Guys,
I hear everything you say, but I feel that this is worth exploring. I've always considered myself squarely in the objectivist camp, and pooh poohed the subjectivists with low DF amps, and dismissed current drive. However, I think that Hawksford makes a pretty good case here:
http://www.essex.ac.uk/dces/researc...J12 Distortion reduction MC current drive.pdf
I DO have to do something about my measuring gear though, as the LMS setup is starting to get problematic, I suspect the old PC with ISA slots I'm keeping around for the LMS card is starting to die on me, and I don't really want to pursue another ISA based PC in this day & age, so have to either wait for his new analyzer, or rig up something with a soundcard. My analog only AP Sytem One is not really suitable for acoustic measurements unfortunately. So, any further exploartion will have to wait for a while. OTOH maybe all I need is to build a pre-amp/phantom power box for the LMS mic and use the AP - I could probably just do straight sine sweeps & distortion measurments in the backyard. Decisions, decisions.....
I hear everything you say, but I feel that this is worth exploring. I've always considered myself squarely in the objectivist camp, and pooh poohed the subjectivists with low DF amps, and dismissed current drive. However, I think that Hawksford makes a pretty good case here:
http://www.essex.ac.uk/dces/researc...J12 Distortion reduction MC current drive.pdf
I DO have to do something about my measuring gear though, as the LMS setup is starting to get problematic, I suspect the old PC with ISA slots I'm keeping around for the LMS card is starting to die on me, and I don't really want to pursue another ISA based PC in this day & age, so have to either wait for his new analyzer, or rig up something with a soundcard. My analog only AP Sytem One is not really suitable for acoustic measurements unfortunately. So, any further exploartion will have to wait for a while. OTOH maybe all I need is to build a pre-amp/phantom power box for the LMS mic and use the AP - I could probably just do straight sine sweeps & distortion measurments in the backyard. Decisions, decisions.....
A highpass (or lowpass for that matter) can also have an amplitude peak (call it a resonance if it is peaky enough) in the freq response, depending on the Q...UnixMan said:I don't see highpass behavior... I see resonances. That's "bandpass", isn't it?
And for example a CB speaker is just a 2nd order higpass at the bottom end.
- Klaus
Hi UnixMan.
Have you ever tried driving a woofer via a series resistor which linearises the current flow (current source) ?
Did it not sound weak and lifeless ?
Actually we don't need a microphone to record this because it is so obvious !
Yes mechanical delay + phase shift, but a suddenly starting dynamic electrical waveform begins driving from t=0 irrespectively and before either of these aspects can be observed, so that the radiated waveform and especially all of the harmonics riding on it become distorted during waveform time !
Do a pen/paper sketch of say a suddenly starting wave which comprises 70Hz via a sub, and 210Hz 3rd harmonic via a mid-bass. Then do the same after you have compressed the first cycle of the 70Hz to establish the phase lead you already accept occurs as the sub cone movement takes time to become established !!!
Critical damping has nothing to do with this first dynamic cycle aspect so cannot improve upon it.
Hi Klaus.
In my mind EQed (inverse filtered) output from a voltage amplifier cannot restore the dynamic waveform.
This being due to the EQ arising in signal time, whereas resultant voice coil current flow is electromechanically modified in driver voice coil motion time.
See;-
http://www.linkwitzlab.com/filters.htm#9
Hi LukasLouw,
We need to remember that suddenly changing waveform current cannot provide equal cone displacement during tonebursts (musical transients) due to driver/system specific kinetic-potential energy storage/exchange mechanisms which arise during waveform time.
____________________________________________________
To compensate/control LF driver kinetic energy requirements a dynamic waveform needs additional boost/cut especially during the first half cycle of waveform change, and this would require a ‘tunable’ electronic gyrator the like of which I have not yet seen developed – one designed to match specific driver/system characteristics.
As indicated by Seigfried’s waveforms, this would typically require 2x voltage output drive capability for the first 90 degrees of waveform and thus a much more powerful amplifier - which could destroy drivers.
