Streaming Codes for Variable-Size Messages

• Published in: IEEE transactions on information theory Vol. 68; no. 9; pp. 5823 - 5849
• Main Authors: Rudow, Michael, Rashmi, K. V.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-09-2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Block codes; Codes; Decoding; Encoding; erasure coding; live streaming applications; Lower bounds; Messages; Packet loss; Packet transmission; Streaming codes; Streaming media; Symbols; Upper bound; Upper bounds; Variability; video conferencing; Videoconferencing
• ISSN: 0018-9448; 1557-9654
• DOI: 10.1109/TIT.2022.3170895

Live communication is ubiquitous, and frequently must contend with reliability issues due to packet loss during transmission. The effect of packet losses can be alleviated by using erasure codes, which aid in recovering lost packets. Streaming codes are a class of codes designed for the live communication setting, which encode a stream of message packets arriving sequentially for transmission over a packet-loss channel. The existing study of streaming codes considers settings where the sizes of the message packets to be transmitted are all fixed. However, message packets occur with unpredictable and variable sizes in many applications, such as videoconferencing. In this paper, we present a generalized model for streaming codes that incorporates message packets of variable sizes. We show that the variability in the sizes of message packets induces a new trade-off between the rate and the decoding delay under lossless transmission. Moreover, the variability in the sizes of message packets impacts the optimal rate of transmission. To address this, we introduce algorithms to compute upper and lower bounds on the optimal rate for any given sequence of sizes of message packets. We then design an explicit streaming code for the proposed model. We empirically evaluate the code construction over a live video trace for several representative parameter settings, and show that the rate of the construction is approximately 90% of an upper bound and 5%-48% higher than naively using the existing streaming codes.