mige0 said:If you woiuld like to know to *what degree* the changes you measured are also caused by Thermal Distortion you *have to* know
1. source impedance
2. real / imaginary part of speaker impedance
3. a TD reference measurement for scaling TD at different power injection at a given frequency
- and possibly my thermal model you now can download
Quantifying TD effects is now pretty precise - once you've referenced the model by measurement.
I don't think its any new for you, that all this factors play together.
What role source impedance plays with respect to TD is also outlined in short in my paper for current versus voltage amplifiers.
Michael
You are missing the point. This is not directly related to what you are doing. These tests were perfromed to look at the effect of changes in VC Z of the transfer function of a passive crossover. The point was that when the results were examined it became apparent that changes in Re cound not account for the change in the response while changes in Le do. The point is that the speaker's frequency response changes pretty significantly in the crossover region (I call 1.5dB significant) with signal level. You would not see this effect if the passive crossover was replaced with an active one, or if the driver Le was not modulated with excursion. It is not thermal distortion.
By the way, you may which to look at this article by Hawsford. (DISTORTION REDUCTION IN MOVING-COIL LOUDSPEAKER SYSTEMS USING CURRENT-DRIVE TECHNOLOGY).
John,
even passive x-over parts can reduce distortion and thermal compression in dynamic drivers by taking a step away from "pure voltage drive".
I assume the x-over you measured or simulated was of high order? Did you check with a shallower slope like a single coil as a series element for example?
/Peter
even passive x-over parts can reduce distortion and thermal compression in dynamic drivers by taking a step away from "pure voltage drive".
I assume the x-over you measured or simulated was of high order? Did you check with a shallower slope like a single coil as a series element for example?
/Peter
Pan said:
You can make the simulations yourself. Just set up any passive filter and load it with a Z representative of a driver and then parametrically change Re and Le and see what happens. Passives are supposed to be loaded by a static load. We don't have that with speakers and the result is that the transfer function of the filter will change as the driver Z changes. It does not change if the filter is active. That doesn't mean actives fix thermally related problems but they do remove dynamic changes in the filter transfer function from the picture, provided the source output impedance is much smaller than the driver Z.
This has long been an argument for active filters and current source drive. Current source drive is actually an inverse though, with driver Z much smaller then the source Z.
john k... said:
As soon as you used "s" you were in the frequency domain.
It is possible to define a z(t) for an individual component, but as soon as there is a circuit this approach fails and one is forced into the frequency domain in order to avoid convolutions at each circuit leg, etc. It IS probably doable, but I'M not about to go through all this math! Just go into the frequency domain and work there. Thats why we have transform theory!!
I wanted to do everything in the time domain because then you could conceiveably get Re(t) at every time step for extremely high resolution. Having to go into the frequency domain means that you can only do this in blocks of the FFT with all the resolution issues that are inherent therein. This may or may not end up being a problem. I haven't decided yet.
Earl,
Some further thoughts.
If I have a load in series with a shunt where the shunt resistance is much smaller than the load then if Vs(t) is the voltage across the shunt and Vl(t) is the voltage across the load, I(t) = Vs(t)/Rs, where Rs is the shunt resistance. Thus, we can compute G(t) = Vl(t)/I(t).
Now, obviously if the load is a pure and constant resistance then G(t) = const = Rload. An FFt of G(t) would show a single, real DC value representative of Rload. If the load remains purely resistive but time varying then the FFt of G(t) would show harmonics as well as the DC component, but they would still all be real components. If the load were reactive and constant it follows that the FFt of G would have a single imaginary DC component. If the load were reactive and time dependent, then the FFt of G would have harmonics, but with only imaginary components. If the load were complex and constant the FFt of G would have only real and imaginary DC components. Now, make the load complex and time dependent and we should see harmonics which may have both real and imaginary components. Here the real components would represent changes in the resistive part of the load and the imaginary components those due to changes in the reactive parts of the load. If the FFt of G showed a single imaginary DC component and numerous real components this would be a case where the reactive component of the load was constant while the resistive component had some time variance. Etc...
