When considering designing an amplifier without an inductor, there are two main stability considerations that must be kept in mind. The first is the global feedback loop stability and the second is the local stability of the output stage.
For global feedback stability, the direct loading of the output stage by a substantial shunt capacitance will add a pole to the open-loop gain and decrease stability. In very simplified terms, if that added pole is an octave above the ULGF, you will lose about 23 degrees of phase margin. An amp that was designed with, say 60 degrees of phase margin in the first place will certainly tolerate this. If the amp's ULGF was 1MHz, an output capacitor that adds a pole at 2MHz should therefore not be a big problem. Where this pole lands depends on the high-frequency open-loop output impedance and the value of the shunt capacitance.
A really beefy output stage with low HF output impedance can overcome a fairly large capacitive load without the pole moving down too close to the ULGF. Of course, we must define what that reasonable maximum load capacitance (without series resistance) is. Let's say it is 0.1uF. A beefy output stage with 1 ohm HF output impedance will create a pole at 1.6MHz with a 0.1uF load. We can see that this is tolerable. REALLY beefy output stages with 4-5 paralleled pairs of fast output transistors can have an open-loop HF output impedance well less than an ohm in the low MHz range.
I do not consider a 1uF load with no series resistance a reasonable load. It is easy to see that even just 1 ohm in series with such a capacitance will greatly mitigate its effect, however.
The second issue of local output stage stability must also be considered. We all know that emitter followers like to oscillate when loaded with a capacitance. We also know that output Triples are more inclined to parasitic oscillations as well. Thus, even if there were no global feedback we would have to deal with this issue. A beefy, well-designed output stage with well-chosen base stopper resistors and perhaps some base Zobel networks can be made to be stable locally with reasonable values of capacitive load, like 0.1uF. This can be tricky, and somewhat layout-dependent, but can be tackled with great care by some skilled designers. I would be reluctant to assert this for capacitances much above 0.1uF that do not have any resistance in series.
Thus, some designers can get away without a coil if the load is remotely reasonable and great care and skill is applied AND lots of rigorous testing is done. However, this is not for the average DIY designer or the majority of commercial designers, as often the rigorous testing needed to assure the reliability of a commercial product that will be made in quantity is simply not done for things like capacitive load stability. As mentioned, it is not just the pure bench capacitive load that must be tested, but also combinations involving different speaker cables and lengths. Will you be stable with an un-terminated length of any speaker cable on the market up to a length of 25 feet? Not just its capacitance, but its transmission line reflection issues (yes, they do indeed act like transmission lines beginning in the low MHz region - see figure 18.4 in my book).
Cheers,
Bob
For global feedback stability, the direct loading of the output stage by a substantial shunt capacitance will add a pole to the open-loop gain and decrease stability. In very simplified terms, if that added pole is an octave above the ULGF, you will lose about 23 degrees of phase margin. An amp that was designed with, say 60 degrees of phase margin in the first place will certainly tolerate this. If the amp's ULGF was 1MHz, an output capacitor that adds a pole at 2MHz should therefore not be a big problem. Where this pole lands depends on the high-frequency open-loop output impedance and the value of the shunt capacitance.
A really beefy output stage with low HF output impedance can overcome a fairly large capacitive load without the pole moving down too close to the ULGF. Of course, we must define what that reasonable maximum load capacitance (without series resistance) is. Let's say it is 0.1uF. A beefy output stage with 1 ohm HF output impedance will create a pole at 1.6MHz with a 0.1uF load. We can see that this is tolerable. REALLY beefy output stages with 4-5 paralleled pairs of fast output transistors can have an open-loop HF output impedance well less than an ohm in the low MHz range.
I do not consider a 1uF load with no series resistance a reasonable load. It is easy to see that even just 1 ohm in series with such a capacitance will greatly mitigate its effect, however.
