edit time over and I've got to expand this. I do not know if there is something in my system that makes it so short in distance, the transition wihtin about one step and it could be more of a gradual slide with some other system. I also do not know how other people perceive it with their speaker systems, although some friends I've made to listen for it say it's quite easily audible and at least one has found it with his home stereo as well. Griesinger listening test with LLD in a small room using loudspeakers seems to suggest that people agree about LLD about at same distance, which lets me believe the transition could be quite short in many home stereo setups and with most people. If this is all true I think it's very powerful concept to help relate to each others perception over internets discussion forums, if everyone knew theirs and which one everybody is talking about, I think it would reduce a lot of confusion from discussion.
ps. I've been able to mellow out the transition a bit, which also makes it a bit further into the room just by attenuating lateral early reflections by experimenting with toe-in and utilizing some acoustic panels. Still, the effect is pretty much the same as before and quite close where it was, very easy to spot.
Last edited:
You're not the only one, dive into some books about psycho-acoustics, and you will find the same conclusions.
But it's great to read that you came to the same conclusion! 🙂
Great to read that people are actually getting their hands dirty and experience these kind of tests!
Especially high Q (small band) dips are almost impossible.
Absolutely fascinating how our brain fills in those gaps.
Which in a sense is kinda nice, because very often these kind of problems are related to destructive interference issues.
Meaning that we can EQ forever, but the issue will always be there.
It also tells something else that doesn't seem to be obvious.
Because this also means that it falls into the "audibility-very-hard" category.
Or in other words, we have to reeeeally focus to hear any differences.
Keyword here is differences, NOT issues/problems.
Meaning that we simply can prioritize these issues much lower on our list!!!
And my apologize if I use that word again (because it seems people don't like it), but this is exactly what I mean with pixel peeping.
Small scale dips are just almost impossible to catch.
I would bet all my money that the very vast majority of well trained listeners wouldn't be able to hear them unless you told them/showed them beforehand.
Than still it will be a question of bias, I think in a double blind AB or ABX test it will still be extremely hard.
and especially @vineethkumar01 thanks for your measurements!!
This confirms what I predicted on the previous page (which was getting a bit lost in the noise, sorry for that).
Please, also notice how much obvious these things are in a normalized graph.
Btw, this can be predicted beforehand by doing a preliminary analyses on the bafflestep and diffraction simulations.
Although this is also always a good example that function follows form.
Often we just simply want to have a certain size cabinet.
This effect is often also less with a proper waveguide btw! 🙂
So now people maybe understand why so many are fan of waveguides.
It fixes so many problems at once! Or at least makes the compromise a little better.
Thank you for all the interesting discussions...
@tktran303 and @PKAudio were discussing the directivity of the SB15NBAC driver in this cabinet. I have used the very similar SB15CAC driver as a midrange in a tower speaker with a 10" wide baffle. Here is a comparison:
One thing to keep in mind is that the overall directivity of this speaker is governed by the mid driver up to about 1.6k, and between 1.6k and 3.2k it is a blend of the mid and tweeter. Above 3.2k it is all tweeter. So the bunching at 2k is offset greatly by the very smooth directivity of the tweeter in the 1.6k-3.2k range.
Again thanks for all the discussion, it will take me a while to digest all of it. 🙂
j.
@tktran303 and @PKAudio were discussing the directivity of the SB15NBAC driver in this cabinet. I have used the very similar SB15CAC driver as a midrange in a tower speaker with a 10" wide baffle. Here is a comparison:
One thing to keep in mind is that the overall directivity of this speaker is governed by the mid driver up to about 1.6k, and between 1.6k and 3.2k it is a blend of the mid and tweeter. Above 3.2k it is all tweeter. So the bunching at 2k is offset greatly by the very smooth directivity of the tweeter in the 1.6k-3.2k range.
Again thanks for all the discussion, it will take me a while to digest all of it. 🙂
j.
@hifijim - I've been reading the recent few days posts and I'm lost as well. (Clearly I have a lot to learn still) So I feel a bit silly asking these basic questiosn, but could you help me understand how you do your measurements based on the bullets below:
- How far is your microphone from driver being measured for Nearfield, Farfield, and Directivity measurements?
- Do you use an audio reference or loopback for calculating delays? If you use Audio Reference, do you keep the reference at the same location for NF and FF measurements?
- Do you measure in your listening room, or do you use a quasi-anechoic measuring location (with fewer reflections) for making EQ and XO decisions?
- What IR Windows and/or Smoothing do you use, and do you change them or keep the same settings for LF, MF and HF?
I'll piggyback on Sixto's questions with one of my own. This is for Jim and any others using VCad. Are you measuring the full 360 degrees in both horizontal and vertical? I thought I saw someone mention once using some sort of simulation of the vertical. That confused me.
Augerpro: I do not measure Vertical responses. Just Horizontals. You need to do two things:
1. In Preferences check "include Verticals". This way Vituix takes Horizontals and fill them in to Verticals. Of course this not entirely accurate, but Kimmo once posted somewhere that the error is quite small, and I consider it better than no verticals at all.
