Destructively Driving Modes

[Nyal Mellor]

I looooooooooove this thread ! Learning a lot ..

KlausR thanks for the papers ..

Another good article - Multiple Subwoofers, Benefits and Setup

And this paper is also relevant re: target curves “Listener Preferences for In-Room Loudspeaker and Headphone Target Responses”, S. Olive, T. Welti and E. McMullin, AES convention paper 8994, Oct 2013.

 

[KlausR.]

Hi,

And this paper is also relevant re: target curves “Listener Preferences for In-Room Loudspeaker and Headphone Target Responses”, S. Olive, T. Welti and E. McMullin, AES convention paper 8994, Oct 2013.

The main results:

1. The preferred in-room loudspeaker target response
is not flat but has a bass boost of about 6.6 dB
below 105 Hz and a treble cut of -2.4 dB above 2.5
kHz.

3. The preferred bass and treble levels of the target
function for loudspeakers and headphones varied
among individual listeners (see Fig 14). For
loudspeaker playback, the range of preferred bass
and treble levels was 17 dB [from - 3 to 14.1 dB]
and 11 dB [from -8.8 to 2.4 dB], respectively.

5. Listening experience had an influence on the
preferred bass and treble levels. The preferred bass
and treble levels were higher for untrained listeners
than the trained listeners for both headphone and
loudspeaker target responses.

None of the individual target responses was flat throughout. People like bass.

If interested, mail me for a copy.

Klaus

 

[andy_c]

In another forum, Michael asked this question:

This is good info! Thanks. I wonder how one could get ahold of the mathematical modeling used in the article cited?
J. Zhang and A. C. Sanderson, JADE: Adaptive Differential Evolution with Optional External Archive, IEEE Transactions on Evolutionary Computation, 13(5): 945 - 958, October 2009.

The article can be found here. That article isn't about audio at all though, but about a general-purpose global optimization algorithm. This type of algorithm attempts to minimize what's called the objective function of the problem to be solved. The objective function is a function that takes as argument a vector (array) of parameter values to be adjusted (such as the Q, center frequency and boost/cut at the center frequency of a whole bunch of PEQs) and calculates a single number, ideally but never actually zero, representing some aggregate measure of the error to be minimized. In the case of a single listening position, the error is the RMS value over frequency of the response deviation from flatness at that position. When using multiple listening positions, the total RMS error is computed by taking the RMS error over all positions to give a single error value. Weighting can be used to give the main listening position higher emphasis than the others if desired.

In the case of the tutorial on my web page, the system uses four subwoofers and four listening positions. Each sub channel has 4 PEQs, one LPF, one delay and one gain per sub channel, times four channels. Each PEQ has 3 parameters (Q, center frequency and boost/cut value). The LPF has one parameter (cutoff frequency) and the delay and gain have one parameter each. Thus each channel has 4*3 + 1 + 1 + 1 = 15 parameters. Therefore there are a total of 60 parameters to be adjusted (15 parameters per sub channel times four sub channels). The objective function thus takes a vector of 60 parameters and the optimization algorithm adjusts them to minimize the total RMS error as computed by the objective function. The role of the optimization algorithm in the above article is to figure out how to adjust these parameters to keep improving the solution.

 

[dallasjustice]

In another forum, Michael asked this question:


The article can be found here. That article isn't about audio at all though, but about a general-purpose global optimization algorithm. This type of algorithm attempts to minimize what's called the objective function of the problem to be solved. The objective function is a function that takes as argument a vector (array) of parameter values to be adjusted (such as the Q, center frequency and boost/cut at the center frequency of a whole bunch of PEQs) and calculates a single number, ideally but never actually zero, representing some aggregate measure of the error to be minimized. In the case of a single listening position, the error is the RMS value over frequency of the response deviation from flatness at that position. When using multiple listening positions, the total RMS error is computed by taking the RMS error over all positions to give a single error value. Weighting can be used to give the main listening position higher emphasis than the others if desired.

In the case of the tutorial on my web page, the system uses four subwoofers and four listening positions. Each sub channel has 4 PEQs, one LPF, one delay and one gain per sub channel, times four channels. Each PEQ has 3 parameters (Q, center frequency and boost/cut value). The LPF has one parameter (cutoff frequency) and the delay and gain have one parameter each. Thus each channel has 4*3 + 1 + 1 + 1 = 15 parameters. Therefore there are a total of 60 parameters to be adjusted (15 parameters per sub channel times four sub channels). The objective function thus takes a vector of 60 parameters and the optimization algorithm adjusts them to minimize the total RMS error as computed by the objective function. The role of the optimization algorithm in the above article is to figure out how to adjust these parameters to keep improving the solution.

Where do I download your software? The link is broken on the website. What does the software do?

 

[andy_c]

Where do I download your software? The link is broken on the website.

