System and method for modeling discrete data sequences

Abstract

An encoder and decoder system for compressing discrete data sequences is disclosed. The encoder includes a self-affine map generator and a backward adaptive map generator which model source vectors formed from a discrete data sequence. The self-affine map generator uses known modeling techniques to determine map parameters for mapping functions. The mapping functions are evaluated by computing a collage error and the mapping function corresponding to the smallest collage error is selected. The backward adaptive map generator determines map parameters for mapping functions which map vectors previously generated to represent the discrete data sequence to the source vector. A distance measurement is computed for each target vector/mapping function combination and the mapping function corresponding to the smallest distance measurement is selected. A map evaluator determines whether the mapping function generated by the self-affine map generator or the backward adaptive map generator best represents the discrete data sequence. The inventive decoder includes an iterative decoder and a backward decoder. The iterative decoder uses a deterministic or random iteration algorithm to generate a synthesized vector from a mapping function received from an encoder. The backward decoder retrieves a vector from a sample buffer at the decoder which corresponds to the vector used to generate the mapping function. The sample buffer contains previously decoded vectors. The mapping function is applied to the retrieved vector once to generate the synthesized vector representing a source vector at the encoder. The decoder selects the affine decoder or the backward decoder based upon a token from the encoder which identifies the process which generated the mapping function.