Room modes

[alaios]

Good morning all.

When we sit relatively close to speakers and not listen at loud volumes, do we excite all room modes?
I know this is a very weird question but I am trying to understand if room issues become prevalent when you play loud.

Regards,
Alex

 

[amirm]

Yes, you always do. Your seating position changes the modes that exists but otherwise, the low frequencies are omni-directional and will load up the room.

 

[dominikz]

I know this is a very weird question but I am trying to understand if room issues become prevalent when you play loud.

Room modes are excited no matter the level you play at.
The room resonances may be somewhat less audibly offending at low levels due to the fact they are located in low frequencies where humans have lower hearing sensitivity (check out "equal loudness contours"), but we're talking about levels where you barely perceive bass so that is definitely not a solution

To illustrate, here's how my Neumann KH120A measure at my desk at only ~70cm listening distance, and compared to anechoic "listening window" (LW) response:

You can clearly see the chaos the room introduces below ~1kHz (and especially below ~200Hz)
The wide dip between 60-100Hz as well at the sharp peak at 130Hz are both very audible at any listening level.

Luckily both can be corrected by integrating a subwoofer and using only 3 bands of PEQ, resulting in IMHO pretty amazing sound:

 

[Owl]

I was wondering why the general consensus is to not try and eliminate frequency nulls, but spikes or peaks can be effectively cut down some. Does it have anything to do with the DSP algorithm, some programs may be better than others, or is the result just horrible sounding.

 

[Emlin]

I was wondering why the general consensus is to not try and eliminate frequency nulls, but spikes or peaks can be effectively cut down some. Does it have anything to do with the DSP algorithm, some programs may be better than others, or is the result just horrible sounding.

Filling in the the nulls by simple increase in the volume doesn't work as you will increase the cancellation causing them proportionately, leaving you with no nett difference.

 

[boxerfan88]

Yeah, EQ can’t really fix nulls.

Me think the best way to fix null is to absorb offending frequencies so that the reflected signal that causes cancellation is lower. This of course is much easier said than done.

 

[DVDdoug]

Filling in the the nulls by simple increase in the volume doesn't work as you will increase the cancellation causing them proportionately, leaving you with no net difference.

Right. But you don't get 100% cancelation. You can make minor corrections but a +6dB boost requires 4 times the power (perhaps reasonable) and +12dB is 16 times the power (usually unreasonable). You can possibly do that at low volumes but at "normal volume" you'll overdrive your amp & speakers.

Me think the best way to fix null is to absorb offending frequencies so that the reflected signal that causes cancellation is lower. This of course is much easier said than done.

"Normal" sound absorption doesn't do much at bass frequencies. Bass traps can (partially) trap the reflected bass, smoothing-out the dips and the bumps.

 

[Emlin]

Right. But you don't get 100% cancelation. You can make minor corrections but a +6dB boost requires 4 times the power (perhaps reasonable) and +12dB is 16 times the power (usually unreasonable). You can possibly do that at low volumes but at "normal volume" you'll overdrive your amp & speakers.


"Normal" sound absorption doesn't do much at bass frequencies. Bass traps can (partially) trap the reflected bass, smoothing-out the dips and the bumps.

Proportionately, the cancellation will be the same, as I said.

 

[neRok]

Me think the best way to fix null is to absorb offending frequencies

The null could be speaker to wall, or listening position to wall, and thus the null can be affected by simply changing position. But doing so will also affect every other multiple of the relevant mode. And modes to do with the ceiling/floor aren't as flexible because often our chairs/desks/etc are fixed height.

 

[dominikz]

Proportionately, the cancellation will be the same, as I said.

Not necessarily, as @DVDdoug wrote, you'd rarely get a 100% cancellation, meaning that you can indeed usually retrieve some energy by boosting with EQ; here's one example from my own experiments (notice the null around 65Hz before and after EQ):

(Source and more details here)

From the same thread, this is how the corresponding EQ correction filter looks (green trace):

However, even if the null can be filled there are still issues with adding so much boost at a low frequency null (in this case approx. +10dB at approx. 65Hz):

• This will eat up a significant amount of amplifier headroom, which is especially problematic is the amp is anyway close to clipping (again in line with what @DVDdoug wrote)
• Some nulls may not be as pronounced in other listening positions, so boosting to get a flat MLP response can result in introducing a resonance at other listening positions

This is why I'd always first recommend to minimize nulls with positioning and use of subwoofer(s); and to only address peaks/resonances with EQ (i.e. only use PEQ cuts, and avoid using EQ boosts). Room treatment may also help, but I suspect that avoiding narrow nulls at low frequencies via absorption would not be trivial (or even doable) for most people.

 

[ZolaIII]

Filling in the the nulls by simple increase in the volume doesn't work as you will increase the cancellation causing them proportionately, leaving you with no nett difference.

