gedlee said:
Sorry no more specs available of that 1.5" no-name compression driver - it only serves for the prove of concept here.
nickmckinney said:
Good point - I was aware that aluminium formers have its pro's and cons especially for the glue points to spider and cone.
If you don't intend it for 7/24 PA usage it might be not that critically - but sure - that would be no good advice for a manufacturer like you.
Michael
mige0 said:
Actually - the most important spec now *is* available for that driver (at least for the nude VC and at low precision)
TD (10W/10kHz) = 0.65 / 3 / 1.01
Meaning, for a measurement taken at 10W power injection at 10kHz there is
- a decrease of SPL of 35% at thermal equilibrium
- a 3 second time constant and
- a 1% overshot at thermal equilibrium
We have to be aware that TD figures may differ significantly due to the complex cooling mechanism for different values of power injection and for different frequencies as well.
Hence - I think, condition of measurement should be part of a Thermal Distortion figure to keep things clear.
First figure "SPL decrease"
- is easy to estimate and gives an intuitive handle on how deep SPL falls at infinitely applying measurement condition. Its always lower than "1"
Second figure "time constant in seconds"
- is easy to measure and gives us kind of intuitive handle on the subject in a more real time frame (than "eternity" for reaching equilibrium )
Third figure "maximum overshot"
- is easy to measure too and also kind of intuitive handle on the subject
Its obviously in the range between "1" and "2"
For a thermal mass tending towards zero there is 100% of overshot - pushing TD to the double of what usual "power compression" figures tell us.
For an infinite thermal VC mass there would be no overshot occurring at all – absolutely regardless of power injection magnitude
Does this make sense for anybody?
the last (?) boring simu:
An externally hosted image should be here but it was not working when we last tested it.
On log scale we can see most easily that injecting 10 times the power results in
10 times the Thermal Distortion.
This makes it easy to convert and compare TD figures / measurements taken at different power injection levels – no need to do any "normalising".
Michael
If I haven't made any gross mistake I guess the topic of Thermal Distortion is basically covered now .
Regarding NTD Earl, – its already in the TD numbers and once someone wraps (convolutes) the RMS around the input plus including the thermal low pass filter we actually could calculate NTD figures from (L)TD.
Michael
Regarding NTD Earl, – its already in the TD numbers and once someone wraps (convolutes) the RMS around the input plus including the thermal low pass filter we actually could calculate NTD figures from (L)TD.
Michael
soongsc said:
They are spread over my past posting - Soongcs.
I'll be back soon with more precise measurements (slightly correcting figures given so far) and also with a pretty interesting comparison to quite a different tweeter.
The SEAS Excel Millenium T25CF002 –
It is used in NAO and ORION and many commercial speakers - is well documented and many in the DIY community love 'em (me too)
I'm currently working on a measurement tutorial to outline measurement procedure and calculation of the TD figures in more detail, taking the Millenium as example.
Michael
Are those actually thermal related distortions? Or are they just normal distortion measurements. What I only see in this thread seem to be just amplitude change over time possibly due to resistance change. The rate of change which really cannot be converted to distortion levels of any significance.mige0 said:
soongsc said:
I'm still not sure of what benefit the distinction between "normal distortion" and "real distortion" could be – besides it seems to be a mental concept more usual in other areas than mine.
The same question was under discussion with Earl some postings back.
The most significant distinction between non-linear and linear "distortion" – as far as I have found out – is, that distortion from linear systems usually don't show side bands in the frequency domain.
I'm really not specialised in this detached-high-level-professors-tech-speech (I quite often found more confusing than clarifying by smearing the very roots of issues)
- but my findings are that *pure* amplitude modulation can obviously create side bands as well
http://www.diyaudio.com/forums/showthread.php?postid=1648180#post1648180
http://members.aon.at/kinotechnik/diyaudio/dipol/space/BDMD/Introducing BDMD.pdf
so for me this distinction is a rather academic one anyway.
Besides that – I already have outlined that the NTD figures are fully contained in my TD figures and John Kreskovsky already has presented a quick first step math for the proof.
