Reactive Near-Field to 3-Meter Far-Field Transformation Based on Deep Convolutional Neural Networks and Plane Wave Spectrum

In response to the pressing demand for accurately transforming reactive near field into 3-meter far field for unknown radiation sources in electromagnetic interference (EMI) testing, this article introduces an innovative hybrid method that integrates a deep convolutional neural network (DCNN) with the plane wave spectrum. By leveraging plane wave expansion, the spatial magnetic near-field data are converted into the spectrum domain, serving as the input for the DCNN model. DCNN’s output is the 3-meter electric far field. In comparison with the method that directly utilizes the spatial magnetic near field as the DCNN input, the proposed method notably improves the generalization of the DCNN model. It enables the output of DCNN (3-meter far field) insensitive to variations in the near-field scanning height. A randomly distributed dipole array is used to generate the pairs of spectral near field and 3-meter far field for training the DCNN. Moreover, the proposed method shows the exceptional robustness against noise interference, even for the signal-to-noise ratio (SNR) is as low as −5 dB. Both numerical and measurement examples demonstrate that the trained DCNN can accurately predict the 3-meter far field based on magnetic reactive near fields acquired at various scanning heights and from different radiation sources.