The funny thing about harmonics is that even the ones you do not hear effect the ones you do. So you need to capture it all DAMIT
What's that? You say you want to try another ABX comparison?
OK sure lets do it, Two tracks ( the same song) One with a low pass at 20k 96db per octave, the other one with out any lowpass recorded at 24bit 192k Does that sound like a fair test?
Let me know if you have trouble getting a hold of a 96db per octave lowpass I can do it here and upload the files already done.
Let me know if you can find a host for the files.
Chris Church
OK, you do that. I'll be interested to see how you propose to get 96 dB per octave past 20 kHz. How far down do you plan to be at 20 kHz? Wait... don't tell me... You want it to be flat to within 1 dB out to 20 kHz and be -96 dB by 40 kHz.
There are many ways to do it but I can use a digital crossover to get the slope but it got me thinking I think I have to be carefull about this because if I use a digital crossover it will change the sound of the processed signal and make the two samples less like each other. Any ideas out there?
Chris, let me put this in perspective for you. You get about 6 dB per octave rolloff in the stopband of a lowpass filter per pole in the transfer function for the filter. With a single pole, you'll be down about 3 dB at the cutoff frequency. When you go to multiple poles, you generally do not put all of your poles at the same frequency, but if you did, you'd get about 3 dB attenuation per pole at the cutoff frequency. By the time you get 96 (!!) dB per octave, you're going to have 16 poles in the filter. If all the poles are co-located, then you'll have 48 dB of attenuation at the cutoff frequency. Something tells me that filtered and unfiltered versions will be easily recognizable if you have 48 dB of attenuation at the intended cutoff frequency. In order to get any sort of flatness in the passband and still get a steep stopband attenuation curve, you'll have to resort to Chebyschev or elliptical designs. With at least 16 poles in the transfer function, some of the poles will be extremely high Q poles which means your parts tolerances are going to be obscenely critical. In order to get an analog filter that was flat to 1 dB in the passband and had 96 dB attenuation one octave into the stop band, you'll probably spend the rest of your life designing, re-designing and tweeking and still not get the job done.
In my opinion, the better approach would be to use a digital domain filter to boost the high frequency content of the signal from a microphone whose transfer function is well known and to do so in such a way that the resulting transfer function is flat out to the highest frequency for which you have calibration data for your mic. So if you have 10 dB of rolloff in your mic's frequency response at 25kHz, that portion of the spectrum should be boosted by 10 dB. The transfer function of your digital filter should ideally be the reciprocal of the complex conjugate of the frequency response of your microphone. This will increase the quantization noise in the ultrasonic portion of the spectrum, but it's the only way I know of the simulate what a truly flat broadband microphone would sound like, at least with any reasonable amount of effort. I'd offer to design a filter like this, but I no longer have access to DSP filter design tools and I don't feel like spending enough time to do it by hand when I know darn well that the only result is to modify signals that I can't hear anyway.
I'd also like to point out that it is not ideal to simply correct the amplitude in the stopband. It would be better to flatten the spectrum by using a filter whose transfer function is the reciprocal of the complex conjugate of the frequency response of the device whose output you wish to flatten. By using the reciprocal of the complex conjugate of the frequency of the mic used to do the recording, you not only correct the amplitude response, but you also get as close as possible to a linear phase (constant delay) response. If you only correct the amplitude response then there may be artifacts from the uncompensated phase distortion that can appear as subharmonics within the audible portion of the spectrum and, as you should expect, you can probably hear arifacts if they are induced within the audible portion of the audio spectrum.
If you really want to do this, I'd recommend using FFT processing to transform the time domain waveform into the frequency domain, apply the required amplitude and phase equalization and inverse transform the result to get back to the time domain. This will be an extremely processor intensive process, but it should yield relevant results.