AAC | ADPCM | Convolution | DAFx | FFT | IIR | Mixer | MP3 | MPEG | MPEG-4

Hi Folks!
              Just needed some information on some of the schemes that
are used to implement DeQuantization of audio samples.
The scheme that I know is DeQuantization using a look-up table of 8192
by storing the value of y^(4/3) in the table and using y as the index
to access these values.

           Is it possible to eliminate the Look-up table totally
without much computational effort ..since calculating value of y^(4/3)
using binomial approx is computationally heavy .......instead can it
be done by using something like approximating the curve using a line
or something like that...(i.e.simple slope calculation)...

      Can someone who has worked AAC throw light on the same?

Thanks in advance,
Siddharth
	

(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )

Please refer the FAAD2 code.
It used only 1k memory for that look-up table.
For your reference.
	--- Siddharth Nag <siddharth.nag@sidd...> wrote:

---------------------------------
Hi Folks!
              Just needed some information on some of the schemes that
are used to implement DeQuantization of audio samples.
The scheme that I know is DeQuantization using a look-up table of 8192
by storing the value of y^(4/3) in the table and using y as the index
to access these values.

           Is it possible to eliminate the Look-up table totally
without much computational effort ..since calculating value of y^(4/3)
using binomial approx is computationally heavy .......instead can it
be done by using something like approximating the curve using a line
or something like that...(i.e.simple slope calculation)...

      Can someone who has worked AAC throw light on the same?

Thanks in advance,
Siddharth
	   To
	_______________________________________________________________________
Yahoo!©_¼¯¹q¤l«H½c
§K¶O®e¶q250MB¡A«H¥ó¦b¦h¤]¤£©È
http://tw.promo.yahoo.com/mail_new/index.html
	

(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )

Hi Sidharth,

 If memory is your problem in keeping a look up table, use the following
approach.

 Keep a look up table of say 1024 elements for x = y ^ (4/3).

 For values more than 1023, use the following.

 x = (16/16) y ^ (4/3)
   = (16/ 2^4 *) y ^ (4/3)
   = 16 * ( (y/8) ^ (4/3))

( (y/8) ^ (4/3))is available in the 1024 element look up table and
multiply by 16 to get an approximate value.

 The only issue is you lose accuracy and there are methods where you can
get exact accuracy by using interpolating the nearest two values
obtained by above method.

Regards,
Tony

-----Original Message-----
From: audiodsp@audi... [mailto:audiodsp@audi...] On
Behalf Of Siddharth Nag
Sent: Thursday, April 28, 2005 5:26 PM
To: audiodsp@audi...
Subject: [audiodsp] Query on DeQuantization of MPEG-4 AAC Decoder

Hi Folks!
              Just needed some information on some of the schemes that
are used to implement DeQuantization of audio samples.
The scheme that I know is DeQuantization using a look-up table of 8192
by storing the value of y^(4/3) in the table and using y as the index
to access these values.

           Is it possible to eliminate the Look-up table totally
without much computational effort ..since calculating value of y^(4/3)
using binomial approx is computationally heavy .......instead can it
be done by using something like approximating the curve using a line
or something like that...(i.e.simple slope calculation)...

      Can someone who has worked AAC throw light on the same?

Thanks in advance,
Siddharth
	

(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )