CIF: Small: Machine Learning for Wireless Propagation Channels

点击翻译按钮获取中文摘要

Wireless communications is an essential part of modern society, and with the introduction of 5G, will become even more important. One of the key challenges of wireless communications is that the signals are distorted on their way from the transmitter to the receiver, i.e., by the propagation channel. In order to make communications efficient, the system needs to know and correct these distortions at the receiver. The current way of getting this information is through sending special, known signals, from which the distortion information can be extracted. However this is highly inefficient as useful (payload) data cannot be sent during this time. These challenges are further exacerbated if the distortion information needs to be known at the transmitter, or when prediction of what the distortions will be in the future is needed, or in a different frequency band, or at different locations. This project explores the use of powerful identification and prediction techniques in Machine Learning (ML) to make such predictions. The project team will combine knowledge about the physics of propagation channels and extensive measurement data to train the ML algorithms together with novel ML training approaches to tackle this problem.Importantly, the structure of wireless channels is very different from structures of other types of data (images, documents, etc.) to which ML has been applied. Thus application of ML to wireless channels cannot simply apply modern training algorithms (e.g. deep learning) known from image classification and other well-known ML applications, but rather requires incorporation of the special properties of wireless propagation and suitable rethinking of the ML approaches. The project will thus develop new methodologies, and provide new insights, both from the perspective of ML and that of propagation channels and their impact on system design. The project will also establish an extensive database of channel measurement and ray tracing data that will be used to train and evaluate the proposed algorithms which will be made available to other US researchers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.