An investigation of reversible variable length codes

Ritesh Nagpal, Purdue University

Abstract

Reversible variable length coding (RVLC) is a state-of-the-art entropy encoding scheme. These codes are investigated in detail. Two algorithms are proposed for constructing RVLCs using Huffman codes. One of the proposed algorithms falls into the category of non-generic algorithms and is developed using the virus-victim analogy. The virus-victim analogy is proposed to understand the affix (simultaneously prefix and suffix) properties of RVLCs. This non-generic algorithm can be adjusted to match several VLC-design requirements, like maximum codeword length, error correction capabilities and implementation complexity. It, however, does not outperform existing construction algorithms in terms of average codeword length. The second proposed algorithm is a generic algorithm and employs an improved codeword selection mechanism. This algorithm is based on graph models and multilevel optimization of the average codeword length. The properties of these graph models are discussed and approximations are suggested for a simple construction algorithm. The same approach can be applied to the optimization of other parameters of reversible variable length codes. Recent advancements in error correcting RVLCs are also studied.

Degree

M.S.E.C.E.

Advisors

Bell, Purdue University.

Subject Area

Electrical engineering

COinS