AIGC for Wireless Data: The Case of RFID-Based Human Activity Recognition

Although great advances have been made in machine learning (ML) based wireless communications and networking, the performance of most ML-based schemes is heavily dependent on the availability of large amounts of high quality radio frequency (RF) data, which are more challenging and costly to obtain than other forms of data. To address this challenge, we propose to leverage diffusion models to generate high quality RF data, and develop a novel lightweight AIGC model for RF sensing, termed RFID-ACCDM (Activity Class Conditional Diffusion Model). RFID-ACCDM can synthesize large amounts of RF data at low cost, conditioned on a particular activity class. The high quality of RFID-ACCDM generated data is demonstrated by metrics of Structural Similarity Index (SSIM) and Frechet Inception Distance (FID), as well as a representative downstream task of human activity recognition (HAR), where the model trained with sufficient synthesized data outperforms the model trained by real data.