As for " levels of any significance " – well this again is the distinction between quality and quantity.
I'm already happy having established a more detailed view and some useful and clarifying explanations on the thermal behaviour of speakers than I have found anywhere else.
Michael
When we start talking about source of distortion, then the way distortion measurement is conducted is very important to determine that analysis can show distortion figures that in some way relate with measured distortion. For example, if one measures distortion while the coil is fairly cool and compare figures throughout the process of rising temperature, then if thermal related distortion should show a relationship with temperature or rate-of-change of temperature.
After having
- calculated thermal effects of voice coils from plain physical properties
- spotted and pointed out some common mental mistakes regarding thermal theory
- presented a equivalent electric schematic to model the thermal effects in speakers
- proven the thermal modelling for speakers by measurements -
I now felt confident to set up an extended simulation to model signal damage due to thermal effects in speakers
Its quite tricky to set up a model to simulate TD on the level of wave shape distortion.
Basically I took the thermal model validated before and arranged additional blocks to mimic the convolution of the undistorted signal with the thermal effects plus adding some normalising and defining global variables to simplify evaluation.
The three traces show how wave shape gets increasingly distorted when varying critical constants.
What we can see here is, that there is considerable damage of the signal *if* we exaggerate the effect heavily.
Kind of surprise is that we clearly see that signal distortion seems to be lowest at the very beginning of voice coil heat up – at least visually .
We already can see some very interesting effects - but before jumping into speculations I'd prefer to validate my extended thermal model by some measurements first
Sure – you'll be save and results should definitely be comparable if you follow the procedure I am working on
Michael
- calculated thermal effects of voice coils from plain physical properties
- spotted and pointed out some common mental mistakes regarding thermal theory
- presented a equivalent electric schematic to model the thermal effects in speakers
- proven the thermal modelling for speakers by measurements -
I now felt confident to set up an extended simulation to model signal damage due to thermal effects in speakers
Its quite tricky to set up a model to simulate TD on the level of wave shape distortion.
Basically I took the thermal model validated before and arranged additional blocks to mimic the convolution of the undistorted signal with the thermal effects plus adding some normalising and defining global variables to simplify evaluation.
An externally hosted image should be here but it was not working when we last tested it.
The three traces show how wave shape gets increasingly distorted when varying critical constants.
What we can see here is, that there is considerable damage of the signal *if* we exaggerate the effect heavily.
Kind of surprise is that we clearly see that signal distortion seems to be lowest at the very beginning of voice coil heat up – at least visually .
We already can see some very interesting effects - but before jumping into speculations I'd prefer to validate my extended thermal model by some measurements first
soongsc said:
Sure – you'll be save and results should definitely be comparable if you follow the procedure I am working on
Michael
gedlee said:
Yes, I knew -
– its the juicy stuff - not that *I* haven't been interested but its harder to - get the thermal model straight
- do the proof of concept on a qualitative level measurement (I sadly have failed in my first attempt to nail it)
- transpose conclusions and generalisations to real world speakers
- validate thermal modelling under real world conditions
all of the above I already have presented for the amplitude envelope "medium term" distortion.
With the next measurements of the Millenium tweeter we will see that there are basically three - quite different - effects happen from thermal effects.
All of them can be fully characterised by my Thermal Distortion figures as proposed.
For the third effect - not shown yet - there will be the need to refine the thermal model and TD figures as well.
Michael
To come closer to "the real thing" I show Thermal Distortion measurement of the SEAS Excel Millennium T25CF002 - the one found in NAO and ORION and many many other DIY and commercial projects.
The Seas Millennium did set kind of standard for dome tweeters some years ago and it might be useful to choose this tweeter to have a common basis to compare with.
In addition to that, this tweeter allows to spot a thermal behaviour not yet shown in this thread.
What we see above is the measurement of the Millennium at 5W power injection at 10kHz for 10 seconds
What we see above are five traces of 0.25, 0.5, 1, 2, 4 seconds time constant – scaled to visually match the measurement before.
