ReFormer: Generating Radio Fakes for Data Augmentation

TL;DR

ReFormer introduces a decoder-only transformer framework that generates RF data by operating in a discrete latent space learned with VQVAE, enabling scalable, prompt-driven data augmentation for wireless systems. By autogressively modeling the discrete latent tokens, the approach produces high-fidelity RF fakes that preserve modulation-specific characteristics, with the nano-GPT variant delivering greater diversity and classification relevance than the MONAI alternative. The study demonstrates strong codebook utilization, robust evaluation via TopP&R, and perceptual consonance in I/Q constellations, underscoring the practicality of transformer-based RF data synthesis for training and benchmarking. Overall, ReFormer offers a simple, scalable alternative to diffusion and GAN-based methods for RF data augmentation, with potential extensions to include channel context and broader RF dimensions.

Abstract

We present ReFormer, a generative AI (GAI) model that can efficiently generate synthetic radio-frequency (RF) data, or RF fakes, statistically similar to the data it was trained on, or with modified statistics, in order to augment datasets collected in real-world experiments. For applications like this, adaptability and scalability are important issues. This is why ReFormer leverages transformer-based autoregressive generation, trained on learned discrete representations of RF signals. By using prompts, such GAI can be made to generate the data which complies with specific constraints or conditions, particularly useful for training channel estimation and modeling. It may also leverage the data from a source system to generate training data for a target system. We show how different transformer architectures and other design choices affect the quality of generated RF fakes, evaluated using metrics such as precision and recall, classification accuracy and signal constellation diagrams.

Figures (8)

• Figure 1: ReFormer approach: VQVAE latent token sequences $Z_Q$ are used to train decoder-only transformer (DoT), making it capable to generate fake sequences $\widecheck{Z_Q},$ which are turned into RF fakes via $D(\widecheck{z_q}).$
• Figure 2: Original Signal Constellations.
• Figure 3: VQVAE Reconstructions Constellations.
• Figure 4: Reconstruction examples for GPT based transformer. The I/Q constellations for each class closely resemble the original signals.
• Figure 5: Reconstruction examples for MONAI transformer. GPT based Transformer creates better-quality I/Q constellations.