Differences

This shows you the differences between two versions of the page.

--- atrac3p:atrac_spectrum_anatomy [2010/03/20 12:57] – megadiscman
+++ atrac3p:atrac_spectrum_anatomy [2010/11/16 18:37] (current) – Fix spectral width for 44.1kHz sample rate. megadiscman
@@ Line 3: / Line 3: @@
 ===== ATRAC3+ spectrum layout =====
-The spectrum (0..22.5kHz) for 44.1kHz sample rate is divided into 16 bands of equal spectral width of 1406,25Hz. The division into bands of same width is a direct consequence of the DCT synthesis scheme. Some of these 16 bands are further devided into bandlets, as at low frequencies the same spectral width occupies a higher relative spectrum range (i.e. a higher pitch interval). The bandlet data has to be combined into band data before the IDCT synthesis can be performed, so all processing steps in time domain work on band basis or even more coarse, while frequency-domain processing often is performed on the bandlet level.
+The spectrum (0..22.05kHz) for 44.1kHz sample rate is divided into 16 bands of equal spectral width of 1378.125Hz. Some of these 16 bands are further divided into quantization units (all quantization units of the same bands have the same spectral), as at low frequencies the same spectral width occupies a higher relative spectrum range (i.e. a higher pitch interval). All quantization units' data has to be combined into the total band data before the IMDCT synthesis can be performed, so all processing steps in time domain work on band basis or even more coarse, while the spectral data itself is encoded on the level of quantization units.
-The selection of spectral dither noise amplitude uses an even more coarse scheme of dither groups that contain several bandlets. Borders of dither groups coincide with borders of bands except for the first dither group that leaves out the first two bandlets of the first band.
+The selection of spectral dither noise amplitude uses a scheme of dither groups that contain several quantization units. Borders of dither groups coincide with borders of bands except for the first dither group that leaves out the first two quantization units of the first band.
 The layout is like this
@@ Line 11: / Line 11: @@
 | dither group |none||  0  ||||||  1  ||||||||  2  ||||||  3  |||||  4  |||||
 | band         |  0  ||||||||  1  ||||  2  ||||  3  ||  4  ||  5  ||  6  |  7  |  8  |  9  |  10  |  11  |  12  |  13  |  14  |  15  |
-| bandlet      | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 |
+| quantization unit | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 |
+One ATRAC3+ block contains 128 spectral samples per band, yielding 2048 samples in total. The lapped structure of the IMDCT, working on 4096 input samples at a time, causes the the transform window size to be cited as 4096 even though the block size is just 2048 samples.
 ===== Data in an ATRAC3+ substream block by level =====
   * Per substream data:
-    * highest non-zero bandlet (used to calculate highest non-zero band and highest non-zero dither group)
+    * highest used quantization unit (used to calculate highest non-zero band and highest non-zero dither group)
     * Mute flag
+    * Time-domain dither info
   * Per channel data
   * Per dither-group data
@@ Line 25: / Line 28: @@
     * Envelopes
     * Tone data
-    * Time-domain dither info
+  * Per quantization unit data
-  * Per bandlet data
     * Spectral quantizers
     * Spectral levels