[Nordenstam]

Originally Posted by DanDan

I believe what we are seeing here is an illustration of how different measurement methods yield different results.

Almost. We used the same measurement method! This goes to show that the same measurements are up to different interpretations depending on the selected observation tool.

Originally Posted by DanDan

I note that with third octave smoothing there is a 1dB difference, while with no smoothing I can achieve something like your 10dB.

Excellent! You've repeated the tests with the same results. Thanks for confirming it.

Originally Posted by DanDan

Third octaves are no accident. Historically they were chosen for testing as they give results which are closely related to aurally detectable anomalies.

Add 0-20ms windowing to simulate the ear/bran integration time and it's getting closer to what the average Joe hear most of the time.

Yet, the sharp null is still there. The ear/brain hears about 24 frequency bands (and this corresponds somewhat with the 30 bands in 1/3 octave graphs across 10 octaves). What this means is that a noise or several overlapping sounds within a particular frequency band will sound more like the 1/3 octave indicates, due to the lack of discrimination within each of the 24 critical bands of the hearing system. The smoothing corresponds to that lack of discrimination of closely spaced simultaneous sounds. It doesn't mean that we only hear 24 different frequencies. The actual dips in the room are more like a 1/24 octave graph shows it to be. A slowly sweeping sine or low distortion instrument playing through single notes will clearly reveal that the dips are really sharp Q's by nature.

There are also a lot of other things happening at other frequencies, besides the big low end dips. The sharp low end dips are small parts of the total problem the reflection is causing. Also did the same tests using 200Hz to 20kHz sweeps. Thought it may come in handy to illustrate that it's not a sub 200Hz problem per se. It's not even worth posting the graphs. The only difference to be seen in the full bandwidth and limited bandwidth measurements is that the information below 200Hz is missing in the latter graphs. The rest looks the same in the FR graphs and the difference is also very very small in the ETC's. The sweep/impulse is linear, having equal amounts of energy across the spectrum. The sub 200Hz info only contains 1/120 linear part of the full 24000Hz spectrum! The RMS level difference in the after-treatment-files, including and excluding sub 200Hz info, is 0.007dB. Comb filtering is a broadband problem.

Originally Posted by DanDan

That is why I chose to use the third octave test in the first instance. (Plus, a bit of laziness) Note also the OP's 'broad dip centred on...'

I bet a cute cat picture there wasn't 1/3 octave smoothing on his measurement.

Originally Posted by DanDan

IMHO this illustration of the dangers in measurement and interpretation thereof has been fascinating. ..

However in this case, I believe the third octave viewpoint is closer to audibility.
In which case I will not be placing panels on the floor, or in front of my speakers.

Treating the floor may seem odd. That's a valid objection! We're so used to hearing a reflection from the ground. It's probably somewhat easier to ignore than other reflections. Speaker polar dispersion patterns also comes into play. It is however an early reflection like any other early reflections, with the same cause and consequences. Some solve the floor reflection issue by appropriate placement of a console, setting it to block the floor reflection.

My console is made as small and open as possible so as to not interfere with the sound field. Have a small mono speaker in the middle and the console is enough to block the floor reflection from that one. But it doesn't do anything to the sound from the mains so I've been using floor treatment. Have gone back and forth between having absorbers on the floor or not and ended up using it. It's not occupying any space I use anyway. Am actually employing those crappy pieces of foam for this purpose. Ghasp! Spread out a bit to cover more than the 40x40cm's used in this test, but still, it obviously doesn't do much to the low end. The reason for using foam is because of the stepping-on-it issue. Thanks for giving the incentive to do something about it! Will look into replacing it with some rockwool pieces. That said, the 6" foam does work well enough to alleviate the otherwise obvious flanger effect from the floor. It's easy to hear it when playing pink noise on a single speaker. Just did that right now, again. A small sanity check for my own sake! Moving the foam back and forth from the correct spot makes it very obvious that the flanger comes and goes. It can be more gone with a better absorber, but the foam is just good enough to take it from annoying to not particularly distractive.


