Adaptive Semantic Video Conferencing for Ofdm Systems

ABSTRACT

Video conferencing has become more common than ever due to the COVID-19 pandemic, which makes high-resolution video transmission a pressing issue. Although semantic video conferencing (SVC) has achieved a great success to improve the transmission efficiency by only transmitting some keypoints to represent changed expressions, its performance can still be improved by adapting to varying channel scenarios, which is lack of study when designing the whole SVC in the end-to-end manner. In this paper, we first establish a standard SVC-OFDM system. Then, the receiver part of the SVC is added with an adaptive network called Switch-SVC for varying channels and improve the accuracy of the received keypoints. Some parameters in Switch-SVC are trained online so that the receiver can adapt to the current environment. Simulation results show that the proposed method can greatly improve the keypoint reconstruction performance compared to the traditional SVC-OFDM receiver without online training.