Waterfalls

[macrae11]

Hi John

I wanted to take some time to read through and attempt to fully understand the papers that you presented however there are still some things that I'm not convinced of from your arguments. In your first article that you posted I don't see it directly referring to integration time. It does mention 50 ms, as that is the rough time where humans start to get out of the Haas effect. (Actually the Haas effect is around 30-40 ms) In attempting to model human hearing, they eliminated peaks that were closer than 50 ms, because they are stating that they aren't heard and there for aren't relevant. I do find it interesting that they don't keep the first detected peak, even though that could potentially be the peak that humans would “hear”.

Also they state that sounds with slow attacks are not represented well by this algorithm they created. Since almost all sounds that fall in the frequencies we're discussing have a slow attack time, I think this article is more or less irrelevant in this discussion.

The second and third articles seem to have less to do with our discussion albeit they had some interesting reading. They are quite long however, and I can't state that I explicitly understood every single paragraph that was written. If there is a specific section to support your case, then perhaps you could point it out to me.

In the third article it seems to me to state that temporal integration doesn't have anything to do with a delay of sensation at various frequencies but rather a variation of sensation based on the duration of tones.

According my understanding of your hypothesis of “gate time” there would be a noticeable delay of lower frequencies from higher frequencies. I'm just guess here, but if you're saying that there's a 50ms “gate time” at 40Hz, than it might be logical to assume that there is a 100ms gate time at 20Hz, and a 25ms gate time at 80Hz?

Using this assumption, I consider the proposition quite preposterous. From listening experience I can with great certainty say, there is no noticeable delay of lower frequencies as my understanding of gate time seems to imply.

If I am changing your words to mean something different than what you intended, please point out my flaws in reasoning.


In regards to the first part of your post, you claim that humans need to hear an entire cycle, or at least a half cycle of a waveform in order to identify it. To my understanding I don't believe this is true. I tried to find some research to this effect, but haven't found what I was looking for as of yet.

I do propose a real world situation though. If humans need to hear an entire cycle of a waveform would they be able to hear low frequencies in headphones. Obviously the size of waveforms are far to large to reside in the space between a headphone and an eardrum. Now I know your response to this, is that the waves would go through their pressure variances, from compression to rarefaction in that space without the full waveform being present at any one time. This still presents the problem, that the waves of lower frequencies would be heard later in reference to higher frequencies, resulting in a glissando effect with everything we heard. Hearing high frequencies and then having the low frequencies be heard later in time. In order to hear 20Hz there would have to be a delay of approximately 50ms causing a discreet delay. (I think this might be a more accurate measurement to use for “gate time” since a 20Hz wave has a wavelength of approximately 17 metres, which would take approximately 50ms to propogate.)

I don't believe there is such a delay although I am going to conduct my own listening tests to determine if I am correct or not.

As it is, I'm going to be on location all of next week, so I probably won't be able to respond to any of your responses, although I will try to read them, and then respond next weeknd.

Thanks for this interesting and entertaining debate. I am certainly learning some new things, and stretching my brain in ways that it hasn't been stretched since I was in school.

Cheers

Andrew

[terry j]

macrae11 wrote:

Hi John

In regards to the first part of your post, you claim that humans need to hear an entire cycle, or at least a half cycle of a waveform in order to identify it. To my understanding I don't believe this is true. I tried to find some research to this effect, but haven't found what I was looking for as of yet.

Andrew

Hi Andrew, in no way am I John!! so I too am interested in his response.

I would however be very surprised that we could, by definition, recognise the frequency of a signal before one complete cycle.

(mental construct here) If a signal starts at the traditional '0' point, and starts rising then it will rise to a certain level, (talking sine waves for ease here) start to fall towards the bottom of the wave (in picture terms). OK, perhaps this might be the earliest point that we can say we have sufficient information for us to predict what the remainder of the signal will be, which is of course an inverse of the preceding. In other words, perhaps it is possible after all to only need half a cycle to recognise the signal.

Maybe you are right? Huhh, talked myself around to admitting the possiblity you are correct in the space of a half baked post!

Will wait for John!

[JohnM]

Hello Andrew,

macrae11 wrote:

According my understanding of your hypothesis of “gate time” there would be a noticeable delay of lower frequencies from higher frequencies. I'm just guess here, but if you're saying that there's a 50ms “gate time” at 40Hz, than it might be logical to assume that there is a 100ms gate time at 20Hz, and a 25ms gate time at 80Hz?