The important thing to keep in mind here is that all of the different measurement viewpoints describe the same physical phenomena. All the views have a direct cause, effect and consequential change in the displayed measurement value. The measurements seen in this thread and the peculiarities of the different frequency response graph options holds true for measuring any reflections. Rotating the room 90 degrees sideways, making the floor reflection a sidewall reflection, or 180' to make it a ceiling reflection, would give the same aural sensation of a flanger or not, the same measurement results and the same viewpoint dilemmas when observing the change. This exercise exemplifies measurements for any early reflection point treatment.

Each measurement view comes with its own set of particularities and caveats. Let me once again expound on this topic. It's essential to understanding measurements! What looks like a gross anomaly on a 1/24 octave graph doesn't look half bad in 1/3 octave. The change is much less again when viewing the total deciBel level in the (full range, >10 octave) SPL meter or RMS measurements. All of those views are indirect responses to the underlying cause. The cause of the actual change that is really happening, physically in the room, is that the level of a particular (relatively loud) time delayed copy of a sound is changed(plus some incidental reflections being trapped as well). That's the only thing going on in real life. All the different measurement methods will do is to show this more or less directly. Some measurements are vague and indirect, some goes straight to the core of the actual physical change. The FR graphs are vague in the sense that they are looking at the comb filter pattern resulting from superpositioning of a vast number of reflections. With the typical default views spanning 500ms, hundreds of reflections are typically included in the measurement. Makes it rather iffy to see the direct cause->consequence of the energy contribution from a single reflection amongst the hundreds of them.

When I measure something, I always try to find the most appropriate measurement tool. Preferably one that shows the value I'm trying to measure directly. There are always many ways to go about viewing the same practical events and they all have their pros and cons. Ie, an oscilloscope shows the frequency of repeating waveforms by having a calibrated time grid on the screen, thereby making it possible to see how much time each cycle take and calculate the frequency using 1/cycletime. It is, however, much quicker to see the cycles per seconds displayed directly in a frequency counter. That oscilloscope example is 100 times easier than trying to reverse-engineer the comb filtering pattern resulting from the superpositioning of hundreds of reflections. That's of course why one typically concentrate on a single dip in the resulting frequency response and try to see the difference there. It can show that things are moving in the right direction, but it doesn't show how much the total energy level of the reflection is diminished. Concentrating on a specific part of the comb filter response also creates a danger of overlooking the change that happens in the rest of the spectra, as seen in this thread.


Here are some typical ways to measure the before/after reflection point treatment we've been doing:


Frequency response graph, no smoothing, 500ms window: shows about 15dB difference in the first comb filter dip, an impressivly sharp dip at 4500Hz approaching 30dB difference, plus a lot of other differences all over the frequency spectra. Definitely a broadband issue!


Frequency response graph, 1/3 octave smoothing, 500ms window: about 1dB difference in the largest response dips and some smaller differences in higher frequencies.

Same graph with 1/2 octave smoothing gives 0.5dB difference in the largest dip change, 0.4dB with 1/1 octave smoothing. I'd say the non-smoothed response is easier when it comes to determine if things are moving in the right direction. Anyhow, the FR graphs gives an indirect view of what's going on. What's really happening is still the same old change in level of a particular early reflection(plus some later stray reflections). The early reflection will cause comb filtering that will affect the whole frequency spectra. To evaluate the total effect, all the frequency response changes have to be taken into account. It's very hard, if not impossible, to guesstimate the total change in all of the peaks and dips that changes across the full spectra. So we may as well stick to full bandwidth measurements to see what the total change is during some time interval. The interesting thing is what time interval we chose to look at.

