Differences

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

--- atrac3p:quantization_unit_dequantization [2010/07/23 16:22] – created megadiscman
+++ atrac3p:quantization_unit_dequantization [2010/10/30 08:25] (current) – Quantitative discussion of dequantization. megadiscman
@@ Line 1: / Line 1: @@
 ===== Dequantization =====
-The quantization units are dequantized by normalization of the integer coefficients depending on the quantization factor and multiplied by the band level (logarithmic scale, +3 makes a factor of two).
+The level of a bandlet chooses the available range. The maximum level, 63, will choose -65536..65536; lowering the level by one will decrease the range by a factor of 2^(1/3), so decreasing the level by 3 will shrink the range by a factor of 2. As an example, the level 15 will choose the range -1..1. The range is then divided into equally sized intervals according to the [[quantization precision table|quantization precision]], and the center of each of these intervals will be the value represented by the quantized coefficient.
+For example for a precision code of 2, there will be five intervals. If the level is 15, the intervals will be assigned to:
+  - -1 to -3/5 (centered at -4/5)
+  - -3/5 to -1/5 (centered at -2/5)
+  - -1/5 to +1/5 (centered at 0)
+  - +1/5 to +3/5 (centered at +2/5)
+  - +3/5 to +1   (centered at +4/5)
+As the quantized coefficients are -2,-1,0,1,2, the interval centers are obviously accessed by coeff*2/5 in this case.