Thanks Dcibel. I am not familiar with VituixCad. will try it out.
I am planning to take another set of measurements, will try out the t=0 timing referenced. Just so I understand the process and what i am supposed to do...
Thanks for the screenshot of Holms Response. Should i take measurement of each driver on it's own axis? So if I start with the tweeter
1. Measure on Tweeter Axis 2. Set the Time zero locked and click on use 3. Measure Tweeter again? on tweeter axis - this is step 5 in the Holms Impulse help - how do i know it works? 4. Measure upper Woofer on Upper Woofer axis 5. Measure lower Woofer on lower Woofer axis
No need to measure all three together? The NF still needs to be merged with the individual Woofer responses? So upper NF merge with Upper FF Time locked Woofer and Lower NF merged with Lower FF Woofer. Would the time lock affect the merge?
You know, if you've already headed down the single channel 3-measurement offset determination path, maybe it's best to complete a design with that path before entering the more "advanced" measurements. But, since you have the gear and you can compare measurements, here's some more explanation.
With a single channel measurement, the input to the computer has little reference of when the signal left the speaker, so the "time of flight" from the speaker to the mic input is mostly lost. The only reference you have is an impulse peak indicating that the sound did hit the mic, and everything that follows. So the start of the FFT window for a single channel measurement is generally based on a peak detection method, using the peak of the impulse as reference. This gets you something close to minimum phase, but adding the HBT "tails" determines minimum phase of the response to greater accuracy. You can determine offsets using a single channel system using that 3-measurement method, but you must retain a single mic location for that to work, meaning your "on-axis" midwoofer measurement is actually slightly off-axis.
A dual channel measurement will use a feedback signal from the amplifier output as a reference impulse, so that the time of flight can be accurately determined. Using this method, you can easily lock the start of the FFT window to a per-determined distance from the mic, such as the baffle surface, and as long as you keep the same mic distance, all measurements will have the time of flight from baffle surface to driver included. This is a great way to do things, but requires a standard mic with 2-channel soundcard, not a USB mic as it is important that the reference channel and mic input come from the same clock source, to keep consistency between reference and mic input. You wouldn't want these signals to be off by more than 1 sample really. An added benefit of a 2-channel system is any non-linearity in the equipment up to the speaker gets effectively calibrated out by the reference signal.
To get some of the benefit of a dual channel system, you can lock the window start so all measurements include a very specific delay before the start of the FFT. To verify that this method will work consistently for you, I would take a measurement as the Holm instruction says, then lock the timing reference and take several subsequent measurements, and observe the phase plots. If the timing is indeed locked, the phase of all subsequent measurements should remain constant.
You'll probably need to use VituixCAD with the right options selected (min phase unchecked) to merge the responses with the additional phase delays included, most other free DIY tools available will just extract minimum phase which will wipe out any excess delay that you are trying to capture. One thing that does appear to be missing from VituixCAD merge tool is the HBT tails, so it looks like you may not benefit from the added phase accuracy of tailing the low and high end of the response.
I was thinking about your MTM measurement, and one easy way to avoid measuring each driver individually, is to use the far field measurement of both drivers playing together, simply subtract 6dB from the resulting measurement and load it into PCD or whatever as a single driver, load twice with the right offsets and this way you can load in the MTM pair with everything you need for off-axis simulation without doing any messing around with individual driver measurements.
If you plan to blend the near field, you will lose the as measured phase. You will be back to extracting min phase of the blended Woofer file, Tweeter file, then using the T+MM measurement to derive the offset.
Hbt and blending will affect phase at the extents, delay can be added back in to match the phase of the far field measured response in the passband to retain the time of flight excess phase. It works, ask me how I know.
Here's far field response, start of FFT arbirarily at 65.2cm to ensure there is some excess delay. As you can see the phase is not "minimum phase".
The near field response, start of FFT here was set to 0.
I then applied the baffle step to the nearfield response, and did a quick and dirty splice at 700Hz.
purple - spliced response
black - HBT "tails"
green - measured phase
blue - HBT minimum phase
As you can see, there is not much agreement in the phase here, I did not start my FFT at correct location for minimum phase data, not by a long shot.
Next, I added a delay of 0.241ms, or 83mm. Now, I would say there is a great deal of agreement in the phase from 700-5kHz+, and the phase of the nearfield splice has been corrected. I would say this is a blended response with the measured phase retained, would you not agree?
With this result, I have two options. Either save the file with the included delay, then in my crossover design I would require zero Z offset, provided I did all my other measurements at the same mic distance and with an FFT start of 65.2cm, or I keep the minimum phase data, knowing that I need to add a 83mm offset in the design to account for the extra delay.
Excellent discussion. I've noticed this "time delay" relationship when I use XSim to switch back and forth between minimum phase for all drivers and raw OmniMic "as measured" phase for all drivers. OmniMic's raw "as measured" phase appears to be a type of minimum phase that includes just a little bit of extra time delay for the software to reach the peak of the impulse response. When I set "mod delays" using the "get file" procedure in XSim, the HBT phase is always "faster" and has less time delay when doing the matches.
Yes, Jeff used to call Omnimic phase "quasi minimum phase" or something like that. Omnimic uses the peak of the impulse, so you can expect the HBT phase might not match up. For a single channel measurement system, I think the window time lock that you can do with Holm or REW is a better way for speaker design if you are limited to a single channel measurement like a USB mic. The 3 measurement delay determination is more of a Band-Aid solution for not being able to measure the delay reliably in the first place.
dcibel said
To verify that this method will work consistently for you, I would take a measurement as the Holm instruction says, then lock the timing reference and take several subsequent measurements, and observe the phase plots. If the timing is indeed locked, the phase of all subsequent measurements should remain constant.
Once you've done that, move the mic back a bit and take another measurement, do you see the impulse move ahead in time and excess phase in the response?
dcibel said
If the timing is indeed locked, the phase of all subsequent measurements should remain constant.
Not sure what you mean by "the phase of all subsequent measurements should remain constant."
The woofer phase is not matching with the tweeter phase. I have taken the tweeter measurement, then locked the t=0 and clicked on use and then took tweeter measurement again, that is the blue plot and then the individual woofer measurement the red and green one.
This image shows the initial tweeter measurement (blue), the locked tweeter measurement (based on the initial tweeter measurement) and one of the woofer measurement (t=0 still locked)
dcibel said
If the timing is indeed locked, the phase of all subsequent measurements should remain constant.
Not sure what you mean by "the phase of all subsequent measurements should remain constant."
What I mean is, when you lock the timing reference, then take say 10 consecutive measurements with your mic and speaker untouched, than all 10 plots should show exactly the same phase response. This is to verify that the timing between your soundcard to speaker and mic to PC are accurate enough for repeatable measurements. If there is any error, then there is a potential for error in any measurement with a locked timing reference in Holm, you might as well stick to the 3-measurement offset determination or invest in a 2-channel setup if you want to continue down that path.
There is already some cause for concern as I would expect the blue and red phase to be equal, but they are not.
It's important to understand what happens with respect to the phase as distance is increased or decreased. One easy way to visualize this in Holm is in the response options, under time zero, you can apply an offset by a number of samples. Try adding or subtracting a few samples to see what it does to the phase, this is the equivalent of moving the mic closer or further away from the speaker while keeping the same timing reference.
FWIW @ani_101 I did some searching and you may be SOL to get reliable results with Holm timing lock and a USB mic, but you may try the test I explained above to confirm.
That indicates Holm having timing lock issues if the playback and record device don't share the same clock, as would be the case using a USB mic with integral ADC and a soundcard output for playback. The way I would look at this personally is that if you have to use a standard XLR mic so input and output are from the same soundcard / audio interface for the timing lock to work then you might a well build a proper 2 channel setup and have a proper timing reference, then you'll have to move on to REW, ARTA, or SoundEasy unfortunately.
If you complete the repeatability test I explained above with issues, then I'm sorry for wasting your time taking you down this path, but it's not going to provide accurate results. You'll have to continue with 3-measurement technique to continue using Holm and a USB mic.
Just a note of clarification: whatever FR files you use to
determine the acoustic offset (AC) should be the same files you use for crossover
development. For example, if you use measured files to
determine the AC you should use those same files in your crossover
program. On the other hand, if you
create minimum phase files then those should be used to determine the AC. There’s a good thread HERE on PETT about the
subject. In that thread Jeff B. stated
this: “don't expect the relative offset to be the same using the measured phase
and the extracted phase; it will change considerably.”
As previously mentioned, the advantage of using minimum
phase files is you can model accurate summed responses at different angles and distances. Otherwise you’re
limited to the axis and summation
point of the measurement.
Thanks for the link Ed. I usually take the measurements throughout to pcd. But I'll run then through Jeff's min phase xls and then import to pcd.
Dcibel, no worries, I wanted to see what the single measurement would look like, but would be using the 3 measurement method for now, though would like to understand the single measurement process a bit better.
I just want to say this thread is a treasure trove of practical information. I am learning a lot following along. thanks Ani for being humble enough to ask for help and explanation, and to everyone who's working through this with him.
Just wanted to add something to this thread, as I am just finishing up a big tower project with the woofer at floor level.
Taking a measurement at about 4ft back at tweeter level, I compared the response of what you get when you gate out all reflections, vs leaving the floor reflection in and gating to the second reflection. You can see here that the difference in response from 100-500Hz is quite significant, which is why you can reduce baffle step compensation from the normal 6dB to more like 0-2dB when the woofer is at floor level.
Comparing to the room response (ungated), you can see as well that the gating out all the reflections would be missing a significant factor, as we are relying on the floor re-enforcement to always be there.
@Ed_Perkins said:
So how do you think a spliced near field/far field response would look compared to the un-gated?
That's where it gets a bit complicated. The nearfield response won't include the floor reflection, so if you apply a baffle diffraction sim to the nearfield response like you normally would, you are generating a response that doesn't include that re-enforcement. For my simulation for design, I ended up splicing a nearfield response that didn't have the baffle diffraction applied at all. The measurement above is a real-world measurement of that crossover design. The overall trend of the in-room response is rather flat.