Thanks for pointing that out. I've been having this weird problem off and on with the file getting dropped from my web provider without notice. I re-uploaded it, so it should work now.

What does the software do?

The best way to find out about it is to read the tutorial and the help contents. See also the tutorial PDF file and the tutorial example files. The tutorial PDF file is the same as what's on the web. The tutorial is also in the .CHM help file that comes with the program.

Here's a quick summary of a couple of example optimizations found in the tutorial. My main beta tester, Jag768, has a problematic room with no way to move the subs. His measurements of mains and subs at 4 different listening positions with no EQ on any of the subs is as follows (with no smoothing):

Note the four positions have lots of ripple, and there is lots of variation in response from position to position. These four plots are actually the result of 20 measurements. Each position has 5 measurements, the 4 subs and the main speakers. They are summed together using complex summation (properly taking phase into account) in the software.

After optimization, using individual filters, gains and delays for each sub, the following result is achieved:

Note that the responses are much flatter, and there is quite a bit less response variation from one position to the next.

Finally, EQ that's common to all 4 subs is applied to flatten out the main listening position. This gives the following result.

 

[dallasjustice]

Thanks Andy. So it's a two step procedure. The first step is to obtain the best delays and gain settings for each sub. The second step is to apply EQ? Which hardware works best with the software? I use a Lynx Hilo, Jriver and use Audiolense or Acourate for EQ.

I'm still working on best sub positions. It's wearing me out. I've been pushing subs around all weekend.

Thanks for pointing that out. I've been having this weird problem off and on with the file getting dropped from my web provider without notice. I re-uploaded it, so it should work now.



The best way to find out about it is to read the tutorial and the help contents. See also the tutorial PDF file and the tutorial example files. Here's a quick summary of an example optimization. My main beta tester, Jag768, has a problematic room with no way to move the subs. His measurements of mains and subs at 4 different listening positions with no EQ on any of the subs is as follows (with no smoothing):

Note the four positions have lots of ripple, and there is lots of variation in response from position to position. These four plots are actually the result of 20 measurements. Each position has 5 measurements, the 4 subs and the main speakers. They are summed together using complex summation (properly taking phase into account) in the software.

After optimization, using individual filters, gains and delays for each sub, the following result is achieved:

Note that the responses are much flatter, and there is quite a bit less response variation from one position to the next.

Finally, EQ that's common to all 4 subs is applied to flatten out the main listening position. This gives the following result.

 

[andy_c]

Thanks Andy. So it's a two step procedure. The first step is to obtain the best delays and gain settings for each sub. The second step is to apply EQ? Which hardware works best with the software? I use a Lynx Hilo, Jriver and use Audiolense or Acourate for EQ.

I'm still working on best sub positions. It's wearing me out. I've been pushing subs around all weekend.

The first step shown above (whose result is the second graph) adjusts individual gain, delay and EQ individually for each sub, the whole kit and kaboodle at once. The optimization is a weighted RMS sum of all listening positions, so the main listening position may get compromised a bit because the optimization takes into account all listening positions simultaneously. The second step, which is optional and which in your case might use Acourate or Audiolense instead (neither of which I'm familiar with) flattens out the main listening position only, using EQ that's common to all subs (not individual per-sub).

The program assumes and only works with IIR filters, not FIR. In the case of the first step above, the following rather bizarre filters result for the four subs (delays not shown, but they will appear in the filter report per the tutorial). This is a Geddes arrangement, so there is no high-pass filtering for the mains. The subs end up with different cutoff frequencies for each low-pass filter, the gains are different, and the PEQs are staggered and different for each channel. The program allows user-specified values for the allowable range of all parameters of all filters, so if e.g. you didn't want such a large span of gain values, you could give them tighter limits.

 

[Rodney Gold]

Dallas.. do you want to optimise at the sweet spot only?.
. thats what I do..I dont worry about optimising bass all over the room as there is no HT involved

 

[dallasjustice]

Dallas.. do you want to optimise at the sweet spot only?.
. thats what I do..I dont worry about optimising bass all over the room as there is no HT involved

Rodney, all I'm doing now is to design a subwoofer setup which is best for MLP without EQ. The Fazenda paper talks about a setup he called "Welti Optimized." I have measure numerous setups in my room including the traditional midwall Welti with four subs and the Welit midwall with two subs. In the end, the most interesting result comes from the source sink/Welti optimized setup using only two midwall subs in opposite polarity. This totally destroys the odd order a axials and the 24hz 1st. However, the even order looks choppy. So, the response flips phase at 47hz. My next test will be to use 2 rear wall subs with a crossover between them. One in positive polarity and one in negative polarity. The idea is have Welti optimized setup which is effective for both odd and even order modes. We will see.