The second issue of local output stage stability must also be considered. We all know that emitter followers like to oscillate when loaded with a capacitance. We also know that output Triples are more inclined to parasitic oscillations as well. Thus, even if there were no global feedback we would have to deal with this issue. A beefy, well-designed output stage with well-chosen base stopper resistors and perhaps some base Zobel networks can be made to be stable locally with reasonable values of capacitive load, like 0.1uF. This can be tricky, and somewhat layout-dependent, but can be tackled with great care by some skilled designers. I would be reluctant to assert this for capacitances much above 0.1uF that do not have any resistance in series.
Thus, some designers can get away without a coil if the load is remotely reasonable and great care and skill is applied AND lots of rigorous testing is done. However, this is not for the average DIY designer or the majority of commercial designers, as often the rigorous testing needed to assure the reliability of a commercial product that will be made in quantity is simply not done for things like capacitive load stability. As mentioned, it is not just the pure bench capacitive load that must be tested, but also combinations involving different speaker cables and lengths. Will you be stable with an un-terminated length of any speaker cable on the market up to a length of 25 feet? Not just its capacitance, but its transmission line reflection issues (yes, they do indeed act like transmission lines beginning in the low MHz region - see figure 18.4 in my book).
Cheers,
Bob
Re Cables & Output etc
Some HiFi
manufactures actually design their Amps to be ONLY used with their cables ! If you use others, it ain't pretty.
Now whether this is just a marketing ploy to sell their cables as well ?
Also, some Pro Amps take a return feed from the speaker/s & use this to control Xmax & Thermal OL etc. Maybe they could, or do, use this feed to offset nasties too ?
Some HiFi
manufactures actually design their Amps to be ONLY used with their cables ! If you use others, it ain't pretty.Now whether this is just a marketing ploy to sell their cables as well ?
Also, some Pro Amps take a return feed from the speaker/s & use this to control Xmax & Thermal OL etc. Maybe they could, or do, use this feed to offset nasties too ?
There is a distinction between overshoot that is settling/step behavior and overshoot that is peaking. Peaking may appear to be overshoot except it will worsen as the input increases. Settling behavior does not change with signal level.
An amp can be perfectly stable and have ringing at the output, because its output impedance interacts with load impedance.
In my experience phase margin and ringing are correlated but not directly related.
It is possible to have overshoot due to output impedance that is not ringing. I believe this can happen with TMC if your OPS has low current gain, similar to how TPC will cause a gain lift at RF which is overshoot. This is side-effect typical for multi-pole compensation schemes, though you can get around it if you drive the compensation with a dedicated buffer. I have never seen this done in a real design before except in chips.
Yes, you will see increased overshoot and frequency response peaking as phase margin decreases. Take a look at Figure 4.10 in my book. There you will see a graph of overshoot and peaking as a function of phase margin for an amplifier with dominant pole compensation and four added parasitic poles above the ULGF. At a 30 degree phase margin you will see overshoot of about 80% and peaking of about 7dB.
Your results may vary depending on the particulars of your amplifier.
Cheers,
Bob
Consider the well-intentioned guy who decides to go into the speaker cable business because he thinks he has a good idea. He has enough knowledge to be dangerous, like many in the high-end.
He understands the concept of impedance matching and reflections, so he decides that his cable will have a characteristic impedance of about 6.3 ohms, since an honestly rated 8-ohm speaker usually comes in with a DCR of about 6.3 ohms (you can already see that he has enough knowledge to be dangerous). He then decides to use very high-quality RG58AU with silver plated copper center conductor and braided shield. Maybe a Teflon dielectric if he can get it.
He'll use eight coax's in parallel to get his desired characteristic impedance. He believes in symmetry, so on 4 of them the plus speaker line will travel the center conductor and on the other 4 it will travel the shield. He'll braid the 8 coax's together in a fancy way. He'll use the finest thick-gold plated connectors, not made in China, if possible. Because some people don't like their power amplifiers anywhere near their speakers, he'll make the cable available in a 25-foot length. He'll sell a pair for $500 per foot. $12,500 for a 25-foot pair - not bad. Yada, yada, yada.
