Why are omni-directional microphones preferred to measure speakers? Why not directional?

[nui]

When doing indoor (and even outdoor measurements it seems) to get a quasi-anechoic response from a speaker, we use time gating to eliminate reflections. This technique, especially indoors, has a lower frequency limit.

Could we not use a directional microphone to attenuate first reflections to lower the limit further?
With ambisonic microphones (e.g. from miniDSP, or RODE) it seems we could even emulate microphones with relevant properties in a post processing step, simplifying everything.

This idea seems so simple, so why does it not work?

Disclaimer: I tried to search here and google this and found nothing of relevance. Perhaps I am using the wrong terms...

 

[Tangband]

1. Because omnidirectional microphones are usually more frequency linear and dont behave otherwise at different distances.

2. At only 1 meter or slightly longer distance in a normal room, you are past the critical distance , meaning the microphone gonna take up more reflective sound than direct sound - omnis behave the same regardless of distance and other microphones will give you false results. In a normal room , its impossible to attenuate first reflections without gating when doing measurements .

3. cardioids are distance sensitive and have proximity effects.

 

[napilopez]

When doing indoor (and even outdoor measurements it seems) to get a quasi-anechoic response from a speaker, we use time gating to eliminate reflections. This technique, especially indoors, has a lower frequency limit.

Could we not use a directional microphone to attenuate first reflections to lower the limit further?
With ambisonic microphones (e.g. from miniDSP, or RODE) it seems we could even emulate microphones with relevant properties in a post processing step, simplifying everything.

This idea seems so simple, so why does it not work?

Disclaimer: I tried to search here and google this and found nothing of relevance. Perhaps I am using the wrong terms...

So off the top of my head -- I know much less about microphones than speakers -- I don't think it would quite work the way you're imagining it. I see your logic, but having performed many a quasi-wnechoic measurement, it seems to me using directional microphones would just be more of a headache!

It only takes a small reflection to wreak havoc on the measurement of the quasi-anechoic response. The purpose of the using a time window is generally to completely, 100% remove any influence from your room's loudest reflections.

In theory a hyper-directional microphone might reduce reflections, but I doubt it would be strong enough attenuation that you'd actually be able to use a significantly wider time window.

Worse, to Tangband's points, the directional attenuation will be different at different frequencies and at different distances.

Consider that, like speakers, microphones become more omnidirectional at the lowest frequencies. That largely much negates the goal of extending the time window in quasi-anechoic measurements, since it's the lowest frequencies of the reflections that are hardest to gate out.

The other practical issue is that it also means you'd have to be extra super-precise about positioning the microphone. An omnidirectional mic like the Umik-1 isn't even truly omni-directional. If you're off by a degree or two it will show up in the upper octaves as a high end roll-off. This problem is exacerbated the shorter the measurement distance.

In short, I think all you'd really be doing with a directional microphone is introducing colored, frequency-uneven reflections to the data. Better to just eliminate the reflections completely by using a time window that cuts off before the first reflection ever hits the microphone.

 

[nui]

Are measurement differences at different distances for non-omnis mics because their frequency response is significantly different for different incident angles?

The other practical issue is that it also means you'd have to be extra super-precise about positioning the microphone.

This is where an ambisonic mic could help, given that more assumptions were true, in that we could aim at speakers after taking the raw measurements. I have already seen software that determines direction of sources automatically, so I was dreaming, that one could automate this process entirely.

Better to just eliminate the reflections completely by using a time window that cuts off before the first reflection ever hits the microphone.

I am only asking, because this does not seem to be a complete solution. It only works up to a lower limit.

 

[Blumlein 88]

Here is the directional pattern of a cardioid microphone vs frequency. Notice how it is nearly omni at lower frequencies. Cardioids usually have less even frequency response. They also have proximity effect where at a certain distance and closer the low end response begins to rise. The effect can slightly alter response up to 1 khz and more at lower frequencies. Usually it won't be important until 2 feet or closer, but that varies a bit with microphone design too.

