darkfenriz said:
This is a good question, and I think your reasoning is OK with a couple of caveats.
First, we would probably be talking about a fairly capacitive part of the speaker load to cause additional phase shift beyond the 70 degrees assumed from compensation. If the open loop phase shift did reach 120 degrees, there would first not be a problem with amount of negative feedback, but rather the issue of concern would be the phase margin. In this case the phase margin would still be 60 degrees, which is OK.
The other thing that is much more subtle is the interaction, or lack thereof, between the two poles, the first pole being due to the compensation and the second pole being due to the capacitive impedance against the open loop output impedance.
If the compensation is achieved by Miller compensation, and the open loop output impedance is the output impedance of the VAS divided by the total beta of the output stage, then these two poles do definitely interact in a way that mitigates the effects. This is because the Miller capacitance wrapped around the VAS affects the output impedance of the VAS, making it no longer resistive, and making the open loop output impedance no longer resistive.
If, on the other hand, one used shunt compensation at the base of the VAS and resistive loading to stabilize the gain of the VAS, then these two poles would be much less interactive and the situation you envisioned of getting 120 degrees or more of open loop phase shift would be more likely.
In any case, this sort of thing makes a good case for testing amplifiers with fairly difficult and reactive loads.
Cheers,
Bob
Ok, next question
Thanks for the answer Bob, unfortunately something is wrong with my acrobat reader so I will have to try to read the IIM distortion paper at a later time.
I have another question and this one relates to an something I have observed when looking at damping factor vs. frequency plots and then comparing these to THD vs. frequency plots. In the case of many solid state amplifiers there is a pronounced drop in the damping factor at a particular frequency. This varies a bit from amp to amp but it seems to typically happen from about 500 Hz to about 2Khz. The result is a curve rather than a flat line, which is presumably ideal.
Now the other observation is that at around the same frequency for these amps the THD begins to go up. In some cases the distortion at 20kz is more than 10 times greater than it was at the point where the rise began.
The other interesting thing I have noticed is that generally the less feedback (or no feedback) the less this occurs (or not at all). Many of them have nearly flat damping factors (albeit much lower in absolute terms) and while a lot of transformer coupled amps have a higher distortion at low frequencies (transformer core saturation?) they seem to be making about the same distortion at high frequencies as at mid frequencies.
The question then is what exactly is going on with the negative feedback above this frequency because if it is taken as a fact that the high damping factor and low distortion are a result of the application of negative feedback then it would seem that the negative feedback is becoming less and less effective at higher frequencies.
Is the loop feedback "too slow" above these frequencies to properly correct the signal? Is this a case the correction signal for a high frequency arriving back at the input and no longer correcting the original signal? Or is there something else going on there?
Thanks for the answer Bob, unfortunately something is wrong with my acrobat reader so I will have to try to read the IIM distortion paper at a later time.
I have another question and this one relates to an something I have observed when looking at damping factor vs. frequency plots and then comparing these to THD vs. frequency plots. In the case of many solid state amplifiers there is a pronounced drop in the damping factor at a particular frequency. This varies a bit from amp to amp but it seems to typically happen from about 500 Hz to about 2Khz. The result is a curve rather than a flat line, which is presumably ideal.
Now the other observation is that at around the same frequency for these amps the THD begins to go up. In some cases the distortion at 20kz is more than 10 times greater than it was at the point where the rise began.
The other interesting thing I have noticed is that generally the less feedback (or no feedback) the less this occurs (or not at all). Many of them have nearly flat damping factors (albeit much lower in absolute terms) and while a lot of transformer coupled amps have a higher distortion at low frequencies (transformer core saturation?) they seem to be making about the same distortion at high frequencies as at mid frequencies.
The question then is what exactly is going on with the negative feedback above this frequency because if it is taken as a fact that the high damping factor and low distortion are a result of the application of negative feedback then it would seem that the negative feedback is becoming less and less effective at higher frequencies.
Is the loop feedback "too slow" above these frequencies to properly correct the signal? Is this a case the correction signal for a high frequency arriving back at the input and no longer correcting the original signal? Or is there something else going on there?
