Research Interests

Quantum computers promise to solve certain problems far faster than any supercomputer, but the fragile quantum states they rely on are easily disturbed by noise. Small errors build up quickly and can derail a computation entirely. This project develops new quantum error correcting codes and better decoding algorithms so that quantum computers can keep working reliably even as they scale up.

AI tools that generate text, images, or audio, like large language models and diffusion based art generators, are usually overconfident: they report high certainty in their outputs even when they're wrong. This project asks whether ideas from error correcting codes, the same math used to keep data reliable on hard drives and over wireless links, can make these generative AI models both better at their core task and more honest about what they don't know.

Future 6G wireless networks will need extremely fast, extremely reliable communication for things like remote surgery and autonomous vehicles. This project uses reinforcement learning to speed up the decoders that correct transmission errors, teaching them to process the right pieces of data in the right order rather than a fixed sequence. It also looks at how machine learning models running on connected devices can be trained to tolerate noisy, error prone wireless links directly, instead of relying only on traditional error correction hardware.

Many modern error correcting codes trade off between decoding easily and getting close to the theoretical performance limit. This project showed that chaining together many small, identical coding structures into one long connected code, called spatial coupling, lets simple iterative decoders reach performance very close to that theoretical limit. The resulting codes also correct errors in a distinctive wave like pattern, with reliability spreading in from the ends of the chain toward the center as the decoder runs.

Compressing speech, audio, and video always involves some loss of information, and the goal is to lose as little quality as possible for a given amount of compression. This project applies spatially coupled coding structures, originally developed for correcting transmission errors, to the compression problem itself, showing that these codes can compress data almost as efficiently as the theoretical best possible, with practical encoding delay.

5G and future wireless networks need to support use cases, like industrial automation and vehicle to vehicle safety systems, that demand both very low error rates and very low delay. This project designs generalized low density parity check codes, which replace simple parity checks with stronger error correcting building blocks, tailored specifically for these ultra reliable, low latency requirements.

Turning an error correcting code into a real chip requires simplifying the math so it can run with limited hardware precision, without giving up too much performance. This project develops quasi cyclic LDPC codes that are easy to encode with simple shift register circuits, along with analysis tools that predict how a decoder will behave once its internal messages are rounded to a fixed number of bits. The group also builds and tests real decoder and encoder hardware, including circuits tuned for medical applications like heart rhythm monitoring.

Wireless and other communication links do not stay the same quality over time, so a single fixed error correcting code is rarely the best choice throughout a connection. This project studies puncturing, a technique that lets one code adapt to multiple transmission rates by selectively withholding some coded symbols, and works out exactly what makes a code well suited to this kind of adaptation.

High speed optical networks need error correcting codes that run at very high code rates with hardware efficient decoding. This project studies braided convolutional codes, a family related to turbo codes where two encoders feed outputs into each other over time, and shows they perform well even close to the noise floor while tolerating the kind of rate adjustment optical systems need.