JustMLS, LspCAD & phase glitches

[Mos Fetish]

Hi

I've just started measuring with JustMLS for use in LspCAD 5.25.

I measured my three drivers (it's a 2.5 way) on each driver's axis at 1 metre, same power level.

I also measured all three drivers connected in parallel on the tweeter axis as suggested in the manual. This enabled me to import that measurement as a target that I could align the combined response of the other three drivers against. Despite playing around with x,y,z offsets and scalings, I still had some major deviations.

I noticed that the phase response for my midbass drivers had slight glitches against the trend. That is, say, if the phase was dropping smoothly, it would suddenly jump a few degrees and then resume it's downward trend. I corrected my frd files by hand to smooth out these anomalies and the match between the target & combined driver responses became better, although no spot on.

So, are these phase anomalies normal, and is it ok to correct them to match the phase trend...

...and, are there any traps for novices regarding getting the summed individual responses to match the target response?

You'll appreciate that I don't feel I can trust my xo work if the starting response is way off.

Thanks.

Mos

[LouC]

Typically, a few wild phase shifts indicates your measuring distance may be slightly off (in 2 channel mode) If you adjust it for each driver, then the combined SPL curve will be slightly off. It takes awhile to get everything to come together. After a few sets of measurements, you'll get a lot better.

[Mos Fetish]

Thanks, Lou

My accuracy at setting the measurement difference and axis alignment is pretty meticulous.

Besides, isn't that what the x,y,z values in the driver dialogs are for? Shouldn't I be able to null any anomalies out?

I'm measuring outside using Behringer ECM8000 and the calibration is good.

Gee, I hope I don't keep banging my head against a brick wall.

Mos

[soongsc]

Originally posted by Mos Fetish
Hi

I've just started measuring with JustMLS for use in LspCAD 5.25.

I measured my three drivers (it's a 2.5 way) on each driver's axis at 1 metre, same power level.

I also measured all three drivers connected in parallel on the tweeter axis as suggested in the manual. This enabled me to import that measurement as a target that I could align the combined response of the other three drivers against. Despite playing around with x,y,z offsets and scalings, I still had some major deviations.

I noticed that the phase response for my midbass drivers had slight glitches against the trend. That is, say, if the phase was dropping smoothly, it would suddenly jump a few degrees and then resume it's downward trend. I corrected my frd files by hand to smooth out these anomalies and the match between the target & combined driver responses became better, although no spot on.

So, are these phase anomalies normal, and is it ok to correct them to match the phase trend...

...and, are there any traps for novices regarding getting the summed individual responses to match the target response?

You'll appreciate that I don't feel I can trust my xo work if the starting response is way off.

Thanks.

Mos

Is the anomaly you see during test of each individual driver or three drivers together? Is the crossover in there or not?
Would you care to post some pics as to what you think the trend should be versus what you measured?

[LouC]

There's a lot going on here so take what you see as an estimate -- not the "truth". You need to fiddle with the gate length, and smoothing functions, (I usually use 1/3 or 1/6 octave) . I usually hit the test button a dozen times or so until the graph looks pretty much the same each time. That may mean diddling with the measurement distance entry by a centimeter or two.

When you do the single point setup (measure both drivers in parallel) it does help set the acoustic center (x,y,z) and relative levels. But I've rarely had one line up exactly. (see image) This is from my Nugget2 design. Other measurements are there for comparison

http://www.lonesaguaro.com/speakers/nugget/nugget2.htm

[5th element]

Originally posted by LouC
There's a lot going on here so take what you see as an estimate -- not the "truth". You need to fiddle with the gate length, and smoothing functions, (I usually use 1/3 or 1/6 octave) . I usually hit the test button a dozen times or so until the graph looks pretty much the same each time. That may mean diddling with the measurement distance entry by a centimeter or two.

When you do the single point setup (measure both drivers in parallel) it does help set the acoustic center (x,y,z) and relative levels. But I've rarely had one line up exactly. (see image) This is from my Nugget2 design. Other measurements are there for comparison

http://www.lonesaguaro.com/speakers/nugget/nugget2.htm

Exactly what I do but I get very good matches, if I have measured right between the drivers.

Here is my measurements of a scan speak d2905/97 and a seas w15cy001. The grey is the measured combination of the two drivers, the black is the simulated combination of the two. Pretty close if u ask me. Oh and there is zero smoothing here, not in the measuring or in the program.

