Collaborative Work Pairs Wireless Networking, Machine Learning Experts to Improve Upon 5G, 6G Performance


Illinois CS professor Deepak Vasisht’s previous work indicated more could be done to improve upon a problem found in 5G performance. He then connected with fellow professor Gagandeep Singh and PhD student Zikun Liu to produce a machine learning based system called “FIRE.”

Written by

Two years ago, Illinois Computer Science professors Deepak Vasisht and Gagandeep Singh along with first-year PhD student Zikun Liu began collaborating to solve a critical bottleneck hindering the performance of 5G/6G wireless systems.

Gagandeep Singh in a floral print shirt (left) and Deepak Vasisht in a blue suit.
Gagandeep Singh (left) and Deepak Vasisht

Concretely, they focused on MIMO – or multiple-input and multiple-output – which is a method for multiplying the capacity of a radio link using multiple transmission and receiving antennas to exploit multipath propagation. MIMO is an essential component for 5G, because of its ability to improve the quality of service and support multiple data streams simultaneously. However, for real-world MIMO deployments, there remains a critical bottleneck – estimating the downlink wireless channel from each antenna on the base station to every client device.

This requirement places a big limitation on what 5G and 6G systems can achieve. In fact, up to half of the network throughput can be wasted due to this requirement. The team invented a new technology to eliminates this channel feedback requirement.

Between Vasisht and Liu’s interest in wireless networking and Singh’s interest in artificial intelligence (AI) and machine learning, the three began working on addressing this pressing issue with an end-to-end data-driven approach

“I’ve worked on this problem in the past, resulting in a paper basically doing the same kind of thing – but using a signal processing approach,” Vasisht said. “We found that the accuracy using signal processing approach was good, but not perfect. This means we could not fully utilize multi-antenna systems with our approach.”

Together, the team designed a new approach to model the core wireless signal propagation process based on generative modeling. Wireless signals, from 5G or 6G base stations, travel through their physical environment and experience complex interactions with the surrounding objects like buildings, roads, and cars.

These complex interactions make this problem challenging and hard to model using traditional discriminative models. The approach designed by the team mimics how wireless signals are changed by obstacles that they encounter in the physical world. By doing so, the team realized, they can remove a crucial bottleneck that is responsible for a huge amount of spectrum waste.

Their work has resulted in a published paper, entitled “FIRE: enabling reciprocity for FDD MIMO systems.” It produces a system called “FIRE,” which serves as an “end-to-end machine learning approach to enable accurate channel estimation without requiring any feedback from client devices.”

The group will present the paper at the ACM 2021 Annual International Conference on Mobile Computing and Networking (MobiCom) in New Orleans at the end of March.

PhD student Zikun Liu enjoyed working with both professors Vasisht and Singh on a project that directly tied in with his research interest of wireless networking.
PhD student Zikun Liu enjoyed working with both professors Vasisht and Singh on a project that directly tied in with his research interest of wireless networking.

“It was remarkable the way this group’s efforts came together,” Vasisht said. “COVID restrictions kept Zikun out of the US, but he ended up working with Gagandeep at ETH Zurich where he was finishing up as a PhD student. When we realized that our interests intersected the way they do, we discussed some common topics of interest and how we might work together.”

“After Zikun introduced himself, we tried to have him serve his internship in a way that would include work he was familiar with and passionate about,” Singh said. “During his three months at ETH Zurich, we wanted him to do something that he could build upon as a PhD student here at Illinois CS.”

The opportunity to work on this project spoke to Liu’s exact interests in CS research and served as a good embodiment of why he chose to conduct his PhD studies here.

“I decided to pursue my doctorate at UIUC because I really like the research environment here, and I saw a group of amazing professors and enthusiastic students during a previous summer internship,” Liu said. “With the rapid development of 5G and 6G networks, wireless networking is an important and interesting topic. Techniques to improve the area are benefiting our lives every day. My advisor, professor Vasisht, and I found a lot of intriguing topics under this context, and we believe that exploring this area is important to make our lives more convenient.”

This presented an exciting opportunity to Singh, who said this project represented an opportunity to apply his methods in a new direction.

“This project allows me to show what carefully designed machine learning models can do in an end-to-end setting for improving real-world systems. That’s what made me think about how exciting this is and how much more impact my work can potentially make.”

Even before the group can present their paper at MobiCom, it has made an impression within the industry.

“5G is likely to change the wireless experience. However, wireless network overheads, such as estimating link quality, consume valuable airtime – both in terms of capacity and latency,” said Ranveer Chandra, Managing Director for Research for Industry at Microsoft. “The UIUC FIRE system is an innovative use of AI and signal processing to accurately estimate downlink link quality, and significantly improve the experience over 5G systems. This paper shows the potential of AI in transforming the design of future wireless systems.”

“Our success on this effort is, in part, due to Illinois CS being a very encouraging space for us to collaborate,” Vasisht said. “It’s amazing to work with someone like Gagandeep on machine learning, and I’ve also thought that Zikun has done a wonderful job. As a student, it’s quite challenging to have two advisors.

“I’m thrilled we now have an opportunity to present our work at MobiCom to see what this community thinks about where we can take it from here.”

Share this story

This story was published March 24, 2022.