He's not a complete nerd, and actually has some business sense. He hires a crack marketing guy with a thesaurus and a graphic arts guy to help create great ads. The marketing guy came from the women's cosmetics business where they perpetually come up with scientific-sounding names for cold cream. The graphic arts guy came from the video game business after he previously got laid off when they finished the last installment of Star Wars.
With RG58 coming in at about 25pF/ft, his 25-foot cables have a capacitance of about 5000pF. The one-way time delay is about 40ns. It is a quarter wave at about 6MHz, with a really mean impedance dip there when unterminated, at least at high frequencies.
I bet this could cook a few amplifiers without coils. But it sounds really good to him and his brother-in-law when it doesn't .
Cheers,
Bob
He understands the concept of impedance matching and reflections, so he decides that his cable will have a characteristic impedance of about 6.3 ohms, since an honestly rated 8-ohm speaker usually comes in with a DCR of about 6.3 ohms (you can already see that he has enough knowledge to be dangerous). He then decides to use very high-quality RG58AU with silver plated copper center conductor and braided shield. Maybe a Teflon dielectric if he can get it.
He'll use eight coax's in parallel to get his desired characteristic impedance. He believes in symmetry, so on 4 of them the plus speaker line will travel the center conductor and on the other 4 it will travel the shield. He'll braid the 8 coax's together in a fancy way. He'll use the finest thick-gold plated connectors, not made in China, if possible. Because some people don't like their power amplifiers anywhere near their speakers, he'll make the cable available in a 25-foot length. He'll sell a pair for $500 per foot. $12,500 for a 25-foot pair - not bad. Yada, yada, yada.
He's not a complete nerd, and actually has some business sense. He hires a crack marketing guy with a thesaurus and a graphic arts guy to help create great ads. The marketing guy came from the women's cosmetics business where they perpetually come up with scientific-sounding names for cold cream. The graphic arts guy came from the video game business after he previously got laid off when they finished the last installment of Star Wars.
With RG58 coming in at about 25pF/ft, his 25-foot cables have a capacitance of about 5000pF. The one-way time delay is about 40ns. It is a quarter wave at about 6MHz, with a really mean impedance dip there when unterminated, at least at high frequencies.
I bet this could cook a few amplifiers without coils
. But it sounds really good to him and his brother-in-law when it doesn't .Cheers,
Bob
That was the question I asked JC, he understood what I was talking to test the amplifier load LC (reasonable).
This text was taken from the website the M.Leach:
Some of the so-called "high definition" cables are designed to minimize the series inductance. Because the series inductance per unit length multiplied by the shunt capacitance per unit length is a constant that is equal to the reciprocal of the velocity of light squared, minimizing the inductance maximizes the capacitance. Therefore, these cables can exhibit a high shunt capacitance. For this reason, I generally do not recommend them. I have heard of a certain "high-end" amplifier that is no longer made smoking when these cables are connected to its output, even with no signal input. A former student, who was working for Panasonic, performed a listening test in their "high-end" listening room to see if there was any audible difference between a "high definition" cable and ordinary zip cord. I do not know what power amplifier they used, but he said that the listeners were in agreement that the high-frequency response was better with the zip cord, no doubt because the high capacitance of the "high definition" cable.
The Leach Amp - Background
Note that even with cable "bad" you would not have more than a few nF.
It really is a hard to test to amplifier, for heavy load, you need high capacity in the current output, If you do not have high capability output current, the inductor will isolate the amplifier a heavy load that is not directly related to oscillation, because you have problems also with low resistive loads.
Another test, is to use the method that Bob Cordell used in his Mosfet Error Corretion using a capacitor 1uF in series 1 Ohm, is not the same thing as testing the amp with 1 ohm, capacitor in series with resistor, will load the amp only in high frequency, will show stability problems, the amp probably find kind of load in the real world.