Your understanding of that is incorrect, and attempting to extrapolate that to a simple frequency dependence magnifies the misunderstanding. Perception of sound is not an instantaneous process, it is affected by what is presented to the ear over a period. A demonstration of that is Temporal Masking, in which a loud sound can prevent us perceiving sounds that arrive up to 100ms after it (not surprising) but also sounds that arrived up to 20ms before it. That does not mean that we simply delay everything to see what might come next, but that our perception of what arrives is affected by all the arrivals within a time window. There is no perceived delay.

macrae11 wrote:

In regards to the first part of your post, you claim that humans need to hear an entire cycle, or at least a half cycle of a waveform in order to identify it. To my understanding I don't believe this is true. I tried to find some research to this effect, but haven't found what I was looking for as of yet.

Think about the spectrum of a short segment of a tone. How much of the tone do you think would be needed before the fundamental can be seen on a plot of the tone's spectrum? The inner ear acts somewhat like a spectrum analyser, with different parts of the spiral portion sensitive to different frequencies. Until enough of a tone has arrived at the ear there is no content at the fundamental to be detected, just some spectral content related to the evolving envelope of the sound. As if that was not problem enough, people seem divided between those who even perceive fundamentals at all and those who rely almost entirely on overtones (google "missing fundamentals") but in any case sufficient signal needs to be received to establish the spectral content for the auditory system's pattern matching to work on.

Meanwhile to return to the original question as to whether applying a filter to the direct sound at a low frequency means what we hear sounds correct or not, let's consider a very simple case of a 40Hz tone arriving at the ear accompanied by a single reflection from a wall a few feet away. For convenience let's have a path difference that corresponds to a quarter wave, 6.25ms, and allow the reflection to be as large as the direct signal. The signals arriving at the ear are then the direct sound from the speaker and a quarter wavelength delayed version. The sum of these two, using basic trig identities for sums of sines, is a sine wave at the original frequency with a 1/8th wavelength phase shift and an amplitude of sqrt(2) times the original, so as far as the listener is concerned the tone has been made louder. To get the tone to the level it would have had without the effect of the reflection we need to reduce the level of the original sound by 1/sqrt(2). In doing that the 1/8th wavelength phase shift remains, but the level is corrected and the listener is none the wiser.

In the more general case of an enclosed space there are only specific frequencies, the modal resonances, at which the multiple reflections from the room's surfaces generate a stable standing wave. The effect at those frequencies is to alter the perceived level of those tones according to the amplitude of the standing wave at a given location in the room. Altering the level of the original sound correspondingly gets us back to the level a tone would have had without the room's influence, with the proviso that nothing can be done for locations where the amplitude of the standing wave is zero and more generally it is inadvisable to boost the signal in locations where the standing wave amplitude is lower than the original signal.

[macrae11]

JohnM wrote:

Hello Andrew,

Your understanding of that is incorrect, and attempting to extrapolate that to a simple frequency dependence magnifies the misunderstanding. Perception of sound is not an instantaneous process, it is affected by what is presented to the ear over a period. A demonstration of that is Temporal Masking, in which a loud sound can prevent us perceiving sounds that arrive up to 100ms after it (not surprising) but also sounds that arrived up to 20ms before it. That does not mean that we simply delay everything to see what might come next, but that our perception of what arrives is affected by all the arrivals within a time window. There is no perceived delay.

John, I understand that sound perception is not instantaneous. But in your initial post speaking of gate time, implied that temporal masking was frequency dependent. eg

JohnM wrote:

When you couple that with the effective gate time of the ear at such frequencies, 50ms or so,

This statement you made, seems to me like you are saying that there is different times for the brain to process signal depending on the frequency. This would cause a delay relative to higher frequencies.

JohnM wrote:

Meanwhile to return to the original question as to whether applying a filter to the direct sound at a low frequency means what we hear sounds correct or not, let's consider a very simple case of a 40Hz tone arriving at the ear accompanied by a single reflection from a wall a few feet away. For convenience let's have a path difference that corresponds to a quarter wave, 6.25ms, and allow the reflection to be as large as the direct signal. The signals arriving at the ear are then the direct sound from the speaker and a quarter wavelength delayed version. The sum of these two, using basic trig identities for sums of sines, is a sine wave at the original frequency with a 1/8th wavelength phase shift and an amplitude of sqrt(2) times the original, so as far as the listener is concerned the tone has been made louder. To get the tone to the level it would have had without the effect of the reflection we need to reduce the level of the original sound by 1/sqrt(2). In doing that the 1/8th wavelength phase shift remains, but the level is corrected and the listener is none the wiser.

