Akash Doshi and Manan Gupta Receive Qualcomm Innovation Fellowship

Tuesday, July 06, 2021

WNCG students Akash Doshi and Manan Gupta have been named winners in the 2021 Qualcomm Innovation Fellowship (QIF) North America Program. Doshi and Gupta proposed the project “Federated Generative Learning for Channel Estimation in mmWave and THz systems.” The pair was one of 16 winners chosen from 43 teams of finalists.

“We are extremely honored to have been selected as the recipients of this prestigious fellowship,” Gupta said. “We look forward to working with Qualcomm to develop impactful and innovative wireless technology.”

The QIF started in 2009 to “cultivate new and forward-thinking ideas and continue to further research and development.” Ph.D. students from select schools apply by submitting a research proposal on an innovative idea of their choosing. Winners earn a one-year fellowship and are mentored by Qualcomm engineers to facilitate the success of the proposed research. The program receives over 100 proposals each year.

Doshi and Gupta crafted their proposal out of an interest in addressing key issues for 6G and beyond.

“Utilizing generative models for mmWave and THz channel estimation is a very exciting area at the intersection of wireless communication and machine learning,” Doshi commented. “It has the potential to fundamentally change the way channel state information is fedback and employed to improve the link quality and throughput.”

Future communication systems may operate at a carrier frequency in the range of 100-300 GHz with over 10,000 cross-polarized antenna elements at each transceiver to achieve a range on the order of 50-100 meters. These future architectures could easily have more than a million combined spatial and frequency dimensions. The project aims to develop novel, scalable, and high-performing approaches to channel estimation for these ultra high dimensional communication systems using deep generative models (DGMs). The team intends to create new tools appropriate for channel estimation, which are (i) resilient to measurement errors and outliers, shifting channel distributions, and low-resolution (e.g. 1 bit) measurements; (ii) adapt to changes in data distribution via federated GAN training.

Akash Doshi and Manan Gupta are both Ph.D. students and Graduate Research Assistants at WNCG. They are advised by Prof. Jeffrey Andrews.