Honored Contributorwhen the signed signal pass through multiply or FIR ,the signal width extend.how should i choose to truncate the signal width . generally,we truncate (LSB)Least Significant Bit ,retain MSB. But,in my design ,after peform this operation,all of signal data become zero.can you give me some tips?
Honored ContributorHi,
It sounds like the intermediate value is so low that you loose the significant bits by trunkating on LSB. Maybe you need to analyze the data path theoretically to understand where the significant bits are in each step of the design to maximize the dynamic range. How about using fixed point arithmetics? Thanks JHonored ContributorUsually you would model the filter with fixed point, and then see what kind of accuracy is acceptable. Then you can truncate accordingly. Remember that rounding errors will compound if you have chained filters.
