CAREER:Advances in Universal Data Compression with Applications to Joint Source and Channel Coding

University Of Utah, Salt Lake City UT

Investigators

Abstract

The purpose of this research is to develop several unexplored areas in data compression, as well as to utilize universal data compression techniques in other applications including biological modelling and a novel direction of joint source-channel coding. The research focuses on four topics: (a) design of joint source-channel universal source code based codes, (b) study of universal compression for large and unknown source alphabets, (c) design of advanced universal coding techniques for non-traditional, yet more realistic, data models with practical implementations, and (d) the study of random access lossless compression. Common techniques in joint source-channel coding suffer from severe synchronization problems in bad channel conditions and do not address universality issues when the source statistics are unknown. This research develops techniques to combat these problems, and even attain "free" gain in channel decoding performance for redundant channel information streams. Common compression schemes assume that the data is from a known alphabet, it has a "standard" stationary or constantly changing statistical model, and it consists of a long sequence. However, (a) there exist compression applications with large unknown alphabets, such as text compression where the words constitute the alphabet, (b) most real data sequences are usually neither stationary nor of constantly varying statistics, and (c) random access is necessary in large compressed data bases. The investigator studies these three non-traditional problems. The research work combines the development of rigorous theoretical results including redundancy and description length bounds, with empirical testing, algorithm design with focus on practical low-complexity techniques, and implementation of proposed techniques. Finally, the research also investigates the use of universal compression techniques to segmentation and modelling of biological sequences.