Yes that's right, but this can result in different value for the inductor than for a test "in born" in some cases without the inductor can do better than with the inductor.
BTW: In a while ago you did a test putting a cable and more one capacitor of 100nF, your amplifier oscillated even using an inductor in the output, think you have solved the problem.
You know where is this post? I could not find...
most commercial floor standing electrostatic are driven from a step up transfromer with SRF ~1-2 octaves above 20 kHz - and there is series winding R - they are not pure uF Cload anywhere in frequency on the amp side, especially near typical audio power amp unity loop gain intercept frequency
maybe piezo supertweeters? - but many will have series R in the level matching pad of the crossover
maybe piezo supertweeters? - but many will have series R in the level matching pad of the crossover
Hi jcx,
Thanks for this observation about electrostatics. I've never owned or measured an electrostatic, but I was under the impression that the step-up transformer had some primary DCR that would effectively be in series with the large capacitance reflected back to the primary by the turns ratio.
I think that amplifiers can be designed to work OK without coils if very carefully designed and verified with a great variety of loads, and will be able to be safe with most remotely-reasonable loads. Nevertheless, I think it is wise to employ a coil of small value (0.5 to 2uH) just to be safe and that such a coil, if properly implemented, will be inaudible.
Stability at rest is not enough to evaluate. It is important to make sure that the amplifier never breaks out into a parasitic oscillation burst under dynamic signal V-I conditions that may be hard to replicate under lab conditions.
Cheers,
Bob
I was just reminded of the amazing differences small changes can make in the subjective sound of an amp. I've been playing with a feedforward EC buffer and I adjusted a trimmer by 2.5%. Suddenly, I was thrown to the floor by the impact of the bass. The music flew out of the speakers. Instruments suddenly occupied space.
I have built several amps of my own design and while I always tried to get good measured specs, I always like to play with the sound. Every time I improve imaging, I learn a bit more about my music. For instance, I was just listening to one track and realized the singer was actually several feet from the microphone; on lesser amps it just sounded like he was right next to it. As imaging improves, you can hear more about the room the music is recorded in.
This is very hard to achieve, but I have occasionally found settings where voices became very real, to the point of intimacy. It sounds like the singer is singing TO you, not just singing. You hang on to every word. If the singer isn't into it, you can tell. But when they are, there can be an overwhelming empathy.
Unfortunately these prototypes, like all my circuits so far, always had difficult flaws, since I wan't particularly worried about overload protection, stability (ha! ...) or practicality.
I have built several amps of my own design and while I always tried to get good measured specs, I always like to play with the sound. Every time I improve imaging, I learn a bit more about my music. For instance, I was just listening to one track and realized the singer was actually several feet from the microphone; on lesser amps it just sounded like he was right next to it. As imaging improves, you can hear more about the room the music is recorded in.
This is very hard to achieve, but I have occasionally found settings where voices became very real, to the point of intimacy. It sounds like the singer is singing TO you, not just singing. You hang on to every word. If the singer isn't into it, you can tell. But when they are, there can be an overwhelming empathy.
Unfortunately these prototypes, like all my circuits so far, always had difficult flaws, since I wan't particularly worried about overload protection, stability (ha! ...) or practicality.
Well, pardon my hyperbole. I don't actually need alcohol to feel normal emotions.
Furthermore, it's perfectly in the realm of possibility that a 2.5% change could have caused subjective differences, because this is a feedforward EC buffer and outside the null, can create a very powerful 2nd harmonic. Incidentally, it does this while producing very little H3 and beyond.
Furthermore, it's perfectly in the realm of possibility that a 2.5% change could have caused subjective differences, because this is a feedforward EC buffer and outside the null, can create a very powerful 2nd harmonic. Incidentally, it does this while producing very little H3 and beyond.
Kean, could you show schematic of that EC buffer, and, sorry, I don't know what EC means?
Oh, I think it's error correction?
Damir
