Waterfalls

[JohnM]

Doug Plumb wrote:

NASA or any other company does not ever use waterfall plots to evaluate the response of their systems and the response of their control systems is similar to what we have here. I have never seen a waterfall plot in any serious controls theory, electric circuit theory paper or in any other form. Its never used for a reason - its useless !! They are only in audio, they look pretty but mean little. A well known Phd acoustician has agreed with me on this point and doesn't use them for the same reason.

They can obfusgate the situation when people mis interpret noise or the mathematical distortions in gating as an actual physical affect, as you have above.

Every single peak in the waterfall plot is due to a peak in the frequency response. If there is a section in the waterfall that displays a decaying portion without a peak in the frequency response then its noise or some other affect.

In some non mimimum phase systems you could possibly see a decaying part that does not show as a resonance in the first slice but the resonance must appear in later slices. This is very unlikely and absolutely unlikely in a modal region room response.

That is overstated Doug. As an example, I'm sure you have seen a great many instances where what appeared to be a single peak in the frequency response is the result of two or more modal resonances, it is only in the waterfall or by gating later parts of the impulse response that this becomes readily apparent. Whilst a waterfall plot can be misinterpreted, so can just about any other data representation - not least the frequency response itself, whose appearance in this context is entirely dependent on the window positions and durations used to generate it. It is very easy to produce frequency responses of widely varying appearance from the same impulse response. Waterfalls and similar/related data visualisations which portray combined frequency and time domain behaviour are very widely used in some fields of analysis, e.g. seismic data in oil/gas/mineral exploration.

[Doug Plumb]

frequency response is the result of two or more modal resonances, it is only in the waterfall or by gating later parts of the impulse response that this becomes readily apparent.

I think that in many cases this is room noise that is also reinforced at certain frequencies by the dimensions of the room. Phase interactions between "modal noise" and signal may cause the waterfall to do weird things and be misinterpreted. I don't trust low level waterfall data because it can be corrupted by noise.

If you have a resonance and can't correct it with one filter then you need more than one, this becomes apparent only after one filter is used and you cannot ideally dampen the resonance. This occured in my room at about 140 Hz. There are three sharp resonances right around that area, easily identified with curve fitting.

Oil exploration, etc look almost exclusively at non minimum phase behavior under which conditions a waterfall plot may apply. I think that generally we should encourage users of our kind of software to look only at frequency response because the distortions associated with waterfall plots are usually poorly understood.

[clubfoot]

But isn't this why the room acoustics is so important? One should not have to apply more than one filter to correct a resonance at one frequency if the room is doing its job! My belief is if you have to apply multiple filters for a narrow frequency spike or bump, it's the room or a very poorly designed sub driver or enclosure "ringing". Granted it still has to be corrected for the system to sound "right" but not multiple filters.

My engineering degree is not in acoustics, so this is just my opinion.

[Blaser]

You cannot have a reasonable room treatment if say you have a problem at 30 Hz...

[Doug Plumb]

One should not have to apply more than one filter to correct a resonance at one frequency if the room is doing its job! My belief is if you have to apply multiple filters for a narrow frequency spike or bump, it's the room or a very poorly designed sub driver or enclosure "ringing". Granted it still has to be corrected for the system to sound "right" but not multiple filters.

My degree is in EE not acoustics as well - but even if you design a room you cannot guarentee not getting multiple modes at one frequency or a narrow band because rooms are never perfectly square nor built to precise dimensions. Higher order modes are dense & sensitive to small differences in planned and actual construction.

You don't need multiple filters to correct - you can just put a big notch in for low frequencies if you want.

[clubfoot]

Doug Plumb wrote:

My degree is in EE not acoustics as well - but even if you design a room you cannot guarentee not getting multiple modes at one frequency or a narrow band because rooms are never perfectly square nor built to precise dimensions. Higher order modes are dense & sensitive to small differences in planned and actual construction.

You don't need multiple filters to correct - you can just put a big notch in for low frequencies if you want.

Absolutely,...if the room is not rectangular (non divisible sides) and very well constructed then its influence on the sound becomes unpredictable and more drastic (read treatments) measures have to be applied to "dampen" resonances, or at least tame them to the point where something like a BFD can control them without degradation in sound quality.

I still don't believe that using a big notch filter would be a final answer,...temporarily, yes. But long term you're going to get tired of it and will want to address what is causing that big bump.

So this is where I agree with brucek and the use of waterfall plots because we are dealing with an analog device that behaves like a car's suspension or sub in this case (spring & shock absorber). Remember the big old "American" sedans and how their suspensions behaved,...excellent frequency response most of the time, but throw in a curve or mid corner bump and they became unsettled, and that's why I believe in this instance waterfall plots are useful.