Well good part of the nuls is actually phase mismatch or cancelling so correcting the phase of offending one with all pass filter actually resolves a lot. It's a EQ all right but one affecting only phase correction and not FR contrary to PEQ which influence both but FR mainly. Before anything else you need to address prominent peek causesd by room mode and preferably using all at your disposal (either a bit wider PEQ PK influencing both or accurate PEQ no smoothing and phases correction). Have a nice time.

 

[Justdafactsmaam]

I was wondering why the general consensus is to not try and eliminate frequency nulls, but spikes or peaks can be effectively cut down some. Does it have anything to do with the DSP algorithm, some programs may be better than others, or is the result just horrible sounding.

They can be cut down. The effectiveness is apparently a point of some debate. The peaks are the result of reinforcement from reflected sound which is smearing that signal in the time domain. So what you get is substantially attenuated direct signal with time delayed reinforcement.

It’s better than nothing but not as good as trapping the reflections and getting the direct sound at the right levels to begin with.

 

[HarmonicTHD]

I was wondering why the general consensus is to not try and eliminate frequency nulls, but spikes or peaks can be effectively cut down some. Does it have anything to do with the DSP algorithm, some programs may be better than others, or is the result just horrible sounding.

No. Nulls are created by cancelation of two sound waves simply speaking. Increasing the energy of each wave (boosting) will still lead to the cancellation with the disadvantage that you tax your system even more and run into danger of clipping (amp) or transducer distortion. As you see nulls in the order of -20dB, even if you would be able to fill them it would take 100times the power (watts). Therefore algorithms and best practice don’t do it.

However no rule without exception. I have seen some people carefully manually boosting (ca 3dB, still requires 2x the power) nulls and measuring some improvements. But I can’t tell how valid those are, as I personally don’t do it for the above reasons.

 

[dominikz]

They can be cut down. The effectiveness is apparently a point of some debate. The peaks are the result of reinforcement from reflected sound which is smearing that signal in the time domain. So what you get is substantially attenuated direct signal with time delayed reinforcement.

It’s better than nothing but not as good as trapping the reflections and getting the direct sound at the right levels to begin with.

Just to comment that ringing in the time domain is linked to the existence of a resonance in the frequency domain - something that Fourier transform teaches us.
This means that cutting the resonance with sufficiently precise filters will remove the temporal ringing as well - I guess one could say it's a nice example of having your cake and eating it too
Here's an example of this from dr. Toole's book (image source):

Note how in the upper picture where a sharp/high-resolution PEQ filter is used the temporal ringing is removed, whereas in the lower picture where a wide/low-resolution PEQ filter is used temporal ringing is only slightly attenuated.

In my understanding the difference between room treatment and EQ is mainly:

• Room treatment applies to all sounds produced in the room and not only those played by the audio system (which can be especially important to performance spaces but may be less relevant in small, normally-furnished residential rooms)
• Room treatment can change the ratio of direct to reflected sound energy and address flutter echo
• Carefully planned room treatment can retrieve some of the energy in the room mode and/or SBIR nulls (but this can usually also be done with carefully placed and integrated subs)

No. Nulls are created by cancelation of two sound waves simply speaking. Increasing the energy of each wave (boosting) will still lead to the cancellation with the disadvantage that you tax your system even more and run into danger of clipping (amp) or transducer distortion. As you see nulls in the order of -20dB, even if you would be able to fill them it would take 100times the power (watts). Therefore algorithms and best practice don’t do it.

However no rule without exception. I have seen some people carefully manually boosting (ca 3dB, still requires 2x the power) nulls and measuring some improvements. But I can’t tell how valid those are, as I personally don’t do it for the above reasons.

First let me say that I absolutely agree with you that boosting nulls should be avoided.
But I think it is important to add that it is IME often technically possible to do it; I've provided one example of this in post #10 (including an attempt of an explanation as to why it is generally considered a bad idea).
IMHO this is what might be causing this debate to sometimes come up - boosting nulls can work for a single listening position and low listening levels so some people see it as useful, but it has quite significant limitations for anything else (and might even cause damage) which is why it is generally discouraged and considered bad practice.

 

[Justdafactsmaam]

There is more to the story than that measurement tells us. It needs to be compared to a level matched measurement in an anechoec chamber to really tell us what is going on. Looks to me like the ringing is still there but along with the direct sound is just lower in amplitude.

 

[dominikz]

There is more to the story than that measurement tells us.

Can you clarify what is missing?

It needs to be compared to a level matched measurement in an anechoec chamber to really tell us what is going on.

How would you propose to do this evaluation, and what would you expect to determine from it? Resonances and ringing seem to be quite well understood from theoretical, practical and perceptual perspectives.