 

[Rodney Gold]

I would like some advice or recommendations to play with

8x6mx2.3m room , Giya G1's not high passed and 4 SVS subs.. assume its a blank piece of paper and I can place anything where I want..
What would the best placement of the subs be and how do I set it all up...
I can DSP each sub individually with 6 peqs, change polarity , set distances etc or control all subs all at once and also have DIRAC ..
My bass preference is slightly overblown , but I hate boom or bloom..got to be crisp articulate bass..but I like to feel it viscerally

 

[dallasjustice]

I would like some advice or recommendations to play with

8x6mx2.3m room , Giya G1's not high passed and 4 SVS subs.. assume its a blank piece of paper and I can place anything where I want..
What would the best placement of the subs be and how do I set it all up...
I can DSP each sub individually with 6 peqs, change polarity , set distances etc or control all subs all at once and also have DIRAC ..
My bass preference is slightly overblown , but I hate boom or bloom..got to be crisp articulate bass..but I like to feel it viscerally

Rodney, I'm still exploring here but this is what I've found works in my recantangularish room.

If your room is rectangularish, my recommendation would be the same as Amir's recommenation to Keith. Always start off with only two subs. Place them each on frontwall and rearwall. Each sub should be as close as possible to the wall's midpoint. The subs should be run in mono. I came to this conclusion based on two things. First, the Fazenda study Klaus cited earlier is really on point. It very clearly demonstrates that the "welti optimized" setup is superior compared with several other configurations. Of course, the study couldn't possibly investigate every possible subwoofer configuration, but I think it's the best place to start off. In that study, the "best" setup was the "optimized" part of the Welti configuration in which the front wall and rear wall subs were run in "anti-phase." In my case, this means I run them in opposite polarity. Others call this "active" room mode cancellation (eg. Celestinos). I think that's a good name for it. When subs are run this way, you will see that the odd order modes are totally eliminated once you fine tuned the phase on the rear sub.(you have to account for the flight time for the front walll wave to the rear wall "sink" sub).

The problem with this setup is that the even order modes are choppy. In my case, the bass is unbelievably flat (no EQ) from 15hz to 47hz. This is the second reason for the Welti setup. No other subwoofer setup combination (no EQ) could do this! 47hz represents the even order harmonic for the axial length mode in my room. This is where the out of polarity rear sub is a problem. At the even order mode, the polarity needs to be the same as the frontwall sub. This is why I feel the rear wall "sink" sub should really be 2 different subs with crossover between them.

IOW, the so called "welti optimized" 2 subwoofer setup is still the best solution, but it can be made better if it is "polarity optimized" throughout the entire low frequency bandpass. I believe this can be done, but it requires a more complex crossover array. I've been talking to Uli about this. It can be setup in Acourate. However, I want to first test my theory out. When I have some free time, I believe I can loopback my REW logsweep into Jriver ASIO line in. In Jriver 21, I can setup crossovers between the two rear subs and then measure it with all three subs together. I will post those measurements when it's all setup.

What may end up working best for me is a variation on the Welti midwall 2 sub array. My usually favorite low frequency bandpass is less than 80hz. In summary, I want to split the low frequency bandpass in half between two rear wall subs in opposite polarity using a simple crossover and use one midwall front sub (3 subs total, but only two frequency dependent active subs). Right now, I am just testing for the best sub array before EQ and crossover to R/L. Once I have it setup the way I want, I hope to cross the whole active cancellation subwoofer array over at 80hz to my R/L speakers and apply final EQ/DSP. I'm just taking this one step at a time.

Michael.

 

[Rodney Gold]

Thanks michael.. Im re arranging my room a bit , will try the 2 subs midwall out of phase a little later, then try the front and back walls (heavy subs to move by myself)

 

[dallasjustice]

Thanks michael.. Im re arranging my room a bit , will try the 2 subs midwall out of phase a little later, then try the front and back walls (heavy subs to move by myself)

Don't forget to measure. I might be totally full of ****.

 

[andy_c]

Thanks Andy. So it's a two step procedure. The first step is to obtain the best delays and gain settings for each sub. The second step is to apply EQ?

You can do that if you like as well. The key thing is that the user specifies what filters (if any), gains and delays are to be used. It doesn't try to figure out what filters are needed, but only tries to optimize the parameter values of the filters (or gains or delays) entered by the user.

So, one thing you could do would be to specify that only individual gains and delays are to be used for each sub. You can also specify that the frequency response should only be evaluated at one listening position. Then the software would go through and try out possibly millions of combinations of individual gains and delays for each sub to get the response as flat as possible at the main listening position without any EQ at all.

 

[dallasjustice]

You can do that if you like as well. The key thing is that the user specifies what filters (if any), gains and delays are to be used. It doesn't try to figure out what filters are needed, but only tries to optimize the parameter values of the filters (or gains or delays) entered by the user.

