Abstract:Random noise arising from physical processes is an inherent characteristic of measurements and a limiting factor for most signal processing and data analysis tasks. Given the recent interest in generative adversarial networks (GANs) for data-driven modeling, it is important to determine to what extent GANs can faithfully reproduce noise in target data sets. In this paper, we present an empirical investigation that aims to shed light on this issue for time series. Namely, we assess two general-purpose GANs for time series that are based on the popular deep convolutional GAN (DCGAN) architecture, a direct time-series model and an image-based model that uses a short-time Fourier transform (STFT) data representation. The GAN models are trained and quantitatively evaluated using distributions of simulated noise time series with known ground-truth parameters. Target time series distributions include a broad range of noise types commonly encountered in physical measurements, electronics, and communication systems: band-limited thermal noise, power law noise, shot noise, and impulsive noise. We find that GANs are capable of learning many noise types, although they predictably struggle when the GAN architecture is not well suited to some aspects of the noise, e.g., impulsive time-series with extreme outliers. Our findings provide insights into the capabilities and potential limitations of current approaches to time-series GANs and highlight areas for further research. In addition, our battery of tests provides a useful benchmark to aid the development of deep generative models for time series.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is to evaluate the effectiveness of generative adversarial networks (GANs) in simulating time - series noise. Specifically, the author hopes to understand whether current GAN architectures can faithfully reproduce different types of noise, which are common in physical measurements, electronics, and communication systems. Through this research, the author aims to provide a valuable benchmark for the future development of time - series GAN models and reveal the capabilities and potential limitations of existing methods. ### Main problems: 1. **Evaluating the learning ability of GANs for different types of noise**: The paper selects four widely - existing noise types for research, including band - limited thermal noise, power - law noise, shot noise, and impulse noise. These noise types cover stochastic processes from stationary to non - stationary, from Gaussian to non - Gaussian, and with long - memory characteristics. 2. **Selecting appropriate GAN architectures**: In order to evaluate the performance of different GAN architectures, the author selects two general models based on the deep convolutional GAN (DCGAN) architecture: one is WaveGAN, which directly models time - series, and the other is an image - domain model that represents time - series using the short - time Fourier transform (STFT). 3. **Quantitatively evaluating the performance of GANs**: The author trains and evaluates these GAN models by simulating the distribution of real - noise time - series with known parameters. The goal is to verify whether GANs can accurately learn and generate these noise types, especially their performance in extreme cases (such as outliers in impulse noise). ### Research motivation: - **Requirements for real - world applications**: Noise is an important challenge in signal processing and data analysis, especially in physical measurements, electronics, and communication systems. Understanding the capabilities of GANs in simulating noise can help researchers better understand and deal with noise problems in practical application scenarios. - **Promoting the development of GAN technology**: Although GANs have made significant progress in fields such as images and audio, their applications in time - series data are still limited. Through systematic evaluation, it can provide guidance for future research and help improve the design and optimization of GAN models. ### Conclusion: The author finds that GANs can learn multiple noise types to a certain extent, but when the GAN architecture is not adapted to certain noise characteristics (such as extreme outliers in impulse noise), its performance will be affected. These findings not only provide insights into the capabilities of current time - series GAN models but also point out the direction for further research. ### Formula summary: - **Autocovariance function of fractional Gaussian noise (FGN)**: \[ \text{Cov}(Y_t, Y_{t + k})=\frac{\sigma_Y^2}{2}\left(|k - 1|^{2H}+|k + 1|^{2H}-2|k|^{2H}\right) \] where \(H\) is the Hurst index and \(\sigma_Y^2\) is the variance. - **Probability density function (PDF) of Bernoulli - Gaussian (BG) noise**: \[ f(x)=(1 - p)N(x;0,\sigma_w^2)+pN(x;0,\sigma_w^2+\sigma_i^2) \] where \(N(x;\mu,\sigma^2)\) represents the normal distribution PDF with mean \(\mu\) and variance \(\sigma^2\). Through these formulas and detailed experimental designs, the author comprehensively evaluates the ability of GANs in simulating different types of noise and provides a valuable reference for future research.

Data-Driven Modeling of Noise Time Series with Convolutional Generative Adversarial Networks

Directed Chain Generative Adversarial Networks

Generative Adversarial Networks in Time Series: A Systematic Literature Review

Generative adversarial networks in time series: A survey and taxonomy

Time-series Generative Adversarial Networks

Unsupervised Image Noise Modeling with Self-Consistent GAN

Time-series Transformer Generative Adversarial Networks

Noise Dimension of GAN: An Image Compression Perspective

TTS-GAN: A Transformer-based Time-Series Generative Adversarial Network

Survey on Generative Adversarial Behavior in Artificial Neural Tasks

Enhancing Seismic Data Denoising Through Multiscale Analysis Across Transformer and GAN

DLT-GAN: Dual-Layer Transfer Generative Adversarial Network-Based Time Series Data Augmentation Method

Time Series Simulation by Conditional Generative Adversarial Net

Biosignal Generation and Latent Variable Analysis with Recurrent Generative Adversarial Networks

A Survey of Modern Deep Learning based Generative Adversarial Networks (GANs)

Class Conditional Time Series Generation with Structured Noise Space GAN

GAN-Based Noise Model for Denoising Real Images

Evaluating generation of chaotic time series by convolutional generative adversarial networks

A Novel Generative Adversarial Network for the Removal of Noise and Baseline Drift in Seismic Signals

TCGAN: Convolutional Generative Adversarial Network for Time Series Classification and Clustering.