I have successfully generated the first half cycle boost using a transformer between amplifier and woofer with choke/capacitor tuning as here;-
http://www.diyaudio.com/forums/showthread.php?s=&threadid=130679
Cheers ….. Graham.
PS. I have shown this circuit before, but few have tried it.
Those who did liked what they heard.
Have you ever tried driving a woofer via a series resistor which linearises the current flow (current source) ?
Did it not sound weak and lifeless ?
Actually we don't need a microphone to record this because it is so obvious !
Yes mechanical delay + phase shift, but a suddenly starting dynamic electrical waveform begins driving from t=0 irrespectively and before either of these aspects can be observed, so that the radiated waveform and especially all of the harmonics riding on it become distorted during waveform time !
Do a pen/paper sketch of say a suddenly starting wave which comprises 70Hz via a sub, and 210Hz 3rd harmonic via a mid-bass. Then do the same after you have compressed the first cycle of the 70Hz to establish the phase lead you already accept occurs as the sub cone movement takes time to become established !!!
Critical damping has nothing to do with this first dynamic cycle aspect so cannot improve upon it.
Hi Klaus.
In my mind EQed (inverse filtered) output from a voltage amplifier cannot restore the dynamic waveform.
This being due to the EQ arising in signal time, whereas resultant voice coil current flow is electromechanically modified in driver voice coil motion time.
See;-
http://www.linkwitzlab.com/filters.htm#9
Hi LukasLouw,
We need to remember that suddenly changing waveform current cannot provide equal cone displacement during tonebursts (musical transients) due to driver/system specific kinetic-potential energy storage/exchange mechanisms which arise during waveform time.
____________________________________________________
To compensate/control LF driver kinetic energy requirements a dynamic waveform needs additional boost/cut especially during the first half cycle of waveform change, and this would require a ‘tunable’ electronic gyrator the like of which I have not yet seen developed – one designed to match specific driver/system characteristics.
As indicated by Seigfried’s waveforms, this would typically require 2x voltage output drive capability for the first 90 degrees of waveform and thus a much more powerful amplifier - which could destroy drivers.
I have successfully generated the first half cycle boost using a transformer between amplifier and woofer with choke/capacitor tuning as here;-
http://www.diyaudio.com/forums/showthread.php?s=&threadid=130679
Cheers ….. Graham.
PS. I have shown this circuit before, but few have tried it.
Those who did liked what they heard.
Graham Maynard said:
well, not exactly. But...
I normally use only feedbackless tube amps. Which typically have an output impedance of a few ohm (mine ~4). I have compared my amps with quite many other DIY and commercial amps of just about any existing technology: tube, SS, hybrid, class A, AB, D... (though I never had the chance to try one of yours).
To make a long story short, whenever I tried a low output impedance amp (typically SS, but also tube ones with plenty of NFB) I have found the sound to be dull and lifeless (as well as "unnatural" and uninvolving).
Now, if I have understood correctly, according to what you say I guess it should be the other way 'round... but that was not IME.
Actually, it looks like my own experience have been quite the opposite of yours!
BTW, I'm talking about the overall (full-range) results, not just basses. And I would say that the most relevant effects (before basses goes out of control and starts polluting everything) are in the "speech band" and thereabout.
P.S.: quite some time ago I have also built a "modified GC" with mixed (voltage+current) NFB which could be set to just about any output impedance, from "0" (only voltage NFB) to very high - almost "pure" current drive (a lot of current NFB) by adjusting a trimpot.
I have played with this "thing" on a few different systems (BTW, by chance all of them had "normal" 2-way speakers).
The results of my "experiments" have been as follows: increasing the output impedance up to a few ohm was (IMO) actually improving the overall perceived sound quality (though sadly nowhere near the quality of my tube amps...
). When set to the minimum impedance, the sound got the typical character I was describing (should I say complaining about?
) before. Progressively increasing the output impedance (IMO/IME) improved things, up to a point where the basses begins to go more and more "out of control" and becomes more and more "flappy" and "boomy" as the impedance is further increased (and thus overall sound quality get worse and worse).