Well the first reflection wouldn't be the same amplitude as the direct signal.(or at least if it is the listener has more acoustic issues than a little EQ could ever fix!) The issue here again I don't think is just about amplitude, it's about time.

JohnM wrote:

In the more general case of an enclosed space there are only specific frequencies, the modal resonances, at which the multiple reflections from the room's surfaces generate a stable standing wave. The effect at those frequencies is to alter the perceived level of those tones according to the amplitude of the standing wave at a given location in the room. Altering the level of the original sound correspondingly gets us back to the level a tone would have had without the room's influence, with the proviso that nothing can be done for locations where the amplitude of the standing wave is zero and more generally it is inadvisable to boost the signal in locations where the standing wave amplitude is lower than the original signal.

Here's the real issue(at least the one I've been talking about) modal resonances causing standing waves. Not only do these standing waves cause amplitude differences(which could be fixed with an EQ) but they also cause time differences. Notes ring out differently than they would without the influence of the room, lasting longer than they are supposed to. I don't think EQ can fix this. EQ will only lower the starting point of the frequency fundamental, which will cause it to dip into the noise floor sooner. Thus giving a perceived "correct" duration to the note, but not actually fixing the problem.

[JohnM]

There are substantial frequency dependencies in the way we process sound. That does not mean the differences in how we process content at different frequencies would cause us to perceive a delay between low and high frequencies, it is simply part of how hearing works. We have no other way of perceiving, and frankly for the lowest frequencies how could there be one? How could any instrument determine the frequency of a sound before it has sufficient signal on which to base that determination? At low frequencies a wider time period is used to build our perception.

Altering the level of the reflection in the simple example only changes the amount of the signal boost.

Finally then, back where we started. Yes, modes affect the time domain. They have rates of attack and decay that depend on their bandwidth. They are accurately modelled as 2nd order systems. So are IIR filters. It is fundamentally wrong to say "EQ will only lower the starting point". The filter has a time domain response just as the mode does, both in the build-up of its attenuation as content at its centre frequency begins and the decay of that attenuation. In a properly configured biquad filter the filter's zeroes cancel the poles of the room mode, leaving the net effect of the poles of the filter itself, which decay faster than those of the mode. The mode's slow decay has been replaced by the faster decay of the filter. These topics are discussed in great depth in some papers, for example Meridian's "The Loudspeaker–Room Interface – Controlling Excitation of Room Modes" from AES 23rd International Conference, Copenhagen and "Modal Equalization by Temporal Shaping of Room Response" by Matti Karjalainen at the same conference.

[Wayne A. Pflughaupt]

In a properly configured biquad filter the filter's zeroes cancel the poles of the room mode, leaving the net effect of the poles of the filter itself, which decay faster than those of the mode. The mode's slow decay has been replaced by the faster decay of the filter.

That certainly goes along way towards explaining the faster near-term decay times I've seen with the waterfalls. I expect that at some point the mode's "residual" decay overtakes and swamps the filter's effect, which is why things don't look so good when comparing long-duration windows? (see my graphs in Post #109) (Of course, one could argue by that time the signal is so low it doesn't matter...)

Regards,
Wayne

[Blaser]

blaser wrote:

Hi Wayne,

I do so many measurements I am not sure if I can still find it. Nevertheless, I can do them again with a 600 ms window, no problem. I'll do it on Saturday.

Fortunately I didn't tell which Saturday
Been very busy. Will do it next Tuesday (I hope to have a replacement day off). Wayne has made me curious to find out what happens after 600 ms

[brucek]

Wayne P says in part:

.......From the little I’ve seen, they don’t really have much effect on time domain. I haven’t seen any comprehensive in-room testing...............

.......From what I’ve seen, the improvement waterfalls show for modal (or “time-domain”) EQ filtering has only been apparent in the short-duration 300 ms window. When the window is lengthened to 600 ms, any advantage modal filters showed over other equalizing techniques pretty much vanished..............

.......Some of my esteemed colleges here (and on other Forums) are believers in time-domain equalization, but they haven’t posted or otherwise offered any evidence to back it up.............

