Generative Adversarial Network for Data Augmentation in Photonic-based Microwave Frequency Measurement

Abstract

Deep learning is a powerful tool for enhancing performance and increasing the functionalities of a system. However, it is challenging to use deep learning to enhance hardware-based photonic systems because a large dataset that covers the whole operation range of each device is needed for achieving an accurate model. However, not all devices in a system can be controlled automatically, making the data collection process challenging and time consuming. In this letter, we use an instantaneous microwave frequency measurement (IFM) system to demonstrate the use of generative adversarial network (GAN) in deep learning platform for data augmentation. With GAN, only 75 sets of experimental data are needed to collect manually from the IFM system. The GAN augments the 75 sets of experimental data into 5000 sets of data for training the model, effectively reduces the amount of experimental data needed by 98.75%, and reduces frequency estimation error by 10 times.