My previous message had the biquad filter parameters for 96k based on Scott's values. This one is the even more accurate 192k version of the sox commandline for the inverse RIAA filter.
sox -G -c 2 C:\Users\Public\RIAA\sox_sweep192kto22k.wav C:\Users\Public\RIAA\sox_sweep192kinvRIAAto22k.wav rate 192k biquad 1.0000000000 -1.9312626096 0.9313724325 1.0000000000 -0.8796912298 -0.1023703458
I started posting the wave files and then canceled the upload when I realized that it would take more than 24 hours. That will have to wait until I get home.
sox -G -c 2 C:\Users\Public\RIAA\sox_sweep192kto22k.wav C:\Users\Public\RIAA\sox_sweep192kinvRIAAto22k.wav rate 192k biquad 1.0000000000 -1.9312626096 0.9313724325 1.0000000000 -0.8796912298 -0.1023703458
I started posting the wave files and then canceled the upload when I realized that it would take more than 24 hours. That will have to wait until I get home.
Basically the smaller the frequencies of interest are relative to the sampling rate the better things are. One observation the textbooks don't always make is that as the sampling frequency approaches infinity the z and s domain analyses must converge. Thinking in terms of state variable analysis terms like I = C*dV/dT truly approach the straight line approximation as the sampling period becomes very small, sort of like how differential calculus is introduced.
If you look at Bob Orban's coefficients you will see that at high sampling rates they correspond very closely to the DSP 101 bilinear transforms of the s domain poles and zero. His procedure adds a high frequency zero and fiddles the other time constants around to make a best fit DC - 20KHz. Above I posted my results simply using a "thermal annealing" method to converge to the best least squares fit over 20Hz - 20KHz.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.