Michael
You have shown what I believe that I always knew existed. That there can be a nonlinear interaction between the thermal response and the acoustic response, although you didn't actually show the acoustic link, only an electrical one. However if there is an electrical one then there has to be an acoustic one.
But, like all effects this has to be scaled into our everyday designs in terms of audibility. This is a very long ways off. To do that we need to get some real world numbers on these thermal constants for real drivers. Thats what I want you to take the data on. With that data I hope to be able to extract the required information from two different drivers and compare, perhaps through a simulation, the effects to see if they are audible on music in a controlled experiment.
You have shown what I believe that I always knew existed. That there can be a nonlinear interaction between the thermal response and the acoustic response, although you didn't actually show the acoustic link, only an electrical one. However if there is an electrical one then there has to be an acoustic one.
But, like all effects this has to be scaled into our everyday designs in terms of audibility. This is a very long ways off. To do that we need to get some real world numbers on these thermal constants for real drivers. Thats what I want you to take the data on. With that data I hope to be able to extract the required information from two different drivers and compare, perhaps through a simulation, the effects to see if they are audible on music in a controlled experiment.
Michael,
the nude VC as you call it, is that a VC out of the gap (like lying on your kitchen table) or is it a vc in the gap of a motor?
I follow this with great interest however seing that the 3rd order harmonic sinks down close to the stimuli signal as the frequency is increased I can not do a connection to nonlinear distortion (new spectral components) other than at very low frequencies.
At such low frequncies the nonlinearity from Bl, C and L of a bass driver swamps the thermal effect with a hundred times or so and that is not including teh fact that a bassdriver VC will stand 10-100 times more heat than a tweeter VC.
If my thinking is right here, and with the basis of the VC used in this test, the real effect would be even lower, say 60db (0.1%) below the third order distortion of a typical high performance driver.
/Peter
the nude VC as you call it, is that a VC out of the gap (like lying on your kitchen table) or is it a vc in the gap of a motor?
I follow this with great interest however seing that the 3rd order harmonic sinks down close to the stimuli signal as the frequency is increased I can not do a connection to nonlinear distortion (new spectral components) other than at very low frequencies.
At such low frequncies the nonlinearity from Bl, C and L of a bass driver swamps the thermal effect with a hundred times or so and that is not including teh fact that a bassdriver VC will stand 10-100 times more heat than a tweeter VC.
If my thinking is right here, and with the basis of the VC used in this test, the real effect would be even lower, say 60db (0.1%) below the third order distortion of a typical high performance driver.
/Peter
To be clear on the issue it is this:
The thermal modulation will depend on the power of the signal. In music, this power is a strong function of the time. It is dynamic and the actualy power will have a "spectrum" that is indeed real and will tend to be very very low in frequency, on the order of seconds or lets say .1 Hz to maybe 30-50 Hz tops. Typical music signals are virtually devoid of these frequencies so the actual musical signal is not a source of this kind of effect directly.
BUT - and here is the issue - the thermal envelope of the signal COULD modulate the signal even if there are no actual signals at those frequencies. This is the effect that I am interested in and it does not require actual frequency content at these low frequencies only actual envelope content, which is most certainly going to occur.
How audible is this? I have no idea. But I have looked and looked for reasons why larger well designed speakers always sound more "dynamic" than smaller ones, even if they are well designed. Thus there must be something inherent in bigger versus smaller that causes this. From what tests that I have done there is an enormous difference in the way smaller drivers handle thermal aspects This has led me to wonder if the thermal aspects could be related to something audible.
This requires a test, but a test requires a hypothesis and a test design that can issolate the hypothesized effects. This is not easy to do. If I can create a model that can be used to manipulate this effect then this model could be used in a test.
Actually the model is quite simple as Michael has shown. What is NOT simple is what are the real parameters for this model? To use unreal or arbitrary parameters could easily be a waste of time.
