Using two microphones to remove room modes from bass measurements

[dominikz]

I found this method very interesting so I decided to do a quick and dirty test of it with the tools I had on hand:

• Neumann KH120A loudspeaker - I find this method especially interesting for loudspeaker with non-removable grills such as this one, where you can't really measure the LF driver in the ultra-nearfield (and therefore can't avoid some room influence with only one microphone)
• Cross-Spectrum Labs calibrated Dayton EMM-6 microphone (used as the closer "Measurement microphone")
• Rode NT2A in omni mode (used as the further "Mode microphone")
• REW with multi-input capture license

Of course this test will be limited due to several reasons, main ones that I can identify:

1. The two microphones used to measure are significantly different in construction.
   However their responses are reasonably close (+/-1dB) in the relevant range (40Hz-1kHz):
   
2. The sensitivity of the two microphones is slightly different - though I aligned the responses with preamp gain to within about 0,5dB
3. The large body of the Rode NT2A will cause some undesired reflections in the higher midrange frequencies
4. The distances between the two microphones (and microphone and driver) were only set approximately with a tape measure, similarly the angle of the microphone array axis to the driver was surely not exactly 90°
5. Lastly I didn't apply the LP filter to remove the effect of the comb filter in the summed response

Setting the first microphone 5cm and the second 10cm from the driver, measuring and then using the REW alignment tool I was able to get the following response:

The black like is the summed response and we can see it is much smoother than either of the individual microphone responses - IMO showing that mode cancellation method works quite nicely. We also see the response doesn't contain the baffle step - which is as expected.

Note that responses arithmetically summed in REW don't contain harmonic distortion data of individual sweep measurements - which is something that a measurement done with the Mode Compensator would contain.

As a bonus, I also tried measuring with the first microphone ~15cm and the second ~30cm from the driver, this is what I got:

We see here the mode cancellation was less efficient this time, probably due to worse SNR at the increased distance from the driver.
Also, we see the shape of the response is different - IMO this is due to three reasons: 1) some baffle step may be present at this distance, 2) probably some of the port response is measured here as well (note that this is a front-ported speaker), and 3) comb filtering changes in frequency as we change the microphone distance (note that I'm not compensating for this in these measurements).

Hope this may be interesting to some!

 

[AudioChiemgau]

Well done!
Three remarks:
1) Also the phase frequency responses of the mics needs to match for best performance.
2) Without the comb filter frequency response measurements are limited to approximatively 100 Hz (compare blue shape of the first chart) for 5 cm distance between the mics.
3) Separating the two mics further results in the comb filter effect becoming already effective at lower frequencies, i.e. 30 Hz for 15 cm distance (compare the second chart). Your measurement fits well with the theory (compare the nulls). For larger mic distances one needs to adapt the inverse filter to get correct results again. As can be seen: Correction can be done up to approximately 500 Hz for 15 cm distance.

 

[dominikz]

Well done!
Three remarks:
1) Also the phase frequency responses of the mics needs to match for best performance.
2) Without the comb filter frequency response measurements are limited to approximatively 100 Hz (compare blue shape of the first chart) for 5 cm distance between the mics.
3) Separating the two mics further results in the comb filter effect becoming already effective at lower frequencies, i.e. 30 Hz for 15 cm distance (compare the second chart). Your measurement fits well with the theory (compare the nulls). For larger mic distances one needs to adapt the inverse filter to get correct results again. As can be seen: Correction can be done up to approximately 500 Hz for 15 cm distance.

Thanks a lot for the very helpful comments and for sharing the details of your work - much appreciated!

 

[SDX-LV]

I am now very confused about baffle step. So using Mode-Compensator there is no baffle step compensation needed because the measurement is not nearfield enough?
If I would measure a 6.5" driver without baffle or enclosure - what would baffle compensation be for Mode-Compensator or for nearfield measurement???
Looking forward to more tests by experienced DIY speaker measurement people.

 

[Naturlyd]

I made a quick try as well. The idea was to test if its possible to isolate the frequency response of a subwoofer in a living room and to compare the response from two different placements in the room. Session 1 was with the sub placed 110 cm above floor level, on top of a large speaker in the middle of the room. In session two the sub was measured placed on the floor and closer to one wall. All in all not very symmetric.

As I plan on building new speakers with active cardioid to cancel SBIR I hope this method is useful. When adjusting the system it would be nice to have some data, not just make semi educated guesses along the way. Or maybe just stay confused but on a higher level.

