Vituix has the ability to use a transfer function file as a filter, you can combine this with the mirror Function or other tools to create an inversion.
https://www.audiosciencereview.com/...eadphone-eq-with-vituixcad.27750/post-1081013
Interesting anecdote in post 10.
I'm sure whoever provided the information meant well and may have had a point in a specific circumstance, but as a blanket rule that is just utter nonsense.
It is important to be careful when using quite high Q's as many of those things are quite inaudible. Gain structure does matter if the DSP is not using floating point math. What really matters is the overall result and in this specific situation the 13K and 16K filters work together to flatten a ripple in the listening window. If you see the combined transfer function of the filter the Q of the 13k filter is modified by the broader Q of the higher filter. They work in combination to get the desired response. I don't think the shape of that is anything to worry about.
If there was a sharp null, I would leave it alone but a dip of no more than a few dB shouldn't be a problem to fill as long as it improves more than just the on axis response when you do. Everything in my crossover version was targeted at the listening window with a mind to how it affected the wider off axis curves and making sure the on axis was still reasonable.
Thanks for your thoughts... I suspected as much. Conventional wisdom works, until it doesn't.
I have read a bit more about the mirror function and inversion. The fact that the Preference rating would score higher when very small +/- 0.5 dB ripples are driven down to +/- 0.25 dB is indicative of nature of the Preference rating. I continue to maintain that the Preference rating should be rounded to the nearest integer. That is the limit of its usefulness.
In other news, I have been looking again at the DSP filter I posted in post 709 https://www.diyaudio.com/community/threads/a-3-way-design-study.376620/page-36#post-7143447
The dip in DI at 3.6k is endemic to the SB15CAC driver. With a 3k crossover, it is hard to avoid it. Different filters can change the shape a bit, but it is always there.
The wavecor driver is very promising of course. @fluid 's crossover is really nice.
It will be very interesting to see how these different drivers sound.
I feel much the same, my interest is in using it as an optimization function, how useful it is for that task isn't clear to me right now.
Your own speakers would probably be good test beds if it was something you wanted to try, they fit the paradigm of cone and dome which this seems to work better with.
Running an optimization in Vituix is quite easy, tick all the functions of the equalisation, gain and delay blocks that you want to include, choose the options you want in the Preference rating section and hit optimize.
I think the point is that it isn't useful to try and drill down to the decimal point, but a change from one whole number to another is more significant.
Practically if something scores 6.1 and 6.95 one is essentially a 6 and another a 7 but at 6.1 and 6.2, the predictive power is lost.
Practically if something scores 6.1 and 6.95 one is essentially a 6 and another a 7 but at 6.1 and 6.2, the predictive power is lost.
It has been a while since I listened to a really good paper cone driver. There are characteristics to the SB "CAC" drivers which different from any paper cone driver I recall. I would describe it as a sharpness, or a clean sound. I particularly note it on the decay of transient sounds. Maybe if I listened to the latest paper cone drivers such as SS illuminator, or a Satori MR16P, or of course Purifi, I would hear the same qualities.
Whenever I use the optimizer, I often seem to get something which is not quite as good as what I can do myself, although sometimes it does exactly what I expect. Oftentimes it is a quite strange result. I realize this is my fault, not the tool. I need to properly constrain the region where the optimization should take place, and place other limits on the optimizer.
Back in my professional career, we would sometimes use neural nets to extract relationships from data. For example, the bending moment and torsion on a wing based on Mach, altitude, dynamic pressure, AoA, Nz, roll rate, and a host of other parameters. It was an early form of AI. When training the neural nets, it was always necessary to significantly message the data before training to remove noise and errors. It was sometimes necessary to cluster a group of data points at a similar point in the sky and average them together into a single representative data point. Otherwise the neural nets would try to predict the random noise inherent in measured flight data. Many times the neural network solution was valid within the range of the training data but would blow up if a parameter exceeded a max/min training value. There was a lot of judgment and expertise that went into pre-processing data for net training.
In the same way, I wonder if Vcad optimizer might do a better job if the driver response data was smoothed, such as 1/6 octave. I wonder how often the optimizer is led astray by diffraction ripples and narrow band driver resonances.
Eventually we found that we had to make in-depth studies of the aero data in order to effectively pre-process the neural net training data set. By the time we had done that level of study, we were in a position to establish the relationships ourselves, based on good aero engineering principles. After a while, we walked away from neural nets, and focused on more explicitly derived relationships based on statistical analysis of the data (i.e. data mining).
I find the optimizer most useful when doing passive crossovers (which I suspect is the real purpose). Active crossover development should be less complicated.
j.
Optimizing with the scoring algorithms seems to work better for me and get less out of control than it used to. I was interested to see what changed, and really it wasn't very much at all which is why I am interested to see if it actually sounds any different. It did the opposite of what I was trying to do in a few areas, it makes me wonder whether what I was trying to do was actually better or not.
This is all just frequency response balancing so it can't cover all the aspects of what makes a speaker sound the way it does.