So I need to determine the real set of parameters for real devices. I will do this with a noise signal whose envelope is modulate below the signal content. In this way I can determine how the system reacts to the envelope seperately from how it reacts to the signal. With this data I can build a realistic model of the effect and use it to simulate the effect with real music. Then a blind listening test can be used to determnine if this calibrated real effect is in fact audible or not.
A lot of work, but a worthy endeavor.
The thermal modulation will depend on the power of the signal. In music, this power is a strong function of the time. It is dynamic and the actualy power will have a "spectrum" that is indeed real and will tend to be very very low in frequency, on the order of seconds or lets say .1 Hz to maybe 30-50 Hz tops. Typical music signals are virtually devoid of these frequencies so the actual musical signal is not a source of this kind of effect directly.
BUT - and here is the issue - the thermal envelope of the signal COULD modulate the signal even if there are no actual signals at those frequencies. This is the effect that I am interested in and it does not require actual frequency content at these low frequencies only actual envelope content, which is most certainly going to occur.
How audible is this? I have no idea. But I have looked and looked for reasons why larger well designed speakers always sound more "dynamic" than smaller ones, even if they are well designed. Thus there must be something inherent in bigger versus smaller that causes this. From what tests that I have done there is an enormous difference in the way smaller drivers handle thermal aspects This has led me to wonder if the thermal aspects could be related to something audible.
This requires a test, but a test requires a hypothesis and a test design that can issolate the hypothesized effects. This is not easy to do. If I can create a model that can be used to manipulate this effect then this model could be used in a test.
Actually the model is quite simple as Michael has shown. What is NOT simple is what are the real parameters for this model? To use unreal or arbitrary parameters could easily be a waste of time.
So I need to determine the real set of parameters for real devices. I will do this with a noise signal whose envelope is modulate below the signal content. In this way I can determine how the system reacts to the envelope seperately from how it reacts to the signal. With this data I can build a realistic model of the effect and use it to simulate the effect with real music. Then a blind listening test can be used to determnine if this calibrated real effect is in fact audible or not.
A lot of work, but a worthy endeavor.
Pan said:
Yes – like seen in my former posting:
http://www.diyaudio.com/forums/showthread.php?postid=1669678#post1669678
And *no*, pix wasn't taken at the kitchen table...
Pan said:
I'm happy to see that my thermal model and my concept of "Thermal Distortion" seems to gain some acceptance.
Basically I agree to what you say, but you also have higher inputs than the 1W shown here.
And as Earl already has pointed out - you *can* produce the power injection envelope also by (low frequency) amplitude modulated HF.
To put things into perspective I've performed a quick measurement:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
What we see above – for reference – is how a 1kHz tone 100% *amplitude* modulated by 3 Hz appears like in the frequency domain (measurements taken on the 20ppm power-resistor described above)
Basically we see side bands at 991Hz / 994Hz / 997Hz and 1003Hz / 1006Hz / 1009Hz and no fundamental (1000Hz) as its completely suppressed here.
Also we see a low peak at 6Hz = at double the modulation frequency (and its harmonics).
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
What we see above – with the nude compression driver VC measured - is how Thermal Distortion *increases* the side bands at (988Hz?) / 994Hz and 1006Hz (1012?) only.
The increase is basically of the same value as shown with the 3Hz single tone some postings back.
The low peak at 6Hz (and its harmonics) is not affected and basically looks the same as with the measurement of the 20ppm temp-co power-resistor
Regarding audibility "in general" - lets see what Earl may find out in controlled tests.
As for me - TD is kind of fact, since I started to think about it and wrote my paper. I have to thank Lynn for triggering my interest in this.
The so called "non- linear" part of TD is just *one* impact to audio performance – but even more interesting the deeper we dig into it.