Yes, without BSC applied to the woofer the real world response is rather "flat". If you had applied BSC, the sound would be very bloated as I'm sure 4thtry can attest. My personal preference would probably be about 2dB of compensation.
I don't understand your question about the floor reflection being "identifiable".
There certainly is a null if the driver is high up off the floor, but since it is so close to the reflection source, the null occurs much higher up in frequency, and in the frequency range of the woofer the reflected wave is mostly additive, not subtractive, which is why the need for BSC is so much less.
Have a look at the red line above, additive behaviour from the floor from 500Hz and less, the first null occuring at 550Hz.
@ani_101 said:
Theoretically the floor bounce results in a null, which would be a dip in the response. I say theoretically as I haven't observed it.
Not sure if there would be a dip or it's low enough in the FR that the room dominates. Is one of the dips in the actual response the floor bounce?
My question can also be totally off base....
Ani, I think you need to clearly distinguish between "floor bounce" and "floor boundary reinforcement". The floor boundary reinforcement effect can be viewed as an 0-6dB inverse of the baffle step, extending over a wide range of frequencies in the 80 to 500Hz region. Floor bounce would simply be a dip in your measurement, somewhere in the 100-200Hz area, based on the microphone tip position with respect to the floor and woofer cone. As dcibel points out, since the floor and woofer are very close together, this affects the frequency of the "floor bounce."
Comments
The woofer phase is not matching with the tweeter phase. I have taken the tweeter measurement, then locked the t=0 and clicked on use and then took tweeter measurement again, that is the blue plot and then the individual woofer measurement the red and green one.
This image shows the initial tweeter measurement (blue), the locked tweeter measurement (based on the initial tweeter measurement) and one of the woofer measurement (t=0 still locked)
Just a note of clarification: whatever FR files you use to determine the acoustic offset (AC) should be the same files you use for crossover development. For example, if you use measured files to determine the AC you should use those same files in your crossover program. On the other hand, if you create minimum phase files then those should be used to determine the AC. There’s a good thread HERE on PETT about the subject. In that thread Jeff B. stated this: “don't expect the relative offset to be the same using the measured phase and the extracted phase; it will change considerably.”
As previously mentioned, the advantage of using minimum phase files is you can model accurate summed responses at different angles and distances. Otherwise you’re limited to the axis and summation point of the measurement.
Dcibel, no worries, I wanted to see what the single measurement would look like, but would be using the 3 measurement method for now, though would like to understand the single measurement process a bit better.
Just wanted to add something to this thread, as I am just finishing up a big tower project with the woofer at floor level.
Taking a measurement at about 4ft back at tweeter level, I compared the response of what you get when you gate out all reflections, vs leaving the floor reflection in and gating to the second reflection. You can see here that the difference in response from 100-500Hz is quite significant, which is why you can reduce baffle step compensation from the normal 6dB to more like 0-2dB when the woofer is at floor level.
Comparing to the room response (ungated), you can see as well that the gating out all the reflections would be missing a significant factor, as we are relying on the floor re-enforcement to always be there.
So how do you think a spliced near field/far field response would look compared to the un-gated?
That's where it gets a bit complicated. The nearfield response won't include the floor reflection, so if you apply a baffle diffraction sim to the nearfield response like you normally would, you are generating a response that doesn't include that re-enforcement. For my simulation for design, I ended up splicing a nearfield response that didn't have the baffle diffraction applied at all. The measurement above is a real-world measurement of that crossover design. The overall trend of the in-room response is rather flat.
The in room measurement looks like no BSC is required.
Is the floor bounce/null identifiable above?
Yes, without BSC applied to the woofer the real world response is rather "flat". If you had applied BSC, the sound would be very bloated as I'm sure 4thtry can attest. My personal preference would probably be about 2dB of compensation.
I don't understand your question about the floor reflection being "identifiable".
Theoretically the floor bounce results in a null, which would be a dip in the response. I say theoretically as I haven't observed it.
Not sure if there would be a dip or it's low enough in the FR that the room dominates. Is one of the dips in the actual response the floor bounce?
My question can also be totally off base....
There certainly is a null if the driver is high up off the floor, but since it is so close to the reflection source, the null occurs much higher up in frequency, and in the frequency range of the woofer the reflected wave is mostly additive, not subtractive, which is why the need for BSC is so much less.
Have a look at the red line above, additive behaviour from the floor from 500Hz and less, the first null occuring at 550Hz.
Ani, I think you need to clearly distinguish between "floor bounce" and "floor boundary reinforcement". The floor boundary reinforcement effect can be viewed as an 0-6dB inverse of the baffle step, extending over a wide range of frequencies in the 80 to 500Hz region. Floor bounce would simply be a dip in your measurement, somewhere in the 100-200Hz area, based on the microphone tip position with respect to the floor and woofer cone. As dcibel points out, since the floor and woofer are very close together, this affects the frequency of the "floor bounce."