Number format for sbsamples from IMDCT (Q4.28 to 16 bit PCM samples)

Hi Rob

I am new in using the MAD Library and upon working with the code, I have got 
doubt in the Scaling (1st operand shifted right by 12 and 2nd operand 
shifted right by 16) being done during the mad_f_mul operation.

I like to know about the sbsamples from the IMDCT routine. Is it in Q4.28 
format. If so why we need to scale it down during SYNTHESIS Filter 
operation. I have read the comments in fixed.h telling that

" * The scale factors are constant as this is not used with SSO.
*
* Pre-rounding is required to stay within the limits of compliance.
"
I like to know how this scaling has been taken care in the further 
calculations.

Please find below my doubt.

I have defined as FPM_DEFAULT and doesnt define OPT_SPEED.[OPT_SSO will be 
defined as FPM_DEFAULT is defined]. With that the mad_f_mul (A,B) is defined 
as [((A+(1L<<11))>>12) * ((B+(1L<<15))>>16)]

This has been used during the synth_full operation. When the synth_full 
function calls the dct32() function,

sbsamples<2><2><576> (in Q28.4 format) from IMDCT is passed on to the 
function where it will be multiplied with costab<1..31>.

During this multiplication the mad_f_mul MACRO is used. After that the 
result from dct32 is stored as the filter output. While storing the filter 
output, the multiplied value is scaled down by a factor of 12.

Then windowing[D array values being scaled down by 12] was applied over the 
filter ouput and the PCM output is stored with a scale down of 2.

The operation of SSO is understandable but the intial scaling that happens 
with the mad_f_mul MACRO needs explanation.

Thanks in advance

Balaji Shanmugam

_________________________________________________________________
News, views and events. Hot downloads ‘n pictures. 
http://www.msn.co.in/Cinema/ Get it all at MSN Cinema!