| Summary: | 博士 === 國立臺灣大學 === 電機工程學研究所 === 99 === A variable-bit look-ahead Huffman decoding problem is investigated in this Dissertation. When we used state diagrams to represent a variable-bit look-ahead Huffman decoder, we found that these state diagrams are not unique. Thus, we need to formulate the finding of state diagrams in a given variable-bit look-ahead Huffman decoding tree as a general look-ahead barrel-shift Huffman tree decomposition (LABSHTD) problem, such that many other variable-bit look-ahead Huffman decoding methods are just instances of this LABSHTD problem.
First, an important procedure that helps to build a correct mapping between a Huffman decoding state diagram and a Markov chain model is presented. The mapping makes it feasible to estimate the decoding throughput rate of an instance of the LABSHTD problem. Then, we use the implementation complexity to estimate the hardware resources required by a decoder instance and calculate the decoding throughput rate from the obtained Markov chain, which will be defined as an optimization problem. The objective of optimization is to maximize the decoding throughput rate by exploiting the solution space of a LABSHTD problem. Since the number of candidate solutions to the LABSHTD problem is growing exponentially, we propose an efficient algorithm to search for a heuristic solution that usually yields a very good optimization result in polynomial computation time. Finally, we show that this algorithm can be extended to solve the Arithmetic Decoding problem.
|
|---|