So, one thing you could do would be to specify that only individual gains and delays are to be used for each sub. You can also specify that the frequency response should only be evaluated at one listening position. Then the software would go through and try out possibly millions of combinations of individual gains and delays for each sub to get the response as flat as possible at the main listening position without any EQ at all.

That's very interesting. It seems like your software could be used in conjunction with other EQ type software as well; eg. Audiolense, Acourate, Dirac Live etc. To me, that could be better since I already feel very happy with the full range EQ applied with other software.

One of the missing pieces when it comes to setting up subs with most DSP software is that the software often assumes the subs are already optimally setup for level and locations. When they aren't, then the final EQ applied can't be most effective.


I just wish there were software which could also optimize subwoofer location. That would be very interesting.

 

[Nyal Mellor]

Rodney, I'm still exploring here but this is what I've found works in my recantangularish room.

If your room is rectangularish, my recommendation would be the same as Amir's recommenation to Keith. Always start off with only two subs. Place them each on frontwall and rearwall. Each sub should be as close as possible to the wall's midpoint. The subs should be run in mono. I came to this conclusion based on two things. First, the Fazenda study Klaus cited earlier is really on point. It very clearly demonstrates that the "welti optimized" setup is superior compared with several other configurations. Of course, the study couldn't possibly investigate every possible subwoofer configuration, but I think it's the best place to start off. In that study, the "best" setup was the "optimized" part of the Welti configuration in which the front wall and rear wall subs were run in "anti-phase." In my case, this means I run them in opposite polarity. Others call this "active" room mode cancellation (eg. Celestinos). I think that's a good name for it. When subs are run this way, you will see that the odd order modes are totally eliminated once you fine tuned the phase on the rear sub.(you have to account for the flight time for the front walll wave to the rear wall "sink" sub).

The problem with this setup is that the even order modes are choppy. In my case, the bass is unbelievably flat (no EQ) from 15hz to 47hz. This is the second reason for the Welti setup. No other subwoofer setup combination (no EQ) could do this! 47hz represents the even order harmonic for the axial length mode in my room. This is where the out of polarity rear sub is a problem. At the even order mode, the polarity needs to be the same as the frontwall sub. This is why I feel the rear wall "sink" sub should really be 2 different subs with crossover between them.

IOW, the so called "welti optimized" 2 subwoofer setup is still the best solution, but it can be made better if it is "polarity optimized" throughout the entire low frequency bandpass. I believe this can be done, but it requires a more complex crossover array. I've been talking to Uli about this. It can be setup in Acourate. However, I want to first test my theory out. When I have some free time, I believe I can loopback my REW logsweep into Jriver ASIO line in. In Jriver 21, I can setup crossovers between the two rear subs and then measure it with all three subs together. I will post those measurements when it's all setup.

What may end up working best for me is a variation on the Welti midwall 2 sub array. My usually favorite low frequency bandpass is less than 80hz. In summary, I want to split the low frequency bandpass in half between two rear wall subs in opposite polarity using a simple crossover and use one midwall front sub (3 subs total, but only two frequency dependent active subs). Right now, I am just testing for the best sub array before EQ and crossover to R/L. Once I have it setup the way I want, I hope to cross the whole active cancellation subwoofer array over at 80hz to my R/L speakers and apply final EQ/DSP. I'm just taking this one step at a time.

Michael.

You could introduce a DSP phase shift filter on the rear sub to do a 180 degree phase shift over a certain frequency range...?

 

[Nyal Mellor]

That's very interesting. It seems like your software could be used in conjunction with other EQ type software as well; eg. Audiolense, Acourate, Dirac Live etc. To me, that could be better since I already feel very happy with the full range EQ applied with other software.

One of the missing pieces when it comes to setting up subs with most DSP software is that the software often assumes the subs are already optimally setup for level and locations. When they aren't, then the final EQ applied can't be most effective.


I just wish there were software which could also optimize subwoofer location. That would be very interesting.

There is software, it's just not easy to use...see the BEM graphs I posted earlier. KYDG also has a more sophisticated tool called FLO.

 

[dallasjustice]

You could introduce a DSP phase shift filter on the rear sub to do a 180 degree phase shift over a certain frequency range...?

Yes, maybe, I don't know. My gut tells me that's my backup option. For me, the exercise is to minimize drastic low frequency FIR filters. #becausepreringing

 

[dallasjustice]

There is software, it's just not easy to use...see the BEM graphs I posted earlier. KYDG also has a more sophisticated tool called FLO.

Is that the finite element analysis software about which only Stanford trained physicist may write software apps?

 

[DaveC]

I think this is a result of Welti, Geddes, and LaJeune's work... some good info:

http://jamesromeyn.com/#/home-audio-gear/dsa-1-0-distributed-subwoofer-array-5-pieces-4k-usd/