That's why I was supposing something like a "critical damping" effect. Each of the (few) speakers that I have tried it with have shown some sort of (wide) "sweet spot" (=>"optimal" driving impedance), in the order of a few ohm.
Besides, I remember to have read an old article where such an effect (at least for some driver of the time on an open baffle) was actually demonstrated by a simple and "crude" yet effective measurement.
very interesting!!
BTW, I wonder: could it be that (without NFB) the OPT of tube amps may play a (somewhat) similar role?
Hi Paulo,
Yes our different experiences have led us along different paths.
Amplifiers having output impedance circa 4 ohm do not so much control LF driver cones on "normal" LS systems, as allow their moving masses and system Q to increase output around and above resonance. This can lead to a relative increase in LF output with minimum stored LS system energy due to driver voice coil current being less influenced by motionally induced back-EMF.
Low output impedance amplifiers can lead to greater levels of dynamically stored energy within "normal" enclosured LS systems; and yes indeed this can cause the sound to become - dull, lifeless, unnatural, uninvolving - and can even lead to additional unwanted LS interaction with the listening room. However, is the amplifier the problem, or is the LS/box electromechanically storing amplifier output energy, or both ?
Sadly I see what has become "normal" in the way of boxed LS systems, forcing owners to choose amplifiers which best mitigate the problems those 'boxes' cause, (as with higher output impedance amplifiers making the boxed LS systems store less energy) which then takes them further away from what could be more accurate and articulately dynamic life-like reproduction.
My observations would mirror yours if I too were to use 'boxed' loudspeakers as my reference, but I see cone mass, 'air springs', resonances and reflections all being aspects I wish to avoid for audio reproduction, so the boxes have gone, and with them the need for controlled output impedance amplifiers too.
You remember something about a driver on a baffle?
A low impedance amplifier plus low driver Qes will maintain dynamic control of cone motion; whereas either a high impedance amplifier, or a high Qes driver will lead to driver Qms modifying reproduction by increasing LF SPL but degrading articulation.
You wonder about the non-NFB OPT/tube amps.
Maximum energy transfer arises when the output impedance of an amplifier equals that of the driver. Here the transformers are instruments of mutual coupling, and can extend the efficiency range due to the voltage headroom available at output anodes. Thus when the LS system impedance rises at LF the output tube anodes can swing through higher voltage headroom range and increase output beyond that which would clip at nominal impedance, or with a SS NFB amplifier.
So yes the non/low NFB tube amp driven LS can comfortably output greater LF SPLs, but the 'dynamic' possible via a SS amplifier will then be missing. Also, whether the reproduction is 'clean' must depend as much upon the LS design as the amplifier, for one ought not be designed without consideration for the other. Even the 'best' amp can sound bad with the an unsuitable LS, and vice-versa - depends which path we take - Q and clarity being on opposite sides optimum.
(Jeepers - what a ramble.)
Cheers ..... Graham.
Yes our different experiences have led us along different paths.
Amplifiers having output impedance circa 4 ohm do not so much control LF driver cones on "normal" LS systems, as allow their moving masses and system Q to increase output around and above resonance. This can lead to a relative increase in LF output with minimum stored LS system energy due to driver voice coil current being less influenced by motionally induced back-EMF.
Low output impedance amplifiers can lead to greater levels of dynamically stored energy within "normal" enclosured LS systems; and yes indeed this can cause the sound to become - dull, lifeless, unnatural, uninvolving - and can even lead to additional unwanted LS interaction with the listening room. However, is the amplifier the problem, or is the LS/box electromechanically storing amplifier output energy, or both ?
Sadly I see what has become "normal" in the way of boxed LS systems, forcing owners to choose amplifiers which best mitigate the problems those 'boxes' cause, (as with higher output impedance amplifiers making the boxed LS systems store less energy) which then takes them further away from what could be more accurate and articulately dynamic life-like reproduction.
My observations would mirror yours if I too were to use 'boxed' loudspeakers as my reference, but I see cone mass, 'air springs', resonances and reflections all being aspects I wish to avoid for audio reproduction, so the boxes have gone, and with them the need for controlled output impedance amplifiers too.
