Drone technology is expected to become a $100 Billion market by 2020. The ripples of their increasing popularity can be observed in both industry and government. For instance, the spectrum issues related to drone communications are now starting to be considered by the Federal Communications Commission (FCC) and industry. While the debate on which one to choose has already begun and is expected to go on for several years, many fundamental questions remain open, in particular the reliability of wireless connections between ground stations and drones, which is critical for both safety and applications (e.g., streaming of video from a forest fire), as well as for the topic of coexistence between a drone and terrestrial networks (DroTerNets). The Wireless Devices and Systems (WiDeS) group led by Prof. Molisch has an ongoing research project funded by NSF to investigate these aspects in detail.
(NSF Grant Number CNS-1923601)
Caching at the wireless edge is a promising method for improving the spectral efficiency and quality of video streaming. Caching strategies usually require knowledge of the video demands, as well as the historical preferences of the individual users. This information needs to be combined with information about the wireless network structure and state to develop deterministic or stochastic caching policies. Yet, in many cases it is undesirable to share detailed user proles with the cache and/or wireless network operator. The main motivation for avoiding such sharing is user privacy; furthermore video streaming providers do not want to share user popularity since this is important condential business information. A promising way out of this dilemma is federated learning (FL), which allows localized training and exchange of machine learning models. The current project thus investigates the advanced federated learning technique for an intelligent and privacy-preserving video caching and computing system and to novel federated learning techniques that preserve user privacy while retaining eciency of caching.
Specifically, the goals are to:
(NSF Grant Number CNS-1923807)
Real-time interactive (RTI) services are anticipated to be the "next big thing" in information technology, merging the areas of sensing, computation, and communications that until recently have been treated separately. In traditional systems, these building blocks are not optimized to work with each other, and particularly not in real-time, constraining the type of services that can be offered. RTIs require real-time aggregation of distributed data streams onto edge/cloud compute servers that can process data as soon as it is generated. The goal of this project is to develop a general mathematical framework as well as concrete algorithms to provide such RTI services with guaranteed latencies. The results benefit the US economy by enabling more efficient, more reliable, and more resilient automation schemes, e.g., for smart factories and farms, as well as improved augmented/virtual reality. The interdisciplinary nature of the project benefits the students working on this project; a detailed plan for outreach and increasing participation of underrepresented groups strengthens the impact of the work.
The project designs a general mathematical framework and concrete algorithms for reliable and resilient Real-time interactive (RTI) systems. In a first thrust, the project explores the limits of delivering latency-critical services reliably. The establishment of the reliable network stability region provides bounds on how much offered RTI traffic a network can support under the best of all algorithms. A packet-lifetime-based queuing model serves as the basis of Lyapunov-drift-inspired control algorithms that not only account for the packet delay but also takes cost such as energy consumption into account. A second thrust of the project develops robust NextG network control algorithms that can adapt to dynamic network states. Integration of the approaches in the first two thrusts under a common Lyapunov drift control framework is an avenue to enable reliable RTI service delivery, which is verified by the implementation of the algorithm on a Platform for Advanced Wireless Research (PAWR) community testbed.
When the frequency of electromagnetic radiation matches the energy gap between different atomic or molecular energy states, its absorption can bring the atom/molecule into a higher (excited) state. Such radiation is thus called resonance radiation. Radiation trapping describes the interaction of this resonance interaction with an ensemble of atoms or molecules, e.g., in a gas (or vapor). Assume an externally created photon, with a wavelength matching a resonant atomic transition, is incident on the gas and gets absorbed, exciting an atom to a higher state. Due to natural decay, the photon is reemitted after some time, but – since its wavelength still matches the atomic resonance, the probability is high that is re-absorbed by a nearby atom, re-emitted after some time, absorbed by yet another atom, and so on – until it finally escapes from the gas.
The radiation trapping process has important consequences for the properties of the resonance radiation emerging from the gas. Firstly, the lineshape is distorted: since photons at the center frequency of the absorption line “see” a high absorption coefficient, the probability of reaching the detector is low, while photons in the “wings” of the lineshape can escape more easily. Secondly, the emerging radiation is suffering from both delay dispersion, frequency dispersion (the reemitted frequencies are different from the absorbed frequencies), and spatial dispersion (photons can be re-emitted into any direction, though there can be a nontrivial relationship between directional dispersion and frequency dispersion).
As modern wireless systems are moving to higher and higher frequencies, there are more situations where the operating frequency matches atomic or molecular transitions. This is true for THz signals that mostly interact with water vapor, as well as free-space optical communications in the infrared or visible spectrum, which might interact with a variety of molecules or atoms, from CO2 to various pollutants. These systems might be used for communication, sensing, or both. The project aims to provide an in-depth investigation of both the fundamentals of radiation trapping and its effects on next-generation wireless communications and sensing.
The goal of this project is the development of new approaches to wireless channel prediction by means of machine learning. This prediction can be in time, space, or frequency, and serves to enhance the efficiency and reliability of wireless communications. The special structure and the physics of wireless propagation need to be taken into account in order to optimize performance of such channel predictions; consequently "standard" machine learning methods known, e.g., from image classification and processing, cannot be directly applied.
Particular topics of investigation are:
The goal is not only to investigate the quality of the predictions, but also the impact of those predictions on system operation.