RE: [MP3 ENCODER] Fixed pOint Psychoacoustic model
Some tips : - avoid use of arctan() by rewriting the CW equation. - Keep exp and mantissa in separate regs in the case of thresholds - "curve fit" to compute SNR directly from Tonality SNR = 10 ^ -(6.0 * Tbb + 29.0 * (1 - Tbb))/10.0 Good luck. Mohammed Chalil -Original Message- From: Manjunath D N [mailto:[EMAIL PROTECTED] Sent: Wednesday, March 12, 2003 12:52 PM To: [EMAIL PROTECTED] Subject: [MP3 ENCODER] Fixed pOint Psychoacoustic model Hi, I am trying to convert Lame code to fixed point in order to port it to a fixed point 24 bit dsp as part of my college project.. While the other modules are manageable, the psychoacoustic model is really driving me crazy! The dynamic range is so high that sometimes even the double precision(48 bits) arent enough.. I request all the Lame developers to suggest a solution to this problem..It's been bugging me for almost a month now and nothing has come out.. Is it advisable to simulate floating point ? Thanks , With Regards, Manjunath D.N ___ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder ___ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder
Re: [MP3 ENCODER] Fixed pOint Psychoacoustic model
> The dynamic range is so high that sometimes even the double precision(48 > bits) arent enough.. You could store mantissa in one integer and exponent into another one. This will probably be more than enough regarding dynamic range. The drawback is that in this case you are creating your own datatype. You might also want to reduce the allowed dynamic range. Anyway, you will probably have to reduce quality in some way in order to reduce processor requirements. Regards, Gabriel Bouvigne www.mp3-tech.org personal page: http://gabriel.mp3-tech.org ___ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder
[MP3 ENCODER] Fixed pOint Psychoacoustic model
Hi, I am trying to convert Lame code to fixed point in order to port it to a fixed point 24 bit dsp as part of my college project.. While the other modules are manageable, the psychoacoustic model is really driving me crazy! The dynamic range is so high that sometimes even the double precision(48 bits) arent enough.. I request all the Lame developers to suggest a solution to this problem..It's been bugging me for almost a month now and nothing has come out.. Is it advisable to simulate floating point ? Thanks , With Regards, Manjunath D.N ___ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder
