VituixCAD v2

[Nil L]

Thank you. I think that if you choose these two of all variables, then matching only the impulses of the drivers does not guarantee a good sound.

[Reet]

I might understand wrong, but IR text export looks total mess where start time or time with estimated delay are not known at least in single channel mode. Just time from the beginning of sample buffer to IR peak is known and commented with "//Start time". Then also name is misleading because that is peak time from the beginning of sample buffer. We can just hope that t=0 equals to constant 1s with timing reference, but that could be variable too because user can move IR in time scale with Control window.
I still don't know what to do with this import. Probably the best is to set initial Skip samples to 1s/sampleInterval - 1 and that's it. And fix the problem with generic time series informed by FrancoB on diyaudio.

John has provided this additional information:
”As an addendum to this, in case it isn't obvious: t=0 occurs (-start time) seconds after the start, the start time is relative to t=0. In samples that is (-start time)/(sample interval).”

I guess for this it brings me back to previous statement, start time would need to be read for each individual file on import.

[kimmosto]

"Start time is the time at which the entire data record starts, generally 1 second before the IR peak. The time that corresponds to depends on where the peak is relative to t=0."

”As an addendum to this, in case it isn't obvious: t=0 occurs (-start time) seconds after the start, the start time is relative to t=0. In samples that is (-start time)/(sample interval).”

Generally 1 s before the peak -> depends on where the peak is relative t=0 -> -start time is relative to t=0. That's clear 😀 Looks that I should restore few lines from previous revision and read start time marked with comment text individually from every response and skip different number of samples to get time locked also in VCAD Convert IR to FR. I don't like to do that, because it removes user control to Skip samples field while exporting multiple FRDs at once. That could be a problem in some cases without adequate spare samples before t=0. Spare samples will not show actual flying time anymore so 1000 mm is not 2.907 ms.

t=0 could be exactly the same sample number in all wav and all text exports. Another option would to round start time to the nearest full sample close to some constant time such as 1 s or 100 ms. In addition, measurement with loopback reference should prevent any manipulation of time scale via Control window to keep timing locked to reference for good. Also measurement window should not have any timing settings with loopback timing. Phase in FRD exports and SPL displays should be relative to adjustable Reference time. Not to adjustable t=0, because time scale (t=0) is locked with the reference. These changes would make REW's timing concept compatible with ARTA and CLIO. Hopefully as simple too.

[Reet]

Generally 1 s before the peak -> depends on where the peak is relative t=0 -> -start time is relative to t=0. That's clear 😀 Looks that I should restore few lines from previous revision and read start time marked with comment text individually from every response and skip different number of samples to get time locked also in VCAD Convert IR to FR. I don't like to do that, because it removes user control to Skip samples field while exporting multiple FRDs at once. That could be a problem in some cases without adequate spare samples before t=0. Spare samples will not show actual flying time anymore so 1000 mm is not 2.907 ms.

I'm okay either way if you decide to make a change or not, as there are ways around the limitation of REW's impulse response export feature.

Also measurement window should not have any timing settings with loopback timing. Phase in FRD exports and SPL displays should be relative to adjustable Reference time. Not to adjustable t=0, because time scale (t=0) is locked with the reference. These changes would make REW's timing concept compatible with ARTA and CLIO. Hopefully as simple too.

Agreed, and I made this comment to John not that long ago, but he was not in agreement, suggesting that it makes it too easy to screw up timing across multiple measurements since windowing is easily adjusted on a per-measurement basis, for all the arithmetic functions within REW. T=0 is easily adjusted as well so I don't fully agree with that argument, as well fixed timing is easily selected using "apply windows to all" button. In any case, I don't think t=0 for timing/phase reference in REW is going to change anytime soon unfortunately. Likely a problem of a measurement program that was first designed without timing reference for room EQ, and loopback timing was a later addition as features expanded.