2. Zou need to measure the drivers individually in their height, with the same distance of the mic from the baffle. Then you enter XYZ of the drivers on the baffle. This way Vituix knows the relative distances of the drivers and manipulates the responses (relative SPLs, angles,...).
1. In Preferences check "include Verticals". This way Vituix takes Horizontals and fill them in to Verticals. Of course this not entirely accurate, but Kimmo once posted somewhere that the error is quite small, and I consider it better than no verticals at all.
2. Zou need to measure the drivers individually in their height, with the same distance of the mic from the baffle. Then you enter XYZ of the drivers on the baffle. This way Vituix knows the relative distances of the drivers and manipulates the responses (relative SPLs, angles,...).
Only +/-90° hor and ver, as measuring also the other halfspace would nearly double the measurement points for effort, but gives not that more insight (maybe when you develop a bass cardioid or so it makes sense). Sometimes I measue only 60° hor and 40° ver, because I experience that the power response of that ballon is a nice sufficient target curve for me, here two examples (dotted lines are the power response hor +/- 60° and ver +/- 40° seperately; straight line is balloon sum):
Have never simulated the vertical.
PKAudio> That sounds like the method I saw mentioned once. I'll have to disagree with Kimmo that the error is small. The responses are very different even for a bookshelf speaker. For a floorstander it would be extremely different, though I'm still working on a measurement stand to do those and don't have a plot to illustrate the difference. At least VCad will include driver offset into the sim, so there is that.
Kwesi> I agree that going very far offaxis may not provide much more information. My concern comes from then generating the CEA2034A plots based on that incomplete dataset. I think this makes for better plots than is real. For example your vertical and horizontal range does not even complete the Early Reflection metric in the CEA2034A standard. Without those, you can't correctly determine DI, or SP, or PIR.
To be clear, I am not trying to be critical or nitpicky. I've used a method like Kwesi's for years, and I believe it is a good one. But I think if we are going to start using the CEA2034A in our plots, we should actually be following the standard. Otherwise we are lying to ourselves, and would be better off with our prior methods.
Kwesi> I agree that going very far offaxis may not provide much more information. My concern comes from then generating the CEA2034A plots based on that incomplete dataset. I think this makes for better plots than is real. For example your vertical and horizontal range does not even complete the Early Reflection metric in the CEA2034A standard. Without those, you can't correctly determine DI, or SP, or PIR.
To be clear, I am not trying to be critical or nitpicky. I've used a method like Kwesi's for years, and I believe it is a good one. But I think if we are going to start using the CEA2034A in our plots, we should actually be following the standard. Otherwise we are lying to ourselves, and would be better off with our prior methods.
For nearfield, the mic is very close, about 1/4 inch. For farfield, I use 36" to 40". There are some guidelines based on baffle width, and if I recall correctly it was that mic distance should be at least 3x baffle width, but we need to verify that.
I use a 2-channel audio interface with a USB connection to my laptop. Until recently I used a Behringer UMC202HD, which is sometimes on sale for less than $100. It worked great, no complaints. Recently I upgraded to a Motu M2. My mic is an Audix TM1.
I use ARTA to capture the impulse response for a given scan, whether NF or FF. I then use the VituixCad to convert the impulse response into a Frequency Response. Vcad has functionality that helps set the correct parameters for NF or FF. The key difference is that NF measurements use a very long time gate, or no time gating at all. FF measurements use a time gate to exclude the first reflection.
If Vcad is used correctly, the timing reference for the FF response is the same for all 3 drivers. There is no need to keep the timing reference constant between a NF and FF scan of the same driver, and I am not even sure how that might be accomplished anyway.
During the merge process, the phase of the NF response is matched up to the FF response, and this process accounts for the natural difference in timing reference between NF and FF.
Generally the listening room or my basement work shop. To get a good 4 ms gate, you need about 40" of clear space around the driver and the mic... in other words, the nearest floor, wall, ceiling, or large surface should be more than 40" away from the driver. Same with the mic, more than 40" to the nearest surface.
The time window, or gate, that I use depends on the first reflection. When you let Vcad help you, it usually suggests a 4.5 ms gate for FF scans. There is a preview window that allows us to see the impulse response, and the first big reflection is usually quite apparent. I just set the gate to be just ahead of the first reflection. However, Vcad uses a mathematical process called Tukey windowing which makes the window size less important, so the standard 4.5 ms gate is often good enough, even if it contains some reflections.
It is not important that the woofer, mid, and tweeter all use the same gate. However, the longer the gate, the better the low frequency resolution becomes. It is ironic (perverse) that with large heavy speakers, the woofer is often on the bottom, and it would be the driver which would most benefit from a long gate. Being on the bottom and close to the floor, the woofer often has a very short gate. This is why testing big heavy speakers is so painful, because they must normally be elevated high enough to get the woofer at least 40" off the floor.
At this point, I measure horizontal polar scans, in 15 degree increments, starting at 0 (on axis) and going to 180 degrees. Vcad assumes that the horizontal polar responses are the same left and right. If the speaker has a non-symmetrical baffle, such as an offset tweeter and midrange, then I would definitely scan to both sides, 0 to 180 and 0 to -180.
