[FFmpeg-devel] [PATCH] Simplify IMDCT in dca.c

[Alexander E. Patrakov]

Benjamin Larsson wrote:

> Alexander E. Patrakov wrote:
>> Michael Niedermayer wrote:
>> 
>>> synthesis window: prCoeff
>> 
>> Plot it in gnuplot. It doesn't look like a typical (smooth) synthesis
>> window.
>> 
> 
> Can you test if the PB condition is true ?

It is irrelevant because the "window" length (512) is not the same as the
transform size. However, the description from
http://ccrma.stanford.edu/~jos/sasp/Pseudo_QMF_Cosine_Modulation_Filter.html
fits perfectly.

Property (1) works with M = 512, L = 8, N = 32

Property (2) (where h(n) was extracted from the response of the "black box"
to the excitation in the 0th band, NOT the official data table itself) has
to be adjusted to the different filter gain. This sum of squares behaves as
follows:

0.124995
0.124999
0.125000
0.125000
0.125000
0.125000
0.125000
0.124999
0.124995 = first part is OK, above this we see only 0.125
0.124319
0.118041
0.097375
0.062500
0.027625
0.006959
0.000681
0.000005 = the second part is OK, because below this are all zeroes
0.000000
0.000000
0.000000
0.000000

Property (3) works, and that's how I extracted h(n) from the "black box"
response.

Property (4) is essential for my encoder.

-- 
Alexander E. Patrakov