The kind of strange effects of amplitude modulation - in the frequency domain - I already have outlined with acoustical measurements at the topic of Back Diaphragm Mirror Distortion
http://www.diyaudio.com/forums/showthread.php?postid=1648180#post1648180
http://members.aon.at/kinotechnik/diyaudio/diy_audio/BDMD/Introducing_BDMD.htm
Michael
Or is it just a cry for "I'm hungry for more!"?
Michael,
I think Pan makes some very good points considering other nonlinearities. I think those are some aspects that could be modelled. If after modelling those there is a significant coherence between the measured and modeled aspects, then there might be TD involved.
Note that in your measurements, as your test frequency increases, the harmonics decreases in level. Could this also be due to less excursion bring the driver into the linear range?
I think Pan makes some very good points considering other nonlinearities. I think those are some aspects that could be modelled. If after modelling those there is a significant coherence between the measured and modeled aspects, then there might be TD involved.
Note that in your measurements, as your test frequency increases, the harmonics decreases in level. Could this also be due to less excursion bring the driver into the linear range?
Michael,
in the middle of your post #72 there are two pictures (no 5 and 6 I think) that shows the thermal behaviour of the compression driver overlayed on top of the Seas driver.
Clearly the Seas driver has less thermal compression here right?
But it is hard (if at all) to get a grip on the relative behaviour since 10W was pushed into the compression driver and only 5W to the Seas driver, also the compression driver would be seriously handicapped by being "nude" if that really means out of the gap.
Am I missing something or do I have a clear understanding of this?
/Peter
in the middle of your post #72 there are two pictures (no 5 and 6 I think) that shows the thermal behaviour of the compression driver overlayed on top of the Seas driver.
Clearly the Seas driver has less thermal compression here right?
But it is hard (if at all) to get a grip on the relative behaviour since 10W was pushed into the compression driver and only 5W to the Seas driver, also the compression driver would be seriously handicapped by being "nude" if that really means out of the gap.
Am I missing something or do I have a clear understanding of this?
/Peter
gedlee said:
It shows the "magic" of convolution – and that your wildest dreams (possibly) have come true, Earl
Lets have a look at a new simulation to make things more clear (and to correct some figures I've stated wrong too)
An externally hosted image should be here but it was not working when we last tested it.
What we can see above is a simulation from my enhanced thermal model - set up to meet the measurements of my last posting
Stimulus signal for both – measurement and simulation - are a 1000Hz signal multiplied by (mixed with) a 3Hz signal.
The amplitude modulated signal is displayed in the time and the frequency domain. In the time domain plot we obviously can see the 1000Hz come and go six times a second.
In the frequency domain two traces are shown
- the input signal (RED)
- the thermally distorted signal (BLUE)
BLUE and RED trace are scaled to have equal peak value at 997Hz and 1003Hz.
Clearly Thermal Distortion creates additional side bands by itself (if we trust my model - and my measurement)
The side bands created by Thermal Distortion are located at 991Hz and 1009Hz respectively.
(the frequency values I sadly have stated incorrect in my last posting)
An externally hosted image should be here but it was not working when we last tested it.
For clarity and comparison again the measurement of the nude no-name compressor driver voice coil
This traces (BLUE / RED) should look basically the same in simulation and from measurement.
We can spot some differences, witch – I guess – are an overlay of side bands created by non-linearity of the measurement equipment (amp and soundcard) and also FFT windowing artefacts.
Basically I have performed the amplitude modulation measurements (and the subsequent simulation) to show, how convolution is able to transpose effects - basically happening out of the audio band – right into the audio band.
If you break down to its simplest form - what you said that you would like to try to capture with your especially composed signals – I think its basically the same here – approached from an other side.
Did I made myself clear now, Earl?
Michael
PS
( "En passant" I also have shown how to create "sub harmonics". Not having see the data from Harman guys, I might be wrong of course – and even *if* - not sure if "sub harmonics" is the right term here
When I changed the modulation frequency from 3Hz to 30Hz - guess what – the low frequency peak jumped up at 60Hz – didn't show this nice little experiment – already many plots around.