The user guide for VituixCad says that it is not necessary to do vertical polar responses. Vcad assumes the horizontal and vertical responses are the same, and according to the users guide, this approximation is close enough.
I was quite skeptical of this, so when I built those tower speakers, I went to the trouble of scanning both horizontal and vertical, from +180 to -180. After I completed the project, some members challenged me to recreate the simulation using only the horizontal scans, and I was surprised to find that Vcad did a great job simulating with only the horizontal polar measurements. Definitely close enough that I would have designed the same crossover had I only had the horizontal data to work with. It was quite surprising to me because the tower speakers were so different in the vertical plane than in the horizontal plane. I can find a link to the thread if you like.
j.
I very rarely measure more than 90 degrees.
That's in many cases more than enough to see how well directivity behaves.
Have a look at the attached. Not great for your Elliptical waveguides.
My view is "better something than nothing", full measurement is always better. I do not see this method as something particularly accurate, but rather as insight in what happens vertically. This is definitely better than sims with ideal drivers. And additionally, I always switch the view between with/without verticals. Anyway, ears are always the final
Here is the link to the study I did... I compared actual measured performance to a simulation using horizontal and vertical polar responses for each driver, and also to a simulation using only horizontal polar driver responses.
https://www.diyaudio.com/community/threads/vituixcad.307910/post-7251651
I think most of the vertical polar response of a speaker system is driven by the driver spacing, and the magnitude and phase of their responses through the range where they overlap. The baffle-induced directivity itself is less important in this case, so it is reasonable to approximate the vertical responses of the drivers by using the measured horizontal data.
Now the case of a waveguide tweeter with an elliptical mouth may be a more specific case, because the vertical and horizontal patterns are intentionally different. If I designed a speaker using a non-circular waveguide, or a ribbon tweeter, or a horn with different vertical /horizontal beamwidths, then I would certainly measure both the vertical and horizontal polar responses for that driver. But a cone or dome midrange ? probably not.
I believe that to get a good calculation of power response, we need polar responses out to 180 degrees.
j.
https://www.diyaudio.com/community/threads/vituixcad.307910/post-7251651
I think most of the vertical polar response of a speaker system is driven by the driver spacing, and the magnitude and phase of their responses through the range where they overlap. The baffle-induced directivity itself is less important in this case, so it is reasonable to approximate the vertical responses of the drivers by using the measured horizontal data.
Now the case of a waveguide tweeter with an elliptical mouth may be a more specific case, because the vertical and horizontal patterns are intentionally different. If I designed a speaker using a non-circular waveguide, or a ribbon tweeter, or a horn with different vertical /horizontal beamwidths, then I would certainly measure both the vertical and horizontal polar responses for that driver. But a cone or dome midrange ? probably not.
I believe that to get a good calculation of power response, we need polar responses out to 180 degrees.
j.
Rule of thumb for far-field is 3x longest baffle dimension, which of course makes far-field measurements indoors very difficult.
Given that, and the questionable usefulness of measurements, say with 4ms gating and only 250Hz freq resolution, I can't help but try to measure outdoors...some how.
If you can manage to make outdoor measurements with longer gates, i humbly suggest it will probably make you rethink how to consider minor VCad differences.
Last edited:
I added a simple 1st order BSC and 2k LR4 low pass... First the SB15NBAC from this project
Now the SB15CAC from the other speaker
That's why I do not show weighted target curves like early reflections and PIR or so in my charts above, there is too limited data. Therefor just the summed power of a +/-60° by +/-40° section of the full sphere, and the two separate contributions. And you can see that the concept works generally, to somewhat compensate a horizontal hump with a vertical drop by placing the xover frequency accordingly.
But for a +/- 90° hor + ver dataset I would not hesitate to use and show full CEA 2034 curves.
Last edited:
Omitting the back does throw off the baffle step.
Many overlook the fact that the diffraction has different directivity than the speaker.
I understand your point @mark100 . I have made outdoor ground plane measurements where the nearest surface was 26 feet away, and I realize the limitations of merging a NF response with a FF response gated to 4 ms. It is not the same.
But the NF data can reveal any narrow band issues with the low frequency drivers, and probably the box as well. The merged data, on the other hand, is very useful for calculating active or passive crossover values even if it has low resolution in the region of the merge frequency. If I had a full spectrum high resolution curve of 10 Hz resolution for each driver, I am not sure how that added information would allow me to design a better speaker. It would remove some uncertainty, absolutely yes, and it would save me some time. But would I make different decisions? I am not sure.
But the NF data can reveal any narrow band issues with the low frequency drivers, and probably the box as well. The merged data, on the other hand, is very useful for calculating active or passive crossover values even if it has low resolution in the region of the merge frequency. If I had a full spectrum high resolution curve of 10 Hz resolution for each driver, I am not sure how that added information would allow me to design a better speaker. It would remove some uncertainty, absolutely yes, and it would save me some time. But would I make different decisions? I am not sure.