News

2021

“Distributed Link Sparsification for Scalable Scheduling Using Graph Neural Networks” submitted to ICASSP 2022.

3 minute read

Published:

If you ever being in a crowded café or stadium, you may have experienced slow or broken Internet even with full signal on your smartphone or laptop. This phenomenon boils down to a small scheduling overhead for every connection to the network. This overhead is not a big deal in wired networks, but for wireless networks, the per-connection overhead is typically proportional to the number of devices sharing the medium. That means you can not get $1/N$ of the total bandwidth of wireless networks, where $N$ is the number of active wireless devices. In fact, if you keep increasing $N$, the overhead would grow and eventually take up all the resources, leaving a network congested over-the-air. The network needs to limit the maximum nubmer of admitted users to prevent this from happening. That’s a big problem.

Now, imagine that there will be 10 million wireless connections per $km^2$ by 2030, mostly from Internet-of-Things (IoT) devices. In that scenario, nothing could get through the network even if their total traffic demand is theoretically well below the bandwidth of the network. This is one of the biggest challenges for 5G and beyond wireless networks, termed as ‘massive access’ or ‘massive connectivity’. (ref. [Chen 2021])

In this work, we come up with a scalable scheduling scheme with reduced overhead for wireless multihop networks. Generally, our solution can reduce the overhead by two orders of magnitude while retain almost $70\%$ of the network capacity, and is applicable to both synchronized and random access networks. Wireless multihop networks are infrastructureless, self-organizing wireless networks, which have been used for wireless networks in harsh/hostile environments, e.g. military communications, disaster relief, environmental and border monitoring. In the future wireless ecosystem, wireless multihop networks will also play a bigger role, e.g., to support IoT, connected vehicles, drones, wireless backhaul for 5G and beyond (small cells, mmWave, satellite constellation). (ref. [Akyildiz 2022])

Our manuscript entitled “Link Scheduling Using Graph Neural Networks” is submitted to IEEE Journal on Selected Topics in Signal Processing.

less than 1 minute read

Published:

In this work arXiv, we address the link scheduling problem in wireless multi-hop networks with orthogonal access, by incorporating machine learning over graphs into conventional algorithmic frameworks. Our work proposes both centralized and distributed solutions, which can improve the bandwidth efficiency of wireless multi-hop networks at low computational and communication costs.

Our paper “Distributed Scheduling using Graph Neural Networks” is accepted by IEEE ICASSP 2021

less than 1 minute read

Published:

In our latest paper “Distributed Scheduling using Graph Neural Networks” accepted by IEEE ICASSP 2021, we augment the distributed greedy scheduler with topology-aware node embeddings generated by Graph Convolutional Networks. Our approach can close the sub-optimality gap by half with minimal increase in the local communication complexity (as low as only one additional round of message passing). The deployment of GCN can be distributed while the training is centralized.

Our manuscript “Deep-Waveform: A Learned OFDM Receiver Based on Deep Complex Convolutional Networks” is revised and submitted to IEEE JSAC

less than 1 minute read

Published:

“Deep-Waveform: A Learned OFDM Receiver Based on Deep Complex Convolutional Networks” is revised and submitted to IEEE Journal on Selected Areas in Communications. The revised version includes significant improvement in the design, training approach, and performance of the complex-valued deep neural networks.

2020

Our paper “A City-Wide Experimental Testbed for The Next Generation Wireless Networks” is published on Journal of Ad Hoc Networks

less than 1 minute read

Published:

Our journal paper “A city-wide experimental testbed for the next generation wireless networks” is published in the journal of Ad Hoc Networks. In this paper, we present a city-wide wireless testbed providing researchers and students with realistic radio environments, standardized experimental configurations, reusable datasets, and advanced computational resources. The testbed contains 5 cognitive radio sites deployed on two campuses of the University of Nebraska-Lincoln and a public street in the city of Lincoln, Nebraska. The testbed is equipped with flagship software-defined radio transceivers, over-the-air and underground antenna arrays, and 20Gbps fronthaul connectivity to cloud facility on campus, and provides remote access to sandbox and live setups of wireless experiments. The development is in collaboration with departments of the university, city of Lincoln, and industrial partners. The goal of this testbed is to improve the accessibility and reproducability of wireless experiments in dynamic spectrum access, 5G/6G, vehicular networks, underground wireless communications, and radio frequency machine learning.

Nebraska Experimental Testbed of Things, known as NEXTT, named finalist for NSF program to expand rural broadband

less than 1 minute read

Published:

A team of researchers and wireless technology experts from the University of Nebraska–Lincoln and the city of Lincoln, along with community, industry and university partners, has been selected as a finalist to lead a prestigious National Science Foundation research program focused on studying novel ways to reduce the cost of broadband delivery to rural communities.

2019

2018

2017

Lincoln Named Smart Gigabit Community

less than 1 minute read

Published:

This morning, Lincoln Mayor Chris Beutler announced Lincoln’s new designation as a Smart Gigabit Community (SGC), making it part of a network of SGCs connected by shared high-speed technology infrastructure. Dec. 11, 2017