WinPCD Results Different Than Xsim/VituixCAD

[dcibel]

wogg for proceeding with VituixCAD and 2-channel design method, it's as easy as reading the "how to start working with VituixCAD" section of the help file, and following the measurement guide for your favourite measurement software. Have fun!

[dcibel]

I'd make some comparison of winPCD model to real world results but it won't open any of my FRD files. dlr if you care to troubleshoot this, I've attached one of them.

​

[dlr]

I'll check it out. What version are you using?

[dlr]

The problem is that your file is normalized, has negative SPL. WinPCD only works with absolute SPL.

[dlr]

Got that part right ;)

Mininum phase is not required for "modelling" off-axis data for anything you've described. Only delay between measured data is needed, which doesn't need to be minimum phase. Measured phase from simple impulse response peak detection may not be minimum phase, but somewhere close, and it does not need to be modified via HBT for accurate delay / phase interaction. HBT can include user error from poorly chosen slopes causing error. min phase / HBT steps can be skipped with same results, just different delay value.

There is no error using HBT as long as you use the same SPL model afterwards, a point I've made for years as well. That's the beauty of models. You get very, very precise relative offsets. You cannot create models without minimum-phase. Modeling provides the relative acoustic offset. Without that, off-axis error is introduced and increases with the off-axis angle. The excess-delay is not accurate from the geometry because you leave the driver settings at (0,0,0), but the larger the driver, the farther back the center is located. This is not accounted for without being able to set the z-axis value. You simply cannot get accurate off-axis delay that way.

Why "model" off-axis at all and leave the mic on the shelf? At best you have a model of crossover interaction off-axis, maybe piston model for the driver? What about diffraction, what about driver breakup, what about waveguides and horns and other complex arrangements? Accurate simulation comes from​ minimizing calculated assumptions.

As I said, that cannot be easily handled, although I have (and have had for some time) a way to include diffraction for the off-axis modeling, but as I also said, that aspect is not overly important given that the two most important factors are first arrival (on-axis usually) and power response. The latter is not impacted to any great degree by the baffle shape/driver positioning other than the baffle step that can be handled in creating the driver SPL model in the first place and should be. As such, it doesn't affect the power response calculation. The driver directionality does, but can be reasonably modeled. But unless you make a large series of off-axis measurements though the entire front hemisphere and the software also uses a weighting scheme that precisely matches the measurements that you made, you cannot get an accurate power response calculation. And how many DIYers are going to make that kind of measurement setup given the difficulty?

"3 measurement scheme" with mic at tweeter axis is problematic, one of those measurements is already off-axis, and simulation is locked at mic measurement distance. 1m measurement distance is not the same as 2-3m listening distance, okay for smaller speaker but not for large speakers, with some large tower with woofer down near the floor you can be capturing SPL error simply from the distance differences from mic to driver, that doesn't translate at real listening distance.

Yes, one or more of those measurements are already off-axis, but that is, IMO, a benefit because it is usually the larger drivers, many of which have extreme peaks on-axis that smooths off-axis, even small amounts off-axis, such as hard cone drivers. Granted I have not yet added SPL-with-distance calculations in to WinPCD, but is on my to-do list. It was much lower priority than other work. However, WinPCD does have an option to make some correction for off-axis measurements to adjust them closer to what would be an on-axis measurement for each driver. To my knowledge, WinPCD is the only software that does this.

Better method would be to measure each driver straight on, not only will on-axis be actually on-axis for each driver, but constant mic distance ensures correct SPL as well. But now how to capture the delay with USB mic? 3 measurement process no longer works, so more works is needed, or timing reference (2-channel). Continue with off-axis measurements and now 1m measurement can be represented at 3m listening distance with just some response interpolation and SPL/distance calculation.

SPL variance with the small excess-delay changes is not that important, though I still want to add that to WinPCD. With regard to the USB mic measurements, that will definitely NOT invalidate the 3-measurement scheme. Quite the contrary, in that case the ONLY way to accurately design is with the 3-measurement scheme because, as I've said, measured phase is not required.

With regard to measuring at 1m and extending this to 3M with SPL and excess-phase correction, you overlook what I expect that you believe is very important, diffraction influence. Go to any diffraction modeling software, set up a model at 1m, then change the distance to 3M. You'll see a fair amount of change in SPL bandwidth response. This is due to the fact that the excess-phase between driver and baffle diffraction signature changes with distance and can be significant depending on the baffle shape and driver positions. The 3M point will not be an accurate representation unless you use a minimum-phase model with software that corrects for the change in diffraction signature with design point change. The latter is one of my most desired changes to WinPCD, but that's very complicated, may not happen.

Anyway, it's clear that you are biased towards use of your own software, and I'm biased towards measuring systems that include timing reference and use of VituixCAD for design. We can move on from this discussion, it's clear that neither of us is going to convince the other of a "better way". Don't get me wrong, I started DIY using a Omnimic, following Jeff's instructions, 3 measurement process etc. but have moved on from that having learned more along the way with each new design adventure.

No, I've been doing this since 1995 when I used CALSOD in a DOS-based PC. Software such as that goes back to the 90's if not earlier. I work on WinPCD because I enjoy it and it is useful to many, especially newbies I suspect. As for my main stereo system, I use the Ultimate Equalizer (UE) DSP, though if I work with a passive system (for fun) I use WinPCD to start, but always put those measurement into the UE. I have made measurements of the UE output using my minimum-phase models as input and can prove just how accurate that can be. The UE result is linear phase. I've confirmed it all with measurements. Just as with measurements of passive systems.

