Re: Re: On the need for output coils
As said by Bob.
I mention again a valuable article by Prof. Otala on "Intermodulation at the amplifier-loudspeaker interface" (Wireless World, Nov & Dec. 1980), which at the time was an eye-opener to me. (I have a photo copy; not sure whether this is available otherwise). It deals with this matter of final stage internal impedance.
PMA,
Regarding the effect of measuring networks, yes, it can of course do that. But it is used to improve on the effect that a direct probe connection will have anyway, which it will normally do otherwise why would one use it. One "calibrates" the network on a low impedance output, and the construction should be soldered directly on to the to-be-measured point without wiring. For C compensation I used small trimmer capacitors (10pF), which mostly could be set at < 1pF. A little bit of a task, but the only way out when the probe itself causes effects (and for us valuta challenged blokes who cannot afford a decent x100 probe).
Regards
G.Kleinschmidt said:
As said by Bob.
I mention again a valuable article by Prof. Otala on "Intermodulation at the amplifier-loudspeaker interface" (Wireless World, Nov & Dec. 1980), which at the time was an eye-opener to me. (I have a photo copy; not sure whether this is available otherwise). It deals with this matter of final stage internal impedance.
PMA,
Regarding the effect of measuring networks, yes, it can of course do that. But it is used to improve on the effect that a direct probe connection will have anyway, which it will normally do otherwise why would one use it. One "calibrates" the network on a low impedance output, and the construction should be soldered directly on to the to-be-measured point without wiring. For C compensation I used small trimmer capacitors (10pF), which mostly could be set at < 1pF. A little bit of a task, but the only way out when the probe itself causes effects (and for us valuta challenged blokes who cannot afford a decent x100 probe).
Regards
OK, I googled that Otala paper. You can get in on the net, but the AES want's a crazy $20 for it (like, you can get atleast three good quality girly magazines for that sort of money).
But wait....Uh-oh..... I did find this related paper for free download:
http://www.cordellaudio.com/papers/interface_intermodulation_distortion.pdf
Looks like Otala was telling porkies again. I'm now searching for the 55 page rebuttal signed by 15 prominent audio engineers. It must be out there somewhere.......
Cheers,
Glen
But wait....Uh-oh..... I did find this related paper for free download:
http://www.cordellaudio.com/papers/interface_intermodulation_distortion.pdf
Looks like Otala was telling porkies again. I'm now searching for the 55 page rebuttal signed by 15 prominent audio engineers. It must be out there somewhere.......
Cheers,
Glen
G.Kleinschmidt said:
Dude, that's what usenet is for
.But wait....Uh-oh..... I did find this related paper for free download:
http://www.cordellaudio.com/papers/interface_intermodulation_distortion.pdf
Looks like Otala was telling porkies again. I'm now searching for the 55 page rebuttal signed by 15 prominent audio engineers. It must be out there somewhere.......
If you want the Otala AES articles, feel free to shoot me an email.
He treats the amp as linear, with a nonzero, linear open-loop output impedance, driving a nonlinear load. He further assumes that any distortion components induced into the open-circuit output voltage by the nonlinear load are degradations of the amp's performance relative to the open-loop condition.
For 4000 quatloos, what's the flaw of that argument?
andy_c said:
Dude, that's what usenet is for
If you want the Otala AES articles, feel free to shoot me an email.
He treats the amp as linear, with a nonzero, linear open-loop output impedance, driving a nonlinear load. He further assumes that any distortion components induced into the open-circuit output voltage by the nonlinear load are degradations of the amp's performance.
For 4000 quatloos, what's the flaw of that argument?
Thanks mate. Email to: glenk (at) picknowl.com.au
Cheers,
Glen
andy_c said:
Dude, that's what usenet is for
If you want the Otala AES articles, feel free to shoot me an email.
He treats the amp as linear, with a nonzero, linear open-loop output impedance, driving a nonlinear load. He further assumes that any distortion components induced into the open-circuit output voltage by the nonlinear load are degradations of the amp's performance.
For 4000 quatloos, what's the flaw of that argument?
Andy, is there a typo in what you write, since how can it have a non-linear load and be open-circuit? Did you mean open loop as stated earlier? Or did I miss read something? I don't think I have that article so I can't look it up.
Pete B.
PB2 said:
Oh, I just meant the Thevenin equivalent voltage source of the open-loop amplifier.
