PMA said:
Here are a few thoughts on the subject off the top of my head.
Pole splitting usually refers to the effect had on the poles by Miller effect compensation. At the VAS, you may have two poles, one at the base and one at the collector.
These might be at the same frequency, say 20 kHz, or at least in the same decade. Where they run together, each at 6 dB/octave, the phase will add up too much by the time you reach the gain crossover frequency. With Miller compensation, one is moved down and one is moved up in frequency, typically by a similar factor. This factor could be 100X. In that case, they would then be located at 200 Hz and 2 MHz. We then have a straight 6 dB/octave rolloff out to beyond the NFB gain crossover frequency (assuming it lies below 2 MHz.
Regarding settling time, we must consider both the open loop case and the closed loop case, and the definition of settling time.
In the closed loop case, settling time is improved because the phase margin is improved at the gain crossover frequency. This means less overshoot and ringing. Settling time is very good for a closed loop response that has a 6 dB/octave rolloff. With less than 90 degrees of phase margin, the damping of the closed loop rolloff becomes less and the rolloff slope increases.
In the open loop, the settling time must be defined in the context of the time it takes to settle to the ideal value of the circuit. Since the ideal open loop circuit function is that of an integrator, the value to which we are supposed to settle is a constantly changing one. In any case, an open loop response will settle to the desired integrator response very quickly because the non-ideal second pole will have been pushed out to a very high frequency.
I hope this explanation helps.
Cheers,
Bob
JPV said:
I have not used two-pole compensation much, but it is a popular way to get more feedback at the high end of the audio spectrum.
We used to call this "T" compensation back at Bell Labs in the 1970's, where we built IC op amps that specifically incorporated it internally so that we could obtain more precise active filter responses with op amps of limited GBW. It was called "T" compensation because the Miller capacitor was replaced with two series capacitors and a shunt resistor to ground at their junction, forming a "T".
Genrally, my recommendation is to shoot for a 9 dB/octave open loop rolloff beginning at 20 kHz (where it transitions from the basic standard 6 dB/octave rolloff to the 9 dB/octave slope) and extending out to one octave below the NFb gain crossover frequency. This will generally assure unconditional stability. Even if the gain drops so that the gain crossover decreases into the region where you have the 9 db/octave slope, you will still have about 45 degrees of phase margin.
One thing to watch out for is to not allow the T network to unduly load the output of the VAS. This can be avoided by buffering the network's pick-off point, or by tapping it off of the predriver.
I worry most about any form of output stage crossover distortion, as it is the least "natural" and most difficult to get rid of.
Cheers,
Bob
Bob,
thank you for explanation. I meant closed-loop settling time.
Actually I was thinking about pole-zero "frequency twin", that may result in long settling of the closed loop response. Depending on pole-zero position, the response has 2 parts - fast and slow. There can be quite fast settling to say 90% of output step, followed by long settling (undershoot or overshoot, both possible). The solution is pole-zero cancellation.
Regards,
Pavel
P.S.: this solution is still not optimal, as long settling (under 1% range) may still remain. For fast settling, the frequency twin should not be more distant from fT than 1 decade, though close placement to fT is again not optimal because of possible instability.
thank you for explanation. I meant closed-loop settling time.
Actually I was thinking about pole-zero "frequency twin", that may result in long settling of the closed loop response. Depending on pole-zero position, the response has 2 parts - fast and slow. There can be quite fast settling to say 90% of output step, followed by long settling (undershoot or overshoot, both possible). The solution is pole-zero cancellation.
Regards,
Pavel
P.S.: this solution is still not optimal, as long settling (under 1% range) may still remain. For fast settling, the frequency twin should not be more distant from fT than 1 decade, though close placement to fT is again not optimal because of possible instability.
PMA said:
These are good points. This kind of settling concern, i.e., to the last 1% or 10% has always been a concern in amplifiers used in A2D converter systems. Its a little less clear how important it is to audio systems when the associated frequency-domain errors lie well above the audio band. Something good to ponder, though.
Bob
Bob,
I agree, this may not be of concern for audio. It is just my professional deformation (design of analog part of measuring instruments).
Regards,
Pavel
I agree, this may not be of concern for audio. It is just my professional deformation (design of analog part of measuring instruments).