As far as 2-channel measurements go, I don't understand why there is so much resistance to include a timing reference in the measurements to capture accurate phase relations between multiple measurements and drivers. It's not like a cheap USB audio interface and XLR mic is all that more expensive than a USB mic. It's like some DIY spirit of needing to prove that you can get by with a paper clip and a rubber band while the correct tool is just an arms length away.

There's one big reason why an impulse timing measurement is fraught with error. The most accurate time point is obtained by the impulse rise, but the larger the driver, the less high frequency content is present to determine that impulse point that results in questionable accuracy. The 3-measurement scheme obviates the need to do that. And being condescending to those who don't think as you isn't warranted.

Btw, if you've been paying attention, the 3 measurement process is automated in VituixCAD in aux menu - time align, just load the 3 measurements and hit solve button to find delay.

So you believe in the 3-measurement scheme. Automated in VituixCAD, manual, but included, in WinPCD. I suppose I could automate that as well, but I feel that part of the benefit of WinPCD is the learning process and ease of use. I don't intend WinPCD to ever overtake more elaborate software. But it's adequate to design excellent systems as is.

dlr

[dcibel]

Yes, one or more of those measurements are already off-axis, but that is, IMO, a benefit because it is usually the larger drivers, many of which have extreme peaks on-axis that smooths off-axis, even small amounts off-axis, such as hard cone drivers. Granted I have not yet added SPL-with-distance calculations in to WinPCD, but is on my to-do list. It was much lower priority than other work. However, WinPCD does have an option to make some correction for off-axis measurements to adjust them closer to what would be an on-axis measurement for each driver. To my knowledge, WinPCD is the only software that does this.


SPL variance with the small excess-delay changes is not that important, though I still want to add that to WinPCD. With regard to the USB mic measurements, that will definitely NOT invalidate the 3-measurement scheme. Quite the contrary, in that case the ONLY way to accurately design is with the 3-measurement scheme because, as I've said, measured phase is not required.

This is only highlighting the benefit of using measured data over modelled/calculated information. You say throw away measured phase, I say throw away minimum phase.

With regard to measuring at 1m and extending this to 3M with SPL and excess-phase correction, you overlook what I expect that you believe is very important, diffraction influence. Go to any diffraction modeling software, set up a model at 1m, then change the distance to 3M. You'll see a fair amount of change in SPL bandwidth response. This is due to the fact that the excess-phase between driver and baffle diffraction signature changes with distance and can be significant depending on the baffle shape and driver positions. The 3M point will not be an accurate representation unless you use a minimum-phase model with software that corrects for the change in diffraction signature with design point change. The latter is one of my most desired changes to WinPCD, but that's very complicated, may not happen.

Small change in diffraction is indeed evident between 1m and 3m. Merge process in VituixCAD is not without some limitations, but the basic compensation is to simulate diffraction at 10m+ to ensure full baffle step is applied to the nearfield response, to compensate for 1m measurement being too close. This compromise provides good result to avoid too thin sounding speaker from too near of a measurement.

There's one big reason why an impulse timing measurement is fraught with error. The most accurate time point is obtained by the impulse rise, but the larger the driver, the less high frequency content is present to determine that impulse point that results in questionable accuracy. The 3-measurement scheme obviates the need to do that. And being condescending to those who don't think as you isn't warranted.

If this is in reference to a 2-channel measurement, this is a gross misunderstanding of what goes on with a 2-channel measurement. It has absolutely no dependence on the driver being measure or mic input. output signal that covers full spectrum includes frequency range to 20kHz+ for high impulse rise, and common clock in soundcard ensures per-sample precision of input for repeatable results. The time reference is set by electrical feedback (2nd channel) not the mic input. You can move the mic 1mm and see that in measured phase result with accuracy and repeatability for any measurement.

It is of course a problem for single channel measurements, but 3-measurement process does not need to include minimum phase / HBT to correct it, you will simply end up with a different delay value due to difference in FFT window start. Acoustic timing reference in REW uses high frequency sound for timing reference to ensure sharp impulse, and tweeter is recommended for acoustic reference device.

So you believe in the 3-measurement scheme.

No. But only method I would suggest for USB users trying to use VituixCAD, other than possibility of acoustic timing reference with REW.

I'll change my statement a bit on minimum phase to make things perfectly clear. Minimum phase is not required and can be forgotten with measurement and design process using VituixCAD in mind, apart special case such as combining nearfield and port response as mentioned. This statement is still valid for my discussion wit Wogg, since it was relating to his design process with VituixCAD. My documents and instructions are written with design process using 2-channel measurement and VituixCAD in mind, not WinPCD, and utilizing measured data entirely for crossover simulation, not modeled/calculated results, with exception of diffraction applied to near-field response. Thank you for providing some additional motivation to put this document together to provide better insight into the need for minimum phase.

I'll provide a couple examples to highlight the case. Here is a big 3-way speaker with woofer at floor level, showing difference in response if design (listening) distance is 1m (white) vs 2.5m (blue dashed). The change in SPL for the woofer is not insignificant, so 1m single axis measurement data at tweeter axis would result in bloated sound if designed for flat response. Detrimental for something like a big array.

​

Here is an example speaker design:
​

Here is the exact same design with 1m excess phase applied, as you can see in the phase response it is far from minimum phase, however exact same result is achieved.
​

[Chris Roemer]

If you add +100dB to it, it begins to look useful (maybe)?

