Deep learning based EVM correction for RF receiver of vector signal analyser

Abstract

This Letter presents a novel deep learning approach for optimising the receiver performance with respect to the error vector magnitude (EVM) metric, which was verified and evaluated by applying it to a self-developed proprietary vector signal analyser (VSA). A four-layer neural network was built and trained to estimate and correct the systematic error of the VSA receiver by using a calibrated commercially available vector signal generator as the training source. Experimental results show that the EVM performance of the self-developed VSA is improved and approaches that of a state-of-the-art VSA.