Forums

Re: Query on fmult in G.726

Started by Bhanu Prakash April 30, 2002
Hi all,

I have a doubt in "FMULT" operation in G.726.
The "fmult" is used to compute multiplication of fixed point number and
a floating point number.In this function, the fixed point is first converted
to floating point and
the sign,magnitude and mantissas of both the floating point numbers are
extracted.

The resultant is computed as follows.

WAnS = SRnS ** AnS ; signs of float1 float2
WAnEXP = SRnEXP + AnEXP ; exponents of float1 float2
WAnMANT = ((SRnMANT * AnMANT) + 48) >> 4

Mantissas of the two operands (Q6 format) are multiplied to form the product
mantissa (Q12),
which is then truncated in Q8 format.Thats why, we are right shifting the
result by 4.
But i dont understand why 48 is added to the product of mantissas.

please clarify my doubt.
Best regards,
Prakash



In order to round off properly, you need to add in the equivalent of 1/2 before
you truncate, i.e. in plain C-syntax you would define

int rnd(float x) { return int(x+0.5)}

If you convert the exact value of zero to floating point, you would get

"zero" == sign=0 exp=0 mant2 == 0x20 * 2^0

(follow the fmult code through to verify this statement).

If you convert the value of one to floating point, you would get

"one" == sign=0 exp=1 mant2 == 0x20 * 2^1

A representation of "half" in floating point can be derived by
averaging "one" and "zero":

half = (one + zero)>>1
= (0x20 *2^1 + 0x20 * 2^0)>>1
= (0x40 + 0x20) >> 1
= (0x60) >> 1
= 0x30
= 48 Note: This fmult definition is the same as in G.721, except it is coded as
in hexadecimal 0x30 instead of 48.

I am not sure that this is 100% correct, but it appears to be a plausible
explanation.

cheers,
Mark

Mark Lakata, Staff Engineer 1225 Charleston Road voice 650-567-5170
MIPS Technologies Mountain View CA 94043 fax 650-567-5002

On Tue, 30 Apr 2002, Bhanu Prakash wrote:

> Hi all,
>
> I have a doubt in "FMULT" operation in G.726.
> The "fmult" is used to compute multiplication of fixed point number and
> a floating point number.In this function, the fixed point is first converted
> to floating point and
> the sign,magnitude and mantissas of both the floating point numbers are
> extracted.
>
> The resultant is computed as follows.
>
> WAnS == SRnS ** AnS ; signs of float1 float2
> WAnEXP == SRnEXP + AnEXP ; exponents of float1 float2
> WAnMANT == ((SRnMANT * AnMANT) + 48) >> 4
>
> Mantissas of the two operands (Q6 format) are multiplied to form the product
> mantissa (Q12),
> which is then truncated in Q8 format.Thats why, we are right shifting the
> result by 4.
> But i dont understand why 48 is added to the product of mantissas.
>
> please clarify my doubt.
> Best regards,
> Prakash >
> _____________________________________
> Note: If you do a simple "reply" with your email client, only the author of
this message will receive your answer. You need to do a "reply all" if you want
your answer to be distributed to the entire group.
>
> _____________________________________
> About this discussion group:
>
> To Join:
>
> To Post:
>
> To Leave:
>
> Archives: http://www.yahoogroups.com/group/speechcoding
>
> Other DSP-Related Groups: http://www.dsprelated.com > ">http://docs.yahoo.com/info/terms/ >


Hi prakash ,
i believe the 48 is added for rounding the final prod. But like to have
more clarifications in depth.

vikram

-------Original Message-------

From: Bhanu Prakash
Date: Tuesday, April 30, 2002 21:18:30
To:
Subject: [speechcoding] Re: Query on fmult in G.726

Hi all,

I have a doubt in "FMULT" operation in G.726.
The "fmult" is used to compute multiplication of fixed point number and
a floating point number.In this function, the fixed point is first converted
to floating point and
the sign,magnitude and mantissas of both the floating point numbers are
extracted.

The resultant is computed as follows.

WAnS == SRnS ** AnS ; signs of float1 float2
WAnEXP == SRnEXP + AnEXP ; exponents of float1 float2
WAnMANT == ((SRnMANT * AnMANT) + 48) >> 4

Mantissas of the two operands (Q6 format) are multiplied to form the product
mantissa (Q12),
which is then truncated in Q8 format.Thats why, we are right shifting the
result by 4.
But i dont understand why 48 is added to the product of mantissas.

please clarify my doubt.
Best regards,
Prakash ------------------------ Yahoo! Groups Sponsor ---------------------~-->
Buy Stock for $4
and no minimums.
FREE Money 2002.
http://us.click.yahoo.com/k6cvND/n97DAA/ySSFAA/GP4qlB/TM
---------------------------------~->

_____________________________________
Note: If you do a simple "reply" with your email client, only the author of
this message will receive your answer. You need to do a "reply all" if you
want your answer to be distributed to the entire group.

_____________________________________
About this discussion group:

To Join:

To Post:

To Leave:

Archives: http://www.yahoogroups.com/group/speechcoding

Other DSP-Related Groups: http://www.dsprelated.com ">http://docs.yahoo.com/info/terms/ .


Attachment (not stored)
attachment
Type: image/gif