Adaptive variable-to-variable length codes

In the last several years, adaptive codes for fixed-to-variable length and variable-to-fixed length codes have been described. This paper examines two methods for implementing adaptive variable-to-variable length codes, which have not been considered before due to the difficulty of designing optimum variable-to-variable length codes. The two adaptive methods are based on dual-tree codes, where a source tree parses the input sequence into source words and a code tree assigns each source word a code word. One adaptive method uses a single dual-tree code, and uses an algorithm which requires a complex logic circuit to adjust the shape of the source and code trees. The second method, called state-tree codes, uses a fixed pool of dual-tree codes and a state machine to select which dual-tree code is used. State-tree codes require more memory than the first method, but only a trivial logic circuit is needed to implement the codes, which will result in a very fast circuit