[amrvf]

Hi,

I would like to understand whether to intervene:
I don't succeed in flattening the 153 Hz blade (DEQ2496)
(2 of my Ht satellites about 6 ft convergent, microphone to the center)



Red waterfall toslink input;
Green waterfall DEQ equalization + FBD - 6 dB@158Hz 1/60 oct
Yellow waterfall DEQ equalization + FBD - 30 dB@158Hz 1/60 oct
Orange waterfall DEQ equalization + FBD - 30 dB@158Hz X 3 1/60 oct.

where am I wrong?

thanks

EDIT amrvf:

excuse me,
soundcard not calibrated!!! (not loopback but microphone signal!!!)

[brucek]

I would say the peak is far too narrow to even hear.

Filters of 1/60th of an octave are far too narrow to be useful. Small microphone movements at that frequency would change the result.

Looks like the signal may be room noise.

brucek

[Chrisbee]

I'm back. Thanks for the earlier responses.

Here are some fresh waterfall plots of my 4 x 15" IB in a very non-parallel 27' attic with multilevel floors, an open stairway in the floor. Ceilings, floor and 45 degree walls are all boarded.

Listening position no BFD:


Listening position with BFD including +16dB boost @ 20Hz 120/60.


Nearfield with BFD.


Any useful thoughts?

[clubfoot]

All I can say is WOW

[Ivaols]

Brucek wrote earlier in this thread:

"As a side note, you can see what a completely terrible idea it is to add a gain filter to boost the level of a sub at low frequencies. You do nothing more than emulate a room mode at the gain frequency"

In Chrisbee`s second waterfall in the post above he has gained 20 hz with 16 db. I cant see that this has emulated a room mode at the gain frequency. The room mode seems to be the same as the higher frequencys.

[Chrisbee]

You may find this an interesting comparison: Measured at the listening position:



Here I have replaced my original 32Hz drivers with four brand new 16Hz drivers from AE.

BFD settings have been reset from the massive +16dB @ 20Hz boost required before.

I wanted to continue the new slope into the infrasonic to match my normal listening levels. Fletcher-Munson, Equal loudness etc.

A wide trough existed between the 12Hz room mode and ~25hz. I filled this with another quick and dirty boost filter at 20Hz..

Trial BFD Filters are:

20Hz + 5dB BW35.
25+2Hz -6dB BW20.
40Hz -4dB BW30
50+5Hz -3dB BW20

[JohnM]

Ivaols wrote:

Brucek wrote earlier in this thread:

"As a side note, you can see what a completely terrible idea it is to add a gain filter to boost the level of a sub at low frequencies. You do nothing more than emulate a room mode at the gain frequency"

In Chrisbee`s second waterfall in the post above he has gained 20 hz with 16 db. I cant see that this has emulated a room mode at the gain frequency. The room mode seems to be the same as the higher frequencys.

That is because the filter Chris used was very wide, spanning 2 octaves (BW 120/60). Modes have narrow bandwidths, typically 1/10th of an octave or less. Using narrow filters to boost the response creates the same extended ringing problems as room modes.

[DrWho]

In the first post...

Am I the only one that finds it odd that the decay rate of your BFD is over 100ms? I understand that some shape is expected for a bandwidth limited spectrum, but that long?

[brucek]

Am I the only one that finds it odd that the decay rate of your BFD is over 100ms?

I don't believe that it's the decay rate of the BFD at all.

A loopback cable (with the BFD removed from the circuit) will show exactly the same result.

Isn't the 100ms that you refer to a function of the duration of the impulse response analyzed?

For example, if connect a loopback from line-out to line-in of the soundcard and sweep a measurement to 200Hz, and increase the duration of the gate out to 1000ms (1Hz frequency resolution), I would get the following result.

waterfall 1000ms duration.jpg

But if I decrease the gate to 100msec (10Hz frequency resolution), I would get the following result.

waterfall 100ms duration.jpg

brucek

[DrWho]

Isn't the 100ms that you refer to a function of the duration of the impulse response analyzed?

An impulse, by definition, has a length of 0 (it's instantaneous). This is impossible in a bandwidth limited system, so one would expect to see something a little bit longer...but it shouldn't be longer than say 1ms - certainly not 100ms.

Are you saying that you only did the sweep up to 200Hz? If so, that might explain it. Why aren't you sweeping to the top limit of your soundcard?

Regardless, simply increasing the volume on the BFD is going to make it seem like it takes longer to decay. The fact that your EQ shows the same thing simply verifies that it is achieving an amplitude change...not so much changing the ringing at that frequency.

[brucek]

An impulse, by definition, has a length of 0 (it's instantaneous).