I also think there is more to this in that as the input level goes up there is also an increase in mechanical resistance which means the form of the back EMF generated in the motor will change. This change in mechanical resistance, as well as suspension compliance... which alter the back EMF will not only result directly in distortion of the cone motion, but also in the current thus making the driver impedance look different. This is most obviously seen around Fs and reflected in T/S parameter variation with level. I don’t see a way to separate out changes in current due to VC heating or changes due to changes in the back EMF. However, like the thermal effects, the back EMF effects will decrease in magnitude with increasing frequency.
Some further thoughts.
If I have a load in series with a shunt where the shunt resistance is much smaller than the load then if Vs(t) is the voltage across the shunt and Vl(t) is the voltage across the load, I(t) = Vs(t)/Rs, where Rs is the shunt resistance. Thus, we can compute G(t) = Vl(t)/I(t).
Now, obviously if the load is a pure and constant resistance then G(t) = const = Rload. An FFt of G(t) would show a single, real DC value representative of Rload. If the load remains purely resistive but time varying then the FFt of G(t) would show harmonics as well as the DC component, but they would still all be real components. If the load were reactive and constant it follows that the FFt of G would have a single imaginary DC component. If the load were reactive and time dependent, then the FFt of G would have harmonics, but with only imaginary components. If the load were complex and constant the FFt of G would have only real and imaginary DC components. Now, make the load complex and time dependent and we should see harmonics which may have both real and imaginary components. Here the real components would represent changes in the resistive part of the load and the imaginary components those due to changes in the reactive parts of the load. If the FFt of G showed a single imaginary DC component and numerous real components this would be a case where the reactive component of the load was constant while the resistive component had some time variance. Etc...
I also think there is more to this in that as the input level goes up there is also an increase in mechanical resistance which means the form of the back EMF generated in the motor will change. This change in mechanical resistance, as well as suspension compliance... which alter the back EMF will not only result directly in distortion of the cone motion, but also in the current thus making the driver impedance look different. This is most obviously seen around Fs and reflected in T/S parameter variation with level. I don’t see a way to separate out changes in current due to VC heating or changes due to changes in the back EMF. However, like the thermal effects, the back EMF effects will decrease in magnitude with increasing frequency.
gedlee said:
No. S is a general complex variable. S can be real, + or -, complex or otherwise. It need not be jw.
In fact, while it's been over a decade, as I recall we used FFt of I(t)/V(t) to find the complex, frequency dependent admittance paramters "Y parameters" to be used in circuit simulations for nonlinear devices.
John
I don't usually think of testing useing sine waves, but your example is very intriging as it seems logical and quite appealing. I'll have to think about using a single sine wave, but I always wory about the validity of this WRT to other frequencies. Sweeping the sine wave is an option, but then how do you sort out the frequency from the time. With noise I can do this.
I agree that there is a complicated situation here between nonlinearity of the driver from mechanical sources and that from thermal sources. It would be very difficult to sort out. But, my gut tells me that the mechanical nonlinearities are not audibly significant - as born out by numerous tests that I have done - but there IS something audible in larger speakers versus smaller ones. Could this be the thermal effects? Perhaps. I do think that this situation has not been ironed out. Why do larger speakers and waveguides with huge drivers (compared to a dome tweeter) sound so "dynamic"? Thats an almost universal comment on my designs, but I have to admit to not being completely comfortable that I know why. I have my suspicions as noted here, but there is no hard data.
I don't usually think of testing useing sine waves, but your example is very intriging as it seems logical and quite appealing. I'll have to think about using a single sine wave, but I always wory about the validity of this WRT to other frequencies. Sweeping the sine wave is an option, but then how do you sort out the frequency from the time. With noise I can do this.
I agree that there is a complicated situation here between nonlinearity of the driver from mechanical sources and that from thermal sources. It would be very difficult to sort out. But, my gut tells me that the mechanical nonlinearities are not audibly significant - as born out by numerous tests that I have done - but there IS something audible in larger speakers versus smaller ones. Could this be the thermal effects? Perhaps. I do think that this situation has not been ironed out. Why do larger speakers and waveguides with huge drivers (compared to a dome tweeter) sound so "dynamic"? Thats an almost universal comment on my designs, but I have to admit to not being completely comfortable that I know why. I have my suspicions as noted here, but there is no hard data.
