Ah that explains a lot. I didn't realize it wasn't removing it all I forgot to take that into account.
The DAC itself didn't really misbehave and the warbles were audible only at max volume and the filter did make a difference. But when something is so low volume its hard to make up any difference so I described it as less noisy. The receiver however is the real "problem" but the way you explain the attenuation makes sense. I'm going to play with the filter some more.
Concerning the PCM filter on the DAC itself and based on this experience I think 'Slow' would be preferable over Sharp (graphs here).
Great learning experience! Thanks!
If you're interested, I thought I'd throw out a few additional notes about crossover topographies that might help as you fiddle with the filters (forgive me if I'm telling you things you already know). The JRiver filters are butterworths, which means at the setpoint frequency (in your case 24 KHz) they are -3dB regardless of the slope of the filter. Stacking two butterworths gives you a Linkwitz Riley filter which will (regardless of slope) be -6dB down at the setpoint frequency. Any X dB crossover filter will be -X dB at one octave above the setpoint frequency (i.e. double the frequency), so 48 dB down at 48 KHz in your case.
In order to reach -3dB (or -6 dB) at the setpoint, the low pass filter must begin attenuating at a lower frequency than the setpoint (have a look at the graphs in this wiki article, you'll see what I mean
http://en.wikipedia.org/wiki/Audio_crossover). Generally speaking, higher order, steeper filters of the same topography will need less "lead-in" room. For example, a second order butterworth filter (12dB) will begin attenuating more than an octave before the setpoint, whereas an eighth order filter (48dB) will begin attenuating much less than an octave before the setpoint. The disadvantage of steeper filters is that they can create phase or impulse anomalies.
The goal is (of course) to have none of the ultrasonics one doesn't want, while leaving the actually audible material unchanged and not introducing any "ringing" or phase anomalies. That's hard to do with conventional crossover topographies, because filters that will effectively eliminate all ultrasonics will also affect audible material to some extent (due to the issues outlined above). So you have to balance whether some slight attenuation/phase weirdness in the upper part of the audio band is worth less ultrasonics, or vice versa. For my part, I start my filtering at 18 KHz, because I can't hear much above 15 or 16 KHz, and I find a slight roll down in frequency response above 10 KHz to be less fatiguing to my ears anyway. But that's a personal preference, and you'll need to find the solution that works for you.
On another note, the "trade-offs" I'm describing are less of a concern in the FIR filtering context (i.e. convolution). If you use convolution, you have access to arbitrarily steep phase adjusted filters, so that would be a theoretically "perfect" solution, but convolution is a bit fiddly, and I wouldn't recommend starting to use it just for this. If you're already using it though, it offers the next best thing to a theoretically perfect "brick wall filter."
Author
Topic: Intermodulation Test (Read 10999 times)