I just finished measuring a completed speaker, and am somewhat surprised to see that the measured response of the crossover is quite different from the modelled response. This is the crossover:
And here is the response:
(I know the response isn't pretty either way, but that's besides the point right now.)
The orange line represents the measured response of the complete speaker, the blue line is the modelled one using raw driver measurements (all gated farfield measurements with a 4.5ms window). Up to around 2.5kHz the lines are quite similar, but in the higher frequencies not so much. I am 100% positive I did not make a mistake in the crossover assembly. Measuring equipment was a Behringer UMC-22 audio interface (with loopback for reference in REW), a Dayton EMM-6 mic, and a cheap TDA7498E amp from aliexpress.
What am I doing wrong here?
I have attached the FRD and ZMA and VituixCad files.
And here is the response:
(I know the response isn't pretty either way, but that's besides the point right now.)
The orange line represents the measured response of the complete speaker, the blue line is the modelled one using raw driver measurements (all gated farfield measurements with a 4.5ms window). Up to around 2.5kHz the lines are quite similar, but in the higher frequencies not so much. I am 100% positive I did not make a mistake in the crossover assembly. Measuring equipment was a Behringer UMC-22 audio interface (with loopback for reference in REW), a Dayton EMM-6 mic, and a cheap TDA7498E amp from aliexpress.
What am I doing wrong here?
I have attached the FRD and ZMA and VituixCad files.
Last edited:
I know, but thats not the issue.
Btw, I just discovered an error in my impedance measurements, have to do some over again. This thread can be suspended until I have new measurements.
Post #1 has been fixed.
Please let me know if you can think of something that would explain the humps at 8kHz and 15kHz.
The answer is (not so) simple: if your measurement is right, your model is wrong.
Please do the following to figure out what's going on:
Please do the following to figure out what's going on:
- Measure the response of the individual drivers without xover
- Model the speaker using your individual driver measurements
- Check if / how the result is different
Thanks. But thats exactly what I did. The blue line in de graph in post #1 shows the response of the modelled xover, using the measured response of the individual drivers without xover. The orange line shows the response of the complete speaker, including the real-world xover based on the modelled one (1-on-1).
If you did a full dual channel measurement you should insert the coordinates (x,y) of the drivers in Vcad and measure with the mic on tweeter axis.
^If mic and DUT stays at same location for all measurements, mic at listening height/axis, then leave coordinates to zero. If each driver is measured on its own 0-axis by moving mic (or DUT) then the coordinates need to be manipulated in simulator to position the measurement set of each driver to represent reality.
The coordinates move the "origin" of each measurement set (driver) in respect to origin of the simulator, all the graphs show response at observation point which is listening distance away from the simulator origin (0,0,0). If both drivers are measured from same location their measurement data already include relative position of the drivers and thus coordinates are left to zero for both (kept the same), they already reflect reality.
The coordinates move the "origin" of each measurement set (driver) in respect to origin of the simulator, all the graphs show response at observation point which is listening distance away from the simulator origin (0,0,0). If both drivers are measured from same location their measurement data already include relative position of the drivers and thus coordinates are left to zero for both (kept the same), they already reflect reality.
Last edited:
I threw your crossover and files into XSim and matched your simulation results, so that's not where the problem lies, So I would look at these areas:
1. Double check your crossover wiring (especially the tweeter circuit) and verify all component values (especially the L-pad resistors) match the simulation values.
2. Double check how you do measurements. One thing I noticed was that your measured raw responses for the TCP115 and ND25FA show a difference of a little over 15 dB. Dayton's IB measurements show a difference of a little less than 10 dB. The mic distance and amp setting have to be identical when you do the raw and with crossover measurements. I assume the tweeter is your design axis.
1. Double check your crossover wiring (especially the tweeter circuit) and verify all component values (especially the L-pad resistors) match the simulation values.
2. Double check how you do measurements. One thing I noticed was that your measured raw responses for the TCP115 and ND25FA show a difference of a little over 15 dB. Dayton's IB measurements show a difference of a little less than 10 dB. The mic distance and amp setting have to be identical when you do the raw and with crossover measurements. I assume the tweeter is your design axis.
The source of measurement error is difficult to troubleshoot remotely. There are a hundred possible errors that can be made, and it is hard to guess which error is involved.
It is also possible that it is not one error, but a combination of multiple errors which add together to create a big difference between simulation and measurement.
Can you post some pictures of your measurement set up, particularly the microphone and speaker?
Are you changing mic position or speaker position when measuring the drivers and the system? Are you keeping the drive voltage the same?