Re: Ok, next question
In most cases, what you are seeing is the result of the reduction in negative feedback as frequency increases, and this is due to necessary feedback frequency compensation. Since NFB generally decreases distortion and increases damping factor, these two quantities will tend to follow the change with frequency of the amount of negative feedback. In the case of no-feedback amplifiers, although the distortion and damping factor may be more flat as a function of frequency, it is largely the higher distortion and lower damping factor at low frequencies that is responsible for this. The effect is not so much one of time delay as it is one of the actual amount of negative feedback applied having to be rolled off at higher frequencies. Since there is less of it, it is less effective.
There are, however, other factors that cause THD to rise at high frequencies and for damping factor to decrease at high frequencies. The former can be caused by output stage dynamic distortions and the latter can be caused by the use of indctors in the output.
Cheers,
Bob
morricab said:
In most cases, what you are seeing is the result of the reduction in negative feedback as frequency increases, and this is due to necessary feedback frequency compensation. Since NFB generally decreases distortion and increases damping factor, these two quantities will tend to follow the change with frequency of the amount of negative feedback. In the case of no-feedback amplifiers, although the distortion and damping factor may be more flat as a function of frequency, it is largely the higher distortion and lower damping factor at low frequencies that is responsible for this. The effect is not so much one of time delay as it is one of the actual amount of negative feedback applied having to be rolled off at higher frequencies. Since there is less of it, it is less effective.
There are, however, other factors that cause THD to rise at high frequencies and for damping factor to decrease at high frequencies. The former can be caused by output stage dynamic distortions and the latter can be caused by the use of indctors in the output.
Cheers,
Bob
I have a completely different set of caveats with darkfenriz’ earlier question:
Dynamic drivers have their voice coil resistance in series with any electro-mechanically coupled impedance, since they are passive systems the reflected mechanical impedance cannot cancel the voice coil resistance, therefore the magnitude of a dynamic driver’s impedance cannot be less than its voice coil resistance at audio frequencies
So the basic premise is flawed – dynamic drivers don’t have significant mid frequency impedance dips
The setup for the question could be modified to recognize that bad crossover design could create low impedance dips in a multiway loudspeaker’s terminal impedance but it should be emphasized that the drivers are not responsible least the “back emf” nonsense gets another round of promotion
Further if we use Bob’s Mosfet amp as an example, my sim with AndyC’s components shows Open loop output impedance at ~ 3mOhm magnitude over 1 – 10 KHz, so 4- 6 Ohm voice coil resistance will not impact the loop gain/phase at all, even a really evil 2 Ohm minimum terminal impedance in badly conceived “8 Ohm” multiway loudspeaker crossover wouldn’t have appreciable influence
Disabling the error correction in the sim of Bob’s amp gives ~ 600 mOhm Open loop output Z at “mid-audio frequencies”, again one has to assume the crossover is providing a unreasonable terminal impedance dip to get even a 3 dB change in the loop gain at mid-audio frequencies – where we expect loop gain to be 100s
At frequencies where the magnitude of the loop gain of a (loaded) feedback amplifier is still >100 we describe the loop as “gain stabilized” at those frequencies – the phase may even exceed 180 degrees without rendering the gain ineffective at reducing the magnitude of output impedance or distortion components - which pretty much covers "IIM"
You might object that Bob’s amp is a poor example to argue from when darkfenriz’ posits an amp having high output impedance but Cherry makes the argument that the situation is virtually the same with either CE or CC output stages, see the 2nd paper in the download of mikeks' post #68 (“Ironing Out Distortion” part 2 “common emitter output stages” and the JAES references)
http://www.diyaudio.com/forums/showthread.php?postid=1113916#post1113916
(the 1st part discusses the rising distortion/reduced feedback vs frequency to some extent and again see the JAES refs…)
darkfenriz said:
Dynamic drivers have their voice coil resistance in series with any electro-mechanically coupled impedance, since they are passive systems the reflected mechanical impedance cannot cancel the voice coil resistance, therefore the magnitude of a dynamic driver’s impedance cannot be less than its voice coil resistance at audio frequencies
