Jump to content


Photo

Ideas for new ways to display data


  • Please log in to reply
65 replies to this topic

#1 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 23 December 2015 - 09:06 PM

We are hopefully going to do a bit of a revamp to the site in the next 6 months or so and we have a lot of ideas.

 

One new comparison chart that I want to add is an SPL per cubic liter of cabinet metric. This would be based at the same 12, 1/3rd octave center frequencies as currently used for the burst testing. The burst SPL would be considered for each system at each bandwidth and a value assigned based on the SPL and the total external volume of the system. This way it would give a relative value of power density at each bandwidth. Should be interesting to see small systems compared to huge horns and the like in this manner. I haven't quite worked out the exact comparative formula yet. Since SPL is a log function it will likely require converting to Pa and back. Shouldn't be too bad though. Anyone want to take a stab at a normalized formula?

 

Below is an idea for a different type of distortion graph that was suggested by Nathan Funk at Funk Audio.

Not exactly easy to put together right now and it's just kludged together in excel at the moment but something could be automated to make it faster I'm sure. Basically the distortion captured during the long term sweep measurements from each system are charted at a single frequency as it increases with the increased output of the device during the sweeps rather than over the whole bandwidth during a single sweep. The sweeps themselves are only measured at increases of 5dB typically so a few data points will be used to extrapolate the rest of the data between to fill the curves in. Judging from the rough results here it looks like this is not an issue. It does make comparison easier but also only represents a very tiny slice of the bandwidth. Still I think it is worth adding something like this possibly though.

 

Here are a few charts showing a collection of drivers tested in the standard sealed cabs with the data shown in the manner detailed above at 10, 12.5, 16 and 20Hz.

 

Suggestions? Thoughts?

 

10Hz

Attached File  10Hz Sealed Distortion.png   102.14KB   0 downloads

 

12.5Hz

Attached File  12.5Hz Sealed Distortion.png   111.74KB   0 downloads

 

16Hz

Attached File  16Hz Sealed Distortion.png   105.79KB   0 downloads

 

20Hz

Attached File  20Hz Sealed Distortion.png   88.29KB   0 downloads

 

 



#2 Diogenes

Diogenes

    Member

  • Members
  • PipPip
  • 25 posts

Posted 23 December 2015 - 09:32 PM

Very information dense and it seems it could get out of hand in a hurry unless there's a way to select which drivers to show, but it tells a lot at a glance. Fwiw, I like it, but as an addition to, not substitution for the already pretty clear way you present data.

 

Admit it, you just wanted to show off how awesome that Rockford Fosgate driver is.  ;)



#3 SME

SME

    Super Bass Overlord

  • Members
  • PipPipPipPipPipPipPipPip
  • 966 posts

Posted 23 December 2015 - 10:46 PM

We are hopefully going to do a bit of a revamp to the site in the next 6 months or so and we have a lot of ideas.

 

One new comparison chart that I want to add is an SPL per cubic liter of cabinet metric. This would be based at the same 12, 1/3rd octave center frequencies as currently used for the burst testing. The burst SPL would be considered for each system at each bandwidth and a value assigned based on the SPL and the total external volume of the system. This way it would give a relative value of power density at each bandwidth. Should be interesting to see small systems compared to huge horns and the like in this manner. I haven't quite worked out the exact comparative formula yet. Since SPL is a log function it will likely require converting to Pa and back. Shouldn't be too bad though. Anyone want to take a stab at a normalized formula?

 

Hi Ricci.  This sounds awesome!  I'd be glad to help contribute my mathematical and acoustic knowledge to this problem.  I practically do conversions between dB SPL and Pa in my sleep.  However, I'm not exactly certain what what you are trying to achieve with your formula.  Should efficiency enter into the picture?  Without considering efficiency and assuming unlimited amp power, most sealed box systems are likely to perform almost the same, especially at the bottom end where everything is excursion limited anyway.  I'm thinking you may be more interested in efficiency per liter.  But really, I guess it depends on the application.

 

I really like your ideas for the distortion graphs and see no issues with your approach.  If your plotting system supports B-spline interpolation, you may be able to get curves that are more smooth and more accurate at the same time, but you'll want to verify that the fits make sense.  For example, some kinds of spline fits assume that either the data points or their derivatives (i.e. slopes) go to zero at the ends of the graph, so that's something to watch out for.

 

By the way, the technical term for filling in between data points as you are doing here is called interpolation.  Extrapolation refers instead to extending curves beyond the last point in a data set.  As you can imagine, extrapolated figures are a lot more likely to give erroneous results, which is why they are typically frowned upon.



#4 3ll3d00d

3ll3d00d

    Ultra Member

  • Members
  • PipPipPipPipPipPip
  • 413 posts

Posted 23 December 2015 - 11:21 PM

 

Here are a few charts showing a collection of drivers tested in the standard sealed cabs with the data shown in the manner detailed above at 10, 12.5, 16 and 20Hz.

 

Suggestions? Thoughts?

 

I tend to find line charts visually quite misleading as your eye is drawn to the outliers rather than the data points, not sure if there is a sensible way to avoid this here though.

 

I suspect that a comparative chart would be good though, i.e. pick a driver as the baseline and then show a.n.other driver as the delta to that baseline.

 

You could also turn that into a surface plot to enable the different frequencies to be pulled into 1 chart. I would think this would be particularly useful when comparing two drivers. You could look at the REW eq window for an example of this in action, the waterfall plot in that screen allows you to overlay the before and after waterfall and you can then reasonably easily the key points of difference.

 

A directivity sonogram or speclab captures are some other examples (albeit 2D examples) of such 3d plots.

 

Also who doesn't like 3d charts? We expect an IRIX interface here :D



#5 shadyJ

shadyJ

    Power Member

  • Members
  • PipPipPipPip
  • 82 posts

Posted 24 December 2015 - 01:25 AM

Hello Josh,

 

As you well know, THD is not a great metric for assessing how clean a subwoofer sounds. Still, it's better than nothing. I am not sure that comparing individual harmonics make sense either, I am not sure how useful that would be. This might be ridiculous, but what about using some kind of harmonic-dependent threshold for long-term output, something like CEA-2010? YOU could probably establish a more sensible metric than the Consumer Electronics Association anyway, you would be surprised at how arbitrary the CEA-2010 thresholds are. They are not based on rigorous experiments. It would take more work to create graphs from your own distortion threshold formula, but it would be an interesting project. 



#6 Infrasonic

Infrasonic

    Bass Overlord

  • Members
  • PipPipPipPipPipPipPip
  • 779 posts

Posted 24 December 2015 - 02:48 AM

Hey! That's a pretty sweet new way to quantify these subs performance. SPL per cubic foot. I love it!

 

I can't think of many (any?) suggestions for new tests, I'm sorry to say.

 

Maybe bring back spectral contamination tests but I believe those were originally nixed from DB tests because you said (paraphrasing) "too hard on a driver" of a test to do. IIRC.

 

Erm, that's all I can think of at the moment.

 

Ummm....

 

 

Ship all the systems to me when you're done testing? Yes. Yes. That sounds good.



#7 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 28 December 2015 - 05:05 PM

Very information dense and it seems it could get out of hand in a hurry unless there's a way to select which drivers to show, but it tells a lot at a glance. Fwiw, I like it, but as an addition to, not substitution for the already pretty clear way you present data.

 

Admit it, you just wanted to show off how awesome that Rockford Fosgate driver is.  ;)

 

The charts I showed are just what I whipped up in excel to see how it would look. I had them done a long time ago and kind of forgot about them. I was adding the RF driver and thought I'd go ahead and post something about it. The T3 19 sure does do well though huh? :D  If something like this was added to the forum it would be similar to the comparable graphs we have now. You would only be able to compare 2 systems at a time not 10+ like I had on these.



#8 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 28 December 2015 - 05:14 PM

Hello Josh,

 

As you well know, THD is not a great metric for assessing how clean a subwoofer sounds. Still, it's better than nothing. I am not sure that comparing individual harmonics make sense either, I am not sure how useful that would be. This might be ridiculous, but what about using some kind of harmonic-dependent threshold for long-term output, something like CEA-2010? YOU could probably establish a more sensible metric than the Consumer Electronics Association anyway, you would be surprised at how arbitrary the CEA-2010 thresholds are. They are not based on rigorous experiments. It would take more work to create graphs from your own distortion threshold formula, but it would be an interesting project. 