Clearly the time constant of roughly 1 sec is almost congruent to the amplitude envelope of our 10sec measurement.
But at a closer look we discover that the tail of the measurement does not go asymptotically towards a certain value as predicted by simulation.
To do some further investigation we take a measurement of 100sec rather than 10sec only
Ups – something new here
What we see in the extended measurement of the Seas Millennium above is that there is a heavy thermal mass overlaying the picture,
It introduces a second time constant, much longer than the time constant of the voice coil.
Presumably this is the lumped time constant of the whole motor structure surrounding the voice coil.
Going one step back and doing the same extended measurement for the nude VC of the no-name compression driver reveals –
- nothing !
Even if we zoom in to the same scale as for the Seas Millennium there is obviously no heavy second thermal mass involved
To keep up with reality, it seems to be beneficial to expand the current thermal model to cover this effect as well.
Basically we can add another branch of thermal mass represented by R and C below the existing VC thermal mass .
Playing around to get the variables right is time consuming but in the end my thermal model proves to be both, easily expandable and very close to "the real thing"
Now we already have two effects of Thermal Distortion with different impacts isolated
– voice coil time constant affecting the "medium term" amplitude distortion
– motor time constant affecting the "long term" amplitude distortion
The third fundamental effect of Thermal Distortion - the "shot term" amplitude distortion is harder to validate by measurement.
Based on the enhanced thermal model presented in my last posting we happily know where to search for this kind of Thermal Distortion (*if* it proves to be correct ).
Both time constant and thermal mass have to be extremely low to grasp the "short term" amplitude distortion facet of Thermal Distortion *in the time domain*.
Below my attempt to get a qualitative prove of concept for the enhanced thermal model I set up to simulate TD on the level of wave form distortion.
What we see above is how a Tungsten bulb varies its current at heat up.
The wave form correlation to my simulation is striking – no?
Sure a Tungsten filament isn't exactly a voice coil – but as I already have demonstrated - a VC taken off the gap behaves basically the same as in the gap – not taking TD long term effects and different cooling into account of course.
There are some weird effects – possibly caused by the extreme temperature range involved (its lovely to watch the light come and go) and some expected non-linearity of the material constants – that make this measurements sort of preliminary.
If the proof holds true for real world speakers we will be faced with an additional source of odd order harmonics in loudspeakers as simulation tells us:
But stay cool - given the measurements of the Tungsten bulb
- were taken at a extreme low frequency of 4Hz
- at extreme temperature increase
- and the short term effect of Thermal Distortion decreases by 20dB each decade due to the thermal low pass function of the VC time constant
– well - we'll possibly facing really hard times proving this facet of TD *by measurement* in loudspeakers
Michael
The Seas Millennium did set kind of standard for dome tweeters some years ago and it might be useful to choose this tweeter to have a common basis to compare with.
In addition to that, this tweeter allows to spot a thermal behaviour not yet shown in this thread.
An externally hosted image should be here but it was not working when we last tested it.
What we see above is the measurement of the Millennium at 5W power injection at 10kHz for 10 seconds
An externally hosted image should be here but it was not working when we last tested it.
What we see above are five traces of 0.25, 0.5, 1, 2, 4 seconds time constant – scaled to visually match the measurement before.
Clearly the time constant of roughly 1 sec is almost congruent to the amplitude envelope of our 10sec measurement.
But at a closer look we discover that the tail of the measurement does not go asymptotically towards a certain value as predicted by simulation.
To do some further investigation we take a measurement of 100sec rather than 10sec only
An externally hosted image should be here but it was not working when we last tested it.
Ups – something new here
What we see in the extended measurement of the Seas Millennium above is that there is a heavy thermal mass overlaying the picture,
It introduces a second time constant, much longer than the time constant of the voice coil.
Presumably this is the lumped time constant of the whole motor structure surrounding the voice coil.
Going one step back and doing the same extended measurement for the nude VC of the no-name compression driver reveals –
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.
- nothing !