So the basic premise is flawed – dynamic drivers don’t have significant mid frequency impedance dips
The setup for the question could be modified to recognize that bad crossover design could create low impedance dips in a multiway loudspeaker’s terminal impedance but it should be emphasized that the drivers are not responsible least the “back emf” nonsense gets another round of promotion
Further if we use Bob’s Mosfet amp as an example, my sim with AndyC’s components shows Open loop output impedance at ~ 3mOhm magnitude over 1 – 10 KHz, so 4- 6 Ohm voice coil resistance will not impact the loop gain/phase at all, even a really evil 2 Ohm minimum terminal impedance in badly conceived “8 Ohm” multiway loudspeaker crossover wouldn’t have appreciable influence
Disabling the error correction in the sim of Bob’s amp gives ~ 600 mOhm Open loop output Z at “mid-audio frequencies”, again one has to assume the crossover is providing a unreasonable terminal impedance dip to get even a 3 dB change in the loop gain at mid-audio frequencies – where we expect loop gain to be 100s
At frequencies where the magnitude of the loop gain of a (loaded) feedback amplifier is still >100 we describe the loop as “gain stabilized” at those frequencies – the phase may even exceed 180 degrees without rendering the gain ineffective at reducing the magnitude of output impedance or distortion components - which pretty much covers "IIM"
You might object that Bob’s amp is a poor example to argue from when darkfenriz’ posits an amp having high output impedance but Cherry makes the argument that the situation is virtually the same with either CE or CC output stages, see the 2nd paper in the download of mikeks' post #68 (“Ironing Out Distortion” part 2 “common emitter output stages” and the JAES references)
http://www.diyaudio.com/forums/showthread.php?postid=1113916#post1113916
(the 1st part discusses the rising distortion/reduced feedback vs frequency to some extent and again see the JAES refs…)
jcx said:
These are all very good points as well.
The only case in which back EMF can be a problem that I know of is in woofers at low frequencies under conditions of unusual non-sinusoidal drive that can cause, on a momentary basis, the back EMF from the momentum of the cone to work against the applied voltage in such a way that, momentarily, the voltage that is effectively across the voice coil resistance is greater than the applied voltage.
Beyond just really badly designed speakers with really low steady-state impedance at certain frequencies, this is why it is a good idea for amplifiers to have a good output current reserve margin.
Bob
Bob, what should I do? My WATT 1 loudspeakers drop to 1/2 ohm at 2KHz.
Should I call up the designer and scold him about poor xover design. Maybe sell my speakers and revert to a another speaker that has a better behaved impedance network? Upgrade my WATT 1 to WATT 3's or more to remove the potential impedance loading problem? (Warning, this could cost me several thousand dollars)
Or lastly, design amps that can handle this momentary loading from the get-go, and ignore the problem with the loudspeaker?
Should I call up the designer and scold him about poor xover design. Maybe sell my speakers and revert to a another speaker that has a better behaved impedance network? Upgrade my WATT 1 to WATT 3's or more to remove the potential impedance loading problem? (Warning, this could cost me several thousand dollars)
Or lastly, design amps that can handle this momentary loading from the get-go, and ignore the problem with the loudspeaker?
john curl said:
Holy cow! Obviously, this is a perfect case for amplifiers with a high output current capability. I agree, design the amp to handle this amazing load.
In addition to whether the amp runs out of current, I would think that these speakers would sound quite different depending on the output impedance of the particular amp they are working with. For example, if they were fed by a tube amp with a df of only 5 or 10, imagine the coloration that would result.
I would also call up the designer and scold him, however. It's hard to imagine why a design needs to dip down that low in impedance.
Best regards,
Bob
That's what I do, I have a peak current of 135 A in my latest design. The amp I am using at the moment with the WATTs only has 105A, but what the heck, I might as well live with it. ;-) Seems to work OK, but many other amps, including some mentioned on this website would have one heck of a time living with these loudspeakers.
Unfortunately, many here have little or no idea what a real (worst case) loudspeaker load is. It might be 8uF with a 1/2 ohm of series resistance, a nominal load with a midrange dip due to an xover artifact, or it might be a number of loudspeakers in parallel, even in the same box, that just happen to see a signal that makes them draw more current than the nominal impedance curve suggests, by a factor of 2 or more.