Yes, but REW doesn't use impulses to measure the response. The impulse response is derived from the system's transfer function, which is determined from the system's response to the sweep. In this case I am interested in an end frequency of 200Hz for the response and waterfall plots. An end frequency in REW is set to the highest frequency desired (in this case 200Hz), and the resulting sweep will span from 0Hz to twice the frequency set (with an overall limit of half the soundcard sample rate) to provide accurate measurement for the selected range (in this case 400Hz).

Sorry, I'm not sure what you're disagreeing with, or even what point you're trying to make?

brucek

[DrWho]

I don't believe that it's the decay rate of the BFD at all.

Exactly, so how do you reconcile the measurements to correlate with reality?

What I'm trying to point out is that the EQ does not ring or resonate or whatever you want to call it. It is not changing over time. You are misinterpreting an artifact of the waterfall calculation.

For a given point in the room where reflections arrive within a wavelength of the frequency in question, the system could be simplified as minimum phase. EQ only works in minimum phase situations.

However, you have not shown that the standing waves have been removed. In fact, they are most certainly still there. The easiest way to verify that the standing waves are still there is to locate the nulls of the standing waves...which do not move to different positions in the room or change frequency as EQ is applied to the signal.

I suppose you could just not sit in the nulls, but it's not quite so easy. The point I was making above is that you're only minimum phase for a single location in space. If you move the microphone even a few inches (like the width of your head), your EQ no longer works perfectly. Sure, some might argue that it is a small compromise, but the point is that it's not perfect. Proper acoustical treatment gets rid of the standing wave, which in turn improves the situation at far more locations in the room (it's not perfect either, but the compromises are much less).

One more comment which has to deal with psychoacoustics. When you walk into a room, there is a certain sonic characteristic to that room, even with no music playing. When you use EQ on the music to try and not trigger behaviors in the room, you end up sending conflicting signals to the listener who is partially expecting that bass guitar to sound like it would sound in their room. In other words, when you sit there listening to music, you are partially going to be identifying some attributes of the sound as being the room and not the music....if your music is precompensated for the room, and then your ears compensate what they hear from the room, then you end up with double compensation and the music sounds disconnected. Perception is definitely different for everyone, but the point is that it's much easier when the room isn't imparting crap on the sound because then there is never any amount of the listener trying to filter it out - it just sounds way more natural.

[brucek]

The point I was making above is that you're only minimum phase for a single location in space.

Yes, and I completely agree and have said the same thing in my post #13 that I'll reprint below:

The modal response of a room acts exactly like a 2nd order filter and matches the BFD generated filters in all aspects. At modal frequencies, a room resonates in gain and Q exactly as if you fed a sub signal through a 2nd order parametric filter. This fact allow us to fashion an identical 2nd order filter with the opposite gain and bandwidth that matches the room mode so it will completely disappear (at the point of measurement).

This doesn't apply outside the low frequency range where signals are no longer considered minimum phase, where primary reflections (second order) from the walls, ceiling and floor arrive at the listening position anywhere in the room with a phase shift of quite a bit less than a cycle. So, the effective limit here of about 80Hz-100Hz is reasonable for equalization in most rooms...(an 80HZ signal has a wavelength of about 14ft)............

Proper acoustical treatment gets rid of the standing wave, which in turn improves the situation at far more locations in the room (it's not perfect either, but the compromises are much less).

Mmmm OK, I don't know the size of the treatment that might be required though and the WAF of such treatments at the frequencies in question. Personally, I think EQ is an acceptable and effective route to follow. You and I have different opinions on that I guess. We'll leave it at that.

brucek

[DrWho]

How often does it show up where EQ predictions are made that do not result in the measured outcome? If the system were minimum phase, wouldn't the measurements correlate exactly with the predictions? Whenever this happens, one might argue that making the frequency response as flat as possible doesn't lead to the most accurate reproduction. Of course this notion is nothing new - there are articles dating back into the 70's.

There are also articles that discuss windowing limits for low frequency waterfalls too - it's a tradeoff between time and frequency resolution (can't have both at the same time). I'll see if I can't find that article too, but basically the data is meaningless without a proper window and obviously in this case, the wrong window is being used because it's showing decay rates that cannot be true.

[JohnM]

The main factors determining the shape and behaviour of the waterfall plot are the types and durations of the window functions used. The main window function is a right half window whose type is set by the Low Freq Decay window selection in the Analysis settings (default is Tukey 0.25) and whose duration is governed by the window control below the waterfall graph. The key to the observed behaviour of the waterfall is the window function applied to the left edge, which is not user-selectable. It is a left half Hann window whose duration is half the selected window duration (making for a total duration of 1.5 times the value in the control). This value was arrived at empirically by trying various settings to find a value that makes it easy to distinguish modal effects in measured data. I may make this a user-configurable parameter in a future build, but the current settings work well for the intended purpose of this plot.

[PeteD]

brucek wrote: