摘要 |
An improved approximation of the desired codebook boundary is achieved, when quantizing data from a source with memory, by adding a dither sequence, the samp les of which are contained within the Voronoi region of an unpartitioned lattice or the first level of lattice partitioning used to generate the trellis code. A sequence of data generated by the source with memory is quantized to a nearest sequence of a predetermined trellis code which is based on coset partitioning of a predetermined lattice into translates of a coset lattice . The trellis code sequence is then filtered with an inverse source filter. A dither sample is adde d to each sample of the sequence such that (1) a sum of the dither sample and the filtered trelli s code sequence sample lies on a super-lattice of the coset lattice, and (2) the dither sample i s inside the Voronoi region of the super-lattice.
|