On equalization filters (Dirac Research)

[LionIT]

I was reading this interesting document, On equalization filters by Mathias Johansson (ex Dirac Research AB).

Towards the end of the document this is written:
For a subwoofer channel, however, minimum-phase inversion will typically be sufficient as the combined room and subwoofer transfer function actually is well modeled by a minimum-phase system at such low frequencies.

I can't understand exactly why this is said. The combined subwoofer and room transfer function should have phase excess characteristics that cannot be equalized with a minimal phase filter simply.

What am I missing?

 

[ppataki]

I was also reading some of their earlier white papers
They claim to be using mixed phased filters, meaning up to XHz they apply minimum phase filters and then above XHz they apply linear phase filters (if my understanding is correct)
So I guess he is just re-confirming that in the above quote
If he is right or wrong I cannot tell but other more knowledgeable folks here will comment on that I am sure

 

[ppataki]

@OCA might be able to shed some light on this
Thank you

 

[LionIT]

There is also another very interesting reading of Dirac Research AB, Controlling the impulse responses and the spatial variability in digital loudspeaker-room correction.
Here their mixed phase approach is explained in more detail, but no reference is made to the representation of the low frequencies with a minimum phase.
They concluded this however:
Constrained, robust mixed phase multipoint design. This solution takes the MSEs (mean squared deviation) at all control points into account and minimizes a weighted average of them under constraints. The constraints may include constraints on loudspeaker powers at different frequencies, a constrained modeling delay and constraints on the allowable pre-ringings of the compensated impulse responses at all control points in the targeted area. Please see [1] for a detailed discussion. The resulting compensators are stable and causal IIR filters, with as long impulse responses as needed. The performance of this strategy is good in an average sense in the frequency domain, and it also corrects time-domain properties. It provides a slightly worse MSE than an unconstrained Wiener solution, but the sound quality is much better from a psychoacoustic perspective.

That "...and it also corrects time-domain properties." is hard to understand to me, referred to 1st post sentence.

Maybe @Flak can provide a contribution?

 

[ppataki]

and it also corrects time-domain properties." is hard to understand to me, referred to 1st post sentence.

That means this:

Take a look at the below step response curves

They are from a Nubert nuVero 140 speaker (that I used to own like 5 years ago before I moved to DIY)
3.5-way, 7 drivers per cabinet

The red curve is the step response without Dirac - you can clearly see the high-mid-low drivers' separation in time (look at the graph from 0ms to 1ms)
The purple curve is the same speaker but with Dirac - you can see a perfect time alignment, the drivers are no longer separated in time (just one peak instead of several)
This will result in a much better imaging and sound will be more 'detached' from the speakers (quasi-point-source)

This huge difference is also clearly visible on the phase curves:

With Dirac it became way more linear

 

[LionIT]

That means this:

Take a look at the below step response curves

View attachment 408211

They are from a Nubert nuVero 140 speaker (that I used to own like 5 years ago before I moved to DIY)
3.5-way, 7 drivers per cabinet

The red curve is the step response without Dirac - you can clearly see the high-mid-low drivers' separation in time (look at the graph from 0ms to 1ms)
The purple curve is the same speaker but with Dirac - you can see a perfect time alignment, the drivers are no longer separated in time (just one peak instead of several)
This will result in a much better imaging and sound will be more 'detached' from the speakers (quasi-point-source)

This huge difference is also clearly visible on the phase curves:

View attachment 408212

With Dirac it became way more linear

Thanks.
I think I understand well what you say.
What I don't understand is how excess phase is corrected at low frequencies.
But perhaps, also seeing various measurements online, it is not corrected. Which makes me wonder what the rational is for not correcting it. Not audible?

 

[ppataki]

You are welcome!

Excess phase is somewhat corrected in the low frequencies:

with Dirac:

Down till 150Hz the correction is significant, below that it is minor

 

[LionIT]

Down till 150Hz the correction is significant, below that it is minor

That is exactly the region I mean. Bass in practice.
Other software like Audiolense or Acourate uses FIR to correct that region, and it seems they are considered effective for this, although they do not use a multi-point measure (ref. Mitch Barnett reviews).
I am curious to understand, if so, why Dirac considers it unnecessary to do so (or not convenient).
Are low freq in room effectively minimum phase?
Or perhaps once the low frequency response in magnitude has been sufficiently equalized, the excess phase, even if not beyond certain levels, is inaudible or even preferable?
Or is there a question of unacceptable latency in processing high wavelength?
I saw that there is also a thread here where it is discussed the alleged problem of the excess phase with Dirac BC.
I would like to understand what the reality of things should be ...

 

[ppataki]

That is exactly the region I mean. Bass in practice.
Other software like Audiolense or Acourate uses FIR to correct that region, and it seems they are considered effective for this, although they do not use a multi-point measure (ref. Mitch Barnett reviews).
I am curious to understand, if so, why Dirac considers it unnecessary to do so (or not convenient).
Are low freq in room effectively minimum phase?
Or perhaps once the low frequency response in magnitude has been sufficiently equalized, the excess phase, even if not beyond certain levels, is inaudible or even preferable?
Or is there a question of unacceptable latency in processing high wavelength?
I saw that there is also a thread here where it is discussed the alleged problem of the excess phase with Dirac BC.
I would like to understand what the reality of things should be ...

I will let others comment, I guess this is until my current knowledge extends...

Just one practical remark: purely from a subjective perspective, I totally prefer low group delay in the bass range
I really don't know why DLBC increases it though....to me it is counter-intuitive

 

[NTK]

I will let others comment, I guess this is until my current knowledge extends...

Just one practical remark: purely from a subjective perspective, I totally prefer low group delay in the bass range
I really don't know why DLBC increases it though....to me it is counter-intuitive

Our hearing has very low sensitivity to group delay in bass.
From: https://acris.aalto.fi/ws/portalfil...udspeaker_Group_Delay_Characteristics_AAM.pdf
Note that the tests were done with the most sensitive/revealing signals known to the authors, and thus we are likely even less sensitive with real life materials.

 

[ppataki]

Our hearing has very low sensitivity to group delay in bass.
From: https://acris.aalto.fi/ws/portalfil...udspeaker_Group_Delay_Characteristics_AAM.pdf
Note that the tests were done with the most sensitive/revealing signals known to the authors, and thus we are likely even less sensitive with real life materials.

This is true, however I am talking about a different scale. In this article they say 10ms of GD is fine; however with DLBC you will get in the range of 100ms (as discussed in this thread). That is definitely audible and I would never want to get in that range in my system (hence I am not using DLBC but I integrate my subs manually to minimize GD)