One thing that was quite good was it's ability to add a subtle slope to a flat design. It is much easier for me to eyeball the right EQ to a flat target than it is for a slope even though the slope is what I want and can be added into the target lines.
High order distortion - 4th, 5th and higher - generally mean mechanical rub and buzz. 4th harmonic rise at about 4 kHz -> fundamental at 1 kHz -> tweeter resonance is near -> check left tweeter. Do a manual sine sweep w/o crossover from 500 to 20000 Hz, REW has a generator with “frequency tracks cursor” option - that’s what you need. Do a regular Fr measurement, then turn on REW’s generator and slowly drag cursor across the graph and listen. Use about 5 V output level.
So the distortion issue turned out to be due to the amplifier I had used. I changed the amp and things are fine now.. 
Thanks everyone for the pointers.
I have been able to hear the 3 way (All SB drivers) system for little bit of time now casually. I need to listen more to identify any faults with it. I also need to get the adaptors for mounting my Wavecor driver on the cabinet. That is still a little bit of time away.
In other news, there is something I had left halfway some time back. It was to take the measurements of the ST260 KVAR horn. The accidents due to measurements set up and 3D printed horn getting damaged prevented me from continuing that experiment at the time. Well now I have printed in once more and did the measurements.
(The following content is a bit of repetition of the post I made in the ATH thread but still posting it here for the sake of continuity if this thread)
These are raw horizontal and vertical polar (4ms gated, mic 1m away from horn outer surface) measurements of the ST260 KVAR on Peerless DFM 2544R00 compression driver.
Pic of the measurement set up (This is what I thought is the vertical polar measuring configuration of the horn)
Horizontal polars
Vertical polars
Temporary system set up
I will create a VituixCAD project with all this data loaded soon and post it here..
Thanks
Vineeth
Thanks everyone for the pointers.
I have been able to hear the 3 way (All SB drivers) system for little bit of time now casually. I need to listen more to identify any faults with it. I also need to get the adaptors for mounting my Wavecor driver on the cabinet. That is still a little bit of time away.
In other news, there is something I had left halfway some time back. It was to take the measurements of the ST260 KVAR horn. The accidents due to measurements set up and 3D printed horn getting damaged prevented me from continuing that experiment at the time. Well now I have printed in once more and did the measurements.
(The following content is a bit of repetition of the post I made in the ATH thread but still posting it here for the sake of continuity if this thread)
These are raw horizontal and vertical polar (4ms gated, mic 1m away from horn outer surface) measurements of the ST260 KVAR on Peerless DFM 2544R00 compression driver.
Pic of the measurement set up (This is what I thought is the vertical polar measuring configuration of the horn)
Horizontal polars
Vertical polars
Temporary system set up
I will create a VituixCAD project with all this data loaded soon and post it here..
Thanks
Vineeth
Hi all,
Please find the link to this VituixCAD project in this link:
https://drive.google.com/file/d/1nemISCq4U_ea9lU-ztaZkXfaTB3ZszI7/view?usp=sharing
I have attached it in my google drive since the files sizes are a little big to attach here directly. Please do give a try on the crossover for this one.
I have tried a little bit on crossover as a 3 way system. But it seems like I am missing a bit of directivity matching here (Maybe a 12inch driver is better than this 9inch driver for this horn?).
Requesting help from you all regarding understanding this and creating a good enough crossover for this configuration if possible.
Here are my current results:
@tktran303: I have developed sort of a fascination for horn-based systems of late.
I dont know if it is the looks or some sort of unexplainable (and maybe unrealizable) expectations about possible sound quality. I have to find it out and just get rid of thoughts like 'what if I could make a horn system better than my current 3-way speaker configuration' once and for all so that I can settle in on some configuration for a longer duration of time.
Hence all these experiments with all sorts of speaker types..
Regards
Vineeth
Please find the link to this VituixCAD project in this link:
https://drive.google.com/file/d/1nemISCq4U_ea9lU-ztaZkXfaTB3ZszI7/view?usp=sharing
I have attached it in my google drive since the files sizes are a little big to attach here directly. Please do give a try on the crossover for this one.
I have tried a little bit on crossover as a 3 way system. But it seems like I am missing a bit of directivity matching here (Maybe a 12inch driver is better than this 9inch driver for this horn?).
Requesting help from you all regarding understanding this and creating a good enough crossover for this configuration if possible.
Here are my current results:
@tktran303: I have developed sort of a fascination for horn-based systems of late.
I dont know if it is the looks or some sort of unexplainable (and maybe unrealizable) expectations about possible sound quality. I have to find it out and just get rid of thoughts like 'what if I could make a horn system better than my current 3-way speaker configuration' once and for all so that I can settle in on some configuration for a longer duration of time.
Hence all these experiments with all sorts of speaker types..
Regards
Vineeth
Making a three way out of two of the same woofers is odd, I would take the high pass out of the top woofer and make it 2.5 way even if you feed each driver by a separate amp, otherwise you are just wasting the output of the top woofer.
