A 3 way design study

[vineethkumar01]

Next round of measurements. This time with Wavecor WF120BD03. Please dont ask me why the driver adaptor is red.. (I ran out of other filaments)

I noticed that 3-4 kHz bunching. It is probably due to baffle size and shape I think.
This is what VituixCAD diffraction tool shows for the baffle simultion for this driver:

I designed a preliminary crossover with above data:

This one looks slightly better than my SB15CAC graph I think..

Still struggling with taking vertical polar measurements due to the curved cabinet.. I will try to do something about it.

Please find the VituixCAD project attached with this post

Thanks
Vineeth

 

[hifijim]

Still struggling with taking vertical polar measurements due to the curved cabinet.. I will try to do something about it.

The first thing I would try is to make a soft cushiony shelf using towels and bits of foam. This will conform to the curved shape of the speaker.

The little wavecor driver has nice performance, and your preliminary crossover has a lot of potential.

 

[fluid]

Still struggling with taking vertical polar measurements due to the curved cabinet.. I will try to do something about it.

Sounds like a job for some 3D printed wedges

I have had a play around with the project data. There is something going on with the woofer to mid transition that is making a good phase blend hard without a lot of delay. It doesn't really make sense to me, it's like there is an unintentional phase inversion somewhere, so I gave up on that and stuck to the top two drivers. The waveguide having the listening window drop down from the on axis was somewhat unexpected, I wonder if this has something to do with the missing vertical as the waveguide is not round.

This is aiming to keep the listening window flat with a slight down slope. There is an intentional phase mismatch through delay to smooth the directivity. A fully time aligned LR4 has a power dip. I have played around a bit trying to balance the on vs off axis around the 3K region where the baffle diffraction is still a little bit problematic. The EQ started out simpler but all of them do something to benefit all curves so it seems silly not to use it in an active system.

 

[vineethkumar01]

Thanks a lot @fluid and @hifijim for the ideas regarding ideas for taking vertical measurements..
I will try those out.. I also have to measure the woofer box once to confirm whether the measurements are ok.

Due to work pressure, I haven't been listening to the speakers much.. so more evaluations are to be taken of each configuration.
I had seen something about Trifonov transient perfect crossovers being mentioned by Kimmosto a while back. I tried to see what it is and tried to use it in the all SB 3 way setup (with mid on top and tweeter down).
This is what I got:

The vertical responses and phase plot look like a mess But I saw interesting shapes in the impulse ans step response.I added genereous amounts of delay seeing the speaker pics here:
https://trifonovaudio.wordpress.com/three-way-2/

and Kimmosto's own speaker responses like below:

Are these kind of crossovers any good? (Probably not my implementation of it, but in general)

 

[augerpro]

I think you probably need the correct acoustical alignment to get the transient perfect behavior. Fluid's iteration is looking pretty damn good!

 

[fluid]

Fluid's iteration is looking pretty damn good!

Thanks, your little waveguide does quite well with the smaller driver.

As to the the transient perfect type stuff, unless latency is a problem a simple fully complementary FIR crossover will do that much better.

 

[fluid]

The version I did by hand had an Olive score of 8.44 according to equation 9. I've since let Vituix optimize by score, first for equation 9 which gave a score of 8.7.

Another version was optimized with the custom equation targeting a PIR of -1.3 and this one cane out at 9.22.

It would be interesting to listen to them to see if there is any appreciable difference in how they sound. Given the ripple in the measurements and the cabinet diffraction at 3K, those scores seem as high as is realistic.

It is interesting to see what changed to make the score higher. Just to show that the score can quite fallible, optimizing for equation 10 created a really weird rising response but the score was over 10, I think 11 might the maximum for that one. Vituix preset attached variations under R2 and R3.

 

[vineethkumar01]

I am definitely going to listen to all these variations in crossovers..

Yesterday, while trying to do a sweep at a little higher volume level on the amp, I heard a slightly annoying buzzing noise from the 3D printed adaptor in the middle of the sweep.. It was occuring over some small band of frequencies.. Dont know what exactly caused it but once i heard it, i couldn't forget it..
I am constantly trying to hear that noise in the middle of everything that i am playing on the speaker..

So I am in the process of asking my friend to help me with getting an adaptor made in mdf/wood.

 

[nc535]

Using the optimizer is new to me so I'm having fun, although shooting in the dark to some extent. Running the optimizer on an already optimized response sometimes drives the result away from optimum - to a result with the response rolled off at former XO. I haven't figure that out yet; when this happens, I got back to unoptimized responses for new runs. (If it hurts, don't do it, (unless working out).)

