Soundcard Calibration Query - Subsonic frequency ripple

[Newmarket]

Hi. First of all - thanks for REW. Just getting into using it seriously for hobby purchases (although quite a bit of experience with Audio Test kit eg AP2/Neutrik A2 longer ago than I wish to think about !)
So - I'm taking a Cal reading of my I/O for a baseline. I think I'm doing that correctly. Response looks fairly typical with a rise up from very low frequencies to 20Hz or so then a fairly flat response with the characteristic ripples coming in at higher frequencies until the response falls steeply going to Nyquist. BUT there are some ripples in the response at very low / subsonic frequencies below 20Hz.
So making Cal file then applying that to a measurement of the I/O itself. Result should closely approximate to a flat response ?
It does in that the hf ripple is cancelled and the overall low frequency drop off is flattened. BUT the ripples in the low frequency response persist.
I have used a longer sweep time for the measurement which might have helped. And the ripple is small in overall terms. But there's definitely something going on and it markedly happens at frequencies below 20Hz.
Apologies for lack of numbers and images atm. I am not at home with the kit atm. But really wondering if there is a known explanation. Thinking in terms of latency / buffer size.
System is on XP 32 Bit - don't laugh - so using the REW version that works with that.
Audio I/O is Soundscape Mixtreme using an external ADC/DAC box interfaced via TDIF. (Reason for sticking with the XP is that this system doesn't work with newer Windows. Well, technically it apparently can work with Win7 - but only as long as Memory is not more than 2Gig so not much use).
Any insight welcome. Will post images etc when I get the chance.
Thanks for reading.

 

[John Mulcahy]

LF ripples can be due to the IR right hand window being too short if the device has very extended LF response.

 

[trobbins]

I guess there is also the effect of response drop off, both of the soundcard's DAC on outgoing signal, and the ADC on signal coming back in, and also related to the analog circuitry in those two signal paths. The soundcard's chipset seems to have some influence in my EMU0404 in that it changes the LF digital roll-off with sample rate setting (I get get lower roll off frequency with 44kHz sample than with 192kHz sample). I measured the responses of my soundcard using an external meter, but even that can be onerous unless the meter has a special low frequency response, so a scope is perhaps easier to give some insight on signal response that is trying to be compensated by the calibration curve.

 

[Newmarket]

LF ripples can be due to the IR right hand window being too short if the device has very extended LF response.

Many thanks for that response. I tried it briefly yesterday and it seems too help enormously. Cheers.

 

[Newmarket]

I guess there is also the effect of response drop off, both of the soundcard's DAC on outgoing signal, and the ADC on signal coming back in, and also related to the analog circuitry in those two signal paths. The soundcard's chipset seems to have some influence in my EMU0404 in that it changes the LF digital roll-off with sample rate setting (I get get lower roll off frequency with 44kHz sample than with 192kHz sample). I measured the responses of my soundcard using an external meter, but even that can be onerous unless the meter has a special low frequency response, so a scope is perhaps easier to give some insight on signal response that is trying to be compensated by the calibration curve.

Yes. I understand the low frequency filtering - and the hpf come to that.
My question was around why the compensated response was not flat at very low frequencies (measuring 2Hz - 24kHz @ fs 48kHz).
John's advice on IR window size seems to have sorted it. Still a few things to check as had some non-repeated results. Or could just be "finger trouble".
fwiw the "phantom ripple" itself is relatively small so not particularly important in itself. But I am looking at the frequency
response of various interface boxes, DIs, Buffers etc so it's cleaner if they are not there at all to any significant degree.