Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: FFT analysis window position (Read 2839 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

FFT analysis window position

In mp3 encoder the FFT analysis window should be centered over granule, I dont know over one granule or two granule?

http://www.hydrogenaudio.org/forums/index....howtopic=16196&

In this post, it suggests that 1024 fft analysis window for the Psychoacoustic Model is at the center of the 1152 (two granule) mdct analysis window. But in LAME, I found it centered over one granule. In AAC, FFT analysis window looks like center over two frame(2048). So I dont know which is correct?

FFT analysis window position

Reply #1
Quote
In mp3 encoder the FFT analysis window should be centered over granule, I dont know over one granule or two granule?

It should be centered over one granule.

Quote
In this post, it suggests that 1024 fft analysis window for the Psychoacoustic Model is at the center of the 1152 (two granule) mdct analysis window.

The MDCT is not centered over 2 granules. It is an overlapping transform, with 50% overlap.
Practically it is covering those data:
*the ending part of the previous granule
*the whole part of the current granule
*the starting part of the next granule.

FFT analysis window position

Reply #2
You still have to take into account, the Delay of the Subband filters..

FFT analysis window position

Reply #3
Let dont think about delay now.
If it should be centered over one granule then in AAC it should only need 1024 fft(100% overlap one frame). But it looks like need 2048 fft that covered 2 frame.

FFT analysis window position

Reply #4
In AAC FFT is, indeed, centered over 2048 samples (2 MDCT frames), at least in the ISO docs.

Btw, I talked with a person who worked closely on psychoacoustic optimization of one MP3 encoder - and he told me didn't realise how big the problem is MP3's _____ (insert rant here)  hybrid filterbank until he started working on psymodel.

Basically, one >should< take into account the difference between hybrid window and a FFT window - difference in size of these transformations (and trouble making 1152-point FFT performance-wise) and their spectral images - make this task quite challenging - and most implementators just use centered simple 1024-point FFT.  In these paper FhG claims solving the problem by using modified FFT windows and checking the difference between hyb. and FFT signal.

Check out:

Name: Second-Generation ISO/MPEG-Audio Layer III Coding
Authors:  Herre, Jürgen; Brandenburg, Karlheinz; Eberlein, Ernst; Grill, Bernhard
AES Number: 3939  (January 1995)

 

FFT analysis window position

Reply #5
Thanks for all the reply.

I have another question about short fft analysis window positon.
In mp3, the long fft will be 1024 points, short will be 256 points. In the ISO doc, the short fft will at
:|......256.....|...128...|...128...|...128...|...128...|.....256.....|
...................|..... block 0 ..... | ..... block 2 ..... |
................................|..... block 1 ..... |

but in lame it looks like this
:|......256.....|...128...|...128...|...128...|...128...|.....256.....|
|....192...|..... block 0 .....|..........|...... block 2.....|...192...|
.................................|..... block 1 ..... |

I dont know why lame change it?

 
SimplePortal 1.0.0 RC1 © 2008-2021