 

Actually there is study that has been put into the thresholds established as the limits for CEA-2010. There are various papers available on harmonic distortion, masking and bass frequencies. It is by no means perfect as has been discussed a few times. There is no doubt that even large amounts of distortion up to 15-20% o the first 2 harmonics are not that easily noticed in the bass range. Especially with complex content with upper frequency masking. Mechanical or non harmonic noises are much more distracting to me personally. Still some sort of standard is better than nothing. On that we agree. I'm not going to get to the point of comparing individual harmonics. That would be a lot of work and practically difficult with the current tools available without investing significant time and resources. At the end of the day I don't know if it would be that useful either. However I still feel that distortion is a useful metric to look at and compare and my preference is for as low of distortion as possible.



#9 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 28 December 2015 - 05:32 PM

You could also turn that into a surface plot to enable the different frequencies to be pulled into 1 chart. I would think this would be particularly useful when comparing two drivers. You could look at the REW eq window for an example of this in action, the waterfall plot in that screen allows you to overlay the before and after waterfall and you can then reasonably easily the key points of difference.

 

A directivity sonogram or speclab captures are some other examples (albeit 2D examples) of such 3d plots.

 

Also who doesn't like 3d charts? We expect an IRIX interface here :D

 

Hmmm...I have to admit a 3d plot with frequency, SPL and THD would be great. Not sure how feasible this is but it can't hurt to poke around.

 

 

 

 

Hi Ricci.  This sounds awesome!  I'd be glad to help contribute my mathematical and acoustic knowledge to this problem.  I practically do conversions between dB SPL and Pa in my sleep.  However, I'm not exactly certain what what you are trying to achieve with your formula.  Should efficiency enter into the picture?  Without considering efficiency and assuming unlimited amp power, most sealed box systems are likely to perform almost the same, especially at the bottom end where everything is excursion limited anyway.  I'm thinking you may be more interested in efficiency per liter.  But really, I guess it depends on the application.

 

I really like your ideas for the distortion graphs and see no issues with your approach.  If your plotting system supports B-spline interpolation, you may be able to get curves that are more smooth and more accurate at the same time, but you'll want to verify that the fits make sense.  For example, some kinds of spline fits assume that either the data points or their derivatives (i.e. slopes) go to zero at the ends of the graph, so that's something to watch out for.

 

By the way, the technical term for filling in between data points as you are doing here is called interpolation.  Extrapolation refers instead to extending curves beyond the last point in a data set.  As you can imagine, extrapolated figures are a lot more likely to give erroneous results, which is why they are typically frowned upon.

 

Haha...You got me on a technicality there. Good catch. Say it with me, interpolation...

It should work well. Depending on how the data would be uploaded and displayed it shouldn't be difficult to smooth the data and curve fit it better. I don't think that'll be a big issue.  

 

About the SPL / cubic liter / frequency metric.

I believe it makes sense to limit the bandwidths to those that are commonly driver or port displacement limited and also affected by Hoffman's iron law. I'm thinking 40Hz and lower or possibly 31.5Hz and lower. The basic idea is some sort of metric that gauges how hard each system is pushing up against HIL in the deep bass. At the upper bass frequencies where the limitation is often amplifier related rather than displacement a different formula would be needed.



#10 3ll3d00d

3ll3d00d

    Ultra Member

  • Members
  • PipPipPipPipPipPip
  • 413 posts

Posted 28 December 2015 - 06:47 PM

Hmmm...I have to admit a 3d plot with frequency, SPL and THD would be great. Not sure how feasible this is but it can't hurt to poke around.

A quick google says highcharts doesn't do such plots so I can give you an example using gnuplot if you want to generate some static images. 