BTW - from *acoustical* measurements – these 60Hz can come up to –50dB with respect to the "fundamental" – quite impressive SPL for a compression driver attached to a horn of roughly 1kHz cut off ! )
soongsc said:
...Or is it just a cry for "I'm hungry for more!"?
soongsc said:
I might be able to model *all* effects of a loudspeaker (?) – but for sure not in that life alone (!)
You will have to wait for some of my further reincarnations – I hope they never will happen though.
Besides that – it already has been done to some extent (not for Thermal Distortion though AFAIK) - just search the web.
soongsc said:
Your observation of the overlaid thermal low pass filter is correct – but no – until now, ALL measurements shown *on the level of wave form deformation* were taken with the nude VC - taken out of the gap.
Michael
Pan said:
Sorry Peter - Pix shown in posting #72 were for comparing qualitative behaviour of the decay not to compare for the FIGURES of TD .
As you said, TD is hardly comparable directly from that plots.
If we would like to compare Seas Millennium with the compression driver more directly, we could do a new 10kHz / 10W measurement for both
To make visual comparing easier, I already normalised current fall to begin at "1":
An externally hosted image should be here but it was not working when we last tested it.
Above: Millennium 10W at 10kHz power injection 100sec
An externally hosted image should be here but it was not working when we last tested it.
Above: compression driver 10W at 10kHz power injection 100sec
This time the compression driver is measured *completed* (in contrast to the nude VC until now)
By mere visual comparison we can see that
- *Voice coil* time constant isn't that different
- TD is considerably better for the (ferro fluid cooled) Seas Millennium
After - lets say 10sec - you have reached roughly 10% distortion with the Seas and roughly 23% with the compression driver – at the *same power injection level*.
Keep in mind, that the higher sensitivity of the compression driver might balance its worse TD in real applications.
Michael
Pan said:
Yes - roughly the difference of 23% distortion to 35% distortion for the compression driver - if you look at my last posting and the plots taken earlier.
You shouldn't try the experiment of taking out of the gap the voice coil of the Millennium and measure at 10W – This is a very delicate VC and it would burn down immediately.
Injecting pure thermal 10W is pretty much right at the limit for this driver if *assembled* !
Michael
mige0 said:
Thanks for response Michael.
This is what's really interesting to me, how the "real life" results turn out.
Do you know the sensitivity of the compresssion driver?
You shouldn't try the experiment of taking out of the gap the voice coil of the Millennium and measure at 10W – This is a very delicate VC and it would burn down immediately.
Injecting pure thermal 10W is pretty much right at the limit for this driver if *assembled* !
Michael
Hehe, allready done that. Not with the Millenium but with another alu-dome Seas tweeter. I don't remember what we put into it out of the gap but it started to smoke for sure.. my friend was fast to pull our the wires so we saved the poor VC.
In the gap a good dome tweeter can easily take 50W continous (sine) for 10 seconds without damage and only marginal thermal compression. We even did that with a SS9700 which don't use ferrofluid, but it showed about the double thermal compression than the Seas alu/ferro tweeter.
Actually these tweeters can handle a lot more in short bursts. Good dome tweeters can take 10ms of 1000-3000W.
/Peter
mige0 said:
Ok!
Tough people you are!
The things you do for science.
We took notes, I'll se if I can find them. Series resistor, 0.1ohm, digital scope tracking the current drop over the resistor.
We are planning a bigger study on tweeter thermal and nonlinear distortion.
Our conclusion was that thermal compression is a non issue at typical domestic listening levels and high quality dome tweeters.
Next time we will measure at higher levels and also perform acoustic measurements on high power short burts. Both for checking the total compression and also nonlinear distortion.
/Peter
Looking at all the analysis and test data up to now, it seems whether the effects would be of concern is questionable. But I really like the insight and appreciate what Michael has been doing.
It would be interesting to see how the response would be if the power level is altered randomly over a period of 100 sec.