SB1721oo.fRD

[dcibel]

Is the problem that WinPCD can't handle negative amplitude values? dB value in my measurements is not SPL, but ratio of mic input to reference signal, usually reference signal is set near 0dBFS so measured dB level is negative. Seems to be not a problem for VituixCAD or Xsim.

[dlr]

Correct. WinPCD requires absolute measurements, that is, 0db is silence. Most drivers are supplied with absolute SPL responses, e.g. 90db nominal SPL sensitivity. You won't see many (maybe none?) manufacturers who provide example graphs other than absolute. That is usually some db for 2.83v at 1m.

[dcibel]

I guess problem is that I'm not a driver manufacturer and don't work from traced datasheet charts ;). My SPL measurement is not calibrated, so I find the dB ratio of input channel to reference channel to at least be useful information to me.

[dlr]

Who said work from datasheets? I simply provided that as an example that most manufacturers provide information for sensitivity. That means absolute response, not normalized. Most measurement systems provide absolute output that shows sensitivity if calibrated. If not, it's some nominal sensitivity that is still not in normalized form.

[dlr]

If this is in reference to a 2-channel measurement, this is a gross misunderstanding of what goes on with a 2-channel measurement. It has absolutely no dependence on the driver being measure or mic input. output signal that covers full spectrum includes frequency range to 20kHz+ for high impulse rise, and common clock in soundcard ensures per-sample precision of input for repeatable results. The time reference is set by electrical feedback (2nd channel) not the mic input. You can move the mic 1mm and see that in measured phase result with accuracy and repeatability for any measurement.

I understand 2-channel measurement systems well. I've been using them since 1995 starting with MLSSA. I still use LAUD that is a calibrated system, so it provides absolute SPL accurately. I've used others as well, but prefer LAUD due to the calibrated output. There is dependence on the driver due to the fact that locating the high frequency rise time is problematic at best.

The feedback channel is used to eliminate all upstream influence on the signal to the driver so that the response measurement is purely that for the driver, it's not for timing, although one can see the delay between arrival of the impulse to the driver and arrival at the mic. Timing (and phase) is all a function of where on places the start time marker, it has nothing to do with the electrical feedback channel.

Software without feedback has to assume a perfect signal chain. That is, no output card non-linearities, no low level cable non-linearities, no pre-amp non-linearities, no amp non-linearities and no conductor impedance between amp and driver, i.e. a perfect hardware system.

A 2-channel system, in contrast, eliminates all of the upstream non-linearities by measuring the signal that is applied to the driver and using that signal as the MLS (in that type of measurement) for the cross-correlation to determine the SPL and phase response of the driver. If the system is not a calibrated one, the SPL is nominal, but not absolute. If, however, the system is calibrated (the hardware used, i.e. sound card, not the mic, although the mic comes into play after the FFT of the impulse response to correct for the mic), then the SPL response is absolute (accurate and precise), a primary reason I still prefer LAUD.

Liberty Instruments MLS Measurements (1993)

It is of course a problem for single channel measurements, but 3-measurement process does not need to include minimum phase / HBT to correct it, you will simply end up with a different delay value due to difference in FFT window start. Acoustic timing reference in REW uses high frequency sound for timing reference to ensure sharp impulse, and tweeter is recommended for acoustic reference device.​

I don't see the point of "acoustic reference", of course a tweeter provides the fastest rise time. The problem is in locating the rise in the impulse of the high frequency in a larger driver that is too low in level to be located accurately, if at all. As soon as you move the start time marker for the larger driver (to obtain the optimal SPL response), you've changed the excess-delay from that used for the tweeter (due to its differing start time market), so the two measurements don't then correlate for phase. Unless you then go back to the output file and adjust the phase in the measured phase result to compensate. This is essentially what is done by generating minimum-phase and determining the relative acoustic offset (excess-phase) required to make the two measurements correlate again. This, though, is coupled with specifying the physical location of the two drivers on the baffle and adjusting the z-axis value to the point that makes the second driver output correlate with the first one (usually the tweeter).

No. But only method I would suggest for USB users trying to use VituixCAD, other than possibility of acoustic timing reference with REW.

I'll change my statement a bit on minimum phase to make things perfectly clear. Minimum phase is not required and can be forgotten with measurement and design process using VituixCAD in mind, apart special case such as combining nearfield and port response as mentioned. This statement is still valid for my discussion wit Wogg, since it was relating to his design process with VituixCAD. My documents and instructions are written with design process using 2-channel measurement and VituixCAD in mind, not WinPCD, and utilizing measured data entirely for crossover simulation, not modeled/calculated results, with exception of diffraction applied to near-field response. Thank you for providing some additional motivation to put this document together to provide better insight into the need for minimum phase.

My primary purpose was to counter what appeared as total dismissal of other valid means of using measurements and to challenge some of what I believe is inaccurate descriptions. I certainly take no issue with focusing on how one can use VituixCAD.

dlr

[dcibel]

I understand 2-channel measurement systems well. I've been using them since 1995 starting with MLSSA. I still use LAUD that is a calibrated system, so it provides absolute SPL accurately. I've used others as well, but prefer LAUD due to the calibrated output. There is dependence on the driver due to the fact that locating the high frequency rise time is problematic at best.

The feedback channel is used to eliminate all upstream influence on the signal to the driver so that the response measurement is purely that for the driver, it's not for timing, although one can see the delay between arrival of the impulse to the driver and arrival at the mic. Timing (and phase) is all a function of where on places the start time marker, it has nothing to do with the electrical feedback channel.

