Help treating my listening room

[Matthew J Poes]

Here are how I deduced what I deduced from your system.


This is a group delay comparison. Not the best way to do this, but I compared it with the excess group delay plots and the overall trends are the same, and this allows me to overlay the graphs to make a point. Yours is the red graph, an ideal full range speaker is the purple, and green is one of my systems. As you can see, if there are subwoofers there is always some group delay compared to a full range speakers, same with the addition of EQ. I have plenty of EQ and crossover filters in my system as well, so you can see that whatever you have done is applying a lot more filtering and thus a lot more group delay.


Here is the TS or peak of the impulse at a given frequency. I noticed that your system had this peak in the TS graph. It suggests a lot of extra delay and it matches the group delay plot. I would look into what is causing that. It's probably a lot of boost near the crossover frequency. It could also be the crossover filter. Sometimes linear phase filters of very high order have very high group delay and I could see it causing a peak. If so, my suggestion is this, sometimes high order linear phase filters aren't a good thing.

Here is an example of some wavelets. You can see how useful these can be for analyzing a system. I wasn't a huge fan of these until someone on this forum suggested I use them. The reason I don't like them is that with a minimum phase system behaving as it should you typically can learn everything you need to know from the steady-state response. Nothing shows up in an FFT plot such as a waterfall or FFT spectrogram that isn't already contained within the steady-state at low frequencies. However, I have found that other interesting things show up in the wavelet that I wouldn't see any other way and I find that useful.


This is that full range speaker I mentioned. It's from a clients room and he has a low false acoustic ceiling. He has very few modal problems and as you can see his wavelet forms a near perfect christmas tree shape. Very few ridges and the peak line has a relatively smooth transition to that dogleg shape I mentioned. Ideally there would be less ridges and no holes in the direct response, but otherwise this is really good.


Here is an example from a system I measured that just had a few little issues. I just needed to make some adjustments to the DSP to get the graph looking a little smoother. The overall delay is a little high but not terrible, the goal, as mentioned, would ideally be to smooth that transition a bit. My guess is that most people's systems probably looks something like this.


This is your wavelet. As you can see, it is more similar to the one above than the first, but...it shows a huge amount of delay in the timing of the peak at LF's. This is what was concerning to me.

I hope that helps.

 

[Horacio Lewinski]

Matthew,

Can't thank you enough for taking the time to do all this. A lot to absorb, and I'm on it!

So the two takeaways I'm picking up:

1. Should lower the order of the crossover. How low would you suggest? 2nd-order? Current is 6th-order.
2. Should work out the time delays. I can introduce that in the active xo, using a negative delay.

I looked into the group delay article you mentioned and came up with this graph:

Between tweeter and midrange there is a passive xo at 4kHz. The active xo points are at 80 and 350Hz.

Excess group delay:
12ms @ 8kHz
39ms @ 300Hz (interestingly, the peak is not at 350Hz)
84ms @ 80Hz

Should I then take the peak at 8kHz as reference and delay the 350Hz xo by 39-12=27ms, and the 80Hz xo by 84-12=72ms?

BTW, how do you get the wavelet spectrogram to use frequency on the vertical axis and time on the horizontal?

 

[Horacio Lewinski]

Here are how I deduced what I deduced from your system.

I hope that helps.

Last night I realized I didn't hit "reply" on my previous post and hence you likely didn't receive a notification, so here I go replying

Was re-reading the minimum phase article on REW help and realized the graph I posted yesterday had 1/3 octave smoothing applied, and I understand 1/48 is better when looking at the low end. So 1/48 shown below.

Upon re-reading the article I'm thinking John was estimating the time delay between sub and mains from the average excess GD rather than peaks like I was referring to in my post above. For that purpose I'm guessing 1/6 smoothing is more practical:

And this correlates better with your statement of the approx 100ms delay of the sub, making me think I'm closer to the right track.
Excess Group Delay:
100ms @ 80Hz
45ms @ 315Hz
14ms @ 8.5kHz

So 350Hz xo should be advanced 45-14=31ms, sub should be advanced 100-14=86ms. Is this right?

 

[Matthew J Poes]

Last night I realized I didn't hit "reply" on my previous post and hence you likely didn't receive a notification, so here I go replying

Was re-reading the minimum phase article on REW help and realized the graph I posted yesterday had 1/3 octave smoothing applied, and I understand 1/48 is better when looking at the low end. So 1/48 shown below.
View attachment 8807

Upon re-reading the article I'm thinking John was estimating the time delay between sub and mains from the average excess GD rather than peaks like I was referring to in my post above. For that purpose I'm guessing 1/6 smoothing is more practical:
View attachment 8808

And this correlates better with your statement of the approx 100ms delay of the sub, making me think I'm closer to the right track.
Excess Group Delay:
100ms @ 80Hz
45ms @ 315Hz
14ms @ 8.5kHz

So 350Hz xo should be advanced 45-14=31ms, sub should be advanced 100-14=86ms. Is this right?

