CIF: Small: Coding for Live Delay-constrained Streaming Communication

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Live streaming communication is fast becoming an integral part of our society due to a paradigm shift in the way people communicate, and share and consume content. Video conferencing, watching high-definition sporting events over streaming counterparts to television channels, and broadcasting social live video streams on commercial platforms are now ubiquitous. Such communications often face disruptions since data transmitted from the sender can get lost or delayed in the communication medium and not reach the receiver in time. In order to avoid the resulting glitches due to such losses, the application on the sender side needs to send data in a redundant fashion such that the original stream can be recovered even when some parts of it are lost. Adding such redundancy consumes additional bandwidth which is a scarce and costly resource. This project will develop a fundamental understanding of the minimum amount of additional bandwidth needed to transmit a live stream with a desired quality, and also will design practical and efficient ways of performing such loss-resilient live streaming communication. The findings from this project will be integrated into graduate and undergraduate courses. Directed efforts will be undertaken for promoting diversity in STEM education and in mentoring undergraduate students. Live streaming communication involves sequentially encoding and decoding packets with a strict delay constraint. Due to the strict delay constraint, loss recovery using retransmissions are often not applicable and forward error correction is needed. The overarching goal of this project is to establish the fundamentals of coding for live, delay-constrained streaming communication under a generalized model. Specifically, the project initiates studies in two broad directions: (1) Handling variable arrival-sizes: Existing models on streaming codes consider a "fixed-size" setup, where all arriving source packets have identical sizes, an assumption which often does not hold in practice, thereby degrading performance. This project proposes a generalized model incorporating variability in arrival sizes and investigates the new problem space, including a new tradeoff resulting from the variability, and constructs optimal streaming codes for the new model. (2) Designing streaming codes for correlated stochastic channels: Common sources of packet erasures over the network are bursty in nature, thereby resulting in correlated losses. The existing body of work on live, delay-constrained streaming communication has largely studied correlated losses only under adversarial settings, which result in significant bandwidth overhead. This project investigates constructing streaming codes for stochastic channel models incorporating correlated losses. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.