A neural network approach to smooth calibrated data corrupted from switching errors

Abstract

A reliable calibration of the vector network analyzer is needed in order to characterize microwave devices. For some VNAs, such as HP8510, solving the 8 error model is not enough to accurately compute the device under test (DUT) S-parameters, but also, a correction of the switching errors inherent to the measurement must be performed. In this paper, a neural network is used to find the S-parameters of the DUT free from switching errors instead of calculating the well-known equations used for that matter.