#11 shadyJ

shadyJ

    Power Member

  • Members
  • PipPipPipPip
  • 82 posts

Posted 28 December 2015 - 07:52 PM

Actually there is study that has been put into the thresholds established as the limits for CEA-2010. There are various papers available on harmonic distortion, masking and bass frequencies. It is by no means perfect as has been discussed a few times. There is no doubt that even large amounts of distortion up to 15-20% o the first 2 harmonics are not that easily noticed in the bass range. Especially with complex content with upper frequency masking. Mechanical or non harmonic noises are much more distracting to me personally. Still some sort of standard is better than nothing. On that we agree. I'm not going to get to the point of comparing individual harmonics. That would be a lot of work and practically difficult with the current tools available without investing significant time and resources. At the end of the day I don't know if it would be that useful either. However I still feel that distortion is a useful metric to look at and compare and my preference is for as low of distortion as possible.

The research that CEA-2010-A is based on is D.E.L. Shorter's 1949 study for BBC "An Investigation Into Non-linear Distortion First Interim Report" and Eric Benjamin's and Louis Fielder's 1988 study "Subwoofer performance for accurate reproduction of music." If you go over those papers, you can see that they are not at all adequate for understanding the audibility of distortion in bass frequencies. They were intended as a 'starter' studies to prompt further research, and they were not intended to be the final word on that subject. CEA-2010-A sort of uses Shorter's suggested weighting scheme for its thresholds (obtained by multiplying the nth harmonic by n2/4), but that is largely guesswork as to how humans might be sensitive to distortion in low frequencies- it is not based on anything like comprehensive testing. Benjamin and Fielder's research only used 4 subjects with some Stax headphones if I remember right. Floyd Toole had some interesting comments on it, but it is really just a 'nascent' study for that subject which was not followed up with any serious research. Anyway, I am not trying to be argumentative, I am just saying the research in this area is very sparse, and CEA-2010-A distortion thresholds are essentially based on guesswork as far as I can see, not hard evidence.

 

The idea of the 3d plot of THD vs SPL would be pretty great. That would be a nice visual depiction of the way drivers misbehave at the edges of their performance. Such a plot might be interesting with CEA data too. If I remember right, Ed Mullen has remarked that CEA-2010 fail thresholds do give a pretty good indication of when the driver has reached its xmax. In your experience, would you say that CEA-2010 thresholds tend to correspond to xmax limits?


  • dbstedt and dgage like this

#12 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 28 December 2015 - 09:23 PM

Shady,

yes I would say that it does seem to track the "useful" excursion limits or output limits pretty closely. Sometimes it seems to limit a bit early and other times the subwoofer may have changed in tone or character noticeably indicating some type of audible distortion or non linearity. In general though it seems close so I'd agree with Ed. There are a few other papers out there if you really dig but none of them is what I'd call comprehensive. Really with this type of subjective data, which could vary drastically from individual to individual, how could there be something comprehensive?

 

BTW. It's been planned to remove the CEA-2010 label from everything here for a while, for a few reasons which I don't feel like going into. It's going to take a LOT of work so I haven't done it yet but it will happen. The testing or data, won't change at all, just the labeling. 



#13 shadyJ

shadyJ

    Power Member

  • Members
  • PipPipPipPip
  • 82 posts

Posted 28 December 2015 - 11:21 PM

Whatever other research exists in the audibility of distortion, it must really be obscure, at least in regards to harmonic distortion. When I asked Charlie Hughes, who is chairman of the CEA standards committee, what research was used to establish the distortion thresholds in CEA-2010-A, he said the Shorter paper and the Fielder/Benjamin paper. As for relabeling from CEA-2010, whatever the reason, I am sure they are good reasons. I am looking forward to where ever you take the data-bass website, I know it will be in a sensible direction. 



#14 SME

SME

    Super Bass Overlord

  • Members
  • PipPipPipPipPipPipPipPip
  • 966 posts

Posted 28 December 2015 - 11:34 PM

About the SPL / cubic liter / frequency metric.

I believe it makes sense to limit the bandwidths to those that are commonly driver or port displacement limited and also affected by Hoffman's iron law. I'm thinking 40Hz and lower or possibly 31.5Hz and lower. The basic idea is some sort of metric that gauges how hard each system is pushing up against HIL in the deep bass. At the upper bass frequencies where the limitation is often amplifier related rather than displacement a different formula would be needed.