Have mentioned a 1/4 dB RMS difference in the before/after impulses, but that was wrong. It was measured on normalized files. Bad idea! It skews the relative levels. Without normalization, the before/after difference across the full spectra is 0.116dB. Even less than the 1/1 octave smoothing indicates, because this measurement is "smoothed to ~10/1 octaves"! About the same result would have been seen with an insanely sensitive SPL meter excited by perfect white noise. Those RMS levels was measured across the full 2.7 seconds long impulse response files. It includes all reflections in the room and a lot of noise floor. The level difference observed will grow as the observation window is targeted in on the time point when the actual physical difference occurs. By decreasing the RMS window to the first 50 milliseconds, difference increases from 0.116dB to 0.530dB RMS. Increasing to 0.835dB difference when measuring the first 10ms, 3.659dB with a window from 3ms to 6ms and a 5.662dB RMS difference when looking at the single millisecond from 4 to 5 ms. Recall that the actual reflection is around 4.4ms. Zooming in at the very peak of the highest reflection around 4.4ms, measuring the sample point peak level, shows a difference at that sample point of 10.4dB. The difference at the reflection maxima increases to 10.9dB when viewing the intersample peak level information, which is using an even finer time resolution than the sample point peak level meters. It's not possible to find an even smaller time slice that'll display a larger difference that that. It's the limit as the measurement time span approaches zero. These last measurement, covering a small slice of time around the particular reflection in question, is the ones that provides the most direct information of what's been achieved by treating that particular point in time. I wouldn't use peak level or intersample peak level information to evaluate acoustics though. It's too discriminating for the purpose! The reflection treatment affects a bigger area of time than the nanosecond information provided by digital peak level measurements. There's also the usual caveat that these impulse response measurements doesn't take both kinetic and potential energy into account, like the ETC's does. Which is part of the reason to use ETC's.

Here's the same before/after treatment as in the FR graphs above, seen in the ETC's:


Long time span view, the particular difference in question still obvious.


Short time span, difference very obvious!

The ETC view shows the difference before/after on the particular reflection in question. It's obvious what and how much is happening. It's WYSIWYG! No weird changes due to window and smoothing choices, just a level change displayed in dB. What less can one ask for?

I'll let the readers decide what they deem most easy and appropriate for measuring the effect of treating a particular reflection.


There's also of course the implication that the measurements can be used to find the offending reflections in the first place. I'm sorry if this comes across as another religious post. Think it's important enough to spend time on this. Am also still having geeky excitement about the subject! The excitement is not only because it puts a decibel level meter on each particular reflection point in the room, so to speak. It's first and foremost a door that opens up for the extremely interesting territory of pushing the room response towards some predetermined goal. It's very hard to get anywhere without knowing where one is going!

Found a very nice post in a Norwegian hifi-forum some days ago. A user (Tolsen ) had used the information provided here lately to write a small and easily understood post, using far less words than I'm able to. It described the why's and how's of ETC's in a manner that took a couple of minutes to read. Seems no one in there is having any problem understanding it. It really is both simple to learn and effective to use.

Originally Posted by DanDan

Neither will I live in vain hope of them combating SBIR.

Interesting. What do you do with the SBIR?

Originally Posted by Dange

Just doing a quick calculation I have the reflection, in theory, arriving at 7.8ms (343m/s over a distance of 2.68m) (2.68m from 2 *sqrt(60^2+120^2) )

Minus direct path 120cm.

Originally Posted by Dange

EDIT: I'm with you now (i think) the ETC time is relative to the direct path to the receiver? That would be around 4.40ms in that case

thumbsup

The impulses have been normalized both in level and time (peak at 0dB and 0ms).


...
10.30 pm saturday night.. Have sown this years kitchen garden, repotted some plants and mastered some nice electronic music inbetween writing this. But still, posting here, 10.30pm saturday night.. Hereby officially giving up on the "life" thing.


Have a great weekend everyone!


Best regards,

Andreas Nordenstam

Edit: now with pictures. Think I spent too much time from uploading the attachments to making the original post..