.......If you want to convince us otherwise, show us your in-room waterfalls (long window, please)...........

Well, I am a believer in the theory that EQ'ing operates in the time domain and removes modal resonance at the listening position, and so I finally found a few minutes on this rainy day to try and satisfy Waynes thirst for an example in real life. It was a bit rushed, but I think I succeeded.

I decided my main system would work fine since it has a few big modal resonances that REW takes care of quite nicely.

Since I have my new laptop with USB soundcard integrated with that main system I didn't mind testing it out.

I felt that the best way to show results would be to simply pick a single resonance out of my raw response and let REW recommend its filter(s) and I would enter them and then tweak a bit while watching only the waterfall plot to reduce the ringing out to zero. I would only work on a single resonance to avoid any confusion.

As I have observed in the past, the ringing was reduced in the time domain to basically zero.

I used a 550ms time window instead of Waynes request of 600ms, since the ringing was simply gone by 550ms in the raw response. If I look at longer time out to 1000ms, that doesn't change.

I used our standard graph limits of 45dB-105dB. Once a signal is below 45dB, it's gone.......




RAW FREQUENCY RESPONSE FROM MY LISTENING POSITION
I've circled the resonant peak I chose to work on. REW identifies it as a peak and recommended two filters. One at 55.7Hz and another around 80Hz that I entered but didn't modify. The only filter I modified was the 55.7Hz
Once I entered the REW recommendation, I played only by watching the waterfall. I moved the center frequency to 55.2Hz and the bandwidth by one notch and the gain by a couple db (if I remember correctly) until I was satisfied. I played until the ringing out was gone.
raw response.jpg



RAW WATERFALL RESPONSE
raw response waterfall1.jpg


RAW WATERFALL RESPONSE (with 45Hz-85Hz horizontal axis)
Just another view....
raw response waterfall1 45 to 85.jpg




RAW (green) AND FILTERED (red) FREQUENCY RESPONSE OF ONE PEAK
compare response after 2 filters.jpg


FILTERED WATERFALL RESPONSE
filtered response waterfall1.jpg



FILTERED WATERFALL RESPONSE (with 45Hz-85Hz horizontal axis)
filter response waterfall1 45 to 85.jpg



OVERLAY COMPARISON OF RAW PEAK AND FILTERED WATERFALL (with expanded axis)
A nifty view where you can you see that the green signal that used to ring out to 550ms is now gone.
special view overlay.jpg


I don't know what more I can do to prove it. The peak and the ringing is simply gone without too much effort.
I know this is only valid at my listening position, but it is valid.

I could now go and work on all the other peaks etc and hopefully reduce most of the time domain problems....

You do have to be careful when you start looking down below 45dB level, that you may see some new ringing signal in certain circumstances. This can be caused by low level noise in the room from furnaces and fans and refrigerators etc.

brucek

[Blaser]

Nice job Brucek, this is quite in line with what I found in my experiment as well.
Nevertheless your experiment would be more illustrative if the SPL at 55.7 Hz is the same at 0 ms for both equalized and unequalized measurements. Pls take 2 measurements (equalized and unequalized) and overlay waterfalls with and without eq. so that they start at 55.7 Hz at the same SPL. This will give us a direct idea of the effect of equalization on the room mode regardless of SPL.

I couldn't do my measurements today, I hope I will do them soon.

[brucek]

Pls take 2 measurements (equalized and unequalized) and overlay waterfalls with and without eq

hehehe, no I won't be doing any more work on proving this theory. I've beat it to death.

I've proved it first with the theoretical lab experiment and Wayne said it had to be real world to convince him.

So, I used my real world HT with a real modal room peak and added a filter to bring it down to the real level that I would use and the ringing is 100% gone.

Enough is enough............

brucek

[Blaser]

I can guess what state of mind you're in after that very long debate... LOL

I'll do some more measurements and post them later on to definitely close this issue as far as the 600 ms is concerned.

[penngray]

I don't know what more I can do to prove it. The peak and the ringing is simply gone without too much effort.
I know this is only valid at my listening position, but it is valid.

I could now go and work on all the other peaks etc and hopefully reduce most of the time domain problems....

You do have to be careful when you start looking down below 45dB level, that you may see some new ringing signal in certain circumstances. This can be caused by low level noise in the room from furnaces and fans and refrigerators etc.

This thread is a great!!