 

Indeed, this is where I thought you were going after all.  Consider this.  Take a driver, put it in a box that's rather small relative to the compliance of the driver, and play it at a frequency well below its sealed box resonance.  In the limit of low frequency vs. resonance and small box vs. driver compliance, the efficiency will exhibit a 3 dB increase per doubling of box size.  (This is true as long as these stated limits hold.)  This is full-on HIL territory.  Note that a drop in output of 3 dB can be compensated for by doubling power input (assuming you have the watts and excursion to spare).  Likewise, a drop in output of 3 dB, while representing a change in level of 0.707X (sqrt of 2) still represents a sound power change of 1/2 X.

 

So I think the figure you want to quote is acoustic efficiency per liter of box (in dB/W/L perhaps) in the limit stated above.  Note that sound power can be quoted in dB just like level can be.  This figure will vary the least between boxes of different sizes, so long as the limits hold true, i.e., small box relative to woofer compliance, frequency low relative to resonance.  The best measurement of this figure will tend to be found when the box is made as small as possible.  It would probably also be good to clarify to the reader how that number is most appropriately used.  In other words, don't do an I.B. mount, and expect to see extraordinary efficiency, since at that point the HIL limits don't apply and it's the woofer suspension that determines the deep bass efficiency.

 

What I expect we will see is that the drivers with the highest B*L/Re will have the best numbers here.

 

Edit: The above is wrong.  Doubling the number of liters will not double the acoustic output *level* in dB.  It will double the absolute acoustic output instead.  So I guess you really do need to use an absolute acoustic power number instead of dB acoustic.  So the formula would look something like this (where V is volume in liters):

 

    <acoutic-power-output-per-L> = 10^(<measured-level-in-dB-@-1W> / 10) / V

 

You would need to instruct to users that they do the following conversion to estimate SPL versus box size:

 

    <SPL> = 20*log10(<acoutic-power-output-per-L> * V)

 

Edit 2: I probably got the bit about suspension stiffness versus box volume wrong.  Feel free to correct me.  Clearly, I need to study more.  ;)


Edited by SME, 29 December 2015 - 06:20 AM.


#15 MKtheater

MKtheater

    Bass Overlord

  • Members
  • PipPipPipPipPipPipPip
  • 521 posts

Posted 29 December 2015 - 03:52 PM

How about adding multiple subs to the comparison charts so we can see how many subs it takes to equal another? 



#16 BeastAudio

BeastAudio

    Ultra Member

  • Members
  • PipPipPipPipPipPip
  • 318 posts

Posted 29 December 2015 - 05:09 PM

Hey! That's a pretty sweet new way to quantify these subs performance. SPL per cubic foot. I love it!

 

I can't think of many (any?) suggestions for new tests, I'm sorry to say.

 

Maybe bring back spectral contamination tests but I believe those were originally nixed from DB tests because you said (paraphrasing) "too hard on a driver" of a test to do. IIRC.

 

Erm, that's all I can think of at the moment.

 

Ummm....

 

 

Ship all the systems to me when you're done testing? Yes. Yes. That sounds good.

 

 

Where are you going to put it all then?

 

7e7cecb152e8781ff8563b33a3e9bdadc10f3b5e


  • Infrasonic and Contrasseur like this

#17 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 29 December 2015 - 05:45 PM

SME I may have a chance to look at this briefly on a real computer today. Hopefully.

 

Not sure I follow on the bit with the users doing estimates? Do you mean for systems not tested officially at DB?



#18 Contrasseur

Contrasseur

    Power Member

  • Members
  • PipPipPipPip
  • 71 posts

Posted 30 December 2015 - 03:08 PM

How about efficiency vs frequency? Use the SPL values from the sweeps to calculate acoustic watts out, and divide V^2 by the impedance values to get electrical watts in. Make the X axis frequency and the Y axis acoustic/electrical watts. This will highlight the comparison between weak motors and extra-strong ones, and at which points in the bandwidth.

