| 摘要 |
Integer MDCT is used in lossless audio coding. However, an integer-reversible transformation is split into consecutive lifting steps, each of which introduces considerable rounding errors to the signal. This problem gets more significant the lower the level of the useful signal to be encoded. Without noise shaping the rounding error noise will impact all frequency bins of the transformed signal equally. This is a particular problem for frequency bins in which the actual signal level is low. The invention limits the impact of rounding error noise coming with, or resulting from, each lifting step in the integer-reversible transformation on the data rate of a lossless codec. The filter coefficients of an adaptive noise shaping filter for transform coefficients are adapted in individual lifting steps according to the current time domain signal characteristics. As an alternative, an auto-regressive pre-filter can be added in front of the lossless transformation, targeting at raising the level of frequency regions with low power to decrease the dominance of rounding errors in these areas. Both processings can be combined to further improve the compression ratio of a lossless codec. |