Well the excess group delay gives you the amount of extra delay you have at specific frequencies. While this is based on an in-room measurement and does account for the room, it isn't the same as looking at something like the wavelet spectrogram, which shows the amount of energy flowing into the system over time at a given frequency. I suspect this might be a better measure of how we hear, but that is a guess. None of this has much research behind it in terms of human perception (i.e. tying a wavelet shape against preference). We don't perceive a lot of group delay as bad necessarily if it's at low frequencies. However the common metrics I see mentioned are keeping it to less than 1 cycle (remember that the research into this never bothered to test group delay at frequencies below 100hz like this, let alone below 50hz). I'm using that as my metric, if you will, for what is ideal. I'm really using these graphs to give me clues into ideal integration and how to objectively measure this beyond the steady-state.

I think you can try your experiments with the delay and both remeasure and listen. Can you AB compare? I would do that, how it sounds matters most and group delay is a tricky topic. A perfectly flat group delay plot doesn't necessarily matter as much as a smooth response. If correcting one makes the other worse, that can be something to consider.

In any case, I think the amount of delay you have is a bit excessive and worth fixing, but I honestly was mostly concerned with the frequencies below 100hz. My guess is that the amount of delay you need to apply below 80hz is not 100ms but something closer to 20-30ms. I would start by removing about 20-30ms of extra group delay. The 100ms peak is not as big a deal. To come up with this number, I'm basically looking at the average. Plotting a flat line through the peaks, relative to 0. I'm also looking at the Wavelet and looking at what I think would bring it inline with what I've seen on systems I think sound good (and bring it down to less than 1.5 cycles of delay).

 

[Horacio Lewinski]

Well the excess group delay gives you the amount of extra delay you have at specific frequencies. While this is based on an in-room measurement and does account for the room, it isn't the same as looking at something like the wavelet spectrogram, which shows the amount of energy flowing into the system over time at a given frequency. I suspect this might be a better measure of how we hear, but that is a guess. None of this has much research behind it in terms of human perception (i.e. tying a wavelet shape against preference). We don't perceive a lot of group delay as bad necessarily if it's at low frequencies. However the common metrics I see mentioned are keeping it to less than 1 cycle (remember that the research into this never bothered to test group delay at frequencies below 100hz like this, let alone below 50hz). I'm using that as my metric, if you will, for what is ideal. I'm really using these graphs to give me clues into ideal integration and how to objectively measure this beyond the steady-state.

I think you can try your experiments with the delay and both remeasure and listen. Can you AB compare? I would do that, how it sounds matters most and group delay is a tricky topic. A perfectly flat group delay plot doesn't necessarily matter as much as a smooth response. If correcting one makes the other worse, that can be something to consider.

In any case, I think the amount of delay you have is a bit excessive and worth fixing, but I honestly was mostly concerned with the frequencies below 100hz. My guess is that the amount of delay you need to apply below 80hz is not 100ms but something closer to 20-30ms. I would start by removing about 20-30ms of extra group delay. The 100ms peak is not as big a deal. To come up with this number, I'm basically looking at the average. Plotting a flat line through the peaks, relative to 0. I'm also looking at the Wavelet and looking at what I think would bring it inline with what I've seen on systems I think sound good (and bring it down to less than 1.5 cycles of delay).

Thank you. Right now I have not introduced any delay purposefully, but obviously the system has it (My guess: my Hilo soundcard directly drives the midrange/treble and midbass amps, while for the subs the Hilo outputs thru S/PIDIF to a 2 channel DAC that drives the subs, so maybe that introduces delay).

Acourate allows me to introduce delay, but have never done it so will be learning how to do that and come back.

BTW, how do you set wavelet spectogram to have frequencies on vertical axis and time on horizontal axis? REW Help doesn't show.

Regards

 

[ddude003]

Set Spectrogram, Controls, Mode: Wavelet, Frequency axis: Y axis...

Looking forward to those photos of your DYI treatments... 8^)

If you treat your front wall the way you described you will have treated aprox 1/6th of your surface area... Do you intend to treat the rear wall too? And windows to the left and doors to the right...

PS - Some S/PIDIF controllers, like STmicro, have been known to add up to 35ms of delay if I am remembering correctly... Asymmetric signal paths... Your only as fast as your slowest component.. 8^( Care to share a bit more about your optimized computer running HQPlayer???

