[Nordenstam]

Originally Posted by DanDan

I really don't see how or why ETC devotees repeatedly put forward the notion that it is a superior or the only proper way to view the information. Both types of graph illustrated the floor bounce.
IMHO the frequency response graph showing the comb filtering is the most visually alarming.

Imagine that you're in front of a 100 channel mixing board. Each channel on the board is fed by a single sound sent through a short delay line. The original sound starts at the very left end of the mixer(or right if you're arab, jap etc ). The next channel have a 2ms delay simulating a work surface reflection. Next channel have a 4ms delay simulating side wall reflection. Next channel at 6ms delay simulating the front wall. Then there's 7 ms simulating ceiling, the side wall on the other side at 8ms, rear wall at 20ms, some twofold and threefold reflection in between there. And so on til the hundred channel mixer is filled up with tightly spaced delay lines. Add cross-feeding to make it all more interesting and like it is in real life. Room acoustics is all about changing the channel level fader for each of those reflections. Any and all such virtual rooms can be "treated" by setting the channel faders appropriately.

Now imagine that someone removes the input level displays(PFL) from the board, while also replacing the channel faders with endless rotaries that does not have any markings. There would be no way to get any idea of how the room sounds by looking at the levels of the channel faders. It would also be a problem using the channel level controls to achieve anything useful, depending on what you send into the board and what's being strapped on the master bus. Hooking a frequency response measurement tool across the master output of the mixing board isn't that useful. It doesn't give us information about each channel on the board and neither does it give direct feedback on what happens when one is changing the level on a particular track/room reflection. Observing the master VU meters isn't going to do much good either. Even rather large changes in the room will produce overall level changes so small that it'll hard to measure any change in the total level of all reflections. The only measurement tool that gives back what we had before loosing the channel faders and input level metering, is the time vs level view (ETC). Such a tool gives us an idea of the reflection pattern as is and it lets us observe the actual level change as seen when treating any particular reflection.

Have been playing with impulse responses in the 'puter lately. Although I do not have high thoughts for the frequency response graphs as such, it's interesting to try to see the correlation between individual reflections and the resulting overall frequency response changes. Virtual room treatment is easy to do by editing the level of particular reflections in a sound editor and then re-importing the impulse response into the acoustic measurement software. It's very handy to be able to test treatment changes without doing actual changes to the room. It's of course way simplified, but much better than not having the option! More on this is another thread.

Originally Posted by DanDan

However, no matter which view is chosen, it seems to me vital that perspective is applied.
e.g. Vicious looking comb dips are audibly insignificant if they are narrow enough.
When view controls are applied to lean toward the audibly significant, the 'problem'
appears relatively small. Certainly compared to modal issues.

The 1/3'rd octave corresponds to the ear/brains 24 critical bands. It's about masking for simultaneous sounds, like noise. It does not imply that the ear will smooth a slowly sweeping sinewave to the response seen in 1/3 octave graphs! Most people, the average listeners of psychoacustics, can discriminate a 5 cent note frequency difference. 5 cent is only a few Hertz at high frequencies and way less than that a low frequencies.


Looking at the level of each reflection makes a lot more sense if one wants to compare a particular room to what one may experience as a human listener. Back to this picture:


and the variable threshold argument in this old post.

Mapping the virtual mixer describe above, or an ETC, against the picture above is easy. There is NO such figure of merit for frequency response graphs. They do not describe the room in a manner that can be used to predict or build specific room responses. Neither do they show what happens when treating a particular reflection. The only way to have a totally flat frequency response is to be in an anechoic chamber.


It's easy to get a general idea for how this works for the ear. Load up some piece of music and put an echo FX in line. Adjust the delay time between 2 and 50 milliseconds while trying various dB levels at various delay times. Headphones may make this easier to hear. The echo should be very audible at levels above -10dB. Somewhere between -10 to -20 dB may be good enough. The amplitude change from a single reflection at -20dB is about 2dB RMS. Should still be within audible range. With some sound sources, even 20-30dB down may not be good enough! A reflection at -30dB still gives 0.5dB amplitude changes between dip and peak in the comb filter response. It all depends on the excitation sound used in the particular test. Short transient sounds are the most revealing. Generally speaking, below 15-20dB within the first 20ms or so should be rather transparent in most circumstances.

Often times subtle problems does not reveal themselves until some special track comes along. Then it's like WTF is THAT?? It's why I put the floor absorbers back after living without for a while. One day, one track, there was this obvious flanger when I moved my head back and forth. IIRC, there was some sort of synthetic marracas like sound that made it obvious. Putting the absorber at the floor suppresses the flanger. Measuring confirmed the before/after change in reflection level. That it happens to be at the lower edge of the general curve for perception is not totally surprising. We're supposed to be better than the average listeners!

