Possible problem on Linear Phase vs Minimum Phase XO filter generation

[clarus]

I am setting up a 3-way 2-channel stereo setup using Audiolense XO. I have generated two sets of correction filters using the same measurement and target, one with the option "Minimum Phase Xover" checked, and one unchecked, which I presume implies linear phase. The simulated frequency responses are similar. However, when I'm listening to them, the version without the "Minimum Phase Xover" option will work correctly at sample rates multiple of 48kHz, and the version with the "Minimum Phase Xover" option checked will work correctly at sample rates multiple of 44.1kHz.
Playback is using with the Audiolense convolver 1.6, hence only 1 alc file. When it's not working correctly, you can hear that phases are all wrong, there is no center image for vocal, some of the bass got cancelled etc.

In the filter generation process, I have selected all samples rate, i.e. 44.1K, 48K, 88.2K, 96K, 176.8K, 196KHz. The original measurement was done at 192kHz in Audiolense.
Is there any additional setting I need to set in order to generate both the "linear phase" and the "minimum phase" XO correctly?

 

[juicehifi]

Hi clarus,

The resampling should be oblivious to which crossovers you choose, so I do not know what's going on here.

You can send me the measurement and I will take a look at it.

You will get a more predictable result if you create corrections with only 192k filters included. The convolver resamples the correction filters if the subject sample rate is missing.

In Audiolense, the measurement is resampled before the correction is made. And sometimes, the corrections will be slightly different from fs to fs. But when the correction filters are resampled (in the convolver) you will get practically identical corrections, except for the cut-off.

 

[clarus]

Hi Bernt,
I have sent you the measurement file.

If I understand correctly, whether the resample is done during filter creation, or resample the correction filter by the convolver on the fly, the result should be the same, as they are both processing the audio at the native/incoming sample rate. The only difference I can see is that it will use less CPU resource during playback if resample during recreation vs resample on the fly, and it doesn't have to keep doing the resampling if the sample rate keep changing.

 

[juicehifi]

The measurement looks ok, but you should use a minimum phase target when you make the minimum delay correction. Minimum phase target is btw what I generally recommend for all corrections. When you use a linear phase target you get pre-ringing that may be audible on certain occations.

I checked the resampling in Audiolense, and the resampled filters in the alc files are OK.

The resampling in alc is only done once, and saved for later use.

 

[clarus]

Change everything to minimum phase and no longer have the sample rate related problem, which doesn't really account for the previous issue that I ran into.
If we use minimum phase target and minimum phase xover, that would mean that we wouldn't be able to achieve a zero phase speaker system. Is there any value in achieving that? or is Audiolense designed to support that?

 

[juicehifi]

You need to use minimum phase to achieve lowest possible latency. Zero phase speaker system is most likely not do be desired anyway, due to pre-ringing associated with the high pass and low pass sections of the speaker. Minimum phase target only affects transition band regions and not the pass band of the speaker. Minimum delay crossovers creates slightly worse driver alignment, but the summation usually works practically as well as liniar phase.