GANexo: A Novel Framework for Synthetic Exoplanet Signal Generation, Noise Removal, and Characterization Based on Generative Learning

This work presents GANexo v1.0, a framework for synthetic generation, noise reduction, and characterization of exoplanetary transit signals. GANexo applies a generative approach that incorporates both white and red noise, accurately reflecting real-world observational conditions and effectively modeling atmospheric and instrumental effects commonly encountered during ground-based observations of these signals. The framework integrates a noise reduction stage using an autoencoder, applying control variables from this component. Finally, a characterization phase is implemented through a regression model based on conditional generative adversarial networks (cGANs), demonstrating significant improvements over other generative approaches such as traditional 1D GAN and Wasserstein GANs (wGANs). The results suggest a precise alternative approach for these tasks using generative learning methods.