 

[Horacio Lewinski]

Set Spectrogram, Controls, Mode: Wavelet, Frequency axis: Y axis...

Well, maybe I'm blind...but I don't see that option:

[QUOTELooking forward to those photos of your DYI treatments... 8^)

If you treat your front wall the way you described you will have treated aprox 1/6th of your surface area... Do you intend to treat the rear wall too? And windows to the left and doors to the right...[/QUOTE]

I happen to have just finished the first two: perforated panel absorbers, where the front panel is 12mm thick, the airspace behind it is 145mm, holes are 4mm diameter spaced 100mm between centers, in square configuration. So perforated area is 0.13%, and if I'm interpretting correctly the design resonance frequency is 38Hz. The panel is 61 x 130 x 15cm external dimensions, so I can cover the wall with panels the same size but different design/functionality - a la BBC panels described by Everest. I plan to place these panels behind the subs, between subs and front wall, and also covering behind the B&W woofers.

I glued the sides and back, sealed, added 2" fiberglass panel, added something to keep it in place when vertical, sealed the top cover.
Need to measure and see if they make any difference.

PS - Some S/PIDIF controllers, like STmicro, have been known to add up to 35ms of delay if I am remembering correctly... Asymmetric signal paths... Your only as fast as your slowest component.. 8^( Care to share a bit more about your optimized computer running HQPlayer???

Intel S1200KPR, Xeon E3 1265lv2, Crucial V4 SSD for OS, 8GB ECC RAM, PPA USB card v2, wide range picoPSU, PPA SATA cable for SSD, Streacom FC8 Evo case, lab linear PSU for picoPSU, Windows Server 2012 R2, AudioPhil's Optimizer, Acourate (digital crossovers, room correction), HQPlayer. This is a headless PC.

To control the system I run Roon thru my phone. Roon server is on another computer, powerful but not as optimized, that also holds the music files and Tidal, and connects thru LAN to the HQPlayer PC.

 

[ddude003]

Well d.am.n it... MacOS here... Sorry about that...
This looks like a feature request for John Mulcahy... Request made in the REW forum...

It looks like you are off to a nice start with your DYI treatments... I am interested to see how these turn out in practice for you... Nice work... I think you have a very nice setup... It will sound delicious once you get your room squared away and get the kinks out of your signal processing kit...

I'm guessing that AudioPhil's Optimizer tears Windows a new one to achieve a near soft realtime audio focused OS...

 

[ddude003]

Horacio, the Frequency axis setting is included in the latest beta version for windows...

 

[Horacio Lewinski]

It looks like you are off to a nice start with your DYI treatments... I am interested to see how these turn out in practice for you... Nice work... I think you have a very nice setup... It will sound delicious once you get your room squared away and get the kinks out of your signal processing kit...

I'm guessing that AudioPhil's Optimizer tears Windows a new one to achieve a near soft realtime audio focused OS...

Hehe. At least it wasn't me being an idiot and missing the setting in front of me!

Yes, I have high expectations for the room treatments. But want to make sure they have the effect they should. Up until now I have not been successful.

AudioPhil's Optimizer is a great tool for computer playback. In fact my computer is a clone of AudioPhil's computer from 4 years ago. No unneccessary bells and whistles that one needs to turn off, no moving parts. Works great.

 

[Horacio Lewinski]

Horacio, the Frequency axis setting is included in the latest beta version for windows...

Thank you!! I thought I had the latest, but there you go.

 

[ddude003]

Yea, go figure... I had downloaded what I thought was the latest release onto my windoz box this afternoon and saw the same control box as you... Made a feature request in the REW forum only to find out the work had already been done and in the current beta... By the way John, Nice work and thank you for such an awesome audio tool...

PS - The Beta version works fine so far on my Windoz 10 box...

 

[Matthew J Poes]

Thank you!! I thought I had the latest, but there you go.

sorry I didn't respond to this. I've been on a speaking tour and only just got back.

 

[Matthew J Poes]

Yea, go figure... I had downloaded what I thought was the latest release onto my windoz box this afternoon and saw the same control box as you... Made a feature request in the REW forum only to find out the work had already been done and in the current beta... By the way John, Nice work and thank you for such an awesome audio tool...

PS - The Beta version works fine so far on my Windoz 10 box...

His addition of wavelet analysis and now the ability to change the axis is really great. I don’t really know why but I find it far more natural to interpret the wavelet with the Y axis as frequency and X axis as time. It makes spotting delay and reflections easier.

He has given us such a powerful tool.