This only tells me that you have some misunderstanding of 2-channel measurement. Reference channel determines t=0 point in the measurement, primary function is timing, normalization is secondary benefit. Where one places the start time marker requires consistent t=0 point between multiple measurements. There is no dependence on the the driver due to the fact that time scale is determined by reference channel which will be repeated the same for each measurement, not the mic input from the driver. You say "problematic at best" yet I can take 100 measurements and have them reproduce the exact same response and phase, and even come back to re-measure days later and set my mic at exactly the same distance from speaker using the measured phase from previous measurement as the guide. HBT requires manual setting of response slopes prone to user error, here I will say "problematic at best" ;).

But don't just take my word for it, REW is free and ARTA is free to play with so it's very easy to verify such things for yourself (or anyone reading this), both are capable of dual channel measurement.

Software without feedback has to assume a perfect signal chain. That is, no output card non-linearities, no low level cable non-linearities, no pre-amp non-linearities, no amp non-linearities and no conductor impedance between amp and driver, i.e. a perfect hardware system.

Solution, put amp and cabling within feedback loop. Only difference will be channel matching of left and right channels on your sound card / audio interface. Any modern equipment is more than capable in this regard, except for cheap junk like Behringer UCA202. Behringer UMC202 is okay.

A 2-channel system, in contrast, eliminates all of the upstream non-linearities by measuring the signal that is applied to the driver and using that signal as the MLS (in that type of measurement) for the cross-correlation to determine the SPL and phase response of the driver. If the system is not a calibrated one, the SPL is nominal, but not absolute. If, however, the system is calibrated (the hardware used, i.e. sound card, not the mic, although the mic comes into play after the FFT of the impulse response to correct for the mic), then the SPL response is absolute (accurate and precise), a primary reason I still prefer LAUD.

This is true. SPL calibration is possible, however I find absolute SPL is very unnecessary for measurement process for crossover design. Relative SPL is all you need.

I don't see the point of "acoustic reference", of course a tweeter provides the fastest rise time. The problem is in locating the rise in the impulse of the high frequency in a larger driver that is too low in level to be located accurately, if at all. As soon as you move the start time marker for the larger driver (to obtain the optimal SPL response), you've changed the excess-delay from that used for the tweeter (due to its differing start time market), so the two measurements don't then correlate for phase. Unless you then go back to the output file and adjust the phase in the measured phase result to compensate.

Acoustic timing reference provides same information as electrical feedback loop for setting t=0 point, just using single channel input and secondary audio source for constant high frequency reference prior to measurement sweep. Again, REW is free so you can try this for yourself any time. Take multiple measurements, set IR window and "apply window to all" to carry over common window settings to all measurements. Acoustic reference is much more of a pain for repeatability than simple electrical feedback, so I still very much recommend 2-channel measurements.

[dlr]

This only tells me that you have some misunderstanding of 2-channel measurement. Reference channel determines t=0 point in the measurement, primary function is timing, normalization is secondary benefit. Where one places the start time marker requires consistent t=0 point between multiple measurements. There is no dependence on the the driver due to the fact that time scale is determined by reference channel which will be repeated the same for each measurement, not the mic input from the driver. You say "problematic at best" yet I can take 100 measurements and have them reproduce the exact same response and phase, and even come back to re-measure days later and set my mic at exactly the same distance from speaker using the measured phase from previous measurement as the guide. HBT requires manual setting of response slopes prone to user error, here I will say "problematic at best" ;).

If you want accurate SPL, the feedback channel is key to this. It's used to cross-correlate with the mic measurement in the MLS scheme to determine the impulse response rather than the original MLS signal output to the sound card. It eliminates all upstream variance to the signal. I don't consider that secondary. Setting the start time marker for high frequency content that is in many cases in the noise is what is problematic for larger drivers, hence my reference to the driver. Setting a time marker and getting repeatability is not an indication of accuracy. Now if you want to set a start marker for measuring multiple drivers in sequence, that's fine, but the problem there will be that the optimal start time marker setting for a tweeter will in all likelihood not be optimal for a woofer and vice-versa. Either one driver or the other will have inaccuracy in the SPL result. That's one aspect of the 3-point measurement scheme is more accurate. There is absolutely no need to set any specific start time marker. You set it to what is best for each driver with a summed measurement that inherently has the phase of both drivers in true correlation.

Setting slopes for HBT generation is not problematic, it's rather easy to do (no different than using any tool) and provides the very useful minimum-phase response. You can modify the measurement however is needed prior to the HBT generation, such as extending the top end or splicing the low end response that was generated in whatever manner you care to do it. Could even me modeled low end response with any appropriate baffle step. In the end, you have a minimum-phase response that is then used with to determine the relative acoustic offset. Nothing more is needed at that point.

Solution, put amp and cabling within feedback loop. Only difference will be channel matching of left and right channels on your sound card / audio interface. Any modern equipment is more than capable in this regard, except for cheap junk like Behringer UCA202. Behringer UMC202 is okay.

Why? If the system is capable of 2-channel measurement, this is superfluous unless the sound card is of poor quality. In that case the user has to go to the extra effort to do some sort of calibration.

This is true. SPL calibration is possible, however I find absolute SPL is very unnecessary for measurement process for crossover design. Relative SPL is all you need.

I find it indispensable. Non-calibrated systems require either making all measurements at one time to get the same voltage output so that the relative relationship between drivers results and the voltage must be measured before every measurement session. Phase repeatability is not the issue if one uses minimum-phase. I have had multi-way systems in use for years, switched drivers, then only needed to measure the new driver, no calibration of any kind required. The output of the new driver10+ years apart from the existing ones is still fully compatible. No need to re-measure the other drivers, no need to be concerned with any change in amp/preamp, certainly no need to maintain info on where the start market may have been placed on earlier measurements. But to each his own. Absolute measurements guarantee no errors creep into the measurements.