[kimmosto]

2.0.98.6 (2023-01-10)

Convert IR to FR

• Import of REW's impulse response as text modified:
  Skip samples set to (-startTime-0.002907)/sampleInterval with 'no timing reference',
  -startTime/sampleInterval with 'loopback as timing reference' and 'loopback as cal and timing reference'.
  Skip samples set individually for each IR file.
  Note! See preferences and measurement settings in the latest VituixCAD_Measurement_REW.pdf (2023-01-10).

Fixed few wrong settings in Measurement instructions for REW. For example ‘Make calibration data from loopback response’ in measurement window must be checked to avoid severe timing error due to normalization of peak time. Any resampling, decimating or alignment of impulse peak or t=0 to full sample is not needed (in my opinion). Wrong setting in some of those special preferences could produce tiny error or difference compared to some other app, but it's slow to test every combination.
This package is finally tested. "Time stamps" in REW, ARTA and Convert IR to FR are equal. Level is still mystery i.e. original SPL is not received automatically so user have to play with scaling and some relative level when using IR exports and Convert IR to FR.

Generally, setting up REW for time locked measurements with any timing reference option is about decade more difficult and slow compared to ARTA, and infinitely more difficult than CLIO which does not require any special settings (due to full duplex IO). Just plug and play, and look the result and save if autosave is not activated.

^I agree that development of REW's timing concept has started from wrong end, but changing that is probably (too) big and risky job. Everyone who started with justMLS of LspCAD in dual channel mode knows how simple this task really is. Impulse buffer does not require any adjustable time scale. Just sample buffer with sampling intervals so we can agree that first sample is t=0. Cross-correlation function and other math puts measured IR to impulse buffer. Timing as sample# * interval is relative to reference (t=0). Phase response depends on different delays in the system so we need working method to adjust phase wrapping to show a) close to minimum phase b) difference to timing reference c) difference to some common mechanical point d) difference to other measurement e) without processing delay of DSP etc. Reference time provides easy and quick tool to show and export phase with whatever time reference for any or all measurements without damaging and losing original measured time series locked with known reference.

[kimmosto]

You may also notice that single channel mode is totally removed from Measurement with REW pdf. The reason is that there is short delay between cone and ports or passive radiators due to speed of sound. Controls window for IR graph does not include an option to move IR to give correct sum of cone and ports/passives. Error is not necessarily too big or even bigger than other errors related to processing and merging of NF and FF data. But it's easier for user to keep loopback connection as is (after far field measurements) and continue to near field with the same settings, except Timing offset and Level.

[Reet]

Fixed few wrong settings in Measurement instructions for REW. For example ‘Make calibration data from loopback response’ in measurement window must be checked to avoid severe timing error due to normalization of peak time.

I assume that you are inferring that you've observed timing error when either "merge loopback response to IR" is selected, or when loopback is used for timing only? That's interesting, I will do a little comparison later today. If nothing else I can provide some feedback to John as I'm sure this behaviour is not intended. In any case, I would expect that "merge loopback response to IR" would be preferred for full "dual channel" measurement at the impulse response level, otherwise impulse response is same as "loopback as timing reference", and a separate calibration response is required to normalize the response to the loopback channel.

[Reet]

For quick test on my configuration, at least I can show no timing error regardless of loopback method used.

[kimmosto]

"merge loopback response to IR" is selected, or when loopback is used for timing only?

The first one. Peak was normalized to 2 ms when measuring actual (DSP) speaker with ca 3 ms latency and 'Merge loopback...' was selected. That happened with FF measurement from different distances and NF of cone and port. I'm quite sure that I verified normalisation from REW's IR graph. Not with text exports in Convert IR to FR which can normalize to 2 ms if the file does not contain correct/assumed comment texts. Normalisation to 2 ms did not happen a week ago when I measured plain loopback cables in both channels. It's also possible that some coincident happened with some preference setting. Or my brain tilted temporarily I can make new test if you can't verify this.
'Loopback as timing ref only' was working okay in my tests. Just some small magnitude errors at LF and HF.

[kimmosto]

Quick test. Normalisation still happens. Red is correct with 'Make calibration...' and green is normalized and wrong with 'Merge loopback...'. So my brain is okay. Huh.