 

[ddude003]

I have played with Time Series data most of my career and then some... And played with many ways to visualize it... From left, increasing in time, to right has always seemed the most understandable way to for me as well as many others...

 

[Eric SVL]

Nice system and great discussion. I can't claim to have the knowledge these guys have.

But I need to ask one thing: are you measuring with your mic facing forward as shown in your pic? And if so, are you using the 0 degree calibration file? Typically we measure at 90 degrees, with the mic pointed towards the ceiling.

Carry on

 

[Matthew J Poes]

Nice system and great discussion. I can't claim to have the knowledge these guys have.

But I need to ask one thing: are you measuring with your mic facing forward as shown in your pic? And if so, are you using the 0 degree calibration file? Typically we measure at 90 degrees, with the mic pointed towards the ceiling.

Carry on

Hi Eric,

Actually, that comment needs a bit of qualification. It has been traditional to measure by pointing the mic at the ceiling with the 90-degree calibration. However, I've learned something that has changed my opinion about the importance of this. The main reason for doing it is to capture more room information, the balance of direct and reflected sound is more natural. At low frequencies, this won't matter (and wouldn't cause the delay we see) but at HF's it does matter. However, here is the problem. Most mic's don't have an accurate 90-degree calibration. Many are created artificially with a generic conversion. When these are compared against an accurately created 90-degree calibration, they do not match each other. For these small 1/4" capsules, they over-compensate for the HF loss and cause an artificial HF rise between 5khz and 15khz, but then do not compensate enough above 15khz. That means that you may be better off pointing the mic toward the speakers and using the 0 degree serialized correction if using for EQ.

So then I had the question, does this matter enough to lose that balance of direct and reflected energy? I did some tests using a handful of different mic's I had access to at the time. All were taken in different places in the room with each orientation done back to back. I scoured the data and I just couldn't find a pattern. No obvious difference (outside of the mentioned frequency response differences). I have two mics with Cross-spectrum corrections, so those do have a correct 90-degree correction, and even these showed some HF response differences, but nothing in the time domain that was obviously different. If there was a difference in the direct vs reflected energy I captured between the two orientations, it wasn't significant enough to have any effect on the acoustic treatment decisions I would make.

Since I have a mic that is accurate pointed up, I use it that way for in-room measurements. However, I no longer feel strongly that a mic should be pointed at the ceiling. @John Mulcahy is the one who turned me onto this. I asked him if he felt it was important and he noted that the inaccuracy of the correction would outweigh the benefit of pointing it at the ceiling for EQ. If the focus is low frequencies, below say 1khz, there are no appreciable differences at all.

 

[Horacio Lewinski]

Nice system and great discussion. I can't claim to have the knowledge these guys have.

But I need to ask one thing: are you measuring with your mic facing forward as shown in your pic? And if so, are you using the 0 degree calibration file? Typically we measure at 90 degrees, with the mic pointed towards the ceiling.

Carry on

I did initially measure with the Mic pointing forward, I'm now measuring with the Mic pointing up like in the picture.

FWIW it's a Dayton EMM-6 Mic that only has one calibration file.

I just moved subs from the wall 15cm, measured bare wall and then measured with the perforated absorbers. Here are pictures. Need to run now but tomorrow morning I'll post the measurements.

 

[Matthew J Poes]

I did initially measure with the Mic pointing forward, I'm now measuring with the Mic pointing up like in the picture.
View attachment 8975

FWIW it's a Dayton EMM-6 Mic that only has one calibration file.

I just moved subs from the wall 15cm, measured bare wall and then measured with the perforated absorbers. Here are pictures. Need to run now but tomorrow morning I'll post the measurements.
View attachment 8976 View attachment 8977 View attachment 8978

Below 1khz it is unlikely that the mic orientation will make much difference with regard to response and the correction curve. For anything at or above that point you will want to point the mic at the speaker being modified.

Of course you shouldn’t be applying correction to a speaker above 1khz based on in room measurements anyway. If using room correction that does full frequency response correction out to 20khz I would point the mic at the speaker. The difference is significant. It will likely over-compensate for HF rolloff of the mic.

While not as bad as I had anticipated, DIRAC did apply more HF correction with the mic pointed up and using the 0 degree correction file (ie the scenario you are in).

 

[Eric SVL]

Below 1khz it is unlikely that the mic orientation will make much difference with regard to response and the correction curve. For anything at or above that point you will want to point the mic at the speaker being modified.

Of course you shouldn’t be applying correction to a speaker above 1khz based on in room measurements anyway.

Why? That is the sound we hear.

If using room correction that does full frequency response correction out to 20khz I would point the mic at the speaker. The difference is significant. It will likely over-compensate for HF rolloff of the mic.