It would be interesting to see how the response would be if the power level is altered randomly over a period of 100 sec.
Pan said:
The things you do for science.
/Peter
Pan said:
Wow - interesting. What exactly triggered *your* interest in that topic – if I may ask?
Did you store DSO plots?
mige0 said:
I should add to the list above, that there are also limitations in what the (current) thermal model mimics .
The thermal model mimics an *absolutely* constant power injection as is set by the input signal.
This is not quite true.
As the resistance goes up due to temperature rise, the injected power goes down as we feed the circuit with - roughly - constant *voltage*. "Roughly" because there is a shunt inserted in the circuit – and the amp also has finite source resistance as have cables and connectors.
The measurement set-up would become extremely complicated to ensure constant *power* injection in its very meaning.
What's of no big deal as long as we are measuring time constants and current / SPL drop due to heat up – the amplitude envelope of steady signals or of burst signals in other words - may significantly alter the *exact* pattern of wave shape deformation predicted by simulation.
We filter out the *amplitude* impact (quantity) of series shunt and source impedance by measuring current *and* voltage across the DUT and it shows us slightly different results as predicted by simulation.
But the *qualitative* behaviour - in this respect - isn't modelled *precisely* yet – at least there is room for further improvement and refinement.
For the time being – the question if we should rather trust in measurement or in simulation here remains open to some degree.
Michael
Michael,
a number of years ago I got tired in only collecting info from others and listening to all the confusion and guesswork.
I doun't doubt that there are situations where dome tweeters will have serious problems, that's obvious. The question for me is where is the limit.
For a high power output speaker in a big room a good tweeter can still be fine if crossed at 3k or so and used with a waveguide.
For smaller rooms (typically used in Europe) and normal playback levels a 2k crossing on a flat baffle will not be a problem IMO.
What you do is very interesting though and I was thinking on the constant voltage vs. constant power thing. To me it makes sense to measure (and simulate) with constant voltage since most amp/speaker interfaces is about voltage drive. Or am I thinking wrong here?
(hang on, i will contradict myself below..
Typically amp source impedance and cable impedance is out of the equation but the crossover isn't and that is an interesting thing.
Power compression is worst when there is a voltage drive and low impedance between the amp output and the voice coil, but what happens when we put coils and resistors in series with a driver is that the effect of the heating is decreased. Also the HD/IMD is typically decreased with a coil in series with a speaker driver, at least if there's significant inductance in the area of interest.
I have not measured these effects myself but they are there and others have looked into it.
To sum up the confusion above, examine a raw driver would be best to use constant voltage even though the driver will be used with a passive x-over which skew things towards current drive.
/Peter
a number of years ago I got tired in only collecting info from others and listening to all the confusion and guesswork.
I doun't doubt that there are situations where dome tweeters will have serious problems, that's obvious. The question for me is where is the limit.
For a high power output speaker in a big room a good tweeter can still be fine if crossed at 3k or so and used with a waveguide.
For smaller rooms (typically used in Europe) and normal playback levels a 2k crossing on a flat baffle will not be a problem IMO.
What you do is very interesting though and I was thinking on the constant voltage vs. constant power thing. To me it makes sense to measure (and simulate) with constant voltage since most amp/speaker interfaces is about voltage drive. Or am I thinking wrong here?
(hang on, i will contradict myself below..
Typically amp source impedance and cable impedance is out of the equation but the crossover isn't and that is an interesting thing.
Power compression is worst when there is a voltage drive and low impedance between the amp output and the voice coil, but what happens when we put coils and resistors in series with a driver is that the effect of the heating is decreased. Also the HD/IMD is typically decreased with a coil in series with a speaker driver, at least if there's significant inductance in the area of interest.
I have not measured these effects myself but they are there and others have looked into it.
To sum up the confusion above, examine a raw driver would be best to use constant voltage even though the driver will be used with a passive x-over which skew things towards current drive.
/Peter
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.