[Reet]

Interesting. In my test above, feedback is completely analog so I will test again shortly, adding miniDSP to the loop to provide 1-2ms of processing delay in the feedback loop, hopefully I can reproduce some error on my end.

[Reet]

Quick test. Normalisation still happens. Red is correct with 'Make calibration...' and green is normalized and wrong with 'Merge loopback...'. So my brain is okay. Huh.

Sorry, I can't reproduce this. DSP in the feedback loop or not, timing appears "perfect" on my end regardless of timing method used. I used miniDSP configured for pass through, it provided 1ms additional signal delay, so if there was some error with the timing reference it should have been obvious.

[kimmosto]

So it must be different revision. I have V5.20.14 and you probably have some rc where normalisation is fixed and added "set t=0 at sample 256" option to preferences.

[Reet]

I'm using the latest 5.20.14 early access build from here, looks like it's up to build 23 as of now:

https://www.avnirvana.com/threads/v5-20-14-early-access-build.11154/

[kimmosto]

My earlier was ea12. Just installed ea23. No changes; "Make calibration..." is okay but "Merge loopback..." is not.
But I found the reason:
"Make calibration..." and "Merge loopback..." give the same and correct result when Timing offset is greater than 0.0000 ms for example 2.907 ms.
"Make calibration..." and "Merge loopback..." give different result when Timing offset is 0.0000 ms. "Make calibration..." is correct and "Merge loopback..." is wrong.

This is something at least I don't expect to happen.

[Reet]

I still can't reproduce a problem on my end, with timing offset at 0.0000ms.

[kimmosto]

My previous description was not exact. Error happens with Neumann KH150 if Timing offset is 0 or 1 ms and Merge loopback is selected. But also Merge loopback works perfectly if Timing offset is 2 or 3 ms. Make calibration works okay with all tested Timing offsets 0-3 ms. Latency of KH150 DSP is ca. 2.8 ms. Measurement distance 20 cm.

With JBL LSR305, all tested Timing offset values from 0 to 3 ms give correct result with both Merge loopback and Make calibration. Latency of LSR305 is ca. 0.8 ms.

Looks that difference between system delay (DSP latency + sound flying time) and Timing offset setting should be within some window/tolerance to lock math and get accurate result with Merge loopback option. If delays don't match program does something unexpected.

[Reet]

Great information, I'll share this thread in the REW forum, hopefully John can determine what the problem is and correct it for the final 5.20.14 release.

[JohnM]

Looks that difference between system delay (DSP latency + sound flying time) and Timing offset setting should be within some window/tolerance to lock math and get accurate result with Merge loopback option. If delays don't match program does something unexpected.

I can't reproduce the behaviour you see even with very large delays in the measurement channel and no timing offset. Could you attach screenshots of your Measure dialog and your Analysis preferences please, there must be some combination of settings I'm missing.

[kimmosto]

mdat files:

https://kimmosaunisto.net/Misc/KH150_test.mdat

https://kimmosaunisto.net/Misc/LSR305_test.mdat

[kimmosto]

Scarlett 2i2 does not have perfect reputation so there is small possibility that some tiny crosstalk interferes timing difference detection. I can do more tests tomorrow without speaker and mic signal, with Xilica XP as a DUT. If needed.

[JohnM]

Thanks. The problem happens when the left window is shorter than the measurement delay. I have fixed it for the next build, in the meantime just increase the left window width.

[kimmosto]

Thanks. The problem happens when the left window is shorter than the measurement delay. I have fixed it for the next build, in the meantime just increase the left window width.

Ok, thanks. I was thinking that before fell asleep because length of the left window and problem limit in ms were quite equal.

Anyway, all IR functions using also frequency domain (such as ETC calculation) should be independent from user's time window settings - defaults and final adjustments. Fortunately loopback signal is - or at least should be very clean and flat so finding IR peak and defining possible time window and FFT buffer automatically without user's interference should be easy. For example ETC or automatic search for t=0 or Reference time -> whole single channel measurement can fail much easier because IR peak of some delayed reflection can be higher than direct sound when directive speaker such as a horn is measured to off-axis >>90 deg.