Even if we zoom in to the same scale as for the Seas Millennium there is obviously no heavy second thermal mass involved
To keep up with reality, it seems to be beneficial to expand the current thermal model to cover this effect as well.
An externally hosted image should be here but it was not working when we last tested it.
Basically we can add another branch of thermal mass represented by R and C below the existing VC thermal mass .
Playing around to get the variables right is time consuming but in the end my thermal model proves to be both, easily expandable and very close to "the real thing"
Now we already have two effects of Thermal Distortion with different impacts isolated
– voice coil time constant affecting the "medium term" amplitude distortion
– motor time constant affecting the "long term" amplitude distortion
The third fundamental effect of Thermal Distortion - the "shot term" amplitude distortion is harder to validate by measurement.
Based on the enhanced thermal model presented in my last posting we happily know where to search for this kind of Thermal Distortion (*if* it proves to be correct
).Both time constant and thermal mass have to be extremely low to grasp the "short term" amplitude distortion facet of Thermal Distortion *in the time domain*.
Below my attempt to get a qualitative prove of concept for the enhanced thermal model I set up to simulate TD on the level of wave form distortion.
An externally hosted image should be here but it was not working when we last tested it.
What we see above is how a Tungsten bulb varies its current at heat up.
The wave form correlation to my simulation is striking – no?
Sure a Tungsten filament isn't exactly a voice coil – but as I already have demonstrated - a VC taken off the gap behaves basically the same as in the gap – not taking TD long term effects and different cooling into account of course.
There are some weird effects – possibly caused by the extreme temperature range involved (its lovely to watch the light come and go) and some expected non-linearity of the material constants – that make this measurements sort of preliminary.
If the proof holds true for real world speakers we will be faced with an additional source of odd order harmonics in loudspeakers as simulation tells us:
An externally hosted image should be here but it was not working when we last tested it.
But stay cool - given the measurements of the Tungsten bulb
- were taken at a extreme low frequency of 4Hz
- at extreme temperature increase
- and the short term effect of Thermal Distortion decreases by 20dB each decade due to the thermal low pass function of the VC time constant
– well - we'll possibly facing really hard times proving this facet of TD *by measurement* in loudspeakers
Michael
Michael
The dual time constant and everything that you have shown thus far is all well know and available in a number of AES papers. What is not quantified is any modulation and nonlinear effects of the thermal time constants on the acoustic output. Your data is making it look like this is not going to be an issue. Do you have a small VC woofer that you could test?
If I sent you a signal would you record the voltage and current into a driver from this signal - two channels, voltage on one current on the other. If I had measurement data to process then I might be able to knock out some calculations because setting up the test is the bulk of the work and I don't have that kind of time.
The dual time constant and everything that you have shown thus far is all well know and available in a number of AES papers. What is not quantified is any modulation and nonlinear effects of the thermal time constants on the acoustic output. Your data is making it look like this is not going to be an issue. Do you have a small VC woofer that you could test?
If I sent you a signal would you record the voltage and current into a driver from this signal - two channels, voltage on one current on the other. If I had measurement data to process then I might be able to knock out some calculations because setting up the test is the bulk of the work and I don't have that kind of time.
gedlee said:
Thanks Earl, always good to know that I'm still walking on solid ground.
Anybody among those AES people who was capable to present a model allowing prediction of thermal behaviour of speakers at the level of wave shape distortion?
There definitely is the burden of conducting useful measurements and time and risk on DUT and equipment one has to invest.
I will come back with additional analysis regarding impacts on wave form distortion as soon I've fixed what ended up in smoke from last measurements (and there's life too)
Certainly - *all* my measurements were taken with one eye on the voltage across DUT.
But for the time being, it wouldn't have delivered any additional insights to show that plots as well but only loaded the postings with repetitive garbage.
soongsc said:
Yes, it might be more intuitively the way you suggest.