Acoustic timing reference provides same information as electrical feedback loop for setting t=0 point, just using single channel input and secondary audio source for constant high frequency reference prior to measurement sweep. Again, REW is free so you can try this for yourself any time. Take multiple measurements, set IR window and "apply window to all" to carry over common window settings to all measurements. Acoustic reference is much more of a pain for repeatability than simple electrical feedback, so I still very much recommend 2-channel measurements.

That sound like undesirable complication to me. I have no such requirements. I've used SoundEasy as well, there's no requirement such as that. The one limiting issue I have with SoundEasy is that it's not a calibrated system, SPL is relative, not absolute, so as with most other measurement systems, for repeatability between measurement sessions you have to set/measure the output to the driver.

Those "common window settings" are guaranteed to be optimal for one driver in a system only unless they are time-aligned. Positioning the start window marker for each driver separately is the only way to get optimal results of SPL for all drivers. That requires a bit more effort to then get appropriate minimum-phase measurement files, but it's worth the time to get the most accurate set of measurements. So I'd say it depends to what degree of accuracy you have as your goal.

dlr

[dcibel]

Setting the start time marker for high frequency content that is in many cases in the noise is what is problematic for larger drivers, hence my reference to the driver. Setting a time marker and getting repeatability is not an indication of accuracy.

You are still focusing on mic input from driver. Timing is from feedback reference channel ie output of amplifier. Very wrong understanding above, in addition to previous wrong statement that feedback signal is not for timing.

Now if you want to set a start marker for measuring multiple drivers in sequence, that's fine, but the problem there will be that the optimal start time marker setting for a tweeter will in all likelihood not be optimal for a woofer and vice-versa. Either one driver or the other will have inaccuracy in the SPL result. That's one aspect of the 3-point measurement scheme is more accurate. There is absolutely no need to set any specific start time marker. You set it to what is best for each driver with a summed measurement that inherently has the phase of both drivers in true correlation.

Set start time arbitrarily near start of impulse, with left window of 1-2ms will capture results with perfection. Optimal timing marker can be as simple as 1 button click in VituixCAD IR -> FR tool. Just hit "Far 1" for setting window start time, adjust right window and go.

Setting slopes for HBT generation is not problematic, it's rather easy to do (no different than using any tool) and provides the very useful minimum-phase response. You can modify the measurement however is needed prior to the HBT generation, such as extending the top end or splicing the low end response that was generated in whatever manner you care to do it. Could even me modeled low end response with any appropriate baffle step. In the end, you have a minimum-phase response that is then used with to determine the relative acoustic offset. Nothing more is needed at that point.

Again, HBT and min phase can be skipped with 2-channel measurement with perfect results. Just measure, set window, done.

Why? If the system is capable of 2-channel measurement, this is superfluous unless the sound card is of poor quality. In that case the user has to go to the extra effort to do some sort of calibration.

I am describing a 2-channel system...second channel is feedback loop...If left and right channel have poor frequency response match, feedback normalization will have error. Like I said, most any half decent soundcard will be fine here, just pointless discussion at this point.

[QUOTE=dlr;n1494027]I find it indispensable. Non-calibrated systems require either making all measurements at one time to get the same voltage output so that the relative relationship between drivers results and the voltage must be measured before every measurement session.

Phase repeatability is not the issue if one uses minimum-phase. I have had multi-way systems in use for years, switched drivers, then only needed to measure the new driver, no calibration of any kind required. The output of the new driver10+ years apart from the existing ones is still fully compatible. No need to re-measure the other drivers, no need to be concerned with any change in amp/preamp, certainly no need to maintain info on where the start market may have been placed on earlier measurements. But to each his own. Absolute measurements guarantee no errors creep into the measurements.

Phase repeatability is not an issue if one uses 2-channel measurement. No 3 measurement process to determine delays, literally just measure and set same IR window start time for all measurements. Minimum phase and HBT is not required (sounding like a broken record).

I've already explained downsides of single channel measurement, single point measurement for design. Moving on...

​
That sound like undesirable complication to me. I have no such requirements. I've used SoundEasy as well, there's no requirement such as that. The one limiting issue I have with SoundEasy is that it's not a calibrated system, SPL is relative, not absolute, so as with most other measurement systems, for repeatability between measurement sessions you have to set/measure the output to the driver.

I've used SoundEasy too, it's not special. It doesn't support single channel measurement, so there's that. SPL can be calibrated if you calibrate for SPL...it's only a "not calibrated SPL system" if you don't calibrate it. These are silly arguments.

Setting voltage to speaker with multimeter takes only a minute if you need to come back and re-measure later on with same output. Previous measurement can also be used if you forgot to set or record the voltage. Absolute SPL is a "nice to have", not a requirement.

​
Those "common window settings" are guaranteed to be optimal for one driver in a system only unless they are time-aligned. Positioning the start window marker for each driver separately is the only way to get optimal results of SPL for all drivers. That requires a bit more effort to then get appropriate minimum-phase measurement files, but it's worth the time to get the most accurate set of measurements. So I'd say it depends to what degree of accuracy you have as your goal.

Absolutely incorrect statement. With same measurment distance from speaker to baffle, and same window start time, SPL will be perfect, more perfect in fact than 3-measurement process at tweeter height.

