The Design of Tree and Trellis Data Compression Systems.
Abstract
Recent results in information theory promise the existence of tree and trellis data compression systems operating near the Shannon theoretical bound, but provide little indication of how actually to design such systems. Presented here are several intuitive design approaches and also a general design philosophy based upon the generation of fake processes i.e., finite entropy processes which are close (in the generalized Ornstein distance) to the process one wishes to compress. Most of the design procedures can be used for a wide class of sources. Performance is evaluated, via simulations, for memoryless, autoregressive and moving average Gaussian sources and compared to traditional Data Compression systems. The new schemes typically provide 1-2 dB improvement in performance over the traditional schemes at a rate of 1 bit/symbol. The inevitable increase in complexity is moderate in most cases.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1978
- Accession Number
- ADA057963
Entities
People
- Joseph Linde
- Robert M. Gray
Organizations
- Stanford University