I did do a correction and changed "...are degradations of the amp's performance" to "...are degradations of the amp's performance relative to the open-loop condition". That was definitely a careless error on my part
.G.Kleinschmidt said:OK, I googled that Otala paper. You can get in on the net, but the AES want's a crazy $20 for it (like, you can get atleast three good quality girly magazines for that sort of money).
But wait....Uh-oh..... I did find this related paper for free download:
http://www.cordellaudio.com/papers/interface_intermodulation_distortion.pdf
Looks like Otala was telling porkies again. I'm now searching for the 55 page rebuttal signed by 15 prominent audio engineers. It must be out there somewhere.......
Cheers,
Glen
That is soooo funny! You're being naughty again. When the cat is away the mouse will play?
Bob
andy_c said:
Dude, that's what usenet is for
If you want the Otala AES articles, feel free to shoot me an email.
He treats the amp as linear, with a nonzero, linear open-loop output impedance, driving a nonlinear load. He further assumes that any distortion components induced into the open-circuit output voltage by the nonlinear load are degradations of the amp's performance relative to the open-loop condition.
For 4000 quatloos, what's the flaw of that argument?
Hi Andy,
First, and importantly, Otala proposed a nice test for the Interface Intermodulation distortion (IIM) that he was describing. The test was a variation on the SMPTE-IM test where 60 Hz and 6000Hz in a 4:1 ratio are put through an amplifier. Only for IIM, the 60 Hz signal is reverse-injected into the output of the amplifier under test. This is an interesting and useful test.
However, Otala asserted that an amplifier with a higher feedback factor and higher open-loop output impedance would suffer more IIM compared to one with a low feedback factor and a low open-loop output impedance with the same closed loop output impedance. In other words, an amplifier with 40 dB NFB and 10 ohms open loop output impedance would produce more IIM than an otherwise-similar amplifier with 20 dB of NFB and 1.0 ohm of open loop output impedance (both amplifiers with a closed loop output impedance of 0.1 ohm for a DF = 80).
This turns out not to be the case. I actually built a test amplifier that could operate in either condition and measured it both ways. IIM was essentially unchanged between the two conditions. If you think about the amplifier in terms of input-referred distortion, you can see quite easily that high NFB does not exacerbate IIM. Otala just did not like NFB.
Cheers,
Bob
I agree that his proposed test is a useful one. But I found his articles to be a confused jumble of ideas.
As I see it, there are two separate aspects of this problem. The first is the question of how linear the closed-loop output impedance of the amplifier is. The second is how sensitive the distortion of the amp is to nonlinearity of the load itself.
For example, one could imagine, and model in SPICE if one wished, an amplifier whose open-loop, open-circuit output voltage is a purely linear function of its difference mode input voltage, but whose output impedance is nonlinear. Of course this is unrealistic and idealized, since there will be distortion in the open-loop, open-circuit output voltage in any real amplifier. It seems to me that this is the issue addressed by the test in which the input of the amp is driven by a signal of one frequency, while a signal of another frequency is back-injected into the output.
One could also imagine a purely linear amplifier with an output impedance that's not zero. Of course, such an amplifier, though purely linear, will still produce distortion when driving a nonlinear load impedance. Yet such an amp will produce zero distortion when driven by in input signal of frequency f1, with a back-injected signal of frequency f2 at the output - even if its output impedance were not so small.
Of course, in real life the amplifier distortion into a nonlinear load could be thought of as a combination of these two scenarios. Clearly, voltage feedback will improve the second case above by reducing the output impedance. There can be no doubt of that. But what about the first case above? This gets back to what I was saying in my earlier post.
In Otala's first AES article on this [1], he shows in Figure 9 an amplifier with open-loop, open-circuit output voltage V4, driving a load which consists of some reactive components plus a current source which will back-inject a signal into the amp. He uses as a figure of merit the ratio of the "normal" component of V4 due to the input signal alone, to the component of V4 due to the back injected current. But the voltage V4 is not the output of the amplifier! Part of the distortion of the voltage V4 is what Cherry would refer to as "anti-distortion", namely an internal distortion signal which partly cancels out the distortion at the output that would otherwise occur due to the back-injected signal. Otala views any increase in this internal distortion as a degradation of the amp's performance relative to the open-loop condition. I view this as the fundamental flaw in his argument that caused him to reach the incorrect conclusion you mentioned about the IIM of amplifiers with various combinations of open-loop output impedance and feedback. Specifically, it's a logic flaw to place a figure of merit on the distortion of some internal voltage of the amp. The only relevant distortion figure is that of the output.