But I have a dumb question....you say that filtering removed your problem but doesnt it only remove it at at SPL level, if you increase the SPL wouldnt the time domain problems re-appear?

[brucek]

The filter removed a modal peak down to the level of the rest of the non peak area. Makes no difference after that if you turn up the wholesale level. It all moves up........

brucek

[penngray]

It all moves up........

I know it all moves up but isnt there there still something in the 400 ms range down below 48 dBs that will move up too, creating something audible?

Im just wondering how this correlates back to room treatments and if filtering replaces treatments at all?

[Blaser]

Brucek,

I knew this would be brought out... This is why I asked about level matching yesterday

[thewire]

penngray wrote:

I know it all moves up but isnt there there still something in the 400 ms range down below 48 dBs that will move up too, creating something audible?

Im just wondering how this correlates back to room treatments and if filtering replaces treatments at all?

No because the human hearing is based on what it chooses to focus on. For example if you are in a room of people all talking at different levels, you can focus your attention on what the person in front of you is saying. If our hearing to into account every frequency from every direction and at every audible range we would be distracted as our hearing perception would go into hearing overload. If the ranges and frequencies are closer together, they become more difficult to discern from one another and become more like noise, less like what we define as human sound.

As for replacement room treatments with a BFD that is entirely up to the person with the room. Like putting a fake fish tank in your home with little floating plastic fish. It is entirely an opinion on which works best.

[thewire]

Here is how gnuware.com says it which is better than I could. I only know this stuff mostly from reading medical books and it a little more difficult to translate than bellow.

Chapter 2. Audio Fundamentals -

2.3.2. Temporal masking

In addition to auditory masking, which is dependent on the relationship between frequencies and their relative volumes, there is a second masking that comes into play, based on time rather than on frequency. The idea behind temporal masking is that humans also have trouble hearing distinct sounds that are close to one another in time. For example, if a loud sound and a quiet sound are played simultaneously, you would not be able to hear the quiet sound. If, however, there is sufficient delay between the two sounds, you will hear the second, quieter sound. The key to the success of temporal masking is in determining or quantifying the length of time between the two tones at which the second tone becomes audible, i.e., significant enough to keep it in the bitstream rather than throwing it away. This distance, or threshold, turns out to be around five milliseconds when working with pure tones, though it varies up and down in accordance with different audio passages.

This process also works in reverse; you may not hear a quiet tone if it comes directly before a louder one, so premasking and postmasking both occur and are accounted for in the algorithm.

[Wayne A. Pflughaupt]

penngray wrote:

I know it all moves up but isnt there there still something in the 400 ms range down below 48 dBs that will move up too, creating something audible?

I don't know about "creating something audible," but what's happening below the graph's lower limit certainly needs to be looked at. Stay tuned.


In order to show that a decrease in ringing has been accomplished, a waterfall graph needs to indicate that an improved rate of decay has been initiated. If not, all you've accomplished is merely a reduction in gain at the targeted frequency. Compared to a baseline graph, an improved rate of decay would be seen as increased spacing between the slices, indicating that the signal level is attenuating faster. This is what improved ringing (faster decay) looks like:

Note the significantly faster rate of decay above 140 Hz with the lower graph, which added bass traps to a room: At about 15 slices, the signal has dropped as much as 50 dB at some frequencies, in less than 200 ms, compared to the baseline which shows decay times at twice that rate or more. You simply can't get this kind of "action" with an equalizer. (Note, the reduction in decay time came with no electronic attenuation of the signal, as you would get with equalizer filters. Any decrease in signal peaks you see are merely the effect of absorption from the traps.)

That said, I'm basically satisfied with the explanation John gave in Post #130 as to how an equalizer can make at least some improvement in ringing with a room mode (emphasis added):

JohnM wrote:

In a properly configured biquad filter the filter's zeroes cancel the poles of the room mode, leaving the net effect of the poles of the filter itself, which decay faster than those of the mode. The mode's slow decay has been replaced by the faster decay of the filter.

So once again, an improvement in ringing will show a faster rate of decay, not just a reduction in gain. To put it in hopefully simpler terms for our non-technical readers: A room mode has a slow rate of decay, and canceling it with a suitable (i.e. matching) EQ filter that has little or no decay factor results in a faster rate of decay for the mode - i.e., reduced ringing. It makes sense, and it's readily apparent in the long-term 600 ms window comparing a base vs. equalized graph I presented in