Regards,
Pavel
So for those among us that usually stick to class A, which distortion should be most worrisome?Bob Cordell said:
For those who are interested in NFB issues, I would like to recommend this book:
http://www.amazon.com/Operational-Amplifiers-Edn-Design-Engineers/dp/0750693177
http://www.amazon.com/Operational-Amplifiers-Edn-Design-Engineers/dp/0750693177
i only have the first edition of this book.
if the second edition is anything at all like the first edition,
i strongly second this recommendation.
it is expensive, but well worth it.
mlloyd1
if the second edition is anything at all like the first edition,
i strongly second this recommendation.
it is expensive, but well worth it.
mlloyd1
PMA said:
Nixie said:
This is a good question, and I don't have a quick answer. I guess I would still be concerned about output stage distortions, even though with Class A they might not be crossover distortions. For example, in a Class A BJT amplifier, I'd still be pretty concerned about Beta droop distortion and high frequency distortion caused by insufficient turn-off speed of the output devices (even though we don't turn the transistors totally off in Class A, they still go through a high dI/dt with high level high-frequency signals). We must make sure that they don't slip into some form of Class AB or worse, totem-pole conduction, under dynamic conditions.
I am generally less concerned about VAS and input stage distortions because I know that I can make them pretty much arbitrarily small with good circuits consisting mainly of small-signal transistors.
It would be interesting to see how Nelson would answer this question, since he has far more experience with Class A than I do.
Cheers,
Bob
Bob Cordell said:
I would agree. From an objective and subjective view, the
output stage is the elephant on the dance floor. Class A
operation minimizes problems, but its energy consumption
makes it impractical for most applications.
For the average home amplifier, that's not that much consumption. Take a car burning a gallon of gas. A medium powered amp drawing 1 kW from the mains would run for 30-40 hours on the same amount of energy. It seems to me that the heat is a far bigger issue, and watercooling is one interesting solution.Nelson Pass said:
My plasma speakers supply draws 1800 W from the mains. Now that would have some effect on the electricity bill, but it's still small compared to, say, heating the house in winter (and indeed this offsets the cost somewhat).
No, but my design is derivative of Hill's. The Plasmatronic speakers dissipate 165 W average in the plasma in each channel. I'm doing about 600 W (aiming for the same SPL but much lower crossover, around 400-500, and even lower distortion which is the case in higher idle dissipation). And no need for helium, which I think was really the main reason his design didn't have much commercial success -- who wants to refill helium tanks every couple of months...
Relay Distortion
This is a bit off-topic, but should be of interest in connection with the pursuit of low distortion.
It is often frowned upon in high end circles to incorporate a relay in the signal path to a loudspeaker. However, there are times when this is unavoidable. An example would be a remote switch for listening comparisons. For this reason, I undertook to do some distortion measurements on a variety of relays. All of these relays had contact ratings between 30A and 80A.
The test setup involved applying a 16V 1 kHz test signal to an 8-ohm load (32 Watts), producing a current of 2A rms. The relay under test was placed in the return (ground) leg of the 8-ohm load. A spectrum analyzer was connected across the relay contacts. With this setup, the analyzer’s response at 1 kHz would be reflective of the resistance of the relay, while harmonics in the spectrum would give an idea of the distortion.
The good news is that one of the relays was exceptionally good; the bad news is that some were surprisingly bad. This means that relays for this kind of application need to be selected carefully. DC resistance was not well-correlated with distortion.
Six relays were tested. The results below were the analyzer readings referred to 0 dB being 100 mV across the relay. Since 2 Amps are flowing, 0 dB corresponds to 50 milliohms. Because this voltage is what would normally be in series with the speaker, a 0 dB reading for a distortion component would correspond to 100 mV out of a total of 16 V, or 0.625% distortion (-44 dB). Thus, a distortion component that is down 80 dB on the analyzer corresponds to 0.00006% (-124 dB). Here are the results:
Rly 1st 2nd 3rd 4th 5th 6th 7th
#1 -12 -86 -64 -91 -85 <-90 <-90
#2 -3 -72 -50 -86 -80 <-90 <-90
#3 -6 -87 -67 <-90 <-90 <-90 <-90
#4 -15 -96 -91 <-100 <-100 <-100 <-100
#5 -10 -69 -56 -80 -73 -91 -85
#6 -3 -<-90 -59 GRUNGE
Relay numbers 2 and 6 were notably bad, each with 35 milliohms resistance and third order distortion as bad as -50 dB on the analyzer, corresponding to in-service third order distortion of 0.002% (-94 dB). Relay #6 had a spectrum characterized by quite a bit of grunge above the analyzer grass. Third order distortion predominated on most relays.