I thought that was the idea - because typical rooms are very live, and the rolloff helps tame that down somewhat in the absence of absorption.

 

[Horacio Lewinski]

Measurements from last night: figuring out the time alignment procedure is requiring further learning and measurements and couldn't do it over the week. Maybe this afternoon.

But as I was anxious to see the effect of these perforated absorbers, last night I took a couple of measurements with the subs only: laptop with REW>Tascam>subs summed mono. It was a 1-pass sweep at 1M (21s length). No DSP filters applied. And ran up to 120Hz despite these being crossed at 80Hz in the regular setup.

The absorbers are 15cm deep, so first I moved the subs and rack 15cm away from the front wall and took a measurement. Then added the absorbers and took another measurement. I'm also including a measurement from June 27th, with the subs playing in the same fashion and at the original position (15cm closer to the wall than last night).

Blue is from June 27th, 15cm closer to front wall, no absorbers.
Red is from last night, no absorbers.
Green is from last night, with absorbers.

Here are the two without absorbers. Interestingly the dip at 42Hz is gone. Serendipity, bad measurement before, or bad measurement now? I guess time will tell. But looks better!

This is last night with vs without absorbers. Basically no effect below 65Hz despite the low design frequency.

Here's the wavelet from June 27th:

Note the time axis goes up to 1 second...a lot!

From last night without absorbers:

With absorbers:

The most noticeable improvement to me is the reduction at 80Hz from about 450ms to 250ms, and generally shorter decay times above 80Hz, slight impact between 60 and 80Hz, and none below 60Hz.

What's your reading of these?

 

[Matthew J Poes]

Why? That is the sound we hear.


I thought that was the idea - because typical rooms are very live, and the rolloff helps tame that down somewhat in the absence of absorption.

Hi Eric,

To your first point, because that isn’t really true. What you measure isn’t what you hear. Your ears are not omnidirectional mics. They are forward evolved, binaural, with a big hunk of meat between them. Further, because acoustically the high frequencies are not acting like the low frequencies in a room. Below the transition zone most rooms operate fully minimum phase. Applying eq is generally ok in that range for exactly that reason.

Above the transition zone, say around 500hz or so, the sound operates stochastically. While it’s a mix of reflected and direct sound, it’s mostly direct sound. However the response in any one place won’t be the same. The steady state that you measure is mostly direct sound yet you are eqing and hearing everything. So you might flatten the direct sound at a small single point in time and space while corrupting everything else.

It’s not so hard to prove if the eq you come up with is good or not. Develop the filters and then measure the speaker in free space across 180 degrees sphere and see what you get. It’s very unlikely to come out good.

The exception is when you eq using either very low Q filters or use shelf filters. These generally just change the tonality a bit and are generally ok. Even so, you want to be careful. My own speakers have a completely flat DI and very flat power response across the frontal hemisphere. However, the 10 degree window on axis has a bump in the treble. If you set the speakers up wrong and capture this bump and then eq it out, 170 degrees of the response will now be wrong. The vast majority of speakers don’t have a flat DI or flat listening window/first reflection window and so that makes application of such eq much worse.

Now as to your second point, most rooms aren’t very lively. The RT60 time for average domestic spaces is around .5 seconds. Even highly reflective spaces that are all wood or tile are often still in that range. That’s because they are so small.

However this misses my earlier point. The mic is rolled off in the 90 Degree orientation and that means that If you measure your speaker it will show the speaker as being rolled off when it is not. You aren’t accurately measuring its response. If you eq that then you can end up compensating for the roll off and making the speaker too bright. In fact, it’s true that our desired response curve is a bit rolled off in the highs. However that is mostly due to the natural acoustics of small rooms (the speaker itself should generally be flat anechoically). What this means is that you will falsely believe your response follows the desirable curve when in fact it doesn’t.

I hope this is helpful.

 

[Eric SVL]

Now as to your second point, most rooms aren’t very lively. The RT60 time for average domestic spaces is around .5 seconds. Even highly reflective spaces that are all wood or tile are often still in that range. That’s because they are so small.

Maybe I should clarify what I mean when I say "lively". I mean that the reflected sound that reaches me is still very audible in level, as it has not had a lot of time (distance) to dissipate. A larger room would result in less reflected sound reaching me, especially from the first reflection points. So what I would hear would be mostly direct sound.

But good insights overall, and I appreciate your explanations.

 

[ddude003]

It is not easy to see the difference in the above with and without absorbers so I did some image processing and created a greyscale anaglyph of the two images... Blue/Green channels are before and red is after...

 

[Eric SVL]

That is clever. I presume the red is the underlying, "before" measurement?