Help - Search - Members - Calendar
Full Version: Mdct And Imdct Algorithms
Hydrogenaudio Forums > Lossy Audio Compression > AAC > AAC - General
kennyzero
What are the fastest algorithms being used in AAC encoders right now? Can anyone give me some information about that?

And, the ones used in open-sourced FAAC encoder...what are those algorithms?

Thx much.
biggrin.gif
petracci
Hi,

Just some references you might want to investigate:
  • A fast algorithm for the implementation of filter banks based on TDAC, Duhamel et al, 1991.
  • Fast algorithms for orthogonal and biorthogonal MLT, Malvar, 1998.
  • A new fast algorithm for the unified forward and inverse MDCT/MDST computation, Britanak et al, 2002.
The first item would be a good starting point. The algorithm uses a complex FFT as the modulation kernel. Investigation of the derivation of this algorithm will enable you to study the newer work as well.

The last item is one of the latest algorithms. I did not include integerized algorithms here.

As for the best algorithm, that depends on your requirements. Typically, there is a trade-off between the number and type of computations and the required memory.

Good luck!
Ivan Dimkovic
QUOTE
And, the ones used in open-sourced FAAC encoder...what are those algorithms?


FAAC and FAAD use (I)MDCT implementation with FFT kernel of 1/4 size of N. This is very efficient implementation of (I)MDCT.

You might also want to check U.S. patent application# 2002/0106020 A1 that deals with optimized MDCT for audio coding.
kennyzero
QUOTE(petracci @ Jul 4 2003, 12:15 AM)
Hi,

Just some references you might want to investigate:

  • A fast algorithm for the implementation of filter banks based on TDAC, Duhamel et al, 1991.
  • Fast algorithms for orthogonal and biorthogonal MLT, Malvar, 1998.
  • A new fast algorithm for the unified forward and inverse MDCT/MDST computation, Britanak et al, 2002.
The first item would be a good starting point. The algorithm uses a complex FFT as the modulation kernel. Investigation of the derivation of this algorithm will enable you to study the newer work as well.

The last item is one of the latest algorithms. I did not include integerized algorithms here.

As for the best algorithm, that depends on your requirements. Typically, there is a trade-off between the number and type of computations and the required memory.

Good luck!

thx...where can i find those documents?
Gabriel
Strange patent application. It seems to me to be a patent on a mathematical computation, and not really on a process.
I thought this was not allowed.
kennyzero
QUOTE(Ivan Dimkovic @ Jul 4 2003, 12:40 AM)
FAAC and FAAD use (I)MDCT implementation with FFT kernel of 1/4 size of N. This is very efficient implementation of (I)MDCT.

You might also want to check U.S. patent application# 2002/0106020 A1 that deals with optimized MDCT for audio coding.

thx, i read the homepage and find that very complicated ph34r.gif
petracci
Hi, here goes...

Duhamel, P.; Mahieux, Y.; Petit, J.P.;
A fast algorithm for the implementation of filter banks based on TDAC.
Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on , 14-17 April 1991 Page(s): 2209 -2212 vol.3

Malvar, H.;
Fast algorithms for orthogonal and biorthogonal modulated lapped transforms
Advances in Digital Filtering and Signal Processing, 1998 IEEE Symposium on , 5-6 June 1998 Page(s): 159 -163

Vladimir Britanak and K. R. Rao
A new fast algorithm for the unified forward and inverse MDCT/MDST computation, Elsevier Signal Processing, Volume 82, Issue 3, March 2002, Pages 433-459
Ivan Dimkovic
QUOTE(Gabriel @ Jul 4 2003, 08:50 AM)
Strange patent application. It seems to me to be a patent on a mathematical computation, and not really on a process.
I thought this was not allowed.

Well, it's just a patent application - nobody guarantees that it is going to be granted smile.gif
petracci
QUOTE(Gabriel @ Jul 4 2003, 10:50 AM)
Strange patent application. It seems to me to be a patent on a mathematical computation, and not really on a process.
I thought this was not allowed.

The application is on the efficient implementation of the mDCT, not on the mathematical computation. This is allowed. But there are already many applications/patents on various implementations, it is hard to see how this one differs from the rest.
kennyzero
QUOTE(petracci @ Jul 4 2003, 01:03 AM)
Hi, here goes...

Duhamel, P.; Mahieux, Y.; Petit, J.P.;
A fast algorithm for the implementation of filter banks based on TDAC.
Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on , 14-17 April 1991 Page(s): 2209 -2212 vol.3

Malvar, H.;
Fast algorithms for orthogonal and biorthogonal modulated lapped transforms
Advances in Digital Filtering and Signal Processing, 1998 IEEE Symposium on , 5-6 June 1998 Page(s): 159 -163

Vladimir Britanak and K. R. Rao
A new fast algorithm for the unified forward and inverse MDCT/MDST computation, Elsevier Signal Processing, Volume 82, Issue 3, March 2002, Pages 433-459

i cannot find those documents in google. sad.gif can anyone help? thx
wkwai
[quote=kennyzero,Jul 4 2003, 12:58 AM] [QUOTE=Ivan Dimkovic,Jul 4 2003, 12:40 AM] FAAC and FAAD use (I)MDCT implementation with FFT kernel of 1/4 size of N. This is very efficient implementation of (I)MDCT.

[/quote]
I think the 8 FFT implementation of the 2048 MDCt/ IMDCT is more efficient than one that uses 4 FFT.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.
Invision Power Board © 2001-2008 Invision Power Services, Inc.