Acoustic phase measurements of loudspeakers

[daftcombo]

Hi,

@amirm does an incredible job of measuring speakers, including electric phase & impedance.

I am a bit surprised though not to see any acoustic phase measurements (the kind of measurement one can do with REW on-axis). Is it because it is of no interest?
Couldn't it give a clue about the quality of the crossover(s) implementation?

Cheers.

 

[RayDunzl]

I measure at the listening position.

The JBLs have a steeply sloping line, high at the left, thousands of degrees down at the right. If measured nearfield, it's pretty flat.

Nearfield, maybe a meter distant, and one trace at listening position at ten feet.

1/12th smoothing:

In the same room, the MartinLogan electrostats throw a flat phase response from the panels to the listening position, and have some slope in the woofers. There's a jump around the crossover (180Hz), but the frequency value of the jump sometimes shifts upward among measurements taken at different times.

I haven't tried to measure the ML nearfield. No need.

No smoothing here:

Of course the ML would fail the approved criteria for a good speaker, so, discard anything you see above.

 

[pozz]

I wonder why the MLs produce such a flat response. Reflections at the microphone position should cause wild deviations. Something to do with them being a line source? Another factor that I'm overlooking?

 

[Soniclife]

I wonder why the MLs produce such a flat response. Reflections at the microphone position should cause wild deviations. Something to do with them being a line source? Another factor that I'm overlooking?

My guess, narrow dispersion.

 

[RayDunzl]

Another factor that I'm overlooking?

I think it has to do with delayed reflections.

The JBLs measure flatter the closer you get to the drivers, but their dispersion sprays the room.

The JBL ETC gives about 10 to 15dB difference between direct and reflected SPL (if I read that measurement correctly)

The ML are much more directional, with 25 to 30dB ratio of direct to reflected.

JBL - red, ML - black and foreground

Unsmoothed frequency response:

 

[andreasmaaan]

Hi,

@amirm does an incredible job of measuring speakers, including electric phase & impedance.

I am a bit surprised though not to see any acoustic phase measurements (the kind of measurement one can do with REW on-axis). Is it because it is of no interest?
Couldn't it give a clue about the quality of the crossover(s) implementation?

Cheers.

In many cases, it seems that @amirm posts the step response, which is essentially the same information.

For example, from the Klipsch R-41M review:

 

[daftcombo]

In many cases, it seems that @amirm posts the step response, which is essentially the same information.

For example, from the Klipsch R-41M review:

Thank you, but how do you read that kind of graph?

 

[tuga]

Of course the ML would fail the approved criteria for a good speaker, so, discard anything you see above.

What do you mean by "good speaker" and why is the ML not one?

Do you mean subjectively "good"?
Isn't that...hmm, errr...subjective?

 

[pozz]

@RayDunzl can you post the RT60 graph for the MLs?

 

[andreasmaaan]

Thank you, but how do you read that kind of graph?

You have to make a few educated guesses or inferences based on what you know about the speaker to interpret it. It's definitely not as straightforward as a phase response graph.

There's a good basic explanation here, and a better, more detailed explanation here (about 1/5 of the way down the page). It was also discussed in some depth on ASR here.

Hope that helps a bit?

 

[andreasmaaan]

@daftcombo I'm procrastinating on another task I'm finding tedious, so I thought I'd write a quick intro myself to interpreting step responses

Here I'll try to illustrate with some idealised examples.

Firstly, here's a two-way speaker in a sealed box with an F3 of 40Hz and an LR4 crossover at 2kHz. Let's assume it has a more-or-less flat frequency response, i.e. the outputs from the two drivers sum correctly on the measurement axis.

The first peak at 1ms is the output of the tweeter. It's initial rise is positive, from which we can infer that it is connected in positive polarity.

The second peak at about 1.4ms is the output of the woofer. Its initial rise is also positive, so it must also be connected in positive polarity.

Since both drivers are connected in the same polarity, we know this is not a 2nd or 6th order filter (which would require each driver be connected in opposite polarity in order to sum correctly).

We can see that the output of each driver integrates smoothly with the other. The two drivers are time-aligned.

If this were a first-order crossover and the drivers were time-aligned, we would see only one peak (as the outputs would sum to linear phase). But here, we have two peaks. So we can rule out that it’s a first-order crossover (and we've already ruled out that it's a 2nd or 6th order crossover). From this, we can infer it's a 3rd, 4th, or 8th order crossover.

With an 8th order filter, you’d expect to see more ringing in the tweeter’s output prior to the peak of the woofer’s output than you see here (and the woofer’s output would also be delayed more in time). So from this we can infer that we’re looking at a 3rd or 4th order crossover here.

At this point, it would be hard to know which of these it is, but if we looked at the vertical polar response and saw that the off-axis nulls were located equidistantly from 0°, from that we could infer that it was not 3rd order (since if it were, the lobes would be tilted downward). That leaves us with a 4th order crossover.

To illustrate how this graph might look if the speaker were designed differently, here is the same hypothetical speaker in the same box, but with a 2nd order crossover at 2kHz this time. Note that the woofer's initial rise is now in the negative direction, because it is now connected in negative polarity:

And now the same speaker with an 8th order crossover. The woofer is now back in positive polarity, but note the significant ringing of the tweeter before the peak of the woofer, and also that the woofer's peak arrives later relative to the 4th order crossover (which in turn arrives later relative to the 2nd order crossover):

I realise this is not particularly comprehensive, but hope it helps some

 

[RayDunzl]

can you post the RT60 graph for the MLs?