[Reet]

Latest early access build 24 released today claims to resolve above mentioned timing issue.

https://www.avnirvana.com/threads/v5-20-14-early-access-build.11154/

[kimmosto]

Loopback calibration and timing is still using time window settings in Preferences. Timing and calibration features should not use any default or user settings for time window. Loopback calibration is full range operation so possible visit in frequency domain should happen for example with Left=500ms, Right=1000ms to create correct response for calibration.
This (2) is consequence if Time window settings in Preferences is set for FF measurements in a small room, but user decides to peek response with long time window:

[Reet]

That's unfortunate, I would suspect the problem is with the implementation of "merge cal with IR", which implies that the loopback channel is run through FFT process to frequency response, then back to impulse response applied to the measurement channel for normalization. Window for this process should then be entire length of measurement, it sounds like this isn't the case. I'll just stick to ARTA for loudspeaker measurement for VCAD, due to simplicity and reliability of results.

Unrelated, but for a small feature request, I think it would be great in the IR to FR tool to physically show the DC offset adjustment from "DC offset" checkbox in the impulse response view, and recalculate ETC and dBFS plots. I think it will make it more obvious to show the effect this option has on the measured result. Sometimes there is low frequency background noise, depending on specific measurement time, DC offset can improve or make worse, but it should be clear from ETC graph to see the overall energy prior to the impulse reduced or not.

[Reet]

To avoid cluttering up the VituixCAD thread, I've posted some evaluation of DC offset in a new thread, here:

https://www.htguide.com/forum/mission-possible-diy/development-tools-techniques/927887-dealing-with-dc-offset-in-measurements

[kimmosto]

Unrelated, but for a small feature request, I think it would be great in the IR to FR tool to physically show the DC offset adjustment from "DC offset" checkbox in the impulse response view, and recalculate ETC and dBFS plots.

Drifting of Step response is quite good indicator for DC offset. ETC shows some difference too, but visiting in frequency domain is done with quite short FFT to maintain performance so indication is not necessarily very strong. VCAD does not have software low pass. It just measures average level of IR in the beginning. Next revision calculates average within ca. 0 ms and 1.5 ms before start of impulse. Few other tricks included to filter possible disturbances such as crosstalk with ARTA close to sample 300. DC compensation is not entirely bullet-proof and as fast as current revision especially with REW import due to 88k extra samples, but can be disabled.

[kimmosto]

2.0.98.7 (2023-01-16)

Convert IR to FR

• DC offset detected and compensated before skipping samples and FFT. Compensated DC offset included in IR, Step, dBFS and ETC responses.
• DC offset value shown in text box above DC offset checkbox.
• DC offset checkbox disabled with CLIO crp and mls/mlsi files.

Graphs

• Curve tension reduced to produce smoother traces: Impedance mag, dBFS, ETC, Auxiliary Box resonances.

Main

• Trifonov TP removed from Shape list box.

[Reet]

Thanks! The offset value and visual representation makes it easier to understand the effect the DC offset checkbox has to the measurement, beyond just the difference in low frequency behavior of the frequency response.

[tktran]

Hello Kimmo,

Can you tell me how the Listening Window and Power Response is calculated in VituixCAD2?

The reason I ask is that I was doing some diffraction modelling recently, (Diffraction tool) and noted that the Listening Window, if calculated as the average of 9 measurements according to Harman (On Axis, Hor +10, +20, +30, -10, -20, -30 and Ver -10 and Ver +10) IS different to average of Hor -30, -29, -28... On axis, Hor +1, +2, +3... 30, and Ver -1, -2, -3... -10, Vert +1, +2, +3... +10).

That is, if one takes the average of 9 measurements for the listening window, it is different to the average of 81 measurements.

It follows then, that the calculated Listening Window and Power Response is dependant on how many measurements are taken (ie. step angle 10 degrees vs 1 degree) and when fine-tuning a crossover, how well we measure correlates with what we hear.

I will attach photos to explain when I get home...

