Huffman tree based metric derivation for a low-complexity sequential soft VLC decoding

L. Perros-Meilhac, C. Lamy
2002 IEEE International Conference on Communications. Conference Proceedings. ICC 2002 (Cat. No.02CH37333)  
This paper considers VLC decoding algorithms based on MAP sequence estimation techniques, using residual source redundancy to provide channel error correction. These algorithms rely on soft values available at the entrance of the VLC decoder. We present here a new soft VLC decoding algorithm based on a sequential decoding technique that is very efficient in terms of decoding complexity. The application of the considered soft decoding algorithms to practical decoding of MPEG-4 texture
more » ... packets under the assumption of an unequal protection scheme is investigated. Algorithm performance is evaluated on the well-known "Foreman" video sequence. Simulation results show that the proposed algorithm provides approximately the same performance as all existing soft decoding algorithms while exhibiting a significantly lower complexity. Index Terms-sequential decoding, MAP estimation, soft-input decoding, variable length codes © Step 1. Load the stack with the tree origin node and assign a metric 0 to this path. © Step 2. Compute the metric of the successors of the top path in the stack. © Step 3. Delete the top path from the stack. © Step 4. Insert the new paths in the stack, and reorder paths according to metric values. © Step 5. If the top path ends stop. Otherwise, return to step 2. The adaptation of the stack algorithm to VLC decoding, that we will call as VLC-stack algorithm, has been proposed by Buttigieg [9]. In this case, one codeword of each length generates a successor path to the top path in the stack. The approximated Massey metric presented in the previous section is used to compare paths. It has been shown [9] that the size of the stack can be limited to a quite small value without degrading appreciably the performance, preventing an infinite growth of the stack. B. A reduced complexity VLC-stack algorithm As most sequence estimation-based VLC soft decoding algorithms, the VLC-stack requires the computation of the metric associated to each codeword at each decoding step, making the metric derivation the most costly part of decoding in term of CPU (central processing unit) requirements. As the size of the stack is limited, successors of the top path having a cumulative metric greater than the one of the worst path in the stack will not be kept. In this section, we show that it is possible to examine and select codewords in the increasing order of their metric value, and thus to reduce the complexity of the VLC-stack by deriving only the useful metrics i.e. metrics associated to extended paths which will be inserted in the stack. The method consists of applying a kind of stack algorithm onto the Huffman tree representing the considered VLC table, in order to select the most likely codewords at the considered time. More precisely, the selection of useful codewords is achieved in the following way: © Step 1. Load a stack with the origin node of the Huffman tree and assign a metric 0 to this tree-path.
doi:10.1109/icc.2002.996962 dblp:conf/icc/Perros-MeilhacL02 fatcat:w5lur2sulrhxjkasrz5ogr2gbe