• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Bass arriving out of phase at the listening position

SoundGuy

Active Member
Joined
Aug 13, 2022
Messages
115
Likes
120
The ports aren't blocked. (I think I already tried blocking ports some time ago and concluded it made no difference, but I guess I could check again.)

This is not a speaker defect. I already tried swapping both speakers for another pair and it made exactly zero difference.

OK. You must be in the Bermuda Triangle - good luck ;-)
 
OP
edechamps

edechamps

Addicted to Fun and Learning
Forum Donor
Joined
Nov 21, 2018
Messages
910
Likes
3,620
Location
London, United Kingdom
So this is intriguing:

ir.png


This is a filtered IR at 80 Hz, 1/3 octave, 6th order filter and zero phase filtering. I deliberately set t=0 to what appears to be the first cycle. I included the rear speakers as reference "witnesses".

We can see that, in general, the front right and the rear speakers tend to play roughly in phase and appear to be well-behaved, beating at 12 ms (consistent with 80 Hz filter wavelength) like clockwork. The same clearly cannot be said for the front left speaker. Before 10 ms it's mostly in phase but then it starts drifting. By 30 ms it's completely out of phase.

The same graph without zero phase filtering:

irs.png


There's something really weird happening near 22 ms. The shape of that "kink" is exactly what one would expect if an out-of-phase reflection was coming in and destructively interfering with the direct sound. The other speakers do not show any such kinks.

Here's what I think is happening:
  • During the first 10 ms it's mostly fine.
  • Near 14 ms we can see it's struggling. Out-of-phase reflections appear to be kicking in and partly interfering with the direct sound. However they are not yet loud enough to cancel it completely.
  • At 24 ms out-of-phase reflections seem to completely cancel out the direct sound.
  • By 30 ms the out-of-phase reflections have taken over and the resulting sound is now almost perfectly out of phase with the other speakers.
  • The out-of-phase reflections keep getting louder and peak near 60 ms.
 

Geert

Major Contributor
Joined
Mar 20, 2020
Messages
1,938
Likes
3,526
The moment I saw the layout of your setup my intuition said your issue is caused by reflections and not room modes. The challenge is to get an understanding of which frequency response anomalies are caused by reflections and which by room modes. I once dealt with that by moving the microphone in 1 plane to see the effect on the frequency response. When dealing with reflections this should shift the cancelation frequency, while room modes on the other hand don't shift. By evaluating how the cancelation frequency changes when moving the microphone in different directions you can figure out the source of the reflection. The fact that this room is so small makes it difficult, because you might be moving from one reflection to another.
 
Last edited:

Thomas_A

Major Contributor
Forum Donor
Joined
Jun 20, 2019
Messages
3,422
Likes
2,407
Location
Sweden
So this is intriguing:

View attachment 231958

This is a filtered IR at 80 Hz, 1/3 octave, 6th order filter and zero phase filtering. I deliberately set t=0 to what appears to be the first cycle. I included the rear speakers as reference "witnesses".

We can see that, in general, the front right and the rear speakers tend to play roughly in phase and appear to be well-behaved, beating at 12 ms (consistent with 80 Hz filter wavelength) like clockwork. The same clearly cannot be said for the front left speaker. Before 10 ms it's mostly in phase but then it starts drifting. By 30 ms it's completely out of phase.

The same graph without zero phase filtering:

View attachment 231961

There's something really weird happening near 22 ms. The shape of that "kink" is exactly what one would expect if an out-of-phase reflection was coming in and destructively interfering with the direct sound. The other speakers do not show any such kinks.

Here's what I think is happening:
  • During the first 10 ms it's mostly fine.
  • Near 14 ms we can see it's struggling. Out-of-phase reflections appear to be kicking in and partly interfering with the direct sound. However they are not yet loud enough to cancel it completely.
  • At 24 ms out-of-phase reflections seem to completely cancel out the direct sound.
  • By 30 ms the out-of-phase reflections have taken over and the resulting sound is now almost perfectly out of phase with the other speakers.
  • The out-of-phase reflections keep getting louder and peak near 60 ms.
Is it possible to show the same but measured with the door open?
 

