Abstract: Prediction of wireless coverage is a key task for network operators. Over the years, many approaches, from measurement campaigns, to simple models based on, e.g., Hata-Okumura models, to (more recently) ray tracing have been used. In recent years, the emergence of powerful machine learning tools has opened a new frontier. RadioMaps use training from ray tracing and measurements to create a fast and efficient way to predict pathloss based on geographical information, including building structure, vegetation, and so on. In this talk we will first discuss the fundamental propagation physics, and describe how they impact the spatial scales on which changes might happen, depending on the operating frequencies. We will also describe some characteristics of ray tracers that might influence accuracy. We will then describe a recent ML structure we call PMNet, as an example for a prediction tool. We will  then finally discuss the training requirements, including data augmentation techniques and transfer learning, which allow a generalization of the predictions.

Bio: Andy Molisch received his degrees (Dipl.Ing. 1990, PhD 1994, Habilitation 1999) from the Technical University Vienna, Austria. He spent the next 10 years in industry, at FTW, AT&T (Bell) Laboratories, and Mitsubishi Electric Research Labs (where he rose to Chief Wireless Standards Architect). In 2009 he joined the University of Southern California (USC) in Los Angeles, CA, as Professor, and founded the Wireless Devices and Systems (WiDeS) group. In 2017, he was appointed to the Solomon Golomb – Andrew and Erna Viterbi Chair.

His research interests revolve around wireless propagation channels, wireless systems design, and their interaction. Recently, his main interests have been wireless channel measurement and modeling for 5G and beyond 5G systems, joint communication-caching-computation, hybrid beamforming, UWB/TOA based localization, and novel modulation/multiple access methods. Overall, he has published 5 books (among them the textbook “Wireless Communications”, third edition in 2023), 22 book chapters, 300 journal papers, and 400 conference papers. He is also the inventor of 70 granted (and more than 10 pending) patents, and co-author of some 70 standards contributions. His work has been cited more than 60,000 times, his h-index is >100, and he is a Clarivate Highly Cited Researcher.

Dr. Molisch has been an Editor of a number of journals and special issues, General Chair, Technical Program Committee Chair, or Symposium Chair of multiple international conferences, as well as Chairperson of various international standardization groups. He is a Fellow of the National Academy of Inventors, Fellow of the AAAS, Fellow of the IEEE, Fellow of the IET, an IEEE Distinguished Lecturer, and a member of the Austrian Academy of Sciences. He has received numerous awards, among them the IET Achievement Medal, the Technical Achievement Awards of IEEE Vehicular Technology Society (Evans Avant-Garde Award) and the IEEE Communications Society (Edwin Howard Armstrong Award), and the Technical Field Award of the IEEE for Communications, the Eric Sumner Award.

Abstract: To enable new applications such as integrated sensing and communication, high precision localization, and augmented reality will require precise and rapid characterization of the radio propagation environment. Conventional high-precision methods usually revolve around some form of ray tracing (RT) approaches, which are computationally demanding and require a precise 3D map of objects in the scenario and their dielectric properties. While the computational aspect has in recent years been alleviated with parallel computing and GPUs, the inability to represent poorly described or unseen environments is a particular pain point for RT. Recent research has shown that propagation-tailored deep neural networks can approximate complex propagation models accurately and rapidly. More importantly, they are able to effectively transfer the learned propagation characteristics to previously unseen environments. 

This talk will first describe the radio map generation in the broader context of Predictive Quality of Service (PQoS), which is a set of learning based techniques that help the network and users predict the QoS behavior, as opposed to reacting to an observed QoS, as is the case in the current networks. Then, an overview of the relevant state of art in learning-based (static) radio map generation will be given. Having the ability to predict (and react upon) the changes in propagation environment (i.e., building a dynamic radio map) will be essential to providing high reliability links. To this end, the talk will delve into the topic of dynamic radio map generation via blockage prediction. Finally, the talk will conclude with a summary of open issues in the area of static and dynamic radio map generation.

Bio: Mate Boban received the Diploma degree in informatics from the University of Zagreb, Croatia, and the Ph.D. degree in electrical and computer engineering from Carnegie Mellon University, Pittsburgh, PA, USA, in 2004 and 2012, respectively. He is a Principal Research Engineer with Huawei Munich Research Center, Germany. Before Huawei, he was with NEC Laboratories Europe, Carnegie Mellon University, and Apple. He is an alumni of the Fulbright Scholar program. He has co-chaired several IEEE workshops and conferences and has been involved in European Union funded projects (5G-CAR, DRIVE-C2X, and TEAM) as a Work Package Leader and editor of deliverables. He is actively involved in key industry and standardization bodies dealing with V2X: 3GPP, 5GAA, and ETSI. His current research interests include resource allocation, machine learning applied to wireless communication systems (in particular, V2X), and channel modeling. He coauthored three papers that received the Best Paper Award, at IEEE VTC Spring 2014, IEEE VNC 2014, and EuCAP 2019.