[Reet]

I understand what you mean. If you use the diffraction tool, and export result at 1 degree increments. Power & DI Chart is different if "angle step" in options is set to 0 deg, 10 deg, or 15 deg increments in the options for power & DI. It comes down to the resolution of the data simply, lack of angular resolution being filled in by interpolation does not 100% replace missing information.

[Reet]

I should add, that it's really only a problem for speakers with narrow angular "problems", like the beaming of the woofer simulated above.

[tktran]

Reet
Exactly. Thanks for visual explanation.

For me the question it is more about what's going on above 2-3KHz, on a typical baffle eg. 8-10" wide, and hence- what we are measuring is typically shown in what you show as green, if taking typical 10-15 degrees step. But is the reality of what we are hearing actually the YELLOW? Or are we hearing closer to the WHITE, due to the wavelengths involved?

So
A) for a passive speaker with fixed level. eg commercial design, what is the "correct" tweeter response. ?

B) Perhaps there is no such thing as the "correct" tweeter level; it should it always be matched to speaker placement/room acoustics or tastes (?or something else)

C) Should tweeters be measured with a 1 degree step? If I should, then I will. Oh brother! (time for an automated turntable)

[kimmosto]

Your test material is way too far from reality. Difference between on-axis and LW is less than 5 dB with quite large tractrix horn. DI somewhere 15-20 dB at top octave. There is small difference <1 dB between angle step 5 and 10 deg. I would call that almost insignificant for optimizing, but 5 deg is more accurate of course assuming that measurement gear is accurate.

Conventional speakers have less than 2 dB differnce between ON and LW at top octave. 10"-15" PA coaxials too. Angle step of 1 deg is never needed in practice. If you need that, whole design is crap and you don't need to measure it. I usually measure conventional speakers with 10 deg angle step, and more directive random shotguns with 5 deg. Motorized table helps with 0-355 deg with step of 5 deg. Manual table and wireless keyboard is fully adequate for 0-180 deg with 10 deg step.

[kimmosto]

Example 2. This is cardioid WWMTMWW concept with Fountek NeoCD 3.0 tweeter. Top octave in vertical plane is different with 5 and 10 deg angle steps for sure, but horizontal rules so finally there is not so much difference or error.

[kimmosto]

This speaker might have bigger difference in LW between 5 and 10 deg angle step, but I don't have measurement data with 5 deg step to know and show it. Tweeter is Mundorf AMT25. DI in ver plane only would be 19 dB at 20k. Beam height is quite...too narrow at top octave in ver plane which could cause more than 0.5 dB difference.

[tktran]

Very interesting.

What angles are used to calculate the Power Response?

[kimmosto]

What angles are used to calculate the Power Response?

That requires complex and long answer if "power response is calculated to simulated angles within selected space in selected planes" is not adequate answer. In addition, calculation of weighting factors for each off-axis angle can also be quite complex in case angle step is not constant within 0-180 deg.

Instructions in user manual related to power & DI calculation:

Frequency responses

Mirror missing defines if VituixCAD should mirror missing measurement data:

• Measurement to equal positive angle selected if negative angle is not available (or vice versa)
• Measurement to equal horizontal angle selected if vertical angle is not available (or vice versa)
• Measurement to equal absolute horizontal angle selected if vertical angle is not available (or vice versa).

Frequency responses are interpolated between off-axis angles and planes loaded to drivers when Interpolate is checked. Program selects the closest off-axis angle loaded to driver when Interpolate is unchecked. This could cause some errors and discontinuities, but enables simple tests and projects with few responses only in Directivity chart.

Power response & DI calculation

Intensity on spherical surface is normally selected for common sized single or multiway speakers. Intensity on spherical surface around speaker is calculated from radial measurements in horizontal and vertical planes.

Note! This is simplified formula copied from book by Beranek & Mellow. It is not used in VituixCAD because cannot cope variable angle step and 0/180 deg measurement.

Intensity on cylinder surface is practical selection for long line sources, or if either horizontal or vertical directivity is temporarily interesting - not accurate power response & DI result. Intensity on cylinder surface around speaker is calculated as pressure R.M.S. from radial measurements, typically in a single (horizontal) plane.


