Calculating ADAU1701 coefficients in uP for long double vs float (64bit vs 32bit.)

The question first:

Given that the actual coefficients in the ADAU1701 are only really 28 bits (the first nibble of the first byte is always a zero) how important is it that I calculate coefficients using 64 bit math?

I am using a uP to calculate coefficients using C (not C++). This is needed because I will need to have peaking filters with a dynamic range of +- 15 db on each of 15 bands.

The frequency range is from 18 to 100 Hz on each of the 15 bands.

I have been trying to carry double precision (long double) 64 bit calculations throughout my "CalcCoeff" routine, then convert to 32 bit float number just at the last step. Of course, the To523 further converts that to a 28 bit, zero padded, integer array to send to the DSP.

I was given to understand the precision of the coefficients are more important at lower frequencies, and I also understand intermediate results of previous calculations are also important.

Is there anything to be gained (practically speaking) by using 64 bits math.

Thanks, Mark.

Audio Signal Processors

coefficients

SigmaDSP Audio Processors

64bit

adau1701