You say all above statements like I am new to this hobby. I am not, having designed speakers for years and confirmed accuracy and repeatability of my own measurements and design process with final results. I've done the single channel USB mic and 3 measurement process, and then moved on to 2-channel process and never going back. I have come to the conclusions that 2-channel process outlined in VituixCAD instructions is the best way yet for loudspeaker design. You are welcome to disagree and keep on with your min phase / HBT single point of design process, your arguments are not going to convince me that my methods are inaccurate or flawed. Quite the contrary actually.

I've no interest in discussing this further.

I will leave with one last bit of information. I took a pair of my own measurements and processed for minimum phase, loaded into WinPCD, entered driver diameter, my crossover and adjusted delay to match the expected result that I get from VituixCAD. Had to adjust circuit slightly because program does not support my schematic. Here it is:

​

I then took these response files and added 1000us of delay, so they are now far from minimum phase. Loaded back in WinPCD, and got exact same result apart from phase plot, and a few frequency errors, I assume from resampling errors. Am I missing something in this "min phase is crucial" argument? Also pushed all off-axis buttons, horizontal and vertical directivity, front hemisphere power, polar nodes, all exactly the same.

​

I will now be headed back to my corner with my 2-channel measurements, VituixCAD, and my incorrect / inaccurate / flawed processes. If anyone would like to join me and learn more about 2-channel process and VituixCAD, send me a PM. I'll probably be keeping participation in this forum to a minimum (pun intended).

[dlr]

I've re-read your earlier posts and I see what you mean by timing. It's a trigger, so then yes, it's a trigger for the sampling in measurement systems that us it that way and a good reason for 2-channel. On that I agree. But the more important part is the elimination of all upstream non-linearities for the cross-correlation to determine the true impulse response. At least for my needs.

Much of your other descriptions only apply to using VituixCAD, not all members here will or want to use it, they have other software that benefit from other techniques.

However, you make some good points outside of that. If you make measurements of each driver on its axis, then delays (Edit: delay to the baffle) would be the same for a set window. But if you measure at one point with the same setting, one of the drivers will still not be optimal due to non-optimal window start marker. It might be consider good enough. That's when using minimum-phase is more useful and when splicing using other software which is what many do. Edit: A fixed start that is 1-2 msec prior to the impulse still uses data not generated by the driver, it's non-causal. It will provide an artificially extended low end response. Maybe it's not important in for some crossover designs, but it will be the case.

Using WinPCD you said that you adjusted delay. What do you mean by that? Did you set the (x,y,z) parameters for each driver, then set the z-axis of one driver for the additional delay?

Were these measurements taken at one point or each driver on-axis? If on-axis, how did you then set the driver (x,y,z) values? If not, same question. Edit: What (x,y,z) values did you use?

With regard to SoundEasy, I'm not aware of any way to calibrate it for the standard 2.83v at 1m for sensitivity that then adjusts if the input is not 2.83v. If so, that's news to me. That's what I mean by calibration, for absolute sensitivity. LAUD provides true absolute sensitivity no matter the voltage applied to the driver, so low volume/high volume measurements provide the same SPL response sensitivity. This means that I can measure any driver at any time at any input voltage and the response will still correlate to any other driver measurement taken at any time. Such as my example of a system that was more than 10 years old. It was not necessary to re-measure the original drivers.

dlr

p.s. I will say that you made me have to think carefully about it all. I don't agree with you on all points, but certainly do on some.

[dcibel]

However, you make some good points outside of that. If you make measurements of each driver on its axis, then delays would be the same for a set window. But if you measure at one point with the same setting, one of the drivers will still not be optimal due to non-optimal window start marker. It might be consider good enough. That's when using minimum-phase is more useful and when splicing using other software which is what many do. Edit: A fixed start that is 1-2 msec prior to the impulse still uses data not generated by the driver, it's non-causal. It will provide an artificially extended low end response. Maybe it's not important in for some crossover designs, but it will be the case.

Ok, you reeled me in for 1 more post ;). Glad to hear that you are starting to come around

This brings me back to my original argument from the start on this debate. Min phase process may change the delay value determined by 3-measurment process, however whether the measured response is minimum phase or not is not important, only the delay value is. You can skip min phase / HBT process, arrive at a different delay value than you otherwise would, plunk the resulting driver responses into your crossover design and get the same result.

In the merge process, nearfield response should have delay applied to align phase with far field response, so that the measured far field delay is maintained in tact through merging process. This step is completely automated in VituixCAD merge tool.

Can I ask you what tools you use for min phase, HBT and merging process? I would like include them in my review as I put together my document on minimum phase.

Here's a little video of my moving the window start all over the place behind the impulse, just for visual that window start time primarily affects phase, the specific location of window start is not as critical as one may think:
https://drive.google.com/file/d/1ZsC...usp=share_link

Using WinPCD you said that you adjusted delay. What do you mean by that? Did you set the (x,y,z) parameters for each driver, then set the z-axis of one driver for the additional delay?

I modified the response files to add 1000us of excess delay to their phase response. VituixCAD calculator tool "scale, delay, invert", simply enter 1000us of delay and save result.

Were these measurements taken at one point or each driver on-axis? If on-axis, how did you then set the driver (x,y,z) values? If not, same question.

These measurements were taken on the driver axis. Drivers are physically well aligned in this design, so min phase result with the same physical offsets I entered in VituixCAD came very close to expected result, in this case, woofer y axis at -160mm, other delay values at zero. Point was not to load in perfect correct data, but to show that minimum phase or not, same result will be represented with same relative differences in phase between the two responses.

