报告题目: Limits on time synchronization over networks
报告人：田玉平教授
报告时间：2019年11月27日（星期三）下午3:00-4:30
报告地点：杭州电子科技大学文一人工智能研究院一楼报告厅
报告人简介：田玉平，1986年清华大学自动化系本科毕业，获学士学位；1991年莫斯科动力学院自动化与计算机技术系研究生毕业，获前苏联自动控制专业技术科学副博士学位(PhD)；1996年获俄罗斯自动控制专业技术科学博士学位(ScD)。1992-2018年，在东南大学自动化学院任博士后、副教授（自1994）、教授（1996）、博士生导师（自1998）、复杂系统控制研究所所长，教育部“复杂工程系统测量与控制”重点实验室副主任。2018年12月至今，任杭州电子科技大学自动化学院（人工智能学院）教授，博士生导师。曾受邀在美国加州大学伯克利分校、澳大利亚中昆士兰大学、香港城市大学任访问教授。2001年获教育部“长江学者奖励计划”特聘教授，2004年获国家杰出青年基金。
报告摘要：While the world is connected by different networks, many physical devices, such as computational units, sensors and actuators, are required to work cooperatively over distances with a synchronized clock. However, delays causing by time sensing and information transmission make the accurate clock synchronization impossible. Are there any limits on time synchronization over networks with random delays? We firstly present a unified structural model of consensus-based time synchronization (CBTS) algorithms. Then, we fit different but famous time synchronization algorithms, such as DCTS, ATS, WMTS etc., into the same structural model with just different algorithm parameters. By studying the asymptotic limits of the relative drift estimation error of CBTS algorithms, we discover almost sure and mean-square divergence conditions of CBTS algorithms. Using these conditions we clearly point out that the most existing time synchronization algorithms are divergent, i.e., error-accumulated, either almost surely, or in the mean-square sense. Finally, we provide a fundamental guideline for designing future time synchronization algorithms without error accumulation. Both numerical simulation and physical experiment results are presented to verify the theory.