DUT is the Dynavoice MW10, a cheap active "subwoofer" bought for testing out backwave cancelling. The crossover is bypassed and the bassreflex port is plugged with foam.

The microphones had pretty similar phase and frequency response. One is a Behringer ECM 8000 and the other is a Mic from Studio 6 that resembles the Behringer. It had a 0.6dB lower sensitivity which was compensated for.

Pictures of the setups:




Here are the measurements, no smoothing. Sorry about poor readability of the comments :

This is with the results from the Audiochiemgau method offset to the about the same level as the 5cm distance measurements. The 10cm measurements are also included.

Edit: layout

 

[Naturlyd]

Here the measurements from session 1 with the sub 110cm above floor level and session 2 placed on the floor. No smoothing. Pretty close?

 

[Naturlyd]

Most of you have read the background for this method. And why you need to compensate for the effect of the comb filter above 100Hz. I wonder if I did this correctly.

First I traced the filter shown in red, its from the documentation:

Then I imported it into REW. You can see it being used i the posts earlier. Did I get it right?

 

[ctrl]

I am now very confused about baffle step. So using Mode-Compensator there is no baffle step compensation needed because the measurement is not nearfield enough?

First, with the Mode-Compensator-Method (MCM), you cannot measure complete speakers, only individual drivers.

Even for support in speaker development, the MCM is actually no real improvement. With a 5cm and 10cm microphone measurement distance, the distance is already large enough for the baffle-step effect to begin affecting the measured near-field frequency response (FR) of the driver. This is something one would ideally want to avoid, as it introduces additional errors when later applying baffle-step correction to the measured near-field frequency response of the driver in the cabinet.

Additionally, in bass reflex or similar concepts, the BR, TL port, or PR at 5-10cm dust cap distance can already influence the measured FR, depending on their location (another potential source of error).

This means that with a "true near-field measurement" of the driver with <=1cm, better results are achieved when using the measurement for speaker development.

Here's a small example to illustrate this and an estimate of the sound pressure level (SPL) errors to expect. A free-field simulation without reflections (as this requires less effort than measuring it in reality) of a sealed speaker (CB concept, width 20cm, height 30cm, depth 23cm) with a 6.5'' woofer is "measured" at <1cm, 5cm, 10cm, and 200cm distances.

Then, the SPL of the FR measurements are level adjusted:

The red curve shows the FR of the driver in the CB enclosure at a distance of 2m. This corresponds to the measurement of the speaker in an anechoic chamber or by Klippel NFS.
The "true near-field measurement" with <1cm mic distance to the dust cap of the driver (-1.5cm from baffle) is shown as the blue curve. The two measurements at 5cm and 10cm dust cap distance (3cm and 8cm from baffle) for the Mode-Compensator-Method (MCM) are shown in purple and green, respectively. The result generated by the MCM lies between these two curves.

For speaker development, not much can be done yet, as the baffle-step correction has not been carried out to get as close as possible to the red curve, the free-field frequency response of the driver in the cabinet.

With a diffraction tool (like the diffraction tool from VCAD - which I used here and is based on a 2D simulation of the baffle, so expect some error, since the simulation is based on full 3D like in the real world), one can create a simple baffle-step correction and apply it to the near-field measurements. The goal is to get as close to the free-field 2m measurement of the woofer in the cabinet (red curve) as possible with the baffle step corrected near field measurements:

It becomes apparent here that the "true near-field measurement" (<1cm from dust cap) with baffle-step correction provides better results than MCM. With <1cm measurement distance to the dust cap, in real measurements practically no room reflections are included in the FR measurement. So no need for more complex MCM measurements.
But since the merging of the baffle-step corrected near-field measurement with the gated far-field measurement usually occurs around 300Hz, the additionally error of the MCM is not too bad for a CB concept woofer (with BR, TL or PR the error might be higher).

The MCM with baffle-step correction also cannot be used up to 1kHz, as the error compared to the free-field measurement is simply too large - with 2-3dB error in the 500-1000Hz range (see purple and green curves of the MCM).

 

[ctrl]

If I would measure a 6.5" driver without baffle or enclosure - what would baffle compensation be for Mode-Compensator or for nearfield measurement???
Looking forward to more tests by experienced DIY speaker measurement people.

I haven't implemented any open baffle (OB) projects yet, but I've simulated several OB speakers.