Assuming that you have some impulse response files saved for your previous driver measurements, I encourage you to go through 3 measurement delay determination (skipping min phase / HBT), followed by merging using the merge tool in VituixCAD, and compare the result against the same process with min phase applied before delay determination. You may be surprised to find the same results, which is really my point from the start. I can reproduce this easily with my own measurements, and will make an attempt to articulate this and past discussion in this thread in a coherent document. Stay tuned

With regard to SoundEasy, I'm not aware of any way to calibrate it for the standard 2.83v at 1m for sensitivity that then adjusts if the input is not 2.83v. If so, that's news to me. That's what I mean by calibration, for absolute sensitivity.

It requires SPL calibrator, I opened the manual for V25 and found it on page 16.143 "Proposed Calibration Sequence". ARTA allows for absolute SPL calibration as well using SPL calibrator. Biggest problem with this when using a USB audio interface, if you touch input gain knobs or output volume knob the calibration is invalidated. Volume adjustment should be done using software level control so ratio of mic input to reference input and output voltage remains the same at all times.

For the odd time that I want some absolute SPL I still have my old Omnimic, but for the most part it stays on the shelf these days.

[dlr]

Can I ask you what tools you use for min phase, HBT and merging process? I would like include them in my review as I put together my document on minimum phase.

I most often had used the FRD tools. Later on I used SoundEasy due to ease of use (and no Office requirement). I used it occasionally for crossover work, but don't like the complexity beyond the basics. I originally bought the V3 version to try the passive simulator that can be used to audition a passive design prior to building the crossover. It eliminated all the pain of making a crossover to audition which then often had to be modified. Back then I was still optimizing crossovers with CALSOD, I still found it better than SoundEasy for that. CALSOD has built-in minimum-phase capability as well as many other options, many of which I never used. Later on is when I used the FRD tools, when I abandoned CALSOD. For box design I originally used CALSOD, but later used UniBox. I use SoundEasy for minimum-phase generation now rather than FRD tools. But I am applying diffraction to the close-mic measurements, then splicing near/far field with FRD tools prior to importing into SoundEasy. Works easily and well for that. I'm working primarily with dipoles now, so I use the spreadsheet by John Kreskovsky for that design. That's to decide on the baffle/driver, all else is measurements afterwards. I then apply felt as required to minimize diffraction, so the baffle diffraction in my case is minimal except for the baffle step part. Finally, I import those measurements into the Ultimate Equalizer. My drivers are very close to time-aligned, so I don't add any delay.

Here's a little video of my moving the window start all over the place behind the impulse, just for visual that window start time primarily affects phase, the specific location of window start is not as critical as one may think:
https://drive.google.com/file/d/1ZsC...usp=share_link

Not as critical as other factors, no, but it has an impact.

{quote]I modified the response files to add 1000us of excess delay to their phase response. VituixCAD calculator tool "scale, delay, invert", simply enter 1000us of delay and save result.[/QUOTE]
That is not a surprise for any fixed point in space if added to all units.

These measurements were taken on the driver axis. Drivers are physically well aligned in this design, so min phase result with the same physical offsets I entered in VituixCAD came very close to expected result, in this case, woofer y axis at -160mm, other delay values at zero. Point was not to load in perfect correct data, but to show that minimum phase or not, same result will be represented with same relative differences in phase between the two responses.

What I need to know is what the exact (x,y,z) parameters were that you used for each driver.

It requires SPL calibrator, I opened the manual for V25 and found it on page 16.143 "Proposed Calibration Sequence". ARTA allows for absolute SPL calibration as well using SPL calibrator. Biggest problem with this when using a USB audio interface, if you touch input gain knobs or output volume knob the calibration is invalidated. Volume adjustment should be done using software level control so ratio of mic input to reference input and output voltage remains the same at all times.

This is what is done in LAUD, the reason I still prefer to use it, despite the limited 48K sample rate.

dlr

p.s. You can see the tools and how I used to design a 2-Way at zaph's web site: The Chameleons. I would still do similarly, but some different tools now.

[dcibel]

I most often had used the FRD tools. Later on I used SoundEasy due to ease of use (and no Office requirement). I used it occasionally for crossover work, but don't like the complexity beyond the basics. I originally bought the V3 version to try the passive simulator that can be used to audition a passive design prior to building the crossover. It eliminated all the pain of making a crossover to audition which then often had to be modified.

Thank you.

I have since moved on from SoundEasy to VituixCAD with zero regret. Not important for this discussion but I found SoundEasy was ridden with bugs and the user interface is a disaster, making for a lot of tedious manual processes that could be simple / automatic. I was tired of feeling like a beta tester with each new release, and frustrating describing to the developer what should be basic functionality for the end-user experience. MLS measurement system is the most functional aspect, but even then, alternatives like REW is free and ARTA at low license cost offer the same functionality with vastly improved UI.

On the topic of crossover simulation, I have put together an instruction to combine VituixCAD with APO EQ to simulate absolutely any filter transfer function that can be created in VituixCAD, either with passive components of active blocks. This is done by exporting the transfer function impulse response and processing through APO EQ using impulse response convolution. This offers the ability for anyone to "simulate" a passive crossover in digital domain for free. Instruction is here:
https://drive.google.com/file/d/1Ivl...usp=share_link

Additionally, to provide similar functionality to UE, I have another instruction on how to implement the transfer function block in VituixCAD as a "mirror EQ". This combined with the above instruction for exporting the filter to APO EQ can provide all functionality of UE.
https://drive.google.com/file/d/1Fe4...usp=share_link

Not as critical as other factors, no, but it has an impact.

