additional accuracy to the fixed point FIR ADAU1787 under sigmastudio

Hello nehem,

It is interesting that you feel you need this extra precision. Nobody has ever had any issues with the precision in the many years we have used these processors. In fact, we often have issues with computer programs or Excel not calculating to enough precision. People try to match the exact number they see SigmaStudio calculate but the filter response still looks the same. Generally, this just has not been a problem and changing a few bits of the LSBs will not noticeably change the filter response.

Keep in mind that 23 bits is about 7 decimal places of resolution:

log10(2^23) = 6.92

Please inform me as to what you are experiencing?

Dave T

Hello David,

Sorry for my late response.

I wanted to implement optimal filter with the ADAU1787 , fixed point FIR. This optimal filter has high magnitude(positive and negative) coefficients and with the fixed point limitation, I had to use few gain stages in series, that increased the noise. I have not found an solution , and will use the old LMS method.

Thanks

Hello, are you implementing an adaptive FIR ? If you are doing NLMS, you need to set the initial mu(step size) small enough such that when it is scaled by the input signal energy the result does not cause the updated coefficients to become larger than the fixed point number format