T1: Learning-based Wireless Positioning and Wireless Sensing: from Meter to Centimeter Precision
Presented by: Kai-Ten Feng (National Chiao Tung Univ.) and Po-Hsuan Tseng (National Taipei Univ. of Technology)
Time: 9:00–12:30
Room: Milo 1
Abstract—This tutorial aims at providing the fundamental limits of wireless positioning, including non-line-of-sight path, multi-path attenuation, lack of map information, and time-varying interferences caused by environmental changes and/or people blocking. We discuss how signal processing and machine/deep learning techniques enhance the positioning performance from meter to centimeter precision. The tutorial focuses on the general positioning/sensing problems using channel state information and received signal strength measurements. The described algorithms/implementations can be directly applied to the wireless localization and sensing for the in-car, roadside, and drone landing applications.
Tutorial Objectives
In the tutorial, we will discuss various types of wireless localization and tracking technologies, including ranging/angle-based, database matching-based, and sensor fusion-based techniques.
Secondly, we will examine the performance limiting factors for wireless positioning based on the theoretical limits. We will take Wi-Fi-based technologies as an example in this tutorial and focus on the techniques related to the received signal strength (RSS) and channel state information (CSI). Based on the multi-path channels, we will explain the problem of non-line-of-sight paths. With the help of map information, the surrounding geometric layout provides further prior information to the position estimation. The other unavoidable time-varying Interferences, which are caused by environmental changes and/or people blocking, will limit the performance for wireless positioning.
On top of the signal processing aspect, we examine how to obtain the time-of-arrival (TOA) and the angle-of-arrival (AOA) using the Wi-Fi system. Based on the distance information obtained from TOA or RSS and the angle information from AOA, several recent proposals which achieve the decimeter-level accuracy will be presented. Moreover, by utilizing the broader bandwidths to enhance the time-resolution, the splicing methods which adopt multiple Wi-Fi channels will be discussed.
Finally, we discuss how machine learning/deep learning approaches overcome the positioning problems that cannot be well-modeled. The channel state information-based fingerprinting methods have been proposed for highly accurate positioning down to centimeter-level. The adoption of clustering techniques with database collection can ease the effect of time-varying channels. With the map information, the skeleton-based location tracking is designed for auto-construction of the walkable region and constrains the possible position at next time instant using the generalized Voronoi diagram. Moreover, we will address the device-free wireless positioning and sensing, which could be utilized for high precision drone landing and presence detection for multiple targets.
Tutorial Outline
1. Introduction of Positioning: State of Art Technologies i. Received Signal Strength vs Channel State Information (CSI) 2. Fundamental Limits of Wireless Positioning i. Multi-path Attenuation ii. Non-line-of-sight Paths iii. Lack of Map Information iv. Time-varying Interferences Caused by Environmental Changes and/or People Blocking 3. Signal Processing Aspects of Positioning Technologies i. Avoiding Multi-path to Revive In-building WiFi Localization – CUPID: RSS/AoA/IMU-based methods ii. Decimeter-level Localization Using WiFi – SpotFi: ToA/AoA-based methods iii. CSI Splicing Methods: Enhancing Time-Resolution using Multiple Wi-Fi Channels 4. Machine/Deep Learning Techniques on Wireless Positioning and Wireless Sensing i. Channel State Information-based Positioning Fingerprinting a. Super-resolution-based CSI Fingerprinting b. Deep Neural Network-based CSI Fingerprinting c. Autoencoder-based CSI Hidden Feature Extraction for Spot Localization ii. Time-varying interferences Caused by Environmental Changes and/or People Blocking iii. Map Information-Assisted: Spatial Skeleton-Enhanced Location Tracking iv. Device-free CSI-based Wireless Localization for High Precision Drone Landing Applications v. Device-Free Multiple Presence Detection using CSI Information with Machine Learning Methods
Primary Audience
This tutorial targets on both academic researchers and industrial engineers who are interested in the topics of wireless positioning and sensing. The described algorithms/implementations can be directly applied to the wireless localization and sensing for the in-car, roadside, and drone landing applications. The covered methodologies include conventional model-based and emerging AI-based approaches, which are related to the graduate/post-graduate in the fields of signal processing and computer science.
Novelty
This tutorial explains the fundamental limits of wireless positioning. This tutorial reviews how recent model-based approaches reach their performance limits by using signal processing methodologies. To deal with the difficulties to model the positioning problems, the design of machine learning/deep learning methods and their corresponding advantages are provided in this tutorial. Moreover, the experimental implementations for wireless localization based on machine learning/deep learning methods and the results of field trials will also be demonstrated.
Biography
Kai-Ten Feng received the B.S. degree from the National Taiwan University, Taipei, Taiwan, in 1992, the M.S. degree from the University of Michigan, Ann Arbor, MI, USA, in 1996, and the Ph.D. degree from the University of California—Berkeley, Berkeley, CA, USA, in 2000.,Since August 2011, he has been a Full Professor with the Department of Electrical and Computer Engineering, National Chiao Tung University (NCTU), Hsinchu, Taiwan, where he was an Associate Professor and Assistant Professor from August 2007 to July 2011 and from February 2003 to July 2007, respectively. He served as the Associated Dean of Electrical and Computer Engineering College, NCTU, starting from February 2017. From July 2009 to March 2010, he was a Visiting Research Fellow with the Department of Electrical and Computer Engineering, University of California at Davis. Between 2000 and 2003, he was an In-Vehicle Development Manager/Senior Technologist with OnStar Corporation, a subsidiary of General Motors Corporation, where he worked on the design of future telematics platforms and in-vehicle networks. His research interests include broadband wireless networks, cooperative and cognitive networks, smartphone and embedded system designs, wireless location technologies, and intelligent transportation systems. Dr. Feng was the recipient of the Best Paper Award from the Spring 2006 IEEE Vehicular Technology Conference, which ranked his paper first among the 615 accepted papers. He was also the recipient of the Outstanding Youth Electrical Engineer Award in 2007 from the Chinese Institute of Electrical Engineering, and the Distinguished Researcher Award from NCTU in 2008, 2010, and 2011. Since 2018, he has been serving as the Technical Advisor for IEEE-HKN Honor Society and National Academy of Engineering Grand Challenges Scholars Program at NCTU. He has also served on the technical program committees in various international conferences.
Po-Hsuan Tseng received the B.S. and Ph.D. degrees in communication engineering from the National Chiao Tung University, Hsinchu, Taiwan, in 2005 and 2011, respectively. Since Feb. 2017, he has been an Associate Professor with the Department of Electronic Engineering, National Taipei University of Technology, Taipei, Taiwan, where he was an Assistant Professor from Aug. 2012 to Jan. 2017. His current research focuses on signal processing for networking and communications, including location estimation and tracking, and mobile broadband access system design