The small impact to frequency response in that video is actually from movement of right window along with reference time. If I keep right window at same time location, only phase is affected by reference time location.

To illustrate, here I have moved window start time from just behind the impulse (green dash), to 1ms behind the impulse (blue solid and excess phase shown), but keeping right window at same time location, just before first reflection (window is 1ms longer given the change in reference time location). Response difference is at windowing tail only, at lowest end of window resolution (purple line), difference in response is a fraction of a dB, otherwise insignificant.
​

That is not a surprise for any fixed point in space if added to all units.

So you are agreeing that relative difference in phase between multiple measurements is what's really important?

What I need to know is what the exact (x,y,z) parameters were that you used for each driver.

I indicated above, y axis for woofer -160mm, other offsets at zero. But it doesn't actually matter what offsets you use, difference between summed response depends on relative difference of phase only, whether they are minimum phase or include excess phase or even negative phase is irrelevant. Result will be the same whether minimum phase or with 1000us excess phase as long as same relations between drivers is the same.

To illustrate this can be easy. With 3 measurement process and single channel USB measurement you will arrive at some delay value that needs to be included for correct simulation. This delay value is simply the excess phase that will be added to one of the measurements in addition to the measured or minimum phase provided.

Repeat the same 3 measurement process using 2 channel system and same window start time for each measurement, you will find that delay value determined through this process is zero. Simply put, the delay is already included in the excess phase of the measurement. With this validation, you should conclude that the 3-measurement process can be skipped, measured data can be used as-is with confidence of accurate results. I think this process may be the best way to show the value for 2 channel measurement for capturing accurate timing information with no extra steps. (and that includes steps for min phase and HBT).

Now, with the relative delay problem out of the way, the coordinate system can be used to indicate physical offsets only, not for correction of phase data in the measurement files. In VituixCAD you would measure on each driver's axis and continue with off-axis measurements, accurate simulation wrt to listener is determined by interpolation of off-axis measured data, and calculation of SPL change over distance due to driver distance to listener. Combined result at listening distance at 2-3m can be achieved even though measurement is at 1m. To toot the VituixCAD horn, this is a successful process that is currently being used by industry professionals world wide (see the "supported by" at the bottom of VituixCAD website).

This is what is done in LAUD, the reason I still prefer to use it, despite the limited 48K sample rate.

Not important but REW is a modern free tool that provides same functionality, as well as same SPL calibration capability as SE and ARTA mentioned previously.

This has been an interesting discussion, thanks for persisting and with an open mind.


​

[dlr]

Additionally, to provide similar functionality to UE, I have another instruction on how to implement the transfer function block in VituixCAD as a "mirror EQ". This combined with the above instruction for exporting the filter to APO EQ can provide all functionality of UE.

I said simulator, should have said emulator. You can run SoundEasy (SE) as a DSP crossover emulator, much the way the Ultimate Equalizer (UE) is. It's an active DSP unit for auditioning, what I find the most useful in SE.

However, now I use mostly the UE. I set it up with the crossover target functions I am considering using in a passive system. I then connect it to my amp and audition that crossover, change the crossover based on my impression, then go back to the passive design to see if that is reasonable to do passively. The UE, like any DSP, has no driver impedance anomalies introduced, but that's a separate issue to be handled when designing the crossover passively.

So you are agreeing that relative difference in phase between multiple measurements is what's really important?

Of course, I never debated that differently. What I did and do say is that the 3-measurement method provides the most accurate modeling and is required to get a the best relative positioning of each driver, so I find minimum-phase measurements to be ideal. I've not seen a way to determine relative offsets accurately otherwise. Even measured phase with correlated phases between drivers requires that the model position the driver with all three (x,y,z) values if you want the off-axis to be correct. This would require knowing the true acoustic center, but that is unknown to high degree of accuracy for drivers. The relative offset method obviates the need to know that.

I indicated above, y axis for woofer -160mm, other offsets at zero.

I take that to mean that you left the z-axis at zero. If that is so, then any off-axis result will have some degree of error, usually increasing the farther off-axis for most of the front hemisphere. That's been one of my main points all along. Any system can easily model the single design point with measured phase. I did that any number of times over the years when I wasn't very concerned with the off-axis. It was always a good, easy starting point. But I am more interested in the off-axis, especially as it relates to the power response. I'm not too concerned with 5 or 10 degrees horizontal off-axis change, but that's because I use felt liberally, so the on-axis are both much smoother. But there is little impact of felt on power response.

But it doesn't actually matter what offsets you use, difference between summed response depends on relative difference of phase only, whether they are minimum phase or include excess phase or even negative phase is irrelevant. Result will be the same whether minimum phase or with 1000us excess phase as long as same relations between drivers is the same.

It relates to the off-axis. If there is no z-axis offset to properly position the drivers, then the off-axis response cannot be accurate. Not in any software I've ever used. The horizontal polar response may see little difference due to smaller excess-phase (delay) change with angles, but the vertical will likely be more affected. This, of course, is at individual points examined, but that affects the power response calculation as well. Power response for fixed rotation deltas (I used 5 degrees of rotation within each ring that is also 5 degrees progressively) must have a weighting assigned to each position as each ring represents a different amount of the total integrated area. Now if the software does an integration of very small deltas, the error with no weighting becomes much smaller.

So I'm not focusing on a single design point, the mic measurement point. I should say the on-axis design point.

dlr