I got a pref rating of 9.5 optimizing to the LW. This produced a flat LW curve while the axial rose about 1.5 db from 900 hz to18 khz. I noticed the optimizer cares not a whit about the height of the vertical response near XO. Manually tweaking the delay on the mid, I was able to optimize it with little change in the pref rating. Changing from 22 usec to 2 usec was all it took.

Listening would be the next step. Likely the sound is quite good. If not, you might consider doing something about the diffraction as dealing with it is driving the XO design.

 

[tmuikku]

Yeah, VituixCAD is very powerful and the xo can be optimized to anything quite easily. Basically crossovers can be considered trivial.

All we need to do is build such structure whose measurements are flawless to get flawless end results Or more down to earth approach: we need to practice the measurement procedure so that our measurements are error free.

Beside that we need to listen to a lot of the stuff, variations of the crossover, in order to figure out what sounds better in order to be able to optimize for the right thing.

I haven't done new xo:s for a while now, need new set of measurements. Because, before I did tens of revisions and almost always ended up using one with lower "score". Moreover, after putting the setings into the DSP and listening there was always need to tweak it a bit. Reflecting the changes back to the simulation makes different curves of course, usually dropping the preference rating numbers some. Lesson is, either my measurements have error, or DSP settings have error, or I'm not sure how to relate perceived sound quality to graphs. All these need some work and practice to get better, experience. Now we are back to the beginning, experience makes better sound like in the old days before trivial simulation procedures. Sometimes frustrating, but in the end its just fun time with the hobby

This is a good thread, effort and combined experience of multiple people to perhaps all learn together, keep on keeping on!

 

[fluid]

I got a pref rating of 9.5 optimizing to the LW. This produced a flat LW curve while the axial rose about 1.5 db from 900 hz to18 khz. I noticed the optimizer cares not a whit about the height of the vertical response near XO. Manually tweaking the delay on the mid, I was able to optimize it with little change in the pref rating. Changing from 22 usec to 2 usec was all it took.

If you are still using the Vituix transfer function inversion process that creates a very smooth response in some of the important curves. The algorithm responds very well to this and the difference in the score can be almost a full point over something with a small amount of ripple in it. There is a good example at ASR. I think that is where the algorithm falls down and is not representative of an improvement in sound.

https://www.audiosciencereview.com/...361a-review-powered-monitor.28039/post-973429

Difference of 8.8 to 9.6 with what seems less than 0.5dB of ripple, I can't see how this would sound any different.

It also depends on what algorithm is used, the original equation 9, seems harder to get higher scores out of. That's why I think listening to the different options would be good to work out what it likes and what I agree with.

 

[Juhazi]

Step/impulse response "optimization" seems to make vertical responses rollercoaster. Choose your poison? Harsch type xo can give ok on-axis, but perhaps FIR tweaked symmetrical would be better? I have never even tried learn FIR...

Aslo I'm wondering what really makes sound better - smooth on-axis or smooth PIR. Nearfield it might be on-axis, but typical home hifi PIR?

 

[mark100]

Aslo I'm wondering what really makes sound better - smooth on-axis or smooth PIR. Nearfield it might be on-axis, but typical home hifi PIR?

Maybe the biggest question out there......

I think it depends mainly on two factors.

First is our preference for the ratio of direct to reflected sound.
You know, the age ole imaging & clarity vs ambiance, "the performance is here" vs 'you are there', ....and all the myriad ways that preference gets described.

Second is simply where in the sound field are we listening....inside the critical distance where the direct field is stronger than reflected, or past that distance into where reflected is stronger than direct.

My take is that a preference for direct, and listening within the critical distance, strongly favors smooth on-axis for best sound.
If someone likes reflected/ambiance the most and listens predominantly in the reflected field, it makes sense PIR becomes more important.

Most folks seem to want it all, and perhaps listen more around the critical distance than not.....ie a balance between direct and reflected.
In this case, my current thinking is that listening window smoothness rules...say with 20 degrees of on-ax.

The more i measure polars and then listen, the less concerned i've been getting with off-axis smoothness beyond the listening window.
This has been certainly true for direct field listening, and I am finding it surprisingly true for reflected field too.
heck, I can hear listening window improvements down the hall in adjacent rooms (literally), but can't say better off-axis polars have done much to help at all, anywhere.

 

[hifijim]

If you are still using the Vituix transfer function inversion process that creates a very smooth response in some of the important curves.

I am not familiar with this concept. Could you explain a bit more, or link to some past information?

I use the optimizer occasionally, but I admit my experience with this tool is modest.

