise suppression. A novel source reconstruction method of printed circuit boards based on complex-valued neural network is proposed. It exhibits high accuracy and good robustness against noise in prediction of the electromagnetic radiation.image