 

To make it even more direct-comparison friendly, normalize it to represent electrical watts needed for a certain spl, say 80db or whatever (given zero compression). Since signal processing is so ubiquitous (and often essential) these days, these charts will take unequalized frequency response out of the equation. If you want a given SPL, this chart shows you exactly how much juice a system needs to make it, and at what bandwidth. Now you can see that the 21IPAL can do 100Hz with .02W and the RE XXX needs .5W or whatever. Moving along the X axis you can see how that changes with frequency, so low FS drivers will really shine in the deep bass. All people need to see is the height of the curve, where lower is better. As clear of a comparison as one could have

 

This completely bypasses the problem of comparing a 3.5 Ohm DCR speaker to a 4 Ohm DCR speaker, especially considering neither of them will be 3.5 or 4 ohms in passband. Now instead of having those "basic response" charts to try and compare 1w/1m SPL, we can really see how they measure up in the one thing that they do, which is converting electrical power to acoustical power.



#19 Contrasseur

Contrasseur

    Power Member

  • Members
  • PipPipPipPip
  • 71 posts

Posted 30 December 2015 - 03:11 PM

Dividing this quantity by liter gives you a "Hoffman's Iron Quotient" graph. Now you can see how much systems balance low vs loud, and which ones are most "Hoffman optimized"



#20 Ricci

Ricci

    Super Bass Overlord

  • Moderators
  • 1,071 posts
  • LocationLouisville, KY

Posted 30 December 2015 - 04:40 PM

How about efficiency vs frequency? Use the SPL values from the sweeps to calculate acoustic watts out, and divide V^2 by the impedance values to get electrical watts in. Make the X axis frequency and the Y axis acoustic/electrical watts. This will highlight the comparison between weak motors and extra-strong ones, and at which points in the bandwidth.

 

There are a number of variable factors involved. This type of thing is shown in enclosure simulation software often and it could be backed out of the data I have collected.  A single representation of the calculated efficiency using the impedance curve and the voltage sensitivity could be done relatively easily. What I'd really like to do is a graph showing the results for each system at low, medium and high power though.

 

 

 

To make it even more direct-comparison friendly, normalize it to represent electrical watts needed for a certain spl, say 80db or whatever (given zero compression). Since signal processing is so ubiquitous (and often essential) these days, these charts will take unequalized frequency response out of the equation. If you want a given SPL, this chart shows you exactly how much juice a system needs to make it, and at what bandwidth. Now you can see that the 21IPAL can do 100Hz with .02W and the RE XXX needs .5W or whatever. Moving along the X axis you can see how that changes with frequency, so low FS drivers will really shine in the deep bass. All people need to see is the height of the curve, where lower is better. As clear of a comparison as one could have

 

This completely bypasses the problem of comparing a 3.5 Ohm DCR speaker to a 4 Ohm DCR speaker, especially considering neither of them will be 3.5 or 4 ohms in passband. Now instead of having those "basic response" charts to try and compare 1w/1m SPL, we can really see how they measure up in the one thing that they do, which is converting electrical power to acoustical power.

 

I do like the idea of this.

 

The issues that occur at higher power levels would be difficult to account for without the impedance data to go with it. The impedance curve can change dramatically which greatly affects the power calculation. In the deep bass most of the systems are limited by driver excursion and output compression unrelated to thermal effects. I personally do not put much emphasis on small signal efficiency for bass systems for these reasons. I'd rather look at efficiency at much higher input power. Doesn't have to be right at the limits of the system but something between one quarter to half of the drivers continuous thermal rating seems reasonable for a "medium" or real world use scenario.

 

Really what is needed is the high power impedance curves. I'd like to get back to testing high power impedance but it is a very intensive test, with a different setup, which takes a long time to complete. What would be great is to have the response shape and magnitude, the distortion and the impedance measurement for each applied voltage level. I'm only missing the impedance data. With all of that it is just a matter of using the data to provide whatever information is wanted.

 

The metric I was wanting to show is a little different than just showing the true efficiency. Yes HIL governs the deep bass efficiency of a system but that in turn also limits the theoretical maximum output. In the deep bass the maximum output is usually limited by the driver or vent displacement capabilities these days, rather than the system efficiency since huge amplifier power is cheap. I'm more interested in which systems have more output potential per unit of volume than which ones have more efficiency per unit of volume. Both are important and are linked but perhaps require separate metrics.






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users