Geert

Major Contributor
Joined
Mar 20, 2020
Messages
1,938
Likes
3,526
A little bit of math to support my reflections idea:

What's driving me nuts is the summation - or lack thereof - between the left and right speakers around 90 Hz.

90 Hz = 3.82 m wavelength. A reflection half of that wavelength, 1.91 m, will cause cancelation. That reflection arrives after the direct sound so we also need to take to speaker-listener distance into account, about 1.7 m. So the total reflection path (speaker > boundary > listener) is about 1.9 + 1.7 = 3.6 m long.

1.9 m relative reflection path length = 5.52 ms. In your "same graph without zero phase filtering" graph we effectively see the front left delay (purple) jumps to about 5 ms from 30 ms on ...

The total reflection pathlength of 3.6 m is about the distance from left speaker > left corner/window > listener. Also left speaker > left back corner > listener is an option.

In your experiment with moving the speakers and microphone we see the cancelation frequency changing when moving the microphone left to right (purple and green measurement). The 'Mic 70 cm right' measurement is interesting as it shifts the cancelation frequency 10 Hz down. This makes sense, as moving away from the window increases the reflection pathlength, which lowers the cancelation frequency. The 80 Hz cancelation we see here corresponds to a wavelength of 4.3 m, half wavelength is 2.15 m while the original one (for 90 Hz) was 1,91 m.

The measurement "Mic 70 cm front" shifts the cancelation frequency even further down. Knowing this kind of shift requires the reflection pathlength to increase, this confirms the reflection is running via the left back corner. The front corner is out as a move to the front decreases the length of this reflection. So the solution could be to place an additional bass trap in the back corner. Although according to my experience you need a lot of absorption to tackle such a boundary reflection.

This is just a bit of brainstorming in an attempt to understand what could be happening, but given the size of the room with all these boundaries in close distance it's impossible to predict if this scenario makes sense. You would need to test to verify it, for example by placing a bass trap in the back corner (but don't use the front one because that might already play a role in handling the front reflections).

Worst case this was all nonsense, especially since I did the math on the back of a napkin while attending a meeting ;)
 
Last edited:

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
We should talk (@NTK). I use GMSH and FeniCS/Firedrake/PETSc to solve the Helmotz equation. Example: horn2.py
Original idea was to compute the optimal shape of a waveguide to get smoother directivity. I built the possible shape of the waveguide as a spline with few parameters.
Gradients are computed with JAX or hand made. I have a 2D working version somewhere but stopped before I went 3D. I would need to optimise the Helmotz solver by
doing something similar to what Mathematica does or maybe simply interpolating between frequencies. Clearly GMRES converge much faster for freqency F when the starting point is the previous solution at frequency F2 ~ F.

Project is stalled by lack of time.
Yeah. Sure. I should be able to find a few hours a week (on the average) to work on learning a new skill.

Not being an applied mathematician, my knowledge on the mathematics behind finite element methods is pretty weak. But I do intend to learn more on how to setup the problems in FEniCS and the other various FEM software packages in Python and Julia.

May be we should start a new thread if you want to start a "community project" (even if it is a community of 2 :)).
 
OP
edechamps

edechamps

Addicted to Fun and Learning
Forum Donor
Joined
Nov 21, 2018
Messages
910
Likes
3,620
Location
London, United Kingdom
I ran some experiments with the wardrobe. Original, wardrobe doors open, and wardrobe doors removed.

wardobeleft.png

wardroberight.png

wardrobesum.png


Clearly the individual left and right channels don't like it when I do this - the existing large dips get even bigger. Somewhat surprisingly they seem to get more in phase but that might simply be because whatever is causing both channels to dip is also causing them to get back in phase.

One thing I did in my last set of measurements was to tell REW to measure between 60 Hz and 150 Hz only. This way I don't have to filter the IR. This is the original setup:

bassir.png


This shows yet another perspective on the issue. The traces are almost right on top of each other until about 12ms. Then the wheels start to come off for the left channel.

Is it possible to show the same but measured with the door open?

Opening the door doesn't make any difference. The filtered impulse response looks exactly the same.
 