If you measure a '6.5" driver without baffle or enclosure' in the near field, you're unlikely to succeed in generating a realistic baffle-step correction using standard software diffraction tools like VCAD to determine, for example, the free-field frequency response at a distance of 2m.
For such extreme cases, these tools are not suitable - at least based on my experience (which is solely based on simulations).

Either you simulate the free-field frequency response with a program like AKABAK, ABEC, Comsol,... for low frequencies and stich these simulated FR together with gated measurements or you try to obtain valid measurements down to frequencies where the driver exhibits the typical -6dB/oct behavior (for lower frequencies you can simply extend this behavior).

 

[Naturlyd]

It would be nice if @AudioChiemgau took the time to comment on the simulations @ctrl did in post 28.

@ctrl: Thanks for the simulation. Is the baffle step correction represented by the blue line in the graph similar (same accuracy) as the correction one do in Vcad?

 

[ctrl]

Is the baffle step correction represented by the blue line in the graph similar (same accuracy) as the correction one do in Vcad?

Thanks for pointing this out.
The diffraction tool used is from VCAD. I have amended the text in Post#28 accordingly.

So in the example we have a reasonably accurate 3D (BEM) simulation of the speaker. And then using a standard 2D based diffraction tool for the baffle step correction of the near-field "measurements" as it's standard in the real world when stitching near and far field measurements together.

 

[DavidR]

I haven't implemented any open baffle (OB) projects yet, but I've simulated several OB speakers.

If you measure a '6.5" driver without baffle or enclosure' in the near field, you're unlikely to succeed in generating a realistic baffle-step correction using standard software diffraction tools like VCAD to determine, for example, the free-field frequency response at a distance of 2m.

Bringing up an old thread. Doing so because I bought the Mode Compensator early last year, but only ran a quick test before setting it aside until now. I needed to set up my distortion test rig and ensure that the electronics were up to the task and then spent time learning how to use REW testing audio interfaces and amplifiers. Now I'm on to speakers.

I had no adequate distortion test hardware for speakers. I've got two older measurement systems, but the microphones aren't up to the task. The MCM provides two calibrated microphones via the preamp. The way that the MCM can elminate room resonances (and possibly other distance reflections) was intriguing. My old systems will do calibrated close-mic measurements that adjust the measured SPL for the driver diameter. One is calibrated, but I found that a 1mm position change of an NF measurement made a 1-2 dB change in the resultant SPL response. I considered that unacceptible for accuracy, so I fell back to manually fitting the NF with the FF which is itself wasn't absolutely accurate due to the limits of the MLS measurement, even on my large baffle.

I also plan to use the dual mic feature without the MCM combined output for some other testing. Even without that, the MCM can be used with a single mic for NF and FF measurement, but there's an oddity with single MEMS microphone on NF. More on that later.

I may start a new thread later as I want to start a discussion more than the two-mic method, but I will start here by saying that my testing results are similar to the sims here by ctrl. More than I would have thought which is actually disappointing as I also want to measure completed system drivers individually. The first set of REW measurements are for a Scan-Speak 15W/4531G on an 8" wide dipole baffle, my 3-way midrange driver. The smaller driver I think accentuates the issue with NF measurements, especially on a small baffle. Later I will run tests on a 10" Peerless driver in a closed box I have on hand.

I will eventually be measuring dozens of older drivers sitting on shelves. These will be mounted on my 2m x 2m test baffle, so the MCM will be useful in that it will eliminate all room modes and possibly the floor/ceiling bounce (short 7 foot basement). I'm also curious to see the result of the very large baffle exposed side edges since it's essentially a huge open baffle on the sides, although the rear has a lot of obstruction due to how it's positioned and braced. The MCM results will be combined with my normal FF measurements.

I realized that even though the MCM MEMS microphones are calibred to each specific input of the MCM preamp, the preamp outputs are not. My old calibrated measurement system can't be used to calibrate the MCM. So I went overboard and also bought a calibrated Umik2 thinking I could use that as an alternative to my older systems and easier than setting up the MCM for FF measurements. I calibrated the MCM outputs at its default gain setting for SPL against the Umik2 using REW and got very close agreement at 0.5m and subsequent NF measurements, enough so that I'm satisfied with that. I have a lot of REW captures of the Mode Compensator and Umik2 tests of the 15W.

 

[DavidR]

To start here are measurements of the Mode Compensator overlaid with the Umik2 that shows the close measured responses.