You remember something about a driver on a baffle?
A low impedance amplifier plus low driver Qes will maintain dynamic control of cone motion; whereas either a high impedance amplifier, or a high Qes driver will lead to driver Qms modifying reproduction by increasing LF SPL but degrading articulation.
You wonder about the non-NFB OPT/tube amps.
Maximum energy transfer arises when the output impedance of an amplifier equals that of the driver. Here the transformers are instruments of mutual coupling, and can extend the efficiency range due to the voltage headroom available at output anodes. Thus when the LS system impedance rises at LF the output tube anodes can swing through higher voltage headroom range and increase output beyond that which would clip at nominal impedance, or with a SS NFB amplifier.
So yes the non/low NFB tube amp driven LS can comfortably output greater LF SPLs, but the 'dynamic' possible via a SS amplifier will then be missing. Also, whether the reproduction is 'clean' must depend as much upon the LS design as the amplifier, for one ought not be designed without consideration for the other. Even the 'best' amp can sound bad with the an unsuitable LS, and vice-versa - depends which path we take - Q and clarity being on opposite sides optimum.
(Jeepers - what a ramble.)
Cheers ..... Graham.
Graham, agreed, voltage drive is best suited for low frequency drivers, too many hoops to jump through with current drive.
Paolo, the perceived improvement in sound quality you experienced with lower damping factors in your listening tests is due to the fact that it added unintentional equalization to the system, i.e. some rise in the top end due to speaker inductance, and some increase in bass response around botom end resonance.
I'm sure some of the perceived improvement was due to lower speaker distortion, but unless you take the frequency variations compared to voltage drive out of the picture, your test was not valid.
To compare apples with apples, you should compensate with either an active EQ or flatten speaker impedance when in current drive mode.
Lukas
Paolo, the perceived improvement in sound quality you experienced with lower damping factors in your listening tests is due to the fact that it added unintentional equalization to the system, i.e. some rise in the top end due to speaker inductance, and some increase in bass response around botom end resonance.
I'm sure some of the perceived improvement was due to lower speaker distortion, but unless you take the frequency variations compared to voltage drive out of the picture, your test was not valid.
To compare apples with apples, you should compensate with either an active EQ or flatten speaker impedance when in current drive mode.
Lukas
Hi Paulo,
I have opened a new thread with some comment relating to your observations here.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=130679
Cheers ........ Graham.
I have opened a new thread with some comment relating to your observations here.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=130679
Cheers ........ Graham.
Graham Maynard said:
Hi Graham,
I think you are misinterpreting SL's comments. What he is showing is that by equalizing the driver, (in his example from Q=1.21, Fs = 55 Hz to Q= 0.5, Fs = 19 Hz), the response, in both the frequency and time domain, become identical to that of a driver which has Q = 0.5, Fs = 19 Hz without EQ.
True enough at resonance and below the resonant frequency, but not in the mass controled (mid band) region.
_________________________________
No. It requires the same type of eq as would be required to flatten the steady state response.
An externally hosted image should be here but it was not working when we last tested it.
I thought I would post this picture to illustrate some of what I have read. The top curve is the impedance of a typical driver. The center curve is the response, normalized to the response with true voltage drive (zero source (Re) impedance) for various series resistors. As Re gets large the response approaches that of a current source. Below that is the phase response variation form the pure voltage drive case, with is directly related to the amplitude variation.
Also when the equation governing cone motion is written the force applied to the driver is given as
F = BL x Is.
For a current source the output voltage will assume what ever value is required to assure that the specified current flows through the VC.
When voltage driven,
F = BL x Vs/ (Rv +jwLv) - (BL)^2 x U /( Rs + Rv + jwLv)
where the second term arises due to the back EMF. Here Rv is the voice coil DC resistance, Lv is the VC inductance and U is the cone velocity. If we equate the two,
I = V/ (Rv +jwLv) - BL x U /( Rs + Rv + jwL)
Now, if we assume V is constant vs. frequency we can solve for I delivered by the (voltage amplifier) or, if we choose we can let I be constant and solve for voltage across the voice coil as delivered by a current source amplifier.