 

[Blumlein 88]

What you are asking for is what the Klippel speaker measuring system does. It is using math to compensate for or eliminate the effect of reflections in the room in which it resides. It uses conventional time windowing above something like 2 khz. It does what they call sonic holography below that. Our own hearing can mostly ignore reflections for the first few milliseconds. So it is possible. I'd think it might be possible to use measurements from 3 points and manage to process out reflections. I don't know of such an affordable setup being available. Nor do I know how to make one.

 

[nui]

Yeah, such a cardioid shape looks awful for measurements.

What you are asking for is what the Klippel speaker measuring system does.

Except simpler and more affordable. Hopes and dreams

 

[abdo123]

Just as it is difficult to make a perfectly linear full range directional speaker i would assume it’s just as difficult to do so in a microphone.

 

[thewas]

Just as it is difficult to make a perfectly linear full range directional speaker i would assume it’s just as difficult to do so in a microphone.

Exactly, you would need more mics or even better an array for that, https://en.wikipedia.org/wiki/Acoustic_camera
Another way to get some directivity is using a pair of mics one in front of the other, although this is used typically to measure the sound intensity, https://www.bksv.com/-/media/literature/Product-Data/bp1880.ashx
I used both frequently at my previous job.

 

[nui]

An array of microphones would start to defeat the purpose I had in mind anyway.
And 2 mics in front of each eclipses my understanding of sound. Rhetorical question: How does that help with sound intensity? Please do not answer for me, as I am almost certain I would not get it.

 

[Jukka]

I was thinking this same question now and accidentally found this thread.

My rationale: if all the money put into big anechoid chambers were put into designing a directional measurement microphone instead, would it not yield any useful results?

It's a lot cheaper to build a microphone than an anechoid chamber at least.

 

[No. 5]

If it could be done, yes, it could yield useful results. Let's throw some numbers at it: I estimate that you would need a beam width of about +/- 15 degrees to avoid a typical floor or ceiling reflection at a 2 meter measuring distance, but the attenuation outside that beam width would need to be fast. Like from -6dB to -36dB over 10 degrees or less. And it would need to do that over eight or nine octaves without side lobes getting above -36dB (or better) relative to on-axis in order for the microphone's directivity to take the place of an IR window. And keep in mind that a device needs to be large relative to wavelength to have an effect on directivity, so for 100Hz, that's a 11.3' wavelength. I have thought about this before, and my conclusion at the time was that the "microphone" would have to be an array a dozen or more feet long. You could make a virtual array by using a single microphone and then processing multiple measurements together since the Device Under Test is sort of "static". Also you would still need one anechoic wall because even though it was highly directional, the reflection off whatever wall was behind the speaker being measured would still be a problem.

I'll do a shameless plug to a project I'm working on here that is similar in concept.

By taking multiple measurement points along an axis, the resulting impulse responses can be aligned and summed together, the desired signal will then be amplified and the undesired reflections will be minimized. I'm just using REW for the software and the hardware is just an ECM8000 and associated ancillaries. Below is a Behringer B2030P measured with this method (red), without this method under the same conditions (green), and one measured on Amir's Near Field Scanner (black) . There is no IR gating, smoothing, or splicing on any of the measurements... but there is some environmental noise on mine. There's obviously work still to be done, but the results are pretty good for an investment of $0.

 

[Jukka]

I've learned that directivity at low frequencies is very difficult and needs at least a large microphone or an array of them. Multiple mics can calculate directivity from phase. For example, miniDSP has an 16 mic array that spans 132 x 132 mm. Or a 4 mics, 160 x 180 mm. Not perfect for 20 Hz, but better than a single mic. If you can capture less reflected sound with the same measurement performance, why not? Even with uneven directivity, the end result is not going to be any worse, is it?

Another argument, windowing achieves the same result with better accuracy. Well, windowing is a trade off between accuracy and bass extension. I don't know if any software applies windowing relative to frequency, but it should, because setting windowing for low bass decreases accuracy in the highs.

