Nice concept. As mentioned if you can do this. Get your enclosure as far from the walls as you can. You will have a much lower accurate knee for your response. In a room this large you seriously can do a groundplane measurement to nearly 30 hertz. Literally lying the loudspeaker on it's side in the rough center of the room. Place your mic 2 metres from your enclosure flush on the floor pointing towards your enclosure to get a 1 metre equivalent measurement. A ground plane measurement can be interchangeable with a true anechoic measurement. Ground plane method paper attached.
Mark
The biggest issue I have with all these kind of averaging techniques, is that you assume things go on a predictable way.
While in reality, you don't know if that is the case.
Or there is always the trap of accidentally averaging issues that shouldn't be averaged (aka smoothed out)
Also from a practical side, it's quite a lot of work, while the stitching method (near-field + far-field), in combination with a ground plane measurements as well as some other measurements will give you the same information.
The only difference is that it won't be in a nice fancy all-in-one presentable graph.
Which I personally don't find all that important for speaker design, building and development.
In reality there are only a few systems where this doesn't work for, and only in some specific cases.
In a huge space like that going down to about 80-100Hz isn't that difficult with a proper time-window.
Even a little lower when we allow a little bit of wiggle.
Below 100-120Hz everything is fairly predictable from all the simulation models for most systems.
While in reality, you don't know if that is the case.
Or there is always the trap of accidentally averaging issues that shouldn't be averaged (aka smoothed out)
Also from a practical side, it's quite a lot of work, while the stitching method (near-field + far-field), in combination with a ground plane measurements as well as some other measurements will give you the same information.
The only difference is that it won't be in a nice fancy all-in-one presentable graph.
Which I personally don't find all that important for speaker design, building and development.
In reality there are only a few systems where this doesn't work for, and only in some specific cases.
In a huge space like that going down to about 80-100Hz isn't that difficult with a proper time-window.
Even a little lower when we allow a little bit of wiggle.
Below 100-120Hz everything is fairly predictable from all the simulation models for most systems.
Hello @Dkalsi
I can certainly try and assist, and please feel free to ask if anything I say should be unclear. I'll go over the steps that you asked about, however, I should state that it's much easier to make the measurements in REW and vector average them in REW.
Vector Averaging makes this much easier. One very important thing to remember is that when performing the measurements, the microphone must stay on the same axis relative to the speaker for all measurements.
Another note on this method: since the intent to is to coherently sum the direct sound of the speaker but incoherently sum the reflected energy so that it turns into random noise, almost any way that you can imagine to accomplish that seems to work. So for example, I have found that varying the height of the speaker and microphone works better than making more measurements at a greater distances, as well as doing this outside since there's only one reflective surface to deal with. To that end, using a directional microphone is also helpful. Here are two measurements taken outside, one with an uncalibrated Behringer B-5 cardioid microphone pointed straight up compared to a calibrated omni making a ground plane measurement.
If any of that needs clarification, please feel free to ask.
I can certainly try and assist, and please feel free to ask if anything I say should be unclear. I'll go over the steps that you asked about, however, I should state that it's much easier to make the measurements in REW and vector average them in REW.
Correct.
- You can, but I don't believe that level of precision is necessary (anyone more versed in signal theory can correct me).
- I haven't tried that, but for the time being, let's go with "no" for that.
- You can, but if you do, remember to reduce their respective levels in Audacity so as not to hard clip the results of step 4.
Correct.
Yes, and make sure that you're not clipping the new track.
Yes.
Yes.
If it all works correctly, you shouldn't need to apply any gating.
You should see the measurement of your speaker with minimal reflections. So if everything worked right, this method, ground plane, and NFS should all look the same... in a perfect world anyway.
Yep.
Vector Averaging makes this much easier. One very important thing to remember is that when performing the measurements, the microphone must stay on the same axis relative to the speaker for all measurements.Another note on this method: since the intent to is to coherently sum the direct sound of the speaker but incoherently sum the reflected energy so that it turns into random noise, almost any way that you can imagine to accomplish that seems to work. So for example, I have found that varying the height of the speaker and microphone works better than making more measurements at a greater distances, as well as doing this outside since there's only one reflective surface to deal with. To that end, using a directional microphone is also helpful. Here are two measurements taken outside, one with an uncalibrated Behringer B-5 cardioid microphone pointed straight up compared to a calibrated omni making a ground plane measurement.
If any of that needs clarification, please feel free to ask.
I have proved it out. Averaging versus groundplane. Where we have winter, like where I live near Ottawa, there is a good 5 months that going outside to do a ground plane will drastically mess up the mechanical parameters of the drivers. Kind of like walking slowly into cold water........
Mark
Why would you go outside to do ground plane measurements?
The whole point is that they can be also done also inside, especially in combination with the stitching method.
Sure, outside is better, but that also counts for regular time window measurements.