But the beauty of gaining insight by simulation is that we can analyse thermal effects *isolated* from any other effects happening in speakers (and there are a lot). With my enhanced thermal model analysis can be done at *any* time scale – up from wave form deformation down to long term motor heat up
Analysing the enhanced thermal model is also
– less risky and
– less time consuming
and results allow to draw a lot of conclusions and to state a bunch of useful generalisations even beyond the speaker topic here - some of which I already have outlined.
On top of that – Thermal Distortion on the level of wave form deformation can be seen as signal convolution with a low pass filter – dominated by the time constant of the voice coil - involved.
An externally hosted image should be here but it was not working when we last tested it.
What is shown above is how wave form deformation gets "less of an issue" the higher the frequencies involved – for the same VC
This low pass filter is set at very low frequencies – from mere physical properties of the voice coils used in our speaker – be it a woofer or a tweeter.
I was able to prove that we can measure thermal behaviour of a voice coil out of the gap without too much penalty
Hence we now got a tool into our hands to simulate *and* measure thermal related wave form deformation isolated from mechanical effects in speakers
This is *kind of* progress IMO.
Michael
gedlee said:
Thanks, Earl
First hand for offering your analytical skills.
But even more so as I take it as an offer for support and help - *for me* - rather than for what I'm currently working out.
This is something I very much appreciate, as I find it a new (pretty much hidden up to now) quality in the context of your contributions here.
gedlee said:
To answer your question:
I already have done some measurements that "knock out" interesting results – especially in the frequency domain. These results basically support the prove of my thermal model, but I haven't shown 'em yet.
Partly because I'm awaiting an improved revision for that software I use, in order to show effects even more clearly and pronounced, partly because I have to sort out some issues with my measurement set-up first, to get even more reliable results at that fragile levels I am investigating here.
On top of that, there is a fundamental difficulty to measure an effect precisely *all* your measuring equipment suffers from.
So, I'd like to ask you for some patience, as I'd like to make the most out of the analytical potential I have available first – but certainly would be happy to take you up on your offer as soon as I reach my limits (which possibly might be pretty quick anyway
). ------
*If* the thermal model presented holds true – and for me its more a question of possibly refining one or two things rather than about its qualitative behaviour in general – we have to face the fact, that the common definition for the best possible amp to be a "wire with gain" is no longer true.
- Or to be more precise – we probably will raise some more questions about the Thermal Distortion figure of that "wire with gain"
What all that past effort done to gain understanding of thermal effects in *speakers* boils down to, is the fact that
- heat plus cooling - and subsequent temperature change - effects signal integrity in *any conductor* of other than zero temperature coefficient
Thinking of conductors (others than VC's) having low thermal mass and low time constant – we pretty fast arrive at all the tiny electronic components we use.
It's not out of the blue that I try to established the parallels to convolution technique when it comes to thermal behaviour on the level of wave shape deformation.
Convolution basically describes that a given signal is coined by an other effect – whatsoever.
Watching out for an explanation why we (at least some of us) are able to detect the "sound of materials" even at homeopathic levels imprinted on the signal – well - convolution tells us that we can hear both, the signal *and* the impulse response overlaid .
Charles Hansen also has some juicy stories on that for example.
Michael
As good as it gets - second round to nail Thermal Distortion on the level of wave shape deformation
As said – its hard to measure this effect and to present clear plots .
But its a good exercise anyway, as it turned out that I have to correct my enhanced model by the part that got lost somewhere on the long way of simplifying the thermal model shown earlier.
No big deal, as all plots shown earlier still are valid with the exception of the qualitative graphs illustrating the odd order harmonics originating from Thermal Distortion.
Changes in time domain plots shown are so small, they cant be spotted at all. But it makes a significant difference when it comes to analysis in the frequency domain.
I'll come back to that in detail - but first lets have a look at the measurements
What can be see above is – for reference – how a 1W / 3Hz power injection into two (parallel connected) power resistors (rated at 50W each) look like.
It basically reflects the distortion figures of the amp and the soundcard used at a very low 3Hz frequency
The RED trace is measured voltage
The BLUE trace is measured current
What can be see above is the same 1W power injection into the nude voice coils of the no-name compression driver at 3Hz / 10Hz / 30Hz.