0.5m on-axis (Umik2 and only Mic1 of the MCM):

10cm to Umik2 and to only Mic1 of the MCM (the same was down for the combined M1/M2 MCM output):

I'll come back to the discrepency below 40Hz later.

 

[DavidR]

These measurements demonstrate how sensitive a near-field (close mic) measurement is to distance. They were made on the 15W mounted on an 8" wide dipole baffle. I knew that the absolute SPL changes dramatically with small movements, but I was not expecting the baffle diffraction to show its influence so early. Being a dipole the SPL starts to drop at higher frequency than closed box, but is more gradual as it will asymptote to 6dB/octave, although even the 0.5m measurement is a bit higher than 6dB. I think that is likely due to the driver rear output being limited, that is, restricted by the driver motor structure and the baffle opening, even though it's cut to open up the rear as much as is practical. Testing with the 10" driver I have will be interesting to compare.

I always used1/4" for NF measurements. It's the highest SPL below 300Hz as shown in the graph. Note that these were SPL aligned at 2500Hz 1/octave in REW. I selected that due to the relative linearity of all the measurements in that range. Note that these are not the MCM combined output, they were made with a single mic of the MCM. They would be similar to the Umik2 had I used it to measure.

This graph shows the same measurements, but they were SPL aligned at 160Hz 1/octave. I did this to show that a single mic measurement are similar at the low end, even at 1/4", due to the dipole nature. I'm a bit puzzled by this. The slopes are nearly identical, but it shows that the dipole influence becomes increasingly stronger with distance and not in a linear way. I'll be posting why I think that occurs later.

 

[Keith_W]

Just to be clear what we are looking at: these are quasi-anechoic measurements using the mode compensator to remove the effect of reflections, yes? Would you be able to share the MDAT? I am very curious to examine the impulse response a bit further.

 

[DavidR]

Just to be clear what we are looking at: these are quasi-anechoic measurements using the mode compensator to remove the effect of reflections, yes? Would you be able to share the MDAT? I am very curious to examine the impulse response a bit further.

These first graphs compare the Umik2 to a single mic (Mic1) that was calibrated against the Umik2. Mic1 and Mic2 are supposed to be equivalently calibrated using the preamp. I've confirmed that. I will post graphs later of the MCM combined output that is designed to eliminate room modes. I was a bit surprised, but the 15W evidently did not have enough signal to energize the room modes enough to be a factor until the mic distance was 0.5m. I will be testing a 10" closed box that is probably more appopriate for that aspect. The impulse responses of the Mic1 alone for these are probably not of much interest since it's almost the equivalent of the Umik2. I will be posting graphs of the combined MCM output.

 

[Chrispy]

TLDR but why would two mics "remove" anything particularly? Mask perhaps?

 

[Keith_W]

TLDR but why would two mics "remove" anything particularly? Mask perhaps?

See the Audio Chiemgau page that explains how the Mode Compensator works: here. And here is a test from AudioXPress that confirms that it works: here.

Essentially, two mics are placed a distance apart (about 5-10cm) on-axis from a woofer. When an impulse is played, the two mics capture the output, but the second mic records lower SPL than the first and with a slight delay due to distance. However, the noise floor and pattern of reflections for both mics is the same (because the distance between the mics is very small compared to the dimensions of the room) and can be subtracted out of the measurement.

There was an earlier thread somewhere on ASR where a member did a test with two microphones without using a mode compensator and used REW's trace arithmetic to remove the effect of noise and reflections.

 

[Chrispy]

See the Audio Chiemgau page that explains how the Mode Compensator works: here. And here is a test from AudioXPress that confirms that it works: here.

Essentially, two mics are placed a distance apart (about 5-10cm) on-axis from a woofer. When an impulse is played, the two mics capture the output, but the second mic records lower SPL than the first and with a slight delay due to distance. However, the noise floor and pattern of reflections for both mics is the same (because the distance between the mics is very small compared to the dimensions of the room) and can be subtracted out of the measurement.

There was an earlier thread somewhere on ASR where a member did a test with two microphones without using a mode compensator and used REW's trace arithmetic to remove the effect of noise and reflections.

Just seems more mic positions would be better in any case vs two fixed positions.

 

[Keith_W]

Just seems more mic positions would be better in any case vs two fixed positions.

Why do you think that? If two mics are sufficient to capture the information that you require, there is no need to increase the cost of taking the measurement with more mics.