Data augmentation using a generative adversarial network for a high-precision instantaneous microwave frequency measurement system

Abstract

In this Letter, an unsupervised-learning platform—generative adversarial network (GAN)—is proposed for experimental data augmentation in a deep-learning assisted photonic-based instantaneous microwave frequency measurement (IFM) system. Only 75 sets of experimental data are required and the GAN can augment the small amount of data into 5000 sets of data for training the deep learning model. Furthermore, frequency measurement error of the estimated frequency has improved by an order of magnitude from 50 MHz to 5 MHz. The proposed use of GAN effectively reduces the amount of experimental data needed by 98.75% and reduces measurement error by 10 times.

Adaptive microwave photonic angle-of-arrival estimation based on BiGRU-CNN [Invited]

“…The number of data required is even huger. We can reduce training cost and alleviate the reliance on large numbers of data through transfer learning [19,20] or data augmentation via a generative adversarial network [21] .…”

Photonics and machine learning: a powerful partnership for sensing

“…1(c). Using data argumentation based on GAN [5], the same frequency measurement system results in a significant improvement in accuracy, as shown in Fig. 1(d).…”

“…[4]. To benefit from machine learning's power, we proposed and demonstrated the use of generative adversarial network (GAN) for data argumentation in microwave photonics systems [5]. GAN is an effective solution for data argumentation [6]- [7], especially for photonic systems that are mainly based on hardware.…”