I think your comments about large speakers and wave guides/horns are just the matter of efficiency, not directly but because of lower effects of compression. That, and the lack of clipping. I think many individuals would be surprised at just how often a 100 watt amp is driven into clipping when used with an 86dB/W speaker playing at moderately loud levels. And when we consider that at those levels the compression effects do add up, further reducing efficiency, there is a further loss in max SPL capability, even on a momentary (transient) basis.
I only suggest the sine input for testing because it would clearly show distortion in the current wave fore. If V = sin(wt) is input and I = sin(wt+phi) doesn't come out, then there is some nonlinear impedance transfer function. It may be hard (impossible?) to isolate whether it is from thermal effects or other nonlinear (mechanical or electrical) effects, but it should show that the net driver Z is behaving nonlinearly.
I only suggest the sine input for testing because it would clearly show distortion in the current wave fore. If V = sin(wt) is input and I = sin(wt+phi) doesn't come out, then there is some nonlinear impedance transfer function. It may be hard (impossible?) to isolate whether it is from thermal effects or other nonlinear (mechanical or electrical) effects, but it should show that the net driver Z is behaving nonlinearly.
John
When you write "dV/dt = s x exp(st)" you have assumed a transform into the frequency domain no matter how you define s. Otherwise the equation is not true. It's actually an operator since, as written, its not even correct, it should be d/dt V(t) = s * V(s). You, of course, will say that you have defined V(t) = exp(st), but this is only a finite function and periodic if s lies in the left half of the complex plane, and you must also define it to be zero for t < 0, otherwise it is still not finite. These conditions do make the only way the assumptions can be true is if this is a transform operator from the time domain into the frequency domain.
My intent is not to show the nonlinearity from the thermal aspects, but to find the time constants of the thermal circuits for heating and cooling effects. These are needed if one is going to simulate the thermal effects for an audibility study.
I don't think that I would jump to the conclusion about large efficient speakers that you do. In our audibility studies of nonlinear effects in compression drivers, we knew that the amps were not clipping and still no one could detect very large THD levels. I simply don't think that its as simple as we would like it to be. There is something that we can hear, but all tests that I have done indicate that it is not mechanical nonlinearity or amplifier clipping. That doesn't leave much else but thermal - and of course the diffraction issues that we have identified as significant.
When you write "dV/dt = s x exp(st)" you have assumed a transform into the frequency domain no matter how you define s. Otherwise the equation is not true. It's actually an operator since, as written, its not even correct, it should be d/dt V(t) = s * V(s). You, of course, will say that you have defined V(t) = exp(st), but this is only a finite function and periodic if s lies in the left half of the complex plane, and you must also define it to be zero for t < 0, otherwise it is still not finite. These conditions do make the only way the assumptions can be true is if this is a transform operator from the time domain into the frequency domain.
My intent is not to show the nonlinearity from the thermal aspects, but to find the time constants of the thermal circuits for heating and cooling effects. These are needed if one is going to simulate the thermal effects for an audibility study.
I don't think that I would jump to the conclusion about large efficient speakers that you do. In our audibility studies of nonlinear effects in compression drivers, we knew that the amps were not clipping and still no one could detect very large THD levels. I simply don't think that its as simple as we would like it to be. There is something that we can hear, but all tests that I have done indicate that it is not mechanical nonlinearity or amplifier clipping. That doesn't leave much else but thermal - and of course the diffraction issues that we have identified as significant.
Hi John, all
“I don’t see a way to separate out changes in current due to VC heating or changes due to changes in the back EMF.”
In the old days, I had to make an amplifier that monitored the Rdc of the woofer it was driving. There are several ways to do this but if you can add a small carrier signal, say a low level sine wave at 10Hz (or I bet @ Rmin would be good too), you can use a synchronous rectifier to detect / determine the Rdc (via small R on ground leg) while it is driven with noise or another signal. These can be very sensitive but the issue is how much damping it needs compared to how fast the Rdc is changing.
Under steady state, this works very well and here it wouldn't matter if you could hear it for testing..
I am anxious to hear about time constants of hifi drivers, with woofers of the 3 and 4 inch coil sizes, the time constant is seconds long and is actually several constants the shortest being the coil heating, the longest being the magnet and frame heating up.
I am not clear if a hifi driver could have a time constant so short that there is appreciable modulation proportional to I^2 X Rdc in each half cycle.
While the motor of a dome tweeter is vastly smaller, it also operates at a much higher frequency.
Two last thoughts, keep in mind that many physical things which have damping, also exhibit changes of those properties with changes in temperature AND recent stress.
One very real way to have a change in a drivers operation with changes in level is via this mechanism. For example, drive a woofer at Fs, near Xmax for a while, then measure the Fs. Come back another day, and measure the Fs, it will be higher than it was after exercise. Cone body’s and other things have a potential to have this “mechanical memory” effect.
Also, to the degree the magnetic system’s operating point can be pushed around by the voice coil current (force acting between the two), it can change it’s apparent operation with level.
For example, ceramic magnet would have a less desirable BH curve than Neo in this regard.
Anyway, it is cool you guys are pursuing this.
Best,
Tom Danley
“I don’t see a way to separate out changes in current due to VC heating or changes due to changes in the back EMF.”
In the old days, I had to make an amplifier that monitored the Rdc of the woofer it was driving. There are several ways to do this but if you can add a small carrier signal, say a low level sine wave at 10Hz (or I bet @ Rmin would be good too), you can use a synchronous rectifier to detect / determine the Rdc (via small R on ground leg) while it is driven with noise or another signal. These can be very sensitive but the issue is how much damping it needs compared to how fast the Rdc is changing.
Under steady state, this works very well and here it wouldn't matter if you could hear it for testing..
I am anxious to hear about time constants of hifi drivers, with woofers of the 3 and 4 inch coil sizes, the time constant is seconds long and is actually several constants the shortest being the coil heating, the longest being the magnet and frame heating up.
I am not clear if a hifi driver could have a time constant so short that there is appreciable modulation proportional to I^2 X Rdc in each half cycle.
While the motor of a dome tweeter is vastly smaller, it also operates at a much higher frequency.
Two last thoughts, keep in mind that many physical things which have damping, also exhibit changes of those properties with changes in temperature AND recent stress.
One very real way to have a change in a drivers operation with changes in level is via this mechanism. For example, drive a woofer at Fs, near Xmax for a while, then measure the Fs. Come back another day, and measure the Fs, it will be higher than it was after exercise. Cone body’s and other things have a potential to have this “mechanical memory” effect.
Also, to the degree the magnetic system’s operating point can be pushed around by the voice coil current (force acting between the two), it can change it’s apparent operation with level.
For example, ceramic magnet would have a less desirable BH curve than Neo in this regard.
Anyway, it is cool you guys are pursuing this.
Best,
Tom Danley
gedlee said:
Earl, that is just not true. Please don't make me scan a page for a basic calculus book. It's just the definition of the derivative of an exponential,
F(x) = exp(ax)
dF/dx = d(ax)/dx exp(ax) = a exp(ax).
There is nothing about frequency here. For "a" = real constant
V = exp(at) yields
I = CdV/dt for
I = a C exp(at) = aC V(t),
Z = V/I = 1/(aC)
which is the EE 101 text book definition of the impedance of a capacitance to an exponentially growing voltage.
The conventional definitions of impedance is Z = V/I and from the defining relationships
V = IR,
I = C dV/dt,
and
V = L dI/dt
when the current and voltage can be expressed as an exponential
I = exp(at)
V = exp(at)
we get
Z for resistor = R,
Z for a cap = 1/(aC)
and Z for an inductor = aL,
all straight out f an EE 101 text. We don't get into the frequency domain until "a" contains an imaginary part such as jw.
John
I would agree that for this simple cases this is correct. But for arbitrary signals (waveforms) I don't think that it holds unless you put some contraints on the value of "a". An arbitrary signal would have to be decomposed into into the "kernal" functions - exponentials here. If the signal is of finite amplitude but infinite duration then only complex exponentials are allowed. If the signals are of finite amplitude and finite duration then other restriction on "a" would apply and before long you'd find that to be at all useful you would have developed either a Laplace transform around the kernal exp(st) or the Fourier transform around exp(iwt). Otherwise you might have something mathematically correct but not very useful for any signals in the real world.
At any rate we both understand the situation.