To me this sounds like a viable subject for research, because you could use both. If windowing is less accurate for highs, but analog directivity is better there, it sounds like best of both worlds, so why not?

 

[No. 5]

I've learned that directivity at low frequencies is very difficult and needs at least a large microphone or an array of them. Multiple mics can calculate directivity from phase. For example, miniDSP has an 16 mic array that spans 132 x 132 mm. Or a 4 mics, 160 x 180 mm. Not perfect for 20 Hz, but better than a single mic. If you can capture less reflected sound with the same measurement performance, why not? Even with uneven directivity, the end result is not going to be any worse, is it?

If it's not worse, is it sufficiently better to be worth the effort? We know that we can already get a perfect measurement above 1kHz with an IR window, so a directional microphone would need to show itself effective below that. No need to go down to 20Hz, 100Hz to 200Hz should be sufficient since we can do a ground plane below that.

Let's take a look at how much attenuation you would want to have for it to be worthwhile. Here's a simulation of an almost perfect loudspeaker, the only flaw is a a resonance at 300Hz:

Any measurements would need to show this problem clearly in order for us to fix them. So lets see what an extremely simplified "room" (just a floor and ceiling) does to things:

So the question is, by how much do the reflections need to be attenuated for this directional microphone to be useful? How about -12dB, that should be easy to do:

Better, but we can't really see the issue this speaker has. Lets try -20dB:

Again an improvement, but I wouldn't call this speakers issue "clearly seen", ideally I'd want better. So here's -30dB:

I'd say that's about perfect. So the question is: can you get -20dB or better of directivity down to 200Hz from a device that 132mm square? If you can, that sounds like a useful endeavor.

I don't know if any software applies windowing relative to frequency, but it should, because setting windowing for low bass decreases accuracy in the highs.

Could you explain what you mean? I don't follow why a longer IR window would decrease accuracy at high frequency.

To me this sounds like a viable subject for research, because you could use both. If windowing is less accurate for highs, but analog directivity is better there, it sounds like best of both worlds, so why not?

As someone who is extremely pro DIY and hands on, I say this with nothing but positive intentions: see if you can make it work. Getting a high resolution look at what's going on in a loudspeaker between 200Hz and 1kHz isn't possible for a lot of designers, hobbyists, and reviewers. It would be useful to have a simple way to do it.

 

[Jukka]

Could you explain what you mean? I don't follow why a longer IR window would decrease accuracy at high frequency.

While doing measurements to my speakers and their individual components, I noticed that some of the HF parts changed "wildly" while changing the windowing. I cannot remember which mdat displays this behavior, but I took some screenshots from one measurement. Below is overlaid the same mid-field measurement three times with a different windowing setting. All contours are 1/24 oct smoothed.

The highest frequency overlaps nicely thanks to smoothing, but if you compare longest and shortest window, the long window is generally a lot zig-zaggier. Below is the same without smoothing.

Next is another measurement, near-field for treble with mixed smoothings. Without smoothing the orange contour would be a lot zig-zaggier, but the smaller window contour is quite smooth even without smoothing.

So either I'm doing something wrong, or longer/shorter window loses accuracy.

Anyway, if I had the resources (time) to conduct more research on this myself, I certainly would. I will need to remember this should the opportunity rise.

 

[No. 5]

Ahh, I see what you mean now.

What you are seeing is the effect of an acoustic reflection (or several reflections) summing with the direct sound from your speaker. What's happening is that the IR window is long enough for the first early reflections to be included in the measurement, and unless the reflective surface or object is small, the reflections that are causing the high frequencies to be squiggly are doing the same to the lows. In your first example, it's being caused by something with a pathlength delay of less than 0.008 seconds, so I would guess that it's probably the floor. The really tight squiggles could just be environmental noise.

One interesting thing to note as it relates to this thread is that above 10kHz there are hardly any squiggles because the tweeter is very directional that high in frequency, and even a 1/4" measurement microphone is a little directional that high too, so a lot less energy is spared out to be reflect back into the mic. So point being, high directivity is helpful for measuring in a reflective space.