Your measurement of the TCP115 goes down below 100 Hz. Is this a merging of near field and gated far field responses, or is it a dual-gate, or blended, or frequency-dependent gating measurement ?
j.
It is also possible that it is not one error, but a combination of multiple errors which add together to create a big difference between simulation and measurement.
Can you post some pictures of your measurement set up, particularly the microphone and speaker?
Are you changing mic position or speaker position when measuring the drivers and the system? Are you keeping the drive voltage the same?
Your measurement of the TCP115 goes down below 100 Hz. Is this a merging of near field and gated far field responses, or is it a dual-gate, or blended, or frequency-dependent gating measurement ?
j.
Your tweeter pretends to be a 90 ohm unit. Tweeter zma file is filled with spl data.
Thanks for spotting that! That was also the case for the TCP115 (midbass) Z measurement, which is why I updated the measurements in post #3. I'll come back with a new Z measurement for the tweeter as soon as I can.
- edit -
Using an older tweeter Z measurement from a year ago the blue (modelled) line is way more linear (as intended). Still deviates a lot from the total system measurement though (orange line in post #1).
Last edited:
Hmm, weird. This should be a gated farfield measurement with a lower limit of around 220Hz.
Seems like I messed up multiple measurements. Next weekend I'll redo all measurements, and be more careful this time. I'll also use a different amp to eliminate that as the cause.
Try to use as few of the filter components as possible while at the same time maintaining the desired filtering quality.
Have you dealt with Z offset in the crossover model? I don't use Vituix, so I'm not sure what you have going on there, but the descriptions so far don't sound like you've followed the typical procedure of measuring individual drivers with no crossover, measuring both drivers simultaneously with no crossover, then using those measurements to adjust Z offset in your model to match the combined measured response with no crossover.
And of course you also have to deal with individual driver X and Y locations, and microphone distance in the model as well.
And of course you also have to deal with individual driver X and Y locations, and microphone distance in the model as well.
VituixCad follows a slightly different procedure...measure each driver on its own axis. Using dual channel measurement you don't need to measure both drivers simultaneously and set Z offset.
But, it doesn't seem like the OP has quite followed VituixCAD instructions because it sounds like he measured all on the same axis, and left x, y, z at 0,0,0.
One other thing...one of the powerful features of VituixCAD is taking into account the directivity, early reflections, etc. which requires a sequence of off-axis measurements of each driver (on it's own axis).
To my knowledge, Z offset is only relevant when doing measurements without timing reference. I'm using separate in & output channels of my audio interface as loopback reference, from which REW can calculated the delay and corresponding distance between driver and mic. This calculated distance can be viewed by clicking the "Offset t=0" button (cumulative shift window). As the exact distance for each measurement is calculated, the phase shift between measurements is also correct as long as the mic stays in the exact same place. I used the acoustic center between the tweeter and midbass in these measurements, at 1m away.
This weekend I'll redo all measurements. I'll also measure both drivers individually and both in parallel from the same mic position. If the theory above is correct, the summed response of the individual drivers should mirror the response of both drivers in parallel without any Z-offset correction.
To troubleshoot your test setup, I recommend reducing as many variables as possible. For acoustic measurements, I would not bother with polar responses, I would focus only on a single axis. I would make all measurements on the tweeter axis, and I would not move the speaker or the mic while taking responses. This is the easiest way to get agreement between VituixCad simulation and measured data.
Once I have good agreement with simulation, I would then feel comfortable in doing full polar responses with the mic on each driver's own axis (although I have better results when I use a common axis for mid and tweeter measurements)...
j.
Once I have good agreement with simulation, I would then feel comfortable in doing full polar responses with the mic on each driver's own axis (although I have better results when I use a common axis for mid and tweeter measurements)...
j.
I have so little experience with vituix, any comment there would be meaningless at best.
But I do a lot of acoustic measurements, both on and off-axis, with manual tuning.
Strongly agree with what Jim has posted,
and with no matter what the tuning method that follows (ie working with sims to find optimal, or staying entirely manual)
A single mic position, single axis, a set fixed timing used for all drivers, with a fixed speaker location......works.
Until a foundational tuning of those measurements is correct and repeatable, I don't think i've even made it to first base with anything valid, for further polar analysis. my 2c
That is my understanding too. I measured a speaker midway between tweeter and midrange but was told I should still enter x and y (although they were all centered so x was still zero). But to be more accurate I needed to measure each on its own axis, not touching the volume or other settings, so I redid all my measurements.
Also, in my case I was measuring at 1m but VituixCAD was simulating for 2m which did make a difference, but not as large as what you are getting.