To All-pass or not to All pass

[ErLan]

Hello,

(Still) in the journey towards the optimal set up in my untreated a symmetrical room... Now facing the following dilemma.

Looking at integration of Lt & Right speakers, I face notable dips in the low end. Analyzing the source led to the understanding that it originates from local phase misalignment between Lt & Rt speakers that drive cancellations (probably not a surprise in an untreated room). Trying to relief that I designed a light cascade of AP compensation filters and applied it on the left filter.

The overall integration with those filters seems to improve the integration in the wide sense, but lead to the following observation\dilemma.

1- When looking at the phases of Lt (wo\filter) and Rt signals with FDW <5 (means very close to a direct sound) & FDW>=6 (more reflections + modes):
* FDW < 5 phase matching is perfect,
* FDW >=6 there is a built up gap in the phases (that lead to the Lt-Rt poor frequency image.

2- When looking at the phases of Lt (w\filter) and Rt signals with FDW <5 (means very close to a direct sound) & FDW>=6 (more reflections + modes):
* FDW < 5 phase matching shows cancelation in the integrated Lt + Rt response
* FDW >=6 overall Lt & Rt look fairly aligned, which lead to improved integration in the low end.

Knowing the sensitivity of our ears to the early 80msec period of sound arrival and its importance to imaging (hence smaller FDW) - may lead to the direction where optimizing phase should be based on low FDW ;
On the other hand, our lack of ability to localize bass at the range being focused at (<100Hz) and the nice integration at wide range of FDW >=6 may hint to optimize phase based on this setup.

I will be glad to get you - experts & experience - perspective on that.

(attached also the REW measurements in case someone is looking to play with the raw materials)

Many thanks!

 

[ernestcarl]

It's usually preferred to use a longer window in the bass or modal region and shorter as you go up higher in frequency when it comes to DRC.

FDW 1-3 will give you a sense of the actual phase in the bass region minus much of the reflections effect coming from your room.

But, if you really must view the real direct response either measure nearfield in the middle of the room (does not take into consideration port contribution) well away from boundaries or outdoors using a tall speaker stand.

BTW, the cancellation you see above 1 kHz with your L+R measurement is the result of the IR peaks not being perfectly centered prior summing. Dunno why there's a time gap, but you can match the pair prior using an A+B trace arithmetic operation or "aligned sum" (via Alignment tool) or vector averaging by right clicking and selecting Set t=0 to IR peak".

Well, you didn't include the all-pass filter you created for the left speaker in the attachment.

Taking just a quick look at the inverted excess phase response of both left and right speakers:



I suspect a single 40Hz inverse 2nd order all-pass might already be a potential "improvement" above 40 Hz in terms of just reducing the dips around 50 and 90 Hz. BUT, unfortunately, it also cancels out the bass response below 40 Hz.


*More detailed "correction" could give a better result.

Considering potential side-effects with phase filtering (increase or decrease in GD and/or introduction pre-causal artifacts), I wouldn't bother if it doesn't substantially improve actual bass response via practical listening tests.

 

[ErLan]

Thanks @ernestcarl for the quick reply.

Few questions,

FDW 1-3 will give you a sense of the actual phase in the bass region minus much of the reflections effect coming from your room.

Correct, yet looking at <50Hz frequencies, FDW=3-4 resembles ~80msec - is not it the time that is most relevant for sound perception? if so, should not it be weighted over the 'settle time' (my dilemma ), what is your take on that?

BTW, the cancellation you see above 1 kHz with your L+R measurement is the result of the IR peaks not being perfectly centered prior summing. Dunno why there's a time gap, but you can match the pair prior using an A+B trace arithmetic operation or "aligned sum" (via Alignment tool) or vector averaging by right clicking and selecting Set t=0 to IR peak".

Thanks, since I was more focusing on the low end, I did not align mic position to the exact point (as far as I recall there is a small difference of <5cm, to some extent resembles also life 'imperfections')

I will try your proposal for the all-pass filter (actually, my was 2 cascaded 2nd order\80Hz, but as i found it sensitive to the FDW - I thought to first settle down my understanding a FDW\smoothing configuration and then fine tune it.

 

[raif71]

 

[ppataki]

You might also consider using one or more subwoofer(s) to fix that low end

 

[ernestcarl]

... sensitivity of our ears to the early 80msec period of sound arrival and its importance to imaging (hence smaller FDW) - may lead to the direction where optimizing phase should be based on low FDW

On the other hand, our lack of ability to localize bass at the range being focused at (<100Hz) and the nice integration at wide range of FDW >=6 may hint to optimize phase based on this setup.

looking at <50Hz frequencies, FDW=3-4 resembles ~80msec - is not it the time that is most relevant for sound perception? if so, should not it be weighted over the 'settle time' (my dilemma ), what is your take on that?

The direct sound is very relevant, but your room mangles the response nevertheless. I can't say how much with certainty one psychoacoustically is able to window-out the later arriving energy in the bass region.

For clarity, all that frequency dependent windowing (FDW) does is remove later arriving energy. I don't believe there is a hard and fast rule or demarcation line that says past this amount of time the audibility becomes (mostly) irrelevant.

Spectrogram


Phase

*Hmmmn... probably the reason why phase goes crazy in the HF above with more narrow windowing is the very initial peak is all but decimated when it finally arrives at the mic after applying FDW -- hence loss of HF response with FDW -- there's also so much later arriving reflections that REW "gets confused" when that energy is abruptly truncated:

I avoided this graphic anomaly by exporting the windowed IR first, and thereafter re-importing the measurements back into REW.




*Curiously, expanding the window sizes in the HF range using MTW instead of FDW improves the impulse and phase plots, but the entire spectrogram measurement still gets shifted to the right regardless.

 

[ErLan]

Thanks for the thorough analysis and explainations.
My key take away is that each FDW will drive slightly different behavior of the phase and as such, correction. Converging to the psychoacoustically preferred option can not be determined using a crisp rule (as many other stuff) but rather via. an iterative tuning and listenung work.