Unfortunately, many here have little or no idea what a real (worst case) loudspeaker load is. It might be 8uF with a 1/2 ohm of series resistance, a nominal load with a midrange dip due to an xover artifact, or it might be a number of loudspeakers in parallel, even in the same box, that just happen to see a signal that makes them draw more current than the nominal impedance curve suggests, by a factor of 2 or more.
I was pointing out that that dynamic driver’s electrical terminal impedance can’t drop below its voice coil DC resistance, and generally voice coil resistance is seldom <1/2 of the rated nominal impedance
The discussion was about reflected load influence on amplifier small signal gain – where the impedance is the relevant quantity
And I pointed out that multiway loudspeaker designers can make crossover choices that do give significant impedance dips
You have added a extreme data point that a early Wilson Audio product dipped to 0.5 Ohms – Wilson doesn’t appear to have repeated this design “feature” in subsequent products – can you think of any good reason – like maybe their customer’s reasonable expectations of nominal speaker impedance rating being a guide for selecting an appropriate amp?
I don’t think anyone here has problems with designing an amp for a particular range of loads – with extreme designs welcomed
I’m sure a virtuous/vicious circle has existed in high end audio between extremely unusual loudspeaker impedances and amplifiers that can accommodate them (virtuous or vicious depending on your perspective as consumer or designer/manufacturer)
In a DIY context I would think the best chance of achieving superior system performance would be to multiamp with each purpose designed/sized amp seeing only a single driver’s terminal impedance and active crossover/time delay compensation done with today’s excellent digital processing – particularly given that many of us only have digital source
The discussion was about reflected load influence on amplifier small signal gain – where the impedance is the relevant quantity
And I pointed out that multiway loudspeaker designers can make crossover choices that do give significant impedance dips
You have added a extreme data point that a early Wilson Audio product dipped to 0.5 Ohms – Wilson doesn’t appear to have repeated this design “feature” in subsequent products – can you think of any good reason – like maybe their customer’s reasonable expectations of nominal speaker impedance rating being a guide for selecting an appropriate amp?
I don’t think anyone here has problems with designing an amp for a particular range of loads – with extreme designs welcomed
I’m sure a virtuous/vicious circle has existed in high end audio between extremely unusual loudspeaker impedances and amplifiers that can accommodate them (virtuous or vicious depending on your perspective as consumer or designer/manufacturer)
In a DIY context I would think the best chance of achieving superior system performance would be to multiamp with each purpose designed/sized amp seeing only a single driver’s terminal impedance and active crossover/time delay compensation done with today’s excellent digital processing – particularly given that many of us only have digital source
Peak current requirements, worst case, imply that for a SINGLE loudspeaker cone, a drive signal exists that will give twice the current that the DC measurement of the voice coil implies. Just read Otala's papers on peak amp current in the JAES. Also, think about 3 drivers in parallel. Now, please understand, I know were you are coming from. I would have said the SAME thing 30 years ago, but even then, I was pushing for more current, and better (less audible) protection circuits, because they 'mysteriously' fired when they should not have. ;-)
john curl said:
Yes I think you're right, this is what the abstract says:
"Based on an analysis of the equivalent circuit of a multiway loudspeaker, the possibility of large drive currents is predicted for a class of non-sinusoidal band and amplitude-limited signals. The current builds up as coherent sum of two parts; charging of the driver reactances, and simultaneous current drain by several drivers. The input current of three commercial loudspeaker systems was measured using a signal derived from on the analysis. The results show that a loudspeaker may draw currents three to six times larger than those calculable from the rated speaker impedance. This indicates that certain generally accepted power amplifier design criteria should be reconsidered."
Jan Didden
ingrast said:
Good point, but there is a lot of resistance to change. And there are challenges. To do it right, this would mean that the amplifiers and active crossovers and speakers would probably have to be sold as an integrated unit. People could no longer mate their favorite power amplifier with the speaker. And unless the amplifiers were integrated into the speaker cabinet, you'd have multiple sets of speaker cables to deal with.
Cheers,
Bob