Re: [ecc] The instruction set for the Turbo codec

[mohamed.kilani_ecc@m4x.org]

Yes. In order to make it fast, you need to a have logarithm operation
implemented.
The main idea in Turbo-Coding is to apply multiple iterations of the
same decoding scheme on the result data each time. This data is a
probability distribution and processing it through your turbo decoding
engine uses multiply operations (Conditional Probabilities).
Since multiply is not that fast even with the best optimisation, we
usually use the logarithmic value of these probabilities and add them
in each loop.(This is based on the fact log(a.b) = log(a) + log(b)).

I don't know if you really need an absolute value but at least you
need a way to calculate -a given a since all probabilities are <1 and
taking the log gives a negative value that you do not want to carry
all the way.

Hope that helps 

Dali

----- Original Message ----- 
From: "Mohammad Mobien" <m_mobien@h... > 
To: <ecc@o... > 
Date: Thu, 28 Feb 2002 08:08:29 +0500 
Subject: [ecc] The instruction set for the Turbo codec 

> Plz, If anyone can give me a brief description of the type of 
> instructions required to make Turbo coding/decoding fast. I have 
> read somewhere that we need to give support for logrithm and 
> absolute. I want to know some more about it. 
> Regards 
> 
--
To unsubscribe from ecc mailing list please visit http://www.opencores.org/mailinglists.shtml