Abhishek Singh will present his FPO "Physics-inspired Computation for Next-Generation Wireless Systems: from Theory to Practice" on Tuesday, August 26, 2025 in Friend 108.
The members of his committee are as follows:
Examiners: Kyle Jamieson (adviser), Wyatt Lloyd, and Yasaman Ghasempour
Readers: Margaret Martonosi and Peter McMahon (Cornell University)
Title: Physics-inspired Computation for Next-Generation Wireless Systems: from Theory to Practice
Wireless communication technologies have become an integral part of daily life. Whether it is video conferencing for work meetings or high-quality video streaming for leisure, there is an ever-increasing demand for high-speed cellular internet. Although significant strides have been made in this direction with the onset of 5G, despite several decades of research, several key tasks like Multiple-Input-Multiple-Output (MIMO) signal detection and Precoding are still performed sub-optimally by real-world deployments. The optimal methods for these functionalities are known to be "NP-Hard", and therefore, practical deployments resort to using polynomial-complexity approximations and suffer a loss in performance. Advancements in MIMO detection and precoding can significantly boost the physical layer throughput, which would lead to a drastic improvement in the cellular data-rates. The last decade has seen an emergence of "Physics-inspired" computation as an alternative to conventional computing to solve NP-Hard problems. Physics-inspired computation involves using the dynamics of a physical system to compute the optimal solution to an NP-Hard optimization problem. Typically, such systems are designed to solve the Ising optimization problem, and other problems need to be converted to the Ising problem. These methodologies span a wide range of technologies, ranging from quantum/optical systems to purely algorithmic methods. This thesis tackles MIMO detection and precoding using physics-inspired computation. In Part 1, we develop the theoretical framework required to utilize physics-inspired computation for MIMO detection and precoding functionalities in state-of-the-art wireless communication systems. We show that our techniques can significantly outperform the existing "conventional-computing" based methods used by real-world systems. In Part 2, we bridge the gap between theory and practice; we design and implement Mavis: a GPU-based, highly scalable, physics-inspired MIMO detector and precoder, which can meet the processing throughput and latency requirements of modern 5G systems. We demonstrate that Mavis can meet the processing requirements of 5G systems with large bandwidths (100 MHz) and provide significant throughput improvements.