Image-mmWave radio frequency domain translation with generative models for gesture recognition

In wireless communication and radar systems, Multi-Path Components (MPCs) play a crucial role as they represent different signal propagation paths. However, obtaining real MPC data is both difficult and costly. To address this challenge, our research explores the use of Generative Adversarial Networks (GANs) to generate MPC data. Our primary goal is to mitigate data scarcity and improve diversity in the MPC field. By utilizing 3D human skeletal keypoints data, we can create a diverse range of MPC data. This approach not only enables cost-effective data augmentation but also advances the capabilities of wireless communication and radar systems. Throughout our work, we faced several challenges, including variability in data points per frame after filtering MPC data, and the complexities of designing and training the GAN model. The GAN method also posed specific challenges, such as the replication of data points, which led to an overpowered discriminator. Despite these obstacles, our research has made significant progress, and we’ve identified key areas for future work to further enhance our results