Checkboxes control which planes are included in power response and directivity index calculations; horizontal, vertical or both.


Half space is for half space designs; speakers flush mounted to wall. Angles >90 deg are excluded from power response and DI calculation, and constant 3.01 dB is added to all DI responses. Directivity chart shows angles -90...+90 deg only.

Corner is for quarter space designs; speakers integrated to inner corner of walls. Angles >45 deg in horizontal plane and >90 deg in vertical plane are excluded from power response and DI calculation, and constant 6.02 dB is added to all DI responses. Directivity chart shows hor -45...+45 deg and ver -90...+90 deg only.

Common boxed speakers and dipoles should be measured and simulated to full space with measurement data 0... ±180 deg.


Check Listening window DI to use listening window average as DI reference instead of selected Reference angle.


Angle density of simulated and visualized off-axis directions is selected with Angle step list box. Available options are 0, 5, 10, 15, 20 and 30 deg. Off-axis angles loaded to drivers are simulated when 0 deg is selected. Initial value is 10 deg.

[kimmosto]

2023-01-18: Quick manual to produce measurement data with REW is not available as long as dual channel measurement with 'Use loopback as cal and timing reference' option does not work.

The reason for removing pdf from the server is that cross-correlation function which should compensate measured frequency response with response of loopback signal does not work. Long time window in Preferences fix something, but not everything. Loopback signal without high-pass works, but it's not accurate and adequate. Also loopback could have high-pass filtering for protection, with or without latency by DSP. High-pass in the loopback ruins the result at the moment.
Users are free to tests every ea build in single and dual channel modes with all possible preference and measurement options, and communicate with JohnM about possible bugs, but I don't try to maintain and publish varyingly wrong instructions until REW works perfectly in typical use cases also with calibration reference.

[Reet]

Users are free to tests every ea build in single and dual channel modes with all possible preference and measurement options, and communicate with JohnM about possible bugs, but I don't try to maintain and publish varyingly wrong instructions until REW works perfectly in typical use cases also with calibration reference.

FWIW I have been in contact with John, and believe I've conveyed the main issue in recent posts at the REW forum here:

https://www.avnirvana.com/threads/v5-20-14-early-access-build.11154/post-87620

Hopefully we will see a fully functioning dual channel measurement shortly. Keep in mind that we are working on early access releases, for evaluation of instructions for VituixCAD, "official" REW release should be the only option to consider.

[Reet]

Speaking of which, build 25 posted just moments ago:

Changed: Merging cal data into the IR uses the whole IR rather than the windowed span

[kimmosto]

Connections in Measurements with ARTA.pdf can be used for testing how for example active 1st order Butterworth high-pass at 1 kHz works. I have Xilica XP-4080 with >1ms latency for this kind of tests.






One extra possibility is to add high pass and shelving high pass options to stimulus. Shelving HP with for example 20 dB step would produce at least something also to LF to avoid measuring just noise. 1st and 2nd order options to HP. Filtering in stimulus can be compensated mathematically without requirement for calibration with loopback. HP/LP feature looks like this in CLIO:

[kimmosto]

Keep in mind that we are working on early access releases...

I'm not very familiar with that kind of software production concept. Public releases for testing by anyone; someone or no one. What would the final version have been like if neither of us had tested, assuming someone else had? My daytime job with machine automation involves some concern for people's lives and costs due to crashes preferably closer to zero than a million dollars, but fortunately programming and testing VCAD can be more sloppy. Anyway, my mistake that I started to update measurement instructions assuming that calibration with loopback works when it is published.

[Reet]

I should have really been a "beta" release, sine all the new features were presented for user testing and feedback. However, I believe that the "Use loopback as cal and timing reference" was rolled out in the current official 5.20.13 release, however the issue you've observed had not been caught until now. It didn't come up in any of my measurements due to the specific conditions, and use of default window settings of left 125ms and right 500ms. In any case, latest release at least applies the calibration arithmetic across the entire measurement without windowing, so hopefully there are no further causes for concern. 🤞