You can increase the accuracy by increasing the size of the lookup table from a table of 8bit values to something bigger, as memory permits. Also, another way is to make the table as the encoder runs. At the beginning, you can initialize it if you want or add the values as the need pops up (if the value is in the table, then use it, if not, calculate it and store it for future use), or both.

There's many ways to "skin a car" (figure of speech, btw).

G.

John van Ommen wrote:

I know that Xing implements lookup tables for their encoder, that's why it's so fast. But floating point math is more accurate, and processing power and ram is cheap these days.

Which does your instructor prefer, speed or accuracy? If I were you, I'd design your project in a way that will make your instructor happy.

John

-----Original Message----- From: mad-user-admin@lists.mars.org [mailto:mad-user-admin@lists.mars.org]On Behalf Of Yuxuan Mao Sent: Sunday, February 24, 2002 11:39 AM To: mad-dev@lists.mars.org; mad-user@lists.mars.org Subject: [mad-user] alternative idct implementation question

Hi,

I'm doing a school project on mp3 decoders. Other teams in my class is proposing to do an 8-bit cosine lookup table implementation of the idct. Does anyone know if this will work?

Yuxuan