 

[vineethkumar01]

So, I am still playing around with the first crossover (nc535 optimized my IIR crossover) on the All SB acoustics 3 way config (WO24P+SB15CAC+SB26CDC).
I completed both the speakers and tried to listen to the set up.
Before listening, I took on-axis measurements at 1m away from each speaker.
Here is how the left and right speakers looked like (3.8ms gated responses with no additional smoothing):

Green plot is left speaker. Purple is right speaker (please ignore legends. That is a keyboard issue with my laptop)
I find that they are really close. Even this level of difference I think could be because of mic orientation while measuring/slight changes in distance etc.

However, here is the issue. From the same measurements, I looked at the distortion tab in REW.
This is how they looked like:

Left speaker

Right speaker

The right speaker looks relatively clean. However, for the left speaker distortion rises up as much a 5% in the 1 to 5k region. And that too with odd harmonics.
I went and put my ears against the drivers while music was being played..
I can hear the distortion clearly.
It is like small amounts of clipping, kind of sparking sort of sound along with the material being played. This is not at all audible in the right speaker

Now I am sort of lost about how to debug it.
Is it driver related?Is it amplifier? Is it my audio interface or my computer itself?
I could hear that the distorted sound is predominantly coming from the SB15CAC on the left speaker.
How do I go in a systematic manner about identify the root cause of this issue?

Thanks
Vineeth

 

[nc535]

If you are still using the Vituix transfer function inversion process that creates a very smooth response in some of the important curves. The algorithm responds very well to this and the difference in the score can be almost a full point over something with a small amount of ripple in it. There is a good example at ASR. I think that is where the algorithm falls down and is not representative of an improvement in sound.

https://www.audiosciencereview.com/...361a-review-powered-monitor.28039/post-973429

Difference of 8.8 to 9.6 with what seems less than 0.5dB of ripple, I can't see how this would sound any different.

It also depends on what algorithm is used, the original equation 9, seems harder to get higher scores out of. That's why I think listening to the different options would be good to work out what it likes and what I agree with.

that 9.5 pref rating used your latest posted XO as a starting point for the optimizer, with the preference screen set to LW.

I do agree that when using mirror response eq, the result is unrealistically smooth and discount the results appropriately in my own thinking. A very high rating with mirror EQ at least tells you you've got the directivity and axial vs power response tradeoffs right.

 

[nc535]

So, I am still playing around with the first crossover (nc535 optimized my IIR crossover) on the All SB acoustics 3 way config (WO24P+SB15CAC+SB26CDC).
I completed both the speakers and tried to listen to the set up.
Before listening, I took on-axis measurements at 1m away from each speaker.
Here is how the left and right speakers looked like (3.8ms gated responses with no additional smoothing):
View attachment 1100880
Green plot is left speaker. Purple is right speaker (please ignore legends. That is a keyboard issue with my laptop)
I find that they are really close. Even this level of difference I think could be because of mic orientation while measuring/slight changes in distance etc.

However, here is the issue. From the same measurements, I looked at the distortion tab in REW.
This is how they looked like:

Left speaker
View attachment 1100881

Right speaker
View attachment 1100882

The right speaker looks relatively clean. However, for the left speaker distortion rises up as much a 5% in the 1 to 5k region. And that too with odd harmonics.
I went and put my ears against the drivers while music was being played..
I can hear the distortion clearly.
It is like small amounts of clipping, kind of sparking sort of sound along with the material being played. This is not at all audible in the right speaker

Now I am sort of lost about how to debug it.
Is it driver related?Is it amplifier? Is it my audio interface or my computer itself?
I could hear that the distorted sound is predominantly coming from the SB15CAC on the left speaker.
How do I go in a systematic manner about identify the root cause of this issue?

Thanks
Vineeth

swapping left to right might shed some light

 

[hifijim]

Now I am sort of lost about how to debug it.

Yes swapping left to right would let you quickly determine if it is a speaker problem or something further up the chain such as amp or other electronics.

Assuming you find it is a speaker problem, you should perform harmonic distortion measurements on each driver individually.

 

[hifijim]

@fluid - In your crossover, you use a +2.5 dB PEQ boost at 13k with a Q = 6. I was under the impression that a PEQ boost should have a Q of 3 or less. If the PEQ is a cut (negative dB), the Q may be as sharp as is needed, but a boost should be low Q to avoid sonic degradation or side effects.

I honestly do not remember who/where I came across this guidance.

Have you found that a high Q boost can be implemented without audible side effects?

j.

 

[Juhazi]

Those distortion peaks can come from furniture etc. resonating, or loose coonectin of driver or it's wires. Play sine at that freq and listen closely to the speaker etc. if you can hear disto. Then put your hand on the suspected source to kill mechanical reso etc...