I'd also like to comment on thermal effects. First, as the VC heats up the result is the same as increasing a series resistance so the effect on the response is simlar to that in the figure, lower output primarily in the region where the driver impedance is dominated by the VC DC R. If voltage drive, as the VC heats up Rv will increase and the current will decrease. Since the heat generated is I^2 x Rv in the region where Rv dominates the driver impedance a doubling of Rv could results in cutting I on the order of 1/2 which results in 1/2 the heat generation, a sort of self correcting or limiting situation. However, if current drive, since the current is fixed, if Rv doubles then there is a potential for the heat generated to double, making the VC heat up more, further increasing Rv, making even greater heat generation....... It can easily result in a unstable situation. Of course, heat rejection to the surroundings must be considered as well, but over all, the possibility of VC burn out is increased with current drive.
What is that plot suppose to show in your mind?
Compare the top plot, the un-eq'ed driver, to the bottom plot, the eq'ed driver. The improvement in response is apparent. The wave form distortion in both cases is a result of the systems (both before and after) being minimum phase. If you want to achieve a perfect burst response the eq would need to be carried low enough so that there is insignificant phase shift at 70 Hz and above.
The eq applied in SL's case is not inverse filter, it is just pole shifting to a lower cut off. If the system HP response is given by
T(s) = 1/(1 + So/Qo + So^2)
where So = jw/wo, wo is 2 Pi Fo, and Qo is Qtc of the woofer, then the inverse filter is
I(s) = 1/T(s)
which would yield an Eq'ed response
T'(s) = I(s) x T(s) = 1.0 = perfect reproduction.
What SL did was to eq by
Eq(s) = (1 + So/Qo + So^2)/(1 + Se/Qe + Se^2)
so that
T'(s) = T(s) x Eq(s) = 1/(1 + Se/Qe + Se^2)
Not perfect, but better (for a 70 Hz burst) due to the lower cut off frequency. In the frequency domain this is observed as amplitude and phase errors. If the system is eq'ed to a low enough frequency following SL's type of Eq but say, shifting the cut off to 5 Hz or even lower (in theory to DC), then the response would be further improved.
As I said, I think you are misinterpreting SL's comments. The Eq corrects the system to have the response of an independent system with the same response as the eq'ed system. It doesn’t not make it perfect unless the independent system is perfect.
He is a plot similar to SL's except it is 5 cycles at 70 Hz. The top is Fs = 55 Hz, Q = 1.21, center Fs = 19 Hz, Q = 0.5 and bottom Fs = 2, Q = 0.5.
Compare the top plot, the un-eq'ed driver, to the bottom plot, the eq'ed driver. The improvement in response is apparent. The wave form distortion in both cases is a result of the systems (both before and after) being minimum phase. If you want to achieve a perfect burst response the eq would need to be carried low enough so that there is insignificant phase shift at 70 Hz and above.
The eq applied in SL's case is not inverse filter, it is just pole shifting to a lower cut off. If the system HP response is given by
T(s) = 1/(1 + So/Qo + So^2)
where So = jw/wo, wo is 2 Pi Fo, and Qo is Qtc of the woofer, then the inverse filter is
I(s) = 1/T(s)
which would yield an Eq'ed response
T'(s) = I(s) x T(s) = 1.0 = perfect reproduction.
What SL did was to eq by
Eq(s) = (1 + So/Qo + So^2)/(1 + Se/Qe + Se^2)
so that
T'(s) = T(s) x Eq(s) = 1/(1 + Se/Qe + Se^2)
Not perfect, but better (for a 70 Hz burst) due to the lower cut off frequency. In the frequency domain this is observed as amplitude and phase errors. If the system is eq'ed to a low enough frequency following SL's type of Eq but say, shifting the cut off to 5 Hz or even lower (in theory to DC), then the response would be further improved.
As I said, I think you are misinterpreting SL's comments. The Eq corrects the system to have the response of an independent system with the same response as the eq'ed system. It doesn’t not make it perfect unless the independent system is perfect.