I still do contend that I don't see a way to get Re(t) using an arbitray signal measurement of V(t) and i(t) (for anything but a simple resistor) without taking an FFT to isolate the real and imaginary parts. This is really the crux of my discussion. A pure sine wave could be used where the phase was simply ignored by aligning the two waveforms, but then you only have data at a single frequency. Can't sweep as I said because then you can't modulate. Modulated noise, as I have posted, seems the most logical to me.
I would agree that for this simple cases this is correct. But for arbitrary signals (waveforms) I don't think that it holds unless you put some contraints on the value of "a". An arbitrary signal would have to be decomposed into into the "kernal" functions - exponentials here. If the signal is of finite amplitude but infinite duration then only complex exponentials are allowed. If the signals are of finite amplitude and finite duration then other restriction on "a" would apply and before long you'd find that to be at all useful you would have developed either a Laplace transform around the kernal exp(st) or the Fourier transform around exp(iwt). Otherwise you might have something mathematically correct but not very useful for any signals in the real world.
At any rate we both understand the situation.
I still do contend that I don't see a way to get Re(t) using an arbitray signal measurement of V(t) and i(t) (for anything but a simple resistor) without taking an FFT to isolate the real and imaginary parts. This is really the crux of my discussion. A pure sine wave could be used where the phase was simply ignored by aligning the two waveforms, but then you only have data at a single frequency. Can't sweep as I said because then you can't modulate. Modulated noise, as I have posted, seems the most logical to me.
gedlee said:
Well it's a simple manner to place limits on these. All heat is generated by V^2/R heating and is given by
cp x M x dT/dt = Q,
or the rate of heating is Q/(cp x M). That is the rate of heating follows Q. In the absence of heat rejection the VC temp increase without limits to
T = integral [(Q/cp x M)] dt
For the purpose of estimating a general time constant for heating, evaluate the integral over a long period of time to find the change in temperature and then compute average rate. It should be independent of frequency. Note that if you can consider Q to be the mean value over a cycle and if you assume Re is a linear function of T then T will go like sqrt(t) I think.
Cooling goes something like
Cp x dT/dt = -k (T - Tr) /dx
where Tr is the sink temperature, k is thermal conductivity of the substance in the gap and dx is the with of the gap.
This yields something like
Tr = T + Cp x dx /k x dT/dt
or T = Tr + (To - Tr) exp [-t/(Cp x dx /k)]
where To is the initial Vc temp k is the thermal conductivity of the fluid in the gap and Tr is the temp of the magnet structure or pole. Note that Tr would actually be dependent on how the pole piece of magnet heats up, but the time constant is independent of this. There are some details missing but you should have the idea.
Note, I assume conduction across the gap because I don't think that (at least for a tweeter) the velocity of the fluid in the gap is sufficient high to consider convective heat transfer. It's just conduction through still air or ferro fluid.
Hi Earl
This is the advantage of synchronous detection.
One can measure the Rdc of the driver, while driving it with a large signal, especially if is is noise.
You drive the speaker with a sine wave of a known level (which is then added to the noise which produces the heating). You use a small series resistor to sample current, you use the source sine’s zero crossing point to flip the polarity of the rectifier.
The noise signals, which are not aligned with the test signal are randomly in and out of phase and so null out while the current resulting from the test signal (and so Rdc hot) is detected as it is in full synchrony with the source.
I know you guys are more DSP and math oriented rather than hardware but do look into this approach as applied in that domain.
Best,
Tom
This is the advantage of synchronous detection.
One can measure the Rdc of the driver, while driving it with a large signal, especially if is is noise.
You drive the speaker with a sine wave of a known level (which is then added to the noise which produces the heating). You use a small series resistor to sample current, you use the source sine’s zero crossing point to flip the polarity of the rectifier.
The noise signals, which are not aligned with the test signal are randomly in and out of phase and so null out while the current resulting from the test signal (and so Rdc hot) is detected as it is in full synchrony with the source.
I know you guys are more DSP and math oriented rather than hardware but do look into this approach as applied in that domain.
Best,
Tom
Tom
I would never do anything in hardware that I could do in software. This just seems obvious to me, but I know you hardware guys are die hards. I said the same thing to Marshall Buck and he responded the same way - "that seemed like the easiest way to do it to me!". I guess easiest is whatever you know how to do. But lets face it, hardaare is never as low in cost as software. Thats why Microsoft is so rich and the hardware guys are all strugling.