Yet, it's not a big problem. By all means! One can live fine with the -20dB reflection. The point is that it's an example of measurement methodology that holds true for any and all situations.

Originally Posted by DanDan

Lupo,
I tested traps behind my speakers to see if they would alleviate SBIR.
No visible result. However, the test is worth repeating, in the light of these newly developed view manipulating skills!

Good luck! Have seen some improvement by treating the offending positions with rockwool. Not as much as I'd like in an ideal world, but it's been better than nothing. Though I haven't tested enough to say much about this.

Originally Posted by DanDan

I work around it. I test and move each speaker in all axes, to get the most even bass. I double check the combined response by RMS averaging and testing again with both driven. Basically I am trying, on a result basis, to use some of the problems to alleviate others.
Bass null in my new treated room, help! w/pics
Quite often I end up using a LF boost on the speaker, or moving the speakers closer to the front wall, occasionally almost touching it.

That's nearly as much art as it's science. Takes much hard work to get it right. Cudos to you for digging the trenches!

Have fallen in love with RPG's room optimizer software for this purpose. The starting points provided corresponds surprisingly well to the rectangular rooms I've tried it in, even if not all walls are perfectly reflecting objects. It's awesome to watch the program go through five digit numbers of virtual moving/re-measuring changes that would take weeks to do by hand. Especially when there's more than two speakers involved.

Originally Posted by DanDan

I have seen it mentioned elsewhere (by SAC I think) that SBIR is amenable to correction by Eq. An interesting and largely overlooked point I reckon.

Now that's interesting! Got more info?

SAC..?

Originally Posted by DanDan

I was referring to my original test. The one using a third octave band of pink noise at 125Hz. This yielded a different result. I don't think the lack of difference can be attributed to 'full bandwidth' measurement. The 85dB SPL was way ahead of background and was limited to one third of an octave.

Got a folder on this machine named "test and beep tones". There are now more than 1500 files in the various subfolders, all of them created to test various ideas. Many of which comes around during internet debates. A good percentage of those files are in a folder named "DanDan". I keep checking up on things, mostly to have a quality control and check that my ideas aren't totally off. Comes in handy when doing forum posts. Recommended!

I've already written why it's impossible to make the narrow band noise change large enough that it'll show up in the SPL meter. But I'll repeat your test and show you what sort of numbers it gives in total energy response. You've mentioned 1/3 octave noise, so the 1/3 octave (four notes) between B(123.48Hz) and D(146.84Hz) should be close enough. Synthesized such a noise (with slow roll off bandpass filtering) and ran the noise through the impulse responses from my measurements, giving the response with and without the rockwool on the floor. The RMS level difference in these no treatment vs rockwool measurements are 0.060dB in the left channel and 0.050dB in the right channel. Also made an extremly sharp filtered noise with hardly any energy outside the band, using FFT filters. Much sharper filtering than just about anyone is practically using. This measured 0.269dB and 0.225dB difference L/R respectively. Still outside the resolving abilities of most SPL meters and real world testing scenarios.

Using a sine wave should give the biggest change of them all. The exact frequency of the prominent dip is a bit hard to find, so I tried with various sine waves. With 137Hz sine wave, the before/after difference was 1.311dB and 1.201dB RMS L/R channel respectively. 138 Hz; 1.170 and 1.296 dB. At only half a Hertz away, 138.5Hz, change rolls sharply off to 0.613 and 0.715 dB. At 139 Hz, 0.055 and 0.019 dB. The 1.3dB change seen at the exact right frequency (plus minus 1 Hertz) is a change that may possibly be resolvable by the nice SPL meter you're using. So if you want to use the SPL meter as a figure of merit, make sure to use a sine wave centered on the exact frequency of the offending dip.

If we assume that we can hear a difference of 5 cents of tuning, the corresponding frequency change is to go from 137 Hz (C#2 minus 20 cents) to 137.4Hz (C#2 minus 15 cents). This is well within the actual width of the dip as measured using the sine wave test.

My room isn't your room, but the overall trends seen in this test should be about the same for your room. The reason why there's so little overall change is because it's only changing one of the 100 faders on the acoustics mixing board!


If you still think this is odd, I urge you to redo the tests you've done with the panel on the floor, except that you use the closest sidewall reflection as the figure of merit. You'll find the same lack of discriminating ability using the SPL meter, the same lack of change in the 1/3 octave frequency response graphs. The objections put forth against those measurement views holds true for any and all reflection measurements.