OP
edechamps

edechamps

Addicted to Fun and Learning
Forum Donor
Joined
Nov 21, 2018
Messages
910
Likes
3,620
Location
London, United Kingdom
You would need to test to verify it, for example by placing a bass trap in the back corner (but don't use the front one because that might already play a role in handling the front reflections).

Sure. Setup:

7b.jpg 7a.jpg 7c.jpg

Results:

leftcorner.png

vecrmscorner.png


(no significant difference on the right channel)

There's quite a significant difference, but it looks like it was obtained by killing the left channel :(

Folks, this UK 3-day weekend is coming to an end (RIP Elizabeth), so I'm afraid I won't be able to try anything else in the immediate future. I think I'm going to focus on trying to simulate the room like @NTK did so that I can then simulate various potential solutions: adding tuned bass traps, adding sound sources, or rotating the setup. Thanks for all the suggestions.
 

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
I'll upload my Mathematica scripts in a couple of hours. You can probably run them on an RPi with 8 GB RAM with some patience (I haven't tried).
 

Thomas_A

Major Contributor
Forum Donor
Joined
Jun 20, 2019
Messages
3,422
Likes
2,407
Location
Sweden
I ran some experiments with the wardrobe. Original, wardrobe doors open, and wardrobe doors removed.

View attachment 231986
View attachment 231987
View attachment 231988

Clearly the individual left and right channels don't like it when I do this - the existing large dips get even bigger. Somewhat surprisingly they seem to get more in phase but that might simply be because whatever is causing both channels to dip is also causing them to get back in phase.

One thing I did in my last set of measurements was to tell REW to measure between 60 Hz and 150 Hz only. This way I don't have to filter the IR. This is the original setup:

View attachment 231991

This shows yet another perspective on the issue. The traces are almost right on top of each other until about 12ms. Then the wheels start to come off for the left channel.



Opening the door doesn't make any difference. The filtered impulse response looks exactly the same.
My best guess is wardrobe corner diffraction. Build some temporary damping panels from the wardrobe corner in the opening and test
 

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
I'll upload my Mathematica scripts in a couple of hours. You can probably run them on an RPi with 8 GB RAM with some patience (I haven't tried).
The attached zip file has both the room mode and room frequency response Mathematica notebooks, and their PDFs.

I simulated with the speakers flipped to the other side of the room just for comparison. It seems to match the milder cancellations of the rear speakers as shown in post #102. The notebook in the zip is this configuration.

Room_FR.gif
 

Attachments

  • 3D Room Response Simulation.zip
    1.1 MB · Views: 40

hmt

Senior Member
Joined
Apr 11, 2020
Messages
397
Likes
479
Sure. Setup:

View attachment 232006 View attachment 232009 View attachment 232010

Results:

View attachment 232008
View attachment 232018

(no significant difference on the right channel)

There's quite a significant difference, but it looks like it was obtained by killing the left channel :(

Folks, this UK 3-day weekend is coming to an end (RIP Elizabeth), so I'm afraid I won't be able to try anything else in the immediate future. I think I'm going to focus on trying to simulate the room like @NTK did so that I can then simulate various potential solutions: adding tuned bass traps, adding sound sources, or rotating the setup. Thanks for all the suggestions.

Is it possible to try to put the bass trap like wall to close the small area where the door is? I know that is in no way a solution but it might help to figure out the role of that area.
 

Axo1989

Major Contributor
Joined
Jan 9, 2022
Messages
2,811
Likes
2,808
Location
Sydney
View attachment 231991

This shows yet another perspective on the issue. The traces are almost right on top of each other until about 12ms. Then the wheels start to come off for the left channel.

I find this part pretty fascinating. I tend to imagine room modes as steady state, which they are in equilibrium. But with music playback we have dynamic behaviour, and those phase effects take their time (sometimes too fast to perceive, sometimes likely perceptible) to play out.
 