Again
The RED trace is measured voltage
The BLUE trace is measured current
What we clearly see is that the current shows a HEAVY peak at third harmonics.
Anything else is pretty equal to the reference measurement.
*But* - if we have a closer look at the reference measurement we also can see there is a tiny difference between voltage and current at the third harmonics as well.
This reflects the reduced Thermal Distortion on the level of wave shape deformation for a 20ppm temp-co resistor (and different time constant as well)
Clearly enough we can see *no* additional odd order harmonics.
This brings us back to the enhanced thermal model I've presented
What got lost in the course of setting up – running and trashing - a multitude of models to mimic the thermal behaviour of conductors, was, that the energy injection for a given signal wave has to be modelled as area under the wave form - as square rather than abs – as power calculates by voltage^2 * resistance.
With that correction re-implemented, the simulation shows almost exactly what we measured:
Above we see the simulation set to meet the measurements of the nude VC.
There are no odd order harmonics other than the third one !
What doesn't fit exactly is that in measurements, third order harmonics do *not* drop by 20dB each decade, as predicted by simulation. I guess this is due to the non-linearity of the amp and soundcard – and all measurements had to be performed *very* close to the measurement limits of that equipment anyway.
Everybody is welcomed to proove and validate the results presented.
All measurements are documented in detail and I have provided my thermal model here for free !
Michael
As said – its hard to measure this effect and to present clear plots .
But its a good exercise anyway, as it turned out that I have to correct my enhanced model by the part that got lost somewhere on the long way of simplifying the thermal model shown earlier.
No big deal, as all plots shown earlier still are valid with the exception of the qualitative graphs illustrating the odd order harmonics originating from Thermal Distortion.
Changes in time domain plots shown are so small, they cant be spotted at all. But it makes a significant difference when it comes to analysis in the frequency domain.
I'll come back to that in detail - but first lets have a look at the measurements
An externally hosted image should be here but it was not working when we last tested it.
What can be see above is – for reference – how a 1W / 3Hz power injection into two (parallel connected) power resistors (rated at 50W each) look like.
It basically reflects the distortion figures of the amp and the soundcard used at a very low 3Hz frequency
The RED trace is measured voltage
The BLUE trace is measured current
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 can be see above is the same 1W power injection into the nude voice coils of the no-name compression driver at 3Hz / 10Hz / 30Hz.
Again
The RED trace is measured voltage
The BLUE trace is measured current
What we clearly see is that the current shows a HEAVY peak at third harmonics.
Anything else is pretty equal to the reference measurement.
*But* - if we have a closer look at the reference measurement we also can see there is a tiny difference between voltage and current at the third harmonics as well.
This reflects the reduced Thermal Distortion on the level of wave shape deformation for a 20ppm temp-co resistor (and different time constant as well)
Clearly enough we can see *no* additional odd order harmonics.
This brings us back to the enhanced thermal model I've presented
What got lost in the course of setting up – running and trashing - a multitude of models to mimic the thermal behaviour of conductors, was, that the energy injection for a given signal wave has to be modelled as area under the wave form - as square rather than abs – as power calculates by voltage^2 * resistance.
With that correction re-implemented, the simulation shows almost exactly what we measured:
An externally hosted image should be here but it was not working when we last tested it.
Above we see the simulation set to meet the measurements of the nude VC.
There are no odd order harmonics other than the third one !
What doesn't fit exactly is that in measurements, third order harmonics do *not* drop by 20dB each decade, as predicted by simulation. I guess this is due to the non-linearity of the amp and soundcard – and all measurements had to be performed *very* close to the measurement limits of that equipment anyway.
Everybody is welcomed to proove and validate the results presented.
All measurements are documented in detail and I have provided my thermal model here for free !
Michael
mige0 said:
Michael
Very interesting, but please don't make statements like this last one as they really bother me. All of this is very preliminary and nothing has been confirmed by anyone else. Nothing has been "proven" at all. It's very good work and a good first step, but it's not yet a complete package.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.