Fixed Point Accuracy , How to ?

[omara007]

Hi folks ..

I'm having some problems truncating my fixed points signals .. I'm having a 16-bit system and I want to carry out a 16bit X 16bit process .. and I need the output to fit in a 16bit vector too .. and I'm havinh a problem in determining the accuracy and where I should cut .. given that my format is 0.15 signed numbers. (0.15 means that the binary point is just after the MSB which indicates the sing as well).

The range of number am using is <= .9999999999999 ..

so, what do you suggest ?

 

[yjcheng]

I think you can built a c-simulation model.

 

[omara007]

I think you can built a c-simulation model.

what's the benfit of doing such thing over using VHDL ?

 

[omara007]

The case now is like this :
I have 2 16-bit fixed point numbers .. (15 downto 0) .. the binary point is between the 15th and the 14th .. and the 2 numbers are signed .. I want to multiply them together and fit the result in 16-bit vector with reasonable accuracy .. what shall I do ? .. from where should I truncate the original 32-bit multiplication result ?


The next question will be .. if this result is to be added to another result generated from a similar process (the whole thing can be viewed as a sum of products), what's the best way not to lose accuracy ?