Excess group delay and mismatched phase response

[RenPa]

What do you have you Sub-mains cross over set to? My comment above assumes that the x-over is set to 80Hz or higher.

In my DLBC projects with 3.4.4 xo are at 80Hz for all groups.

 

[ppataki]

I have checked your new measurements, unfortunately all of them have the same huge GD starting from below 100Hz:

If I compare your earlier individual sub measurement with one of these new measurements it shows elevated GD

I can still only say that this is a sub integration issue

I was playing around with the Time Alignment tool in REW with these measurements:
A: G 50 patte 58 mur
B: GAUCHEsvs PRO -gran corri

I changed the delay of B by -9.9ms (NOTE: I spent literally 10 seconds doing this, probably it can be fine-tuned even much better) and simulated the integration of the two and this is what I got:

Frequency response

GD

As you can see the GD now looks way better
To me this proves that DLBC is the culprit here and your system could indeed be integrated in a much better way
If you could somehow change the delay values manually, that sould bring much better results (like the one above)

 

[ppataki]

After spending a few more seconds with the Alignment Tool this is the FR and GD I got

And the Wavelet now looks fantastic:

So indeed it can be done correctly

 

[RenPa]

OK. but if I am right, you aligned only one sub with front left. I have 2 subs. With the same mdat with one measurement at mlp only I aligned the 2 subs first with alignment tool then aligned the sum aligned with front left and I got this

which seems quite similar to DLBC. Or am I wrong somewhere?
Before DLBC I had miniDsp 2x4 HD in the chain.

 

[ppataki]

Can you please re-post the above diagram with these settings?

 

[RenPa]

Can you please re-post the above diagram with these settings?

Of course:

 

[RenPa]

And I made the exercice with the 2 subs with reverse polarities in relation with the satellites which is my setting in my system for Dirac measurements.

 

[ppataki]

Both those above graphs look way better compared to the DLBC wavelet you had
(from post #22)

 

[RenPa]

Both those above graphs look way better compared to the DLBC wavelet you had
(from post #22)

View attachment 407151

OK.
What seems evident for you and some others and me too about the importance of mastering the group delay doesn't seem the case for Dirac. I refuse to think they are stupid. So what explanation can we find?

I don't want to seal my ported subs. But do you think it would help DLBC with GroupDelay if I seal all the satellites? I know to safe answer is try it it but in theory ?

 

[ppataki]

OK.
What seems evident for you and some others and me too about the importance of mastering the group delay doesn't seem the case for Dirac. I refuse to think they are stupid. So what explanation can we find?

I don't want to seal my ported subs. But do you think it would help DLBC with GroupDelay if I seal all the satellites? I know to safe answer is try it it but in theory ?

Have you tried asking Dirac why this is?

Regarding the port stuffing: in the last 5 years I have had almost 20 speaker building DIY projects. I only build sealed subs since they have better transient response vs ported ones. So I would give it a try definitely. On the other hand some other folks here will not agree with this....

 

[OCA]

These are the best measuring picks of pairs in your mdat. That step response growing back after 60ms is very odd, points to a non-zero DC offset in the impulse response.

 

[RenPa]

points to a non-zero DC offset in the impulse response.

Sorry my knowledge doesn't help me to understand what do you mean. If it is important I must to understand before opening a new thicket to Dirac

 

[RenPa]

Have you tried asking Dirac why this is?

No. I opened already a ticket long time ago for Arcam bugs with DLBC firmware update above 3.4.4.
I wish @OCA will explain what he meant by non-zero DC and will see if I can add it to the ticket.
What I can't begin to understand is if the solutions are not so difficult to find when you know the basics (not like me but like OCA or you) why on earth such a big company like Dirac can't do it ?

 

[OCA]

No. I opened already a ticket long time ago for Arcam bugs with DLBC firmware update above 3.4.4.
I wish @OCA will explain what he meant by non-zero DC and will see if I can add it to the ticket.
What I can't begin to understand is if the solutions are not so difficult to find when you know the basics (not like me but like OCA or you) why on earth such a big company like Dirac can't do it ?

It's a bit too technical but step response swings should normally gradually fade out over time and since it's the integral of the impulse response, impulse average must be non-zero. This is not something Dirac would intentionally do and is most likely a bug. That could harm hardware in the long run, too. More info below:

https://www.perplexity.ai/search/non-zero-dc-offset-fft-RcZLWfglRIG4y_pcl3XAgA#1

 

[LIΟN]

As you can see the GD now looks way better
To me this proves that DLBC is the culprit here and your system could indeed be integrated in a much better way
If you could somehow change the delay values manually, that sould bring much better results (like the one above)

I posted an mdat capture of this in another thread, and I think DIRAC is the unintended bug. (Because it's also in known issues)
It may be different for each person, but a user from another community who provided me with dlbc vs dlbmdat said that when DLBC was active, it sounded like a lot of sound (negative, almost reverb, not the good side).

And his auditory feeling is explained by this delay, which of course appears in GD, and the difference of 20ms is something I can't understand in any way.
Of course, some people say that Dirac would have intended it.

 

[ppataki]

I posted an mdat capture of this in another thread, and I think DIRAC is the unintended bug. (Because it's also in known issues)
It may be different for each person, but a user from another community who provided me with dlbc vs dlbmdat said that when DLBC was active, it sounded like a lot of sound (negative, almost reverb, not the good side).

View attachment 407308

View attachment 407310
And his auditory feeling is explained by this delay, which of course appears in GD, and the difference of 20ms is something I can't understand in any way.
Of course, some people say that Dirac would have intended it.

This cannot be normal and intended (at least in my opinion)
We absolutely hear this kind of huge GD values even at the low end of the spectrum
I believe this must be a bug in DLBC as you said
I always strive to minimize GD since it is audible above a certain threshold and I am sure @RenPa is hearing the same

 

[RenPa]

It's a bit too technical but step response swings should normally gradually fade out over time and since it's the integral of the impulse response, impulse average must be non-zero. This is not something Dirac would intentionally do and is most likely a bug. That could harm hardware in the long run, too. More info below:

https://www.perplexity.ai/search/non-zero-dc-offset-fft-RcZLWfglRIG4y_pcl3XAgA#1

OK.
Could help to decide if it worth it to mention to Dirac.
Here is what I think you show but for update 3.4.4 DLBC filter . Front left xo 80Hz. The filter was calculated after 13 measurements with Dirac

And this is the same channel , same xo but with 3.11 DLBC update and only one measurement at mlp:

 

[RenPa]

And here REW measurement for front left + 2subs DLBC off and xo 80Hz and delays in receiver.
The FR is not so cute than with Dirac on:

But GD seems better (as alignment tool predicted)

Remain to ear how it sounds. DLBC must do other choice than prioritize GD over FR response...

 

[GXAlan]

I am not sure how to interpret these, but it’s interesting. These are *before* charts.

This is with a normal premium HT sub. Mix of Power Sound Audio, dual opposed 12” and a 15” passive JBL sub powered through a Mini DSP PWR-250.


These are with a pair of Meyer Sound MM-10’s. Meyer Sound is silly expensive for not a lot of extension, but I cannot tell if this is random luck, the lack of extension, or if this reflects the benefit of their phase correction.

 

[ppataki]

And here REW measurement for front left + 2subs DLBC off and xo 80Hz and delays in receiver.
The FR is not so cute than with Dirac on:

View attachment 407460

But GD seems better (as alignment tool predicted)
View attachment 407461

Remain to ear how it sounds. DLBC must do other choice than prioritize GD over FR response...

This looks how it should (in my opinion)
You can apply some EQ to make it a bit better and you are done (given that your ears like it too )