I hope this clears up what I was trying to say earlier.
[1] Intermodulation Distortion in the Amplifier Loudspeaker Interface, Matti Otala, AES preprint 1336
As I see it, there are two separate aspects of this problem. The first is the question of how linear the closed-loop output impedance of the amplifier is. The second is how sensitive the distortion of the amp is to nonlinearity of the load itself.
For example, one could imagine, and model in SPICE if one wished, an amplifier whose open-loop, open-circuit output voltage is a purely linear function of its difference mode input voltage, but whose output impedance is nonlinear. Of course this is unrealistic and idealized, since there will be distortion in the open-loop, open-circuit output voltage in any real amplifier. It seems to me that this is the issue addressed by the test in which the input of the amp is driven by a signal of one frequency, while a signal of another frequency is back-injected into the output.
One could also imagine a purely linear amplifier with an output impedance that's not zero. Of course, such an amplifier, though purely linear, will still produce distortion when driving a nonlinear load impedance. Yet such an amp will produce zero distortion when driven by in input signal of frequency f1, with a back-injected signal of frequency f2 at the output - even if its output impedance were not so small.
Of course, in real life the amplifier distortion into a nonlinear load could be thought of as a combination of these two scenarios. Clearly, voltage feedback will improve the second case above by reducing the output impedance. There can be no doubt of that. But what about the first case above? This gets back to what I was saying in my earlier post.
In Otala's first AES article on this [1], he shows in Figure 9 an amplifier with open-loop, open-circuit output voltage V4, driving a load which consists of some reactive components plus a current source which will back-inject a signal into the amp. He uses as a figure of merit the ratio of the "normal" component of V4 due to the input signal alone, to the component of V4 due to the back injected current. But the voltage V4 is not the output of the amplifier! Part of the distortion of the voltage V4 is what Cherry would refer to as "anti-distortion", namely an internal distortion signal which partly cancels out the distortion at the output that would otherwise occur due to the back-injected signal. Otala views any increase in this internal distortion as a degradation of the amp's performance relative to the open-loop condition. I view this as the fundamental flaw in his argument that caused him to reach the incorrect conclusion you mentioned about the IIM of amplifiers with various combinations of open-loop output impedance and feedback. Specifically, it's a logic flaw to place a figure of merit on the distortion of some internal voltage of the amp. The only relevant distortion figure is that of the output.
I hope this clears up what I was trying to say earlier.
[1] Intermodulation Distortion in the Amplifier Loudspeaker Interface, Matti Otala, AES preprint 1336
We seem to have steered away somewhat from Bob's intention that we should consider the necessity of an output coil with regard to stability. This was set off by my reference to the Otala article, which I intended simply to accentuate that the power stage internal impedance has something to do with matters, as suggested by Glen. Apology for that.
I nevertheless followed the ensuing discussion with interest. Some statements were not quite borne out by my own findings, but about that later - I still have to read the references shown.
Back to coils, I am busy simulating the normal 3 output configurations (emitter follower, quasi e.f. and full complementary pair) on their own for stability with capacitive load, and indications are that some will do that better than others, but with other disadvantages. I am using the FCP because of its superior distortion characteristics, but indications are that because of the high feedback with the particular internal C's, it is not conducive to certain capacitive loads. A small series inductor appears to be the simplest solution there. As they say on tv: To be continued .....
Regards.
I nevertheless followed the ensuing discussion with interest. Some statements were not quite borne out by my own findings, but about that later - I still have to read the references shown.
Back to coils, I am busy simulating the normal 3 output configurations (emitter follower, quasi e.f. and full complementary pair) on their own for stability with capacitive load, and indications are that some will do that better than others, but with other disadvantages. I am using the FCP because of its superior distortion characteristics, but indications are that because of the high feedback with the particular internal C's, it is not conducive to certain capacitive loads. A small series inductor appears to be the simplest solution there. As they say on tv: To be continued .....
Regards.
Johan Potgieter said:
Johan,
Thanks for getting us back on track. I think that a few SPICE simulations are a very good idea, and can provide some insight here. Indeed, stability analysis and the like is one of my favorite uses for SPICE.
I might suggest that it would be useful to look at the stability issue with both low-ft (2 MHz) and high-ft (30 MHz) output devices. John Curl suggested that output stages with higher ft output devices might be less prone to capacitive load instability, and this may have merit.