[5th element]

And here is the measurements for a scan speak d2905/95 tweeter and a peerles 850476. Again no smoothing and exactly the same way measured.

Now I get really good repeatable results like this, but one thing I HAVE to set by final measuring, after designing a crossover, is the level of the tweeter.

[5th element]

The way I do it is to measure the tweeter first and adjust the measurement distance until I get a phase response that looks like this.
When im talking about measurement distance here, I am refering to the value you set in JustMLS. I always measure at about 60-80cm. Leaving the microphone where it is and adjusting the distance in MLS.

Then I just measure the woofer at exactly the same distance, but lowering the mic position. Keeping all the settings in JustMLS the same.

[Mos Fetish]

Thanks for the feedback so far, everyone.

I've achived a better result by experimenting with just measuring the tweeter and upper bass-mid in my project. I left out the lower bass-mid to simplfy things and limit potential errors.

I measured the tweeter and bass at 91 cm on axis. Same settings for both measurements. For the combined measurement however, I measured on the bass axis and adjusted x,y values for the tweeter. As you can see, I've got a pretty good alignment.

But, here's the interesting part. If you look at the reading just below 5 khz you'll see a small jump in the phase and a corresponding jump in the summed response (black). Now, I checked the individual driver measurements and phase is smooth for both the bass and treble, but at exactly the point where that small spike is, that's where the bass driver's phase flips 180 degrees. This seems like an anomaly.

So, should I invert the phase of the bass to move that 180 deg change?

Plus, I also seemed to get smoother phase responses by checking the 2 sec delay in Just MLS. I can't think why this should matter. Anyone?

I'm going to now try going back to a combined reading on the tweeter axis and see if I can achieve a close result like below. Then I'll try all three drivers again.

I can't help thinking this is nothing but a phase problem, somehow.

Mos

[ergo]

I'm at work so I can not post samples at the moment but I will try to put down some guidelines I use to set up for new speaker project.

*Put the microphone on axis with the tweeter 60cm-80cm measured from baffle.
NB. If you are unsure of the tweeter capability use a capacitor in front of tweeter (15-30uF). I almost always measure without the cap - I recon that if the tweeter is not able to survive a "bshhhht " sound for a short time it's a crappy tweeter anyway.

*measure the tweeter response - make the gating to be no more than 6ms - save the file

*now do not move the microphone at all and measure the woofer response - save the file

*now wire the tweeter and woofer in parallel with same polarity - measure the response (still no moving the mic) - save the file

***********

Now make a new LspCad project.

*set the tweeter to coordinates 0,0,0, ,
*set tweeter impedance file (assumed you measured it previous to spl measurement session- if not you have to do that also)
*set the tweeter frequency response file
* set the tweeter radius/diameter to zero in ver5 or to 0.1 in ver6 of LspCad !!!!

*set the woofer to coordinates x, y , z (depend on you real location of woofer - leave z=0 for now)
*set woofer impedance file (assumed you measured it previous to spl measurement session- if not you have to do that also)
*set the woofer frequency response file
* set the woofer radius/diameter to zero in ver5 or to 0.1 in ver6 of LspCad !!!!

!!! You have to set the radius/diameter to disable the internal offaxis calculation. If you put a real value here then for woofer LspCad would take the small offaxis into account - but in this case it's already in measurement.

!!!Now next important point - set the measurement distance in settings to the same as you used for measurement.!!!

Now enable the tweeter + woofer in parallel frequency response as a reference response.

The graphs should be similar but not quite right. Opening the woofer parameter window and adjusting Z coordinate should at one point get the graphs to mach almost perfecly (note the Z can be with " -" also due to the time of flight with this measurement, but it does not matter)

I have had very very good matching of the curves this way. The small offaxis for the woofer response is in my opinion much smaller mistake than not getting the response to exactly mach. You can do measurement of woofer on tweeter axis and woofer on woofer axis and compare - for woofers upto 6'' there really is not much to worry about.

So I usually use the same measurements for crossover design itself also. I just change the radius/diameter to correct value and the measurement distance to 2m for example.

(If you want to verify some more measure everything once more just put the tweeter in reverse when wiring them in parallel. Then don't forget to reverse the tweeter in LspCad also. Now you get different graph but the Z value should come out the same)

Ergo

PS. I have samples of many projects done this way and it works very well - upto measuring the real crossover in the end and comparing it with simulated one.

one sample