3, 6 - JBL without and with "correction"
9, 12 - ML without and with "correction"

I'd venture RT60 is not particularly speaker dependent...

 

[pozz]

3, 6 - JBL without and with "correction"
9, 12 - ML without and with "correction"

I'd venture RT60 is not particularly speaker dependent...

View attachment 56775

It wouldn't be, really.

That reverb time is definitely standard. With that many reflections bouncing about, I'm still confused how the phase response can be so flat.

 

[RayDunzl]

With that many reflections bouncing about, I'm still confused how the phase response can be so flat.

Well, all I can figure is:

The highest level bounce with the MartinLogans is at 7ms, the reflection off the wall behind, and the second highest level bounce is the distance of speaker to wall behind the mic (mic had some absorption behind it) to the wall behind the speakers to the mic at 27ms.

The JBLs (red) have many reflections, never tried to figure out from where they come, with refelctions (relative to direct) as much as 20dB higher in level.

 

[edechamps]

Is it because it is of no interest?

I am sceptical it will tell you anything about perceived audio quality. I have yet to find any studies that show any significant impact of phase response (or any other time domain behavior, for that matter) on loudspeaker preference in realistic conditions.

However, I am interested in phase response for a different reason: it can help selecting speakers that can be combined together in one system. For example, let's say I want to buy a pair of front speakers and a (different) centre speaker. Ideally, I'd want to use a centre speaker that has a similar phase response to the other speakers, otherwise, they will be out of phase in some parts of the spectrum, and will therefore interfere with each other. That, for me, is a good reason to publish phase data.

 

[pozz]

Well, all I can figure is:

The highest level bounce with the MartinLogans is at 7ms, the reflection off the wall behind, and the second highest level bounce is the distance of speaker to wall behind the mic (mic had some absorption behind it) to the wall behind the speakers to the mic at 27ms.

The JBLs (red) have many reflections, never tried to figure out from where they come, with refelctions (relative to direct) as much as 20dB higher in level.

View attachment 56802

View attachment 56803

I guess the question to ask is what does it take to measure flat phase from the listening position in a regular room. It's not a question I can answer.

My own measurements in my old room with the ADAM S2Vs showed >2000° across the spectrum IIRC, which fits what I've read (not the number, but the general effect of reflections on phase).

 

[RayDunzl]

I guess the question to ask is what does it take to measure flat phase from the listening position in a regular room. It's not a question I can answer.

Well, I can't either.

I "cherry picked" the measure above, if only a little. You'll see below.

---

Passive crossover woofer to panel is 180Hz

The electrical impedance/phase doesn't seem to matter .

Prior iteration - ML Quest Z

That's not flat.

---

Measurements aren't always flat, and the two speakers can add strangely to become flat:

L, R, Both, without DRC:

L, R, Both, with DRC:

Twenty four measures of both speakers with increasing SPL ,measures (I expected more variation):

It was a "compression" test. 46 to 92dB SPL.

Lots of overlap, but amusing variations.

"Flattest" at highest SPL.

My own measurements in my old room with the ADAM S2Vs showed >2000° across the spectrum IIRC, which fits what I've read (not the number, but the general effect of reflections on phase).

That's what the little JBLs do at the listening position.

 

[daftcombo]

Anybody tried to enter their crossover values in RePhase, generate a convolution correction file and listen to the result (and re-measure)?

I did, but can't ABX.

 

[andreasmaaan]

Anybody tried to enter their crossover values in RePhase, generate a convolution correction file and listen to the result (and re-measure)?

Yeh, have done something like this with a pair of 3-way speakers with LR4 crossovers.

I measured the acoustical phase of the speakers and then corrected it with a filter generated in RePhase (I only partially corrected the bottom end roll-off, though), then measured to confirm the filter had worked (which it had).

In that case, I thought I could hear a difference when I first implemented the filter, but then was not able to reliably discern a difference when informally ABX testing myself. I didn't try too hard though - perhaps with more work I could have got to the point of reliably detecting a difference. But if there was anything audible there, I decided it was so subtle as to be unimportant.

I did, but can't ABX.

If you convolve the RePhase filter with a recording, you can ABX it against the original using Foobar.

 

[levimax]

Anybody tried to enter their crossover values in RePhase, generate a convolution correction file and listen to the result (and re-measure)?

I did, but can't ABX.

I use DIY analog active 3 way crossovers so I know they are LR4. I also use Rephase for final EQ adjustments and also have played around with phase alignment. I find the "Linearization Filters" to work more less perfectly and will get you very close to "phase aligned" and with some small manual adjustments and you can get "perfect".

When I listen after the phase is "perfect" of course I hear "tight bass", "more punch", "cleaner vocals". I then do ABX with or without the phase adjustments and I can not tell any difference at all .

Currently I use the "Linearization Filters" only (to get me 95% there) as it is quick and easy and I don't worry about any fine adjustment as I can't hear it anyway but figure why not get "close"