I would never do anything in hardware that I could do in software. This just seems obvious to me, but I know you hardware guys are die hards. I said the same thing to Marshall Buck and he responded the same way - "that seemed like the easiest way to do it to me!". I guess easiest is whatever you know how to do. But lets face it, hardaare is never as low in cost as software. Thats why Microsoft is so rich and the hardware guys are all strugling.
Hi Earl, guys
Well I admit I am semi handicap when it comes to deeper math and have traditionally leaned on hardware based measurements for most things.
I made this suggestion because I believe synchronous detection can be implemented in software too, that was my meaning when refereeing to math / DSP and I suggested that you “look into this approach as applied in that domain”.
Here is a link to one fellow who has done that in conjunction with LabView.
http://www.mrflip.com/papers/LIA/
The basic advantage of this is one can extract the measurement signal while buried in noise or a sea of other signals, one could use the noise signal for heating and the small carrier signal for measuring Rdc simultaneously, so you could measure the Rdc as it changes while heating up or cooling down.
Best,
Tom
Well I admit I am semi handicap when it comes to deeper math and have traditionally leaned on hardware based measurements for most things.
I made this suggestion because I believe synchronous detection can be implemented in software too, that was my meaning when refereeing to math / DSP and I suggested that you “look into this approach as applied in that domain”.
Here is a link to one fellow who has done that in conjunction with LabView.
http://www.mrflip.com/papers/LIA/
The basic advantage of this is one can extract the measurement signal while buried in noise or a sea of other signals, one could use the noise signal for heating and the small carrier signal for measuring Rdc simultaneously, so you could measure the Rdc as it changes while heating up or cooling down.
Best,
Tom
Tom
Of course it could be done in software too. But the point is that once you are using software and a digitally sampled signal then the whole world of signal processing opens up to you and there is an awful lot that can be done. Its not popular at all, but I continue to use noise as a signal because noise is an ideal signal from a number of statistical directions. It has ideal statistics which sine waves and other waveforms don't have. Its so easy to do things with noise compared to other signals. For any given application there is always an ideal signal which optimizes resolution or speed or what ever it is you want to optimize. But for a general purpose signal noise is supreme because it is the most statistically stable. Since I do all kind of different tests I find noise always works, but is usually not optimum. But the fact that it always works keeps me using it.
Of course it could be done in software too. But the point is that once you are using software and a digitally sampled signal then the whole world of signal processing opens up to you and there is an awful lot that can be done. Its not popular at all, but I continue to use noise as a signal because noise is an ideal signal from a number of statistical directions. It has ideal statistics which sine waves and other waveforms don't have. Its so easy to do things with noise compared to other signals. For any given application there is always an ideal signal which optimizes resolution or speed or what ever it is you want to optimize. But for a general purpose signal noise is supreme because it is the most statistically stable. Since I do all kind of different tests I find noise always works, but is usually not optimum. But the fact that it always works keeps me using it.
Thanks for all the effort you put into this. Very much appreciated.
A simple way to extract time constants would be very comfortable to have and ease further analysis, be it by Mathlab or SPICE.
From own experience – and in accordance with Tom - I doubt things can be sorted out *precisely enough* with a single simple measurement like noise.
Even that - after my simulations and measurements shown here - we now know for what to watch out, there are too many imponderabilites involved
- different time constants even for one and the *same* frequency and power injection (radiation, convection, conduction)
- different time constants for any frequency
- different time constants for any power injection
not to speak from all the overlaying mechanical issues
Its not out of the blue that I settled to with steady sinus to get the most clear results from measurements – and though the graphically approach to identify time constant isn't really "scientifically" - its reliable and easy to perform for *everybody*
but lets see...
Earl, have you already created the file with a HP ?
Another good news – I finished my model rev 2.3 and again provide it for free to download at
http://members.aon.at/kinotechnik/diyaudio/diy_audio/TD/SPICE_modelling_TD.htm
Over model rev.2 .0 the updated thermal model provides more stringent results when it comes to wave form deformation
Basically we get a good impression how wave form deformation builds up over six decades of power injection.
