Date：2024/04/02 (二)

Date：14:00~17:00

Venue：電資大樓 工EC2002

Topic 1： RFID-based 3D Human Pose Tracking: Design, Generalization, and Data Augmentation, Speaker: Prof. Shiwen Mao

Topic 2：O-RAN Uplink Performance Improvement for 5G Massive MIMO, Speaker: Prof. Rob Maunder

In the first talk, Prof. Mao introduced the importance of 3D Human Activity Recognition (HAR) in Human-Computer Interaction (HCI), and explored techniques to protect user privacy without the use of video cameras. He first demonstrated the deep learning-based RFID-Pose system, a vision-assisted, real-time 3D human pose estimation system that estimates the spatial rotation angles of each human limb, and utilizes these angles to reconstruct human poses in real-time using forward kinematics. He then discussed the problem of generalization of trained systems when applied to new environments and introduced meta-learning and TARF techniques as solutions, as well as the RFPose-GAN approach to data augmentation using Generative Adversarial Networks (GANs), which aims to reduce the cost of collecting training data.

The second talk was given by Prof. Maunder, who shared his research on "O-RAN Uplink Performance Improvement for 5G Massive MIMO". He provided an in-depth analysis of the major barriers to the adoption of Open RAN in high-performance 5G networks, in particular its lack of support for mMIMO antenna arrays, and detailed the O-RAN consortium's new forwarding protocol and its improvements for mMIMO optimization. He also introduced the O-RAN Alliance's new Uplink Performance Improvement (UPLI) work program and two proposals aimed at improving uplink performance.

These two excellent presentations not only provided in-depth technical analysis, but also brought new research ideas and collaboration opportunities to the students and teachers present.