He is a plot similar to SL's except it is 5 cycles at 70 Hz. The top is Fs = 55 Hz, Q = 1.21, center Fs = 19 Hz, Q = 0.5 and bottom Fs = 2, Q = 0.5.
An externally hosted image should be here but it was not working when we last tested it.
john k... said:
Exactly ! And Fs= 2Hz ?
In our real world that degree of 'perfection' cannot be achieved, and thus I wonder why you keep presenting such theory !
The problem remains waveform distortion during reproduction time, and how best to counter it without causing additional secondary energy storage/release problems to arise, again during music reproduction time.
Do you have a working solution for this residual first half cycle LF waveform error problem, one which can be realistically integrated with the output of other drivers, if yes I'd be interested to see it ?
Cheers ......... Graham.
Forgive me Graham, but when I read your statement I did not see any caveats about what is practical, etc. I read an unconditional statement that inverse filters can not correct the problem. And I was trying to point out that they can. Not trying to be argumentative here. What is or isn't practical is another issue, to me.
The ideal of eq'ing the system flat to low frequency is an attempt to reduce the phase nonlinearity over the bandwidth of reproduction. It is the phase nonlinearity that is the root of the problem. However, there is a better way to attack it. This low frequency problem has been addressed and there is actually digital software available for correcting it. The stored energy isn't so much related to the amplitude variation of the response as it is phase. That is, it is a consequence that the system in minimum phase. If the phase of the 2nd HP response (q = 0.5, Fc = 20 Hz) were corrected to linear phase the first 1/2 cycle problem would disappear. All that would remain is amplitude scaling due to the amplitude response of the system. The Q = 0.5, 20 Hz system has an amplitude error of about -0.7dB at 70 Hz so what you would see is a 70 Hz output which followed the input exactly but the output would be scaled down by a factor of 0.96. If the response was a Q = 0.7 @ 20 Hz then this amplitude error at 70 Hz would be insignificant. Of course, at lower frequency where the response amplitude starts to roll off there would always be some amplitude scaling, though no time error.
The problem with this type of phase correction is that it requires a relatively long delay (maybe a few hundred msec, perhaps more). This isn't too much of a problem for music but would not be acceptable for sinc with a picture as required in HT.
The problem isn’t confined to low frequency. It occurs in every crossover unless the crossover is designed to transient perfect. The last point I would make is that I’m not so sure how important this low frequency problem really is. In room listening at low frequency is more about the energy storage problems of the room which will contaminate the response to a far greater degree than those associated with the woofer system.
Anyway, this is far off the topic of current drive.
The ideal of eq'ing the system flat to low frequency is an attempt to reduce the phase nonlinearity over the bandwidth of reproduction. It is the phase nonlinearity that is the root of the problem. However, there is a better way to attack it. This low frequency problem has been addressed and there is actually digital software available for correcting it. The stored energy isn't so much related to the amplitude variation of the response as it is phase. That is, it is a consequence that the system in minimum phase. If the phase of the 2nd HP response (q = 0.5, Fc = 20 Hz) were corrected to linear phase the first 1/2 cycle problem would disappear. All that would remain is amplitude scaling due to the amplitude response of the system. The Q = 0.5, 20 Hz system has an amplitude error of about -0.7dB at 70 Hz so what you would see is a 70 Hz output which followed the input exactly but the output would be scaled down by a factor of 0.96. If the response was a Q = 0.7 @ 20 Hz then this amplitude error at 70 Hz would be insignificant. Of course, at lower frequency where the response amplitude starts to roll off there would always be some amplitude scaling, though no time error.
The problem with this type of phase correction is that it requires a relatively long delay (maybe a few hundred msec, perhaps more). This isn't too much of a problem for music but would not be acceptable for sinc with a picture as required in HT.
The problem isn’t confined to low frequency. It occurs in every crossover unless the crossover is designed to transient perfect. The last point I would make is that I’m not so sure how important this low frequency problem really is. In room listening at low frequency is more about the energy storage problems of the room which will contaminate the response to a far greater degree than those associated with the woofer system.
Anyway, this is far off the topic of current drive.