I also think that the focus on the high-frequency open-loop ouput impedance of the output stage is especially important to looking at what a capacitive load (or one isolated by a minimal amount of coil inductance) will do in terms of de-stabilizing the global NFB loop.
Your mention of the CFP also brings to light the importance of local stability in the output stage under capacitive loading conditions, especially where there is an explicit local loop as with a CFP (and for EC, for that matter).
The ability to form oscillator topologies with emitter followers and source followers looking into capacitive loads, particularly when there are some stray inductances in the layout, is also something important (and which can be helpfully examined by SPICE). In these situations, it would seem that the presence of emitter or source ballast resistors might tend to mitigate the tendency to instability in some cases, however.
Cheers,
Bob
Output impedance of Hafler DH220
I just measured the output impedance of a Hafler DH220 lateral MOSFET amplifier. In an earlier post, Glen had expressed concern about the HF output impedance (aka effective output inductance) of an amplifier using a MOSFET output stage, so I thought it would be interesting to measure one.
For the measurement I used my HP 3580A spectrum analyzer and its tracking oscillator to plot the output impedance as a function of frequency. I buffered the tracking oscillator output with a wideband power buffer flat to 2 MHz, and applied the buffered signal to the output of the Hafler through a precision 10-ohm resistor. Operating level into the 10 ohm resistor was 100 mV rms.
I applied this signal and the input of the spectrum analyzer to the Haflker output terminals with a Kelvin connection and took the measurement, sweeping out to 50 kHz. I then did the same thing on the "upstream" side of the coil. Here are some of the results:
100 Hz: 0.025 ohms and 0.004 ohms
2 kHz: 0.032 ohms and 0.006 ohms
10 kHz: 0.11 ohms and 0.018 ohms
20 kHz: 0.20 ohms and 0.040 ohms
30 kHz: 0.28 ohms and 0.050 ohms
50 kHz: 0.40 ohms and 0.079 ohms
These numbers suggest that the output inductance with coil is about 1.5 uH.
The output inductance without the output coil is on the order of 0.25 uH.
So, bottom line is that even an ordinary lateral MOSFET amplifier has a pretty low effective output inductance before the coil.
Cheers,
Bob
I just measured the output impedance of a Hafler DH220 lateral MOSFET amplifier. In an earlier post, Glen had expressed concern about the HF output impedance (aka effective output inductance) of an amplifier using a MOSFET output stage, so I thought it would be interesting to measure one.
For the measurement I used my HP 3580A spectrum analyzer and its tracking oscillator to plot the output impedance as a function of frequency. I buffered the tracking oscillator output with a wideband power buffer flat to 2 MHz, and applied the buffered signal to the output of the Hafler through a precision 10-ohm resistor. Operating level into the 10 ohm resistor was 100 mV rms.
I applied this signal and the input of the spectrum analyzer to the Haflker output terminals with a Kelvin connection and took the measurement, sweeping out to 50 kHz. I then did the same thing on the "upstream" side of the coil. Here are some of the results:
100 Hz: 0.025 ohms and 0.004 ohms
2 kHz: 0.032 ohms and 0.006 ohms
10 kHz: 0.11 ohms and 0.018 ohms
20 kHz: 0.20 ohms and 0.040 ohms
30 kHz: 0.28 ohms and 0.050 ohms
50 kHz: 0.40 ohms and 0.079 ohms
These numbers suggest that the output inductance with coil is about 1.5 uH.
The output inductance without the output coil is on the order of 0.25 uH.
So, bottom line is that even an ordinary lateral MOSFET amplifier has a pretty low effective output inductance before the coil.
Cheers,
Bob
Re: Output impedance of Hafler DH220
I said that a low power MOSFET amplifier can easily have an effective output inductance in the uH range. Of course, this depends on the design. Your measurements of the DH220 do not negate my statement.
The DH220 has two pairs of parallel connected output devices and they are driven from a low impedance source consisting of a complementary emitter follower – the latter being particularly significant.
I would not expect such a design to have a particularly high effective output inductance.
Cheers,
Glen
Bob Cordell said:
I said that a low power MOSFET amplifier can easily have an effective output inductance in the uH range. Of course, this depends on the design. Your measurements of the DH220 do not negate my statement.
The DH220 has two pairs of parallel connected output devices and they are driven from a low impedance source consisting of a complementary emitter follower – the latter being particularly significant.
I would not expect such a design to have a particularly high effective output inductance.
Cheers,
Glen