A better directivity match would have the crossover down around 1K, the Satori don't seem to gain directivity evenly and plateau out before rising again.
Whether this is too low for the CD I don't know.
You can smooth the directivity and get a reasonable listening window but those drivers really don't want to join together
The amount of delay needed quite high or a phase inversion like you used, it still makes me wonder if the measurements are right for the woofer, can you check the original impulse responses?
The amount of delay needed quite high or a phase inversion like you used, it still makes me wonder if the measurements are right for the woofer, can you check the original impulse responses?
@fluid: Thanks. You are right fluid.. The woofer measurements had some issues.
I measured woofer polars again the hard way. Took all my strength to lift that box up and keep it on that turning stand.
This time I also ensured that the top woofer at least is 110+ cm up from the floor. Then merged it with newly taken nearfield response.
Measurement setup:
Here is how the 2.5 way crossover that I tried looks like:
Six pack
Please take the updated project I have attached here:
https://drive.google.com/file/d/1afJqIazHHSjD_jMYqQa49OaxLo-DM6bR/view?usp=sharing
I measured woofer polars again the hard way. Took all my strength to lift that box up and keep it on that turning stand.
This time I also ensured that the top woofer at least is 110+ cm up from the floor. Then merged it with newly taken nearfield response.
Measurement setup:
Here is how the 2.5 way crossover that I tried looks like:
Six pack
Please take the updated project I have attached here:
https://drive.google.com/file/d/1afJqIazHHSjD_jMYqQa49OaxLo-DM6bR/view?usp=sharing
I'm glad that is confirmed, the results above look much better now. I noticed the same thing in the Wavecor results too. Which driver was polarity inverted ?@fluid : You are also right in the polarity inversion observation.. The cd horn polarity was inverse compared to the woofer when i measured.. accidentally connected the wrong posts on the amp..
Hi @fluid,
This is how the impulse responses (on axis) of the different drivers look like:
On the top is the Satori WO24P.
On the bottom left is wavecor WF120BD03.
On the bottom right is the latest CD horn combination.
It looks like the WO24P and wavecor have same polarity but the CD horn is reversed compared to them.
There were other issues with my WO24P measurements from earlier, exactly as @hifijim suspected. Mainly the measurements were much more corrupted by reflections due to the lower height at which the woofer was measured. Now with an increased height from the floor, there are significant differences for the same duration of window length (approx 4ms). Nearfield measurements are identical to before.
Here is a comparison between the old and new merged responses of the WO24P
On the left are the old responses
On the right are the new responses I created yesterday.
In addition to the significantly pronounced hump in response from 500Hz to 1kHz, the old measurements are plagued by reflection-induced bumps throughout the bandwidth above and below 1kHz. Also, weird is that deep notch in response around 2kHz in the (30 degree off axis) old response, which might have been caused by some external noise (building construction ongoing near my house) that would have crept in while I was taking measurements for this driver in the past.
Now I sort of also understand why my foam box prototype (with the woofer) sounded much better compared to current boxes. In that experimental build, I measured the woofer at sort of 140cm above the floor, right in the midway height between floor and ceiling and therefore those measurements (which I played around for crossover at the time) might have had the least reflection induced artifacts in the responses that I got with 4ms gating.
Thanks
Vineeth
This is how the impulse responses (on axis) of the different drivers look like:
On the top is the Satori WO24P.
On the bottom left is wavecor WF120BD03.
On the bottom right is the latest CD horn combination.
It looks like the WO24P and wavecor have same polarity but the CD horn is reversed compared to them.
There were other issues with my WO24P measurements from earlier, exactly as @hifijim suspected. Mainly the measurements were much more corrupted by reflections due to the lower height at which the woofer was measured. Now with an increased height from the floor, there are significant differences for the same duration of window length (approx 4ms). Nearfield measurements are identical to before.
Here is a comparison between the old and new merged responses of the WO24P
On the left are the old responses
On the right are the new responses I created yesterday.
In addition to the significantly pronounced hump in response from 500Hz to 1kHz, the old measurements are plagued by reflection-induced bumps throughout the bandwidth above and below 1kHz. Also, weird is that deep notch in response around 2kHz in the (30 degree off axis) old response, which might have been caused by some external noise (building construction ongoing near my house) that would have crept in while I was taking measurements for this driver in the past.
Now I sort of also understand why my foam box prototype (with the woofer) sounded much better compared to current boxes. In that experimental build, I measured the woofer at sort of 140cm above the floor, right in the midway height between floor and ceiling and therefore those measurements (which I played around for crossover at the time) might have had the least reflection induced artifacts in the responses that I got with 4ms gating.
Thanks
Vineeth
Last edited:
The woofer measurements do look much better, it really does show that you need to be careful with how you measure the results can be so different if you get it wrong. Interesting that the woofer and Wavecor are in the same positive polarity I had exactly the same delay difficulty with them as the waveguide.
I imagine revisiting your other combinations in light of these new measurements would indeed give a different impression.
I imagine revisiting your other combinations in light of these new measurements would indeed give a different impression.