Timing issue?

[Mitchco]

I often take loopback measurements to validate FIR filter designs. I feed REW's output signal through JRIver's ASIO input, through JRiver's convolution engine containing FIR correction filters, DAC, analog outputs to inputs to amps, speakers, measurement mic, mic pre, ADC and REW input. The Lynx Hilo converter I use has a consistent delay, the USB driver allows one to configure buffer size. JRiver also has a couple of adjustable buffer settings, along with the 750ms FIR filter delay. The USB driver is ASIO multi-client application friendly, so there are no other audio devices involved.

Every so often, I get a funny timing response (i.e. step response) where one channel looks like it should, but the other channel sometimes varies in amplitude like this:

I have measured the same speakers before with near identical step response and then sometimes it measures like above. Repeating the measurement produces more variability in the responses. Also not sure why they are not starting at 0 amplitude at t=0 either...

Recently I tried using http://www.roomeqwizard.com/betahelp/help_en-GB/html/makingmeasurements.html#fileplayback which worked, but produced similar results to above. It seems some days measures as it should in both channels, and other days similar to the above. Note I don't see any other real differences between channels looking at the other REW graphs... The frequency response is nearly identical between channels.

Any assistance appreciated.

Cheers,
Mitch

 

[John Mulcahy]

Best posting an mdat with some example measurements. The step responses are rising before t=0 (and afterwards) due to an offset in the impulse response. There is an option to apply a 2 Hz high pass to the captured input on the soundcard preferences dialog which might help with that.

 

[Mitchco]

Thanks John. See attached.

 

[John Mulcahy]

Ta. I've made a few measurements on my desktop setup as well and see similar variations. Measuring from 20 Hz uses a higher start frequency for the sweep (10 Hz) and removes the content below 10 Hz during processing, which gives more consistent results. I don't think this is inherent in the measurement process, since a series of sweeps from 0 Hz on an analog loopback behave as expected. It is most likely due to variable low frequency noise (my desktop setup has no output to speak of below 50 Hz so noise is dominant below that). I'll look into it and see if I can find any other explanation for the variation in low frequency content.

 

[Mitchco]

Thanks again John for looking into this. I believe the mdat's I uploaded was measuring from 5 Hz as I have a couple of large subs that make it into single digits and was trying to capture that. The sweeps were calibrated at 83 db SPL at the listening position. You are right, before I had subs, I was measuring a bit higher and had less low frequency content and did not run into this variation as much.

 

[John Mulcahy]

The 5.20 beta versions are a bit more susceptible to this than 5.19. When the measurement start frequency is below 20 Hz the sweep starts with a linear portion from DC to 10 Hz then log from 10 Hz to twice the end frequency, or half the sample rate if lower. In the 5.20 beta versions that linear portion had a fade in added, which reduces the VLF content of the sweep. Potentially helpful for sub drivers, but that reduction in LF energy means LF noise is not suppressed as much as with 5.19. I may revert that for the next 5.20 beta build.

 

[Mitchco]

Hi John, if you revert back in the next beta build, I am happy to retry the measurements and report back. Thanks, Mitch

 

[Mitchco]

Hi John, V5.20 B10 solved the issue:

Thank you!

Mitch