Main conclusion from that heavily exaggerated example to be drawn is that distortion comes in more and more the higher we force thermal load.
Nothing new – in fact – but very intuitively seen on that simu.
The green trace is the resistance rise (and recovery ! ) envelope and the two horizontal traces serve as temperature markers at roughly 100°C and 200°C above ambient.
Introducing temperature markers is especially useful IMO as it gives us a reference that we understand intuitively.
After having provided qualitative understanding with my first model and quantitative understanding and predictability for mid an long term TD effects with model rev 2.0 - I now have hope that with the updated model rev 2.3 I have closed the gap between mid / long term behaviour and wave form deformation on a qualitative level as well.
Measurements (=*real* data
) will show, also for the newly included feature of modelling cooling due to radiation...
John, thanks for the Mills-Hawksford paper. Clever guys !
Michael
A simple way to extract time constants would be very comfortable to have and ease further analysis, be it by Mathlab or SPICE.
From own experience – and in accordance with Tom - I doubt things can be sorted out *precisely enough* with a single simple measurement like noise.
Even that - after my simulations and measurements shown here - we now know for what to watch out, there are too many imponderabilites involved
- different time constants even for one and the *same* frequency and power injection (radiation, convection, conduction)
- different time constants for any frequency
- different time constants for any power injection
not to speak from all the overlaying mechanical issues
Its not out of the blue that I settled to with steady sinus to get the most clear results from measurements – and though the graphically approach to identify time constant isn't really "scientifically" - its reliable and easy to perform for *everybody*
but lets see...
Earl, have you already created the file with a HP ?
Another good news – I finished my model rev 2.3 and again provide it for free to download at
http://members.aon.at/kinotechnik/diyaudio/diy_audio/TD/SPICE_modelling_TD.htm
Over model rev.2 .0 the updated thermal model provides more stringent results when it comes to wave form deformation
An externally hosted image should be here but it was not working when we last tested it.
Basically we get a good impression how wave form deformation builds up over six decades of power injection.
Main conclusion from that heavily exaggerated example to be drawn is that distortion comes in more and more the higher we force thermal load.
Nothing new – in fact – but very intuitively seen on that simu.
The green trace is the resistance rise (and recovery ! ) envelope and the two horizontal traces serve as temperature markers at roughly 100°C and 200°C above ambient.
Introducing temperature markers is especially useful IMO as it gives us a reference that we understand intuitively.
After having provided qualitative understanding with my first model and quantitative understanding and predictability for mid an long term TD effects with model rev 2.0 - I now have hope that with the updated model rev 2.3 I have closed the gap between mid / long term behaviour and wave form deformation on a qualitative level as well.
Measurements (=*real* data
) will show, also for the newly included feature of modelling cooling due to radiation...John, thanks for the Mills-Hawksford paper. Clever guys !
Michael
After the exaggerated waveform deformation in my last posting shown to illustrate Thermal Distortion on the level of wave form deformation – here I now show the according 3Hz measurements done with the nude compression driver voice coil at 1W, 5W and 10W power injection
1W power injection
5W power injection
10W power injection
Below what's been predicted by SPICE simulation with the new model rev 2.3
1W, 5W, 10W power injection
Well not that bad – but actually not as good as I hoped it would be .
Could need some help from you guyes !
What you think may cause the strange shift in measurements around the zero crossing line ?
There clearly is a difference where the falling and the rising shoulder has its discontinuity.
Shouldn't this discontinuity be at *exactly* zero line either way (as also seen in my simu)?
What could cause that shift? Do I need a better amp here?
Voltage measurement over DUT is OK – hence not shown.
Michael
1W power injection
5W power injection
10W power injection
Below what's been predicted by SPICE simulation with the new model rev 2.3
1W, 5W, 10W power injection
Well not that bad
– but actually not as good as I hoped it would be .Could need some help from you guyes !
What you think may cause the strange shift in measurements around the zero crossing line ?
There clearly is a difference where the falling and the rising shoulder has its discontinuity.
Shouldn't this discontinuity be at *exactly* zero line either way (as also seen in my simu)?
What could cause that shift? Do I need a better amp here?
Voltage measurement over DUT is OK – hence not shown.
Michael
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.