Rescue911

Member
Joined
May 27, 2022
Messages
36
Likes
17
Somehow it turns out that I get the impression that the author is just trolling us... :(
Sorry if that's not right.
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Here's what I think is happening:
  • During the first 10 ms it's mostly fine.
  • Near 14 ms we can see it's struggling. Out-of-phase reflections appear to be kicking in and partly interfering with the direct sound. However they are not yet loud enough to cancel it completely.
  • At 24 ms out-of-phase reflections seem to completely cancel out the direct sound.
  • By 30 ms the out-of-phase reflections have taken over and the resulting sound is now almost perfectly out of phase with the other speakers.
  • The out-of-phase reflections keep getting louder and peak near 60 ms.

While I think some of these things may be happening indeed, IMO, all speaker channels are heavily being modified by the acoustics of the room esp. in the bass region.

I wonder how much DSP can “restore” each channel — even if mostly only at at the mic measuring point — as close as possible to what’s theoretically an ideal minimum response. Semi-automated processing ffrom Trinnov or Acourate (by an expert calibrator) should be able to salvage this system better than any manual by-hand EQ could. *Although, we already know that even those companies also say that to get best results with DSP one must deal with improving room acoustics as much as possible first.
 
Last edited:

SoundGuy

Active Member
Joined
Aug 13, 2022
Messages
115
Likes
120
Sure. Setup:

View attachment 232006 View attachment 232009 View attachment 232010

Results:

View attachment 232008
View attachment 232018

(no significant difference on the right channel)

There's quite a significant difference, but it looks like it was obtained by killing the left channel :(

Folks, this UK 3-day weekend is coming to an end (RIP Elizabeth), so I'm afraid I won't be able to try anything else in the immediate future. I think I'm going to focus on trying to simulate the room like @NTK did so that I can then simulate various potential solutions: adding tuned bass traps, adding sound sources, or rotating the setup. Thanks for all the suggestions.
The two strongest boundary effects will be “speakers to wall behind them” and “listener distance to rear wall”.

You get cancellation at 1/4 wavelength of those distances. So simply measure those distances and calculate where your deep nulls will be and see how well that compares to your plot.

You can’t fix this with any EQ. The only solution in this room to minimize these primary boundary effects is to:

1) Make sure that the two distances speakers to wall AND listener to wall are as different as possible (partial fix)
2) Choose a diagonal setup. Rotate your desk, speakers and listening position by 45 Degrees. Ensure your listening position is as far away from a boundary as possible. (This will provide the Most optimal setup you can achieve in this space)

I guarantee that a diagonal setup with listener far from boundaries will measure infinitely better. This is because you will eliminate the symmetry which creates the very specific strong 1/4 wave cancellation between 60 and 120 Hz for both listener and L and R speaker. The room modes you need to reduce are typical mode 2 and mode 3.

Here is a visual explanation of why


More on Room Modes

 
Last edited:

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
I guarantee that a diagonal setup with listener far from boundaries will measure infinitely better. This is because you will eliminate the symmetry which creates the very specific strong 1/4 wave cancellation between 60 and 120 Hz for both listener and L and R speaker. The room modes you need to reduce are typical mode 2 and mode 3.

I don’t know how willing @edechamps is going to be as that large TV is going to take up a lot of real estate — most likely having to completely block the window. But one should at least try it…

Oh, and I’m sure most people here do not think of EQ as an actual “fix” or cure for boundary interference issues at all, but it would still be interesting to know to what extent can DSP help “mitigate” frequency response deviations and improve overall coherence for better summation between all speaker channels in this kind situation if further repositioning changes were simply not an option. Maybe not everyone here thinks much of it, but I’m quite curious about this myself…
 

Thomas_A

Major Contributor
Forum Donor
Joined
Jun 20, 2019
Messages
3,422
Likes
2,407
Location
Sweden
If the TV screen is in line with the drivers, it will act as a prolongation of the baffle. IMO just stuff thick panels on the wall on each side of the TV screen, behind the speakers. (And IMO, rotate the setup 180°, even if it needs rerouting cables around the room.) For the wall behind you, use diffusor material on the wall.
 

abdo123

Master Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
7,425
Likes
7,940
Location
Brussels, Belgium
Thank you all for reminding me the i can use all pass filters to insert delay without having to use FIR.

I realised i have multi subwoofer nulls that this can easily take care of.
 
Top Bottom