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

Area of Science:

• Photonics
• Machine Learning
• Signal Processing

Background:

• Photonic-based instantaneous microwave frequency measurement (IFM) systems often require substantial experimental data for training deep learning models.
• Data acquisition can be time-consuming and resource-intensive, limiting the practical application of these systems.

Purpose of the Study:

• To propose an unsupervised learning platform, specifically a generative adversarial network (GAN), for augmenting experimental data in deep learning-assisted IFM systems.
• To demonstrate the effectiveness of GANs in reducing the need for extensive experimental datasets.
• To improve the accuracy of frequency measurements in IFM systems.

Main Methods:

• An unsupervised learning platform, a generative adversarial network (GAN), was employed for data augmentation.
• A small dataset of 75 experimental data points was augmented to 5000 data points using the GAN.
• The augmented dataset was used to train a deep learning model for microwave frequency measurement.

Main Results:

• The GAN successfully augmented a small dataset (75 samples) into a significantly larger dataset (5000 samples).
• The use of GAN-based data augmentation reduced the required experimental data by 98.75%.
• Frequency measurement error improved by an order of magnitude, decreasing from 50 MHz to 5 MHz, a tenfold reduction.

Conclusions:

• Generative adversarial networks (GANs) are highly effective for experimental data augmentation in photonic-based IFM systems.
• GANs substantially reduce the experimental data requirements for training deep learning models, making IFM systems more accessible.
• The proposed method significantly enhances measurement accuracy, demonstrating a practical advancement in microwave frequency measurement technology.