Looks to me like the ringing is still there but along with the direct sound is just lower in amplitude.

Note that this is an example with filter of only two resolutions: 1/3 octave (wide) and 1/20 octave (sharp) to illustrate the concept.
Notice that 1/3 octave filter only negligibly reduces temporal ringing while the more precise 1/20 octave filter almost fully eliminates it.

Question: What do you expect would be the result if we used an even more precise (e.g. 1/48 octave) filter?

The 1/3 octave filter in this case can't fully 'flatten' the frequency response, so some of the resonance is actually still there - just not visible at this very low resolution.
1/20 octave filter can better match the fine shape of the resonance and can therefore do a better job at 'flattening' it - but possibly still not perfect if the resonance is in reality sharper than the 1/20 octave resolution we used to measure it.

The example from dr. Toole's book in my previous post is IMHO a perfect explanation why sufficient filter precision is required to adequately remove resonances in frequency response and to eliminate the related temporal ringing.

 

[Justdafactsmaam]

What is mostly missing is a tell all measurement. A waterfall plot of a speaker/room with a clear resonance peak before and after EQ.

The result I expect would be that if the EQ flattens the response on the crest of the impulse that the resonant frequency will still take longer to decay than the adjacent frequencies. If the decay is relatively uniform then the EQ truly is the magic bullet.

The measurement you posted from Floyd Toole doesn’t tell us about that. There is no metric on the SPLs in those measurements which is a problem and the unEQed decay time is about 150 milliseconds. I am skeptical that this represents a real substantial resonant peak in the bass region. What bass resonances decay in 150 milliseconds without EQ?

 

[dominikz]

What is mostly missing is a tell all measurement. A waterfall plot of a speaker/room with a clear resonance peak before and after EQ.

Waterfall plot is still a result of Fourier transform (just sampled at specific timeframes and drawn in 3D) so it must show the same thing - it is a mathematical necessity. However, given that waterfall plots trade frequency resolution for time resolution (or vice versa) they can sometimes be misleading.
This is actually illustrated in dr. Toole's book as well - see these examples of the same measurement, presented in waterfall form with a few different resolution variants (link to source with text):

Notice how the plot under (b) has a more precise frequency response shape but makes it seem that resonances never stop ringing in the time domain, while the one under (d) doesn't show precise frequency response shape, but shows that time-domain ringing is not nearly as bad.
Note that all of these are showing exactly the same data - it is just a matter of presentation.

The result I expect would be that if the EQ flattens the response on the crest of the impulse that the resonant frequency will still take longer to decay than the adjacent frequencies. If the decay is relatively uniform then the EQ truly is the magic bullet.

If the frequency domain resonance is completely removed then the associated temporal ringing is removed as well.
Again, note that we can use Fourier transform to calculate the frequency response from the measured impulse response, and Fourier transform theory explains that resonances in frequency domain and ringing in the time domain are in fact the same thing - just that we can draw it in two different ways.

The measurement you posted from Floyd Toole doesn’t tell us about that. There is no metric on the SPLs in those measurements which is a problem and the unEQed decay time is about 150 milliseconds. I am skeptical that this represents a real substantial resonant peak in the bass region. What bass resonances decay in 150 milliseconds without EQ?

The plot looks fine to me - in the time domain response on the right (un-EQed) we can see there are approx. 5 cycles per 100ms which indicates a frequency of about 50Hz, which in turn corresponds to the peak of the resonance in the frequency response on the left.
150ms will fit 7,5 full cycles of a 50Hz tone (50Hz x 0,15s = 7,5 cycles).
The number of cycles it takes for the ringing to die down will be proportional to how distinct the resonance is in the frequency domain - which is exactly why we need corrective filters that are sharp enough to fix it.

 

[ernestcarl]

What is mostly missing is a tell all measurement. A waterfall plot of a speaker/room with a clear resonance peak before and after EQ.

Waterfall:


Spectral Decay:


Morlet CWT:


*Absent equalization, the peaky bass frequency response in a highly dampened room decays more evenly -- EQ works better as well.

 

[janbth]

Interesting thread!

I have been thinking about filling in dips in the frequency response. I agree that this is something one should usually avoid. However, I speculate that this could work under some specific conditions:

1. The null must be caused by the positioning of the speaker or sub - as opposed to the location of the listener.
2. The speaker/sub must not be located exactly in the null, but at some distance from it. This will ensure that the speaker/sub is «somewhat coupled» to the mode with the null.
3. There must be enough headroom.

Again, this is speculation, but might explain why people sometimes get good results from boosting. Also, I guess it would be a challenge to determine that a measured dip in the FR comes from the speaker-room coupling, and not from the listener-room coupling. Some kind of room averaged measurement like MMM, perhaps?