Well unless your space is super limited, but that is obvious I think in all cases.
That is not even easy with averaging methods.
But yeah, fair enough all methods do have their pros and cons this way
Wow - thank you all for responding.
What is not obvious from the picture included is that the ceiling above the DUT is substantially higher than the ceiling above me - and the back wall was not too much further behind me either. This was the best location I could go with to maximize reflection-free time.
I do regret missing the opportunity to take ground-plane measurements however. I've used this method plenty in the past to obtain below 100hz measurement - but I guess I don't know it well enough to be using it above 100hz. My concern has always been whether the "reflected image" of the speaker has the impact of effectively doubling baffle height (if standing) or doubling baffle width (if the speaker is on its side). I don't know if that would impact below 1.0Khz measurements. One issue with this speaker is that the top hat (mid+tweeter) is separated from the bass cabinet (i.e. it sits on top), and due to its triangular geometry, it cannot be laid on its side (without toppling the top over). I guess I could have strapped it to the bass cabinet for the sake of taking measurements. Damn! I won't get a chance to take it back there anytime soon :-(.
@Kravchenko_Audio - thank you for the attached PDFs - will take a look
@aslepekis - I will try the REW-Vector averaging method as well. I will have to try the slide method first, but only because I will have to wait until my brother is available again to help me lift the speaker onto a lift (so we can test it outside while varying height). Quick question about the outside varying-height method - how many measurements would you say is enough (8-maybe? - I suppose the more the better, but just curious) - and how much of an increment (2" at a time?).
Is the delta above 800hz purely due to the use of different mics (and not because of the measurement methods employed)?
What is not obvious from the picture included is that the ceiling above the DUT is substantially higher than the ceiling above me - and the back wall was not too much further behind me either. This was the best location I could go with to maximize reflection-free time.
I do regret missing the opportunity to take ground-plane measurements however. I've used this method plenty in the past to obtain below 100hz measurement - but I guess I don't know it well enough to be using it above 100hz. My concern has always been whether the "reflected image" of the speaker has the impact of effectively doubling baffle height (if standing) or doubling baffle width (if the speaker is on its side). I don't know if that would impact below 1.0Khz measurements. One issue with this speaker is that the top hat (mid+tweeter) is separated from the bass cabinet (i.e. it sits on top), and due to its triangular geometry, it cannot be laid on its side (without toppling the top over). I guess I could have strapped it to the bass cabinet for the sake of taking measurements. Damn! I won't get a chance to take it back there anytime soon :-(.
@Kravchenko_Audio - thank you for the attached PDFs - will take a look
@aslepekis - I will try the REW-Vector averaging method as well. I will have to try the slide method first, but only because I will have to wait until my brother is available again to help me lift the speaker onto a lift (so we can test it outside while varying height). Quick question about the outside varying-height method - how many measurements would you say is enough (8-maybe? - I suppose the more the better, but just curious) - and how much of an increment (2" at a time?).
Is the delta above 800hz purely due to the use of different mics (and not because of the measurement methods employed)?
The best position would be in the length of the room and/or at least in the middle of the room.
It's all about maximizing the distance to any close walls.
So basically where the mic is at this point.
Don't forget that the back wall reflection also counts, as well as the back wall reflection behind the microphone.
Getting the speaker up as high as possible and move yourself plus your desk somewhere in the corner of the room or in this case on an angle compared to the speaker or in the doorway/entrance.
If you have a (crude) floor-plan, that would be a little easier to explain.
In some cases placing the speaker/source ani-parallel to the walls really helps as well.
Just out of curiosity, I took some measurements of an SB-Acoustics NXR17 midwoofer in s sealed cabinet - no crossover.
Typical bookshelf-size cabinet. I took measurements starting at a distance of 90cm - and went further away from the speaker by 10cm for each measurement thereafter - and took 12 measurements (to end at approximately 2 meters from the woofer - 90cm + 110cm = 200cm). Have not built a slide yet - just used a laser level and used a fixed mic height - so the mic was on the same horizontal and vertical axis for each measurement.
1) I then "Align SLP" - arbitrarily used 80db at 1Khz.
2) Took vector average
3) Took dB average
4) Took RMS Average
5) Measured 1cm from woofer dust cap
6) Measured 25cm from baffle
Psychoacoustics smoothing was applied for purpose of making it easier to read.
The thinking behind including 25cm from baffle measurement was to get a semi-near-field measurement where the ratio of direct to reflected sound was significant, and not too close (e.g. 5mm or 10mm from dust capt) and possibly still capture baffle-step/diffraction.
The 10db scale makes it seem like an artificially close result.
The RMS average seems to be a better match than the vector.
Typical bookshelf-size cabinet. I took measurements starting at a distance of 90cm - and went further away from the speaker by 10cm for each measurement thereafter - and took 12 measurements (to end at approximately 2 meters from the woofer - 90cm + 110cm = 200cm). Have not built a slide yet - just used a laser level and used a fixed mic height - so the mic was on the same horizontal and vertical axis for each measurement.
1) I then "Align SLP" - arbitrarily used 80db at 1Khz.
2) Took vector average
3) Took dB average
4) Took RMS Average
5) Measured 1cm from woofer dust cap
6) Measured 25cm from baffle
Psychoacoustics smoothing was applied for purpose of making it easier to read.
The thinking behind including 25cm from baffle measurement was to get a semi-near-field measurement where the ratio of direct to reflected sound was significant, and not too close (e.g. 5mm or 10mm from dust capt) and possibly still capture baffle-step/diffraction.
The 10db scale makes it seem like an artificially close result.
The RMS average seems to be a better match than the vector.
Attachments
@leadcoma - well, in what is included above, I have not yet merged the near field with the far field. I've simply taken a measurement of 25 cm from the baffle. That should capture some baffle loss, while still maintaining a higher direct sound-to-reflected sound ratio.
When I'm typically designing a two-way (simple geometry) bookshelf - I take
The problem arises when you are building a large speaker with a low crossover point - say somewhere between 200hz -800hz.
The larger the speaker, the:
As I became more interested in this hobby, I became more interested in the science behind it (despite my limited technical background) and less interested in the music itself.
I'm now one of those fools that, the only way for me to think a speaker sounds good is for me to also "see" (not hear) that it measures right. What can I say?
Utilimaltey, I'm hoping that the process suggested in this thread will provide me a second option (when ground-plane measurements may not be an easy option) to verify a speaker's performance in the 100hz - 1,000hz range.
When I'm typically designing a two-way (simple geometry) bookshelf - I take
- far-field measurement, then
- take near-field measurements, then
- adjust near-field measurements for predicted diffraction/baffle-loss using VituixCAD, then
- adjust the SPL of the near-field measurements based on a combination of SD of the driver + eyeball with SPL of the fair-field measurement.
- Usually when I adjust based on the SD of the driver (using VituixCAD merger) tool, I typically only have to adjust an additional 0.5db - 1.0db to align with the far-field measurement.
- I then merged the far-field and the (adjusted) near-field.
- Verify with gound-plane measurements, because its easy enough to haul a small bookshelf in and out of the house.
The problem arises when you are building a large speaker with a low crossover point - say somewhere between 200hz -800hz.
The larger the speaker, the:
- Heavier it is to move it in-and-out of the house
- The bulkier it is to move it in-and-out of the house
- For far-field measurement, you need to be further away from the speaker to capture how the enclosure impacts the driver's response
- The further you measure, the lower the gate-time (all other things being equal)
- Low the gate time, lower the resolution in the lower frequencies in the fair-field measurement
- Lower the low-frequency resolution - the more difficult it becomes to
- align SLP, and
- select merge frequency (how much do you want to rely on low-resolution fair-field vs. adjusted nearfield - which one is more reliable?)
As I became more interested in this hobby, I became more interested in the science behind it (despite my limited technical background) and less interested in the music itself.
I'm now one of those fools that, the only way for me to think a speaker sounds good is for me to also "see" (not hear) that it measures right. What can I say?
Utilimaltey, I'm hoping that the process suggested in this thread will provide me a second option (when ground-plane measurements may not be an easy option) to verify a speaker's performance in the 100hz - 1,000hz range.
8 sounds like a good number to me, but you may be able to get good results with less if you math your heights so that the arrival delays from the ground reflection are spaced out sufficiently... and you may keep a better relationship with your brother.
I haven't tested the method with height variation enough to give a comment on what increment is ideal, but I did experimentally find that having at least two of the measurement heights such that the reflection arrival times are devisable by 2 flatten out some of the response variations.A little from column A a little column B. The cardioid does have some rise in its response, but I think the baffle reflection in the ground plane is contributing to a change in the response shape. I've attached the measurement file set so anyone can analyze and make their own conclusions (it also includes a measurement with the omni in the same location as the upturned cardioid).
I applaud your experiments! So far, you and Mark K. are the only other people on this forum to experiment with this method, and I do agree that your RMS average looks like a better fit... but the "sag" between 70Hz and 3.5kHz in the RMS Vs. Vector is interesting to me, there's clearly a lot of the room showing up below a few hundred Hertz. Daniel Krol's original paper also shows more variation at low frequency in small rooms too since room modes aren't quite the same animal as higher frequency reflections.
Having good ground plane data would be nice to compare to also.
What I do is take about 20 measurements in rapid succession. I lat a tape measure on the floor, and move the mic stand in even increments. What seems to happen using the math functions is a removal of the echos. I have had good correlation with groundplane measurement down to 30 hertz this way. The only difference I have had is a broad suck out near where you would usually have a floor and ceiling bounce in-fill. The correlation is plenty close enough to do serious work.