Relay numbers 1, 3, and 5 were of medium performance, with resistance ranging from 25 milliohms down to 13 milliohms, and third order distortion on the analyzer ranging from -56 through -64 dB, corresponding to 0.001% (-100 dB) to 0.0004% (-108 dB).
Relay #4 was the clear winner. This was an 80-Amp automotive relay. It had a resistance of 18 milliohms and third order distortion of -91 dB on the analyzer, corresponding to 0.00002% (-135 dB).
Although Relay #4 was the only “automotive” relay in the test, it would be premature to conclude that an automotive relay will always be better than others for the audio application. Similarly it would be premature to conclude that the higher current rating is responsible for the enhanced performance, based on the variability among the other relays which were rated at 30 Amps. There is also no evidence that contact resistance is a good predictor of distortion performance. I’m tempted to speculate that the specifics of the contact surface chemistry has a large role in the distortion performance, in that two different sets of contacts might exhibit the same contact resistance, but different amounts of nonlinearity in that resistance.
Cheers,
Bob
This is a bit off-topic, but should be of interest in connection with the pursuit of low distortion.
It is often frowned upon in high end circles to incorporate a relay in the signal path to a loudspeaker. However, there are times when this is unavoidable. An example would be a remote switch for listening comparisons. For this reason, I undertook to do some distortion measurements on a variety of relays. All of these relays had contact ratings between 30A and 80A.
The test setup involved applying a 16V 1 kHz test signal to an 8-ohm load (32 Watts), producing a current of 2A rms. The relay under test was placed in the return (ground) leg of the 8-ohm load. A spectrum analyzer was connected across the relay contacts. With this setup, the analyzer’s response at 1 kHz would be reflective of the resistance of the relay, while harmonics in the spectrum would give an idea of the distortion.
The good news is that one of the relays was exceptionally good; the bad news is that some were surprisingly bad. This means that relays for this kind of application need to be selected carefully. DC resistance was not well-correlated with distortion.
Six relays were tested. The results below were the analyzer readings referred to 0 dB being 100 mV across the relay. Since 2 Amps are flowing, 0 dB corresponds to 50 milliohms. Because this voltage is what would normally be in series with the speaker, a 0 dB reading for a distortion component would correspond to 100 mV out of a total of 16 V, or 0.625% distortion (-44 dB). Thus, a distortion component that is down 80 dB on the analyzer corresponds to 0.00006% (-124 dB). Here are the results:
Rly 1st 2nd 3rd 4th 5th 6th 7th
#1 -12 -86 -64 -91 -85 <-90 <-90
#2 -3 -72 -50 -86 -80 <-90 <-90
#3 -6 -87 -67 <-90 <-90 <-90 <-90
#4 -15 -96 -91 <-100 <-100 <-100 <-100
#5 -10 -69 -56 -80 -73 -91 -85
#6 -3 -<-90 -59 GRUNGE
Relay numbers 2 and 6 were notably bad, each with 35 milliohms resistance and third order distortion as bad as -50 dB on the analyzer, corresponding to in-service third order distortion of 0.002% (-94 dB). Relay #6 had a spectrum characterized by quite a bit of grunge above the analyzer grass. Third order distortion predominated on most relays.
Relay numbers 1, 3, and 5 were of medium performance, with resistance ranging from 25 milliohms down to 13 milliohms, and third order distortion on the analyzer ranging from -56 through -64 dB, corresponding to 0.001% (-100 dB) to 0.0004% (-108 dB).
Relay #4 was the clear winner. This was an 80-Amp automotive relay. It had a resistance of 18 milliohms and third order distortion of -91 dB on the analyzer, corresponding to 0.00002% (-135 dB).
Although Relay #4 was the only “automotive” relay in the test, it would be premature to conclude that an automotive relay will always be better than others for the audio application. Similarly it would be premature to conclude that the higher current rating is responsible for the enhanced performance, based on the variability among the other relays which were rated at 30 Amps. There is also no evidence that contact resistance is a good predictor of distortion performance. I’m tempted to speculate that the specifics of the contact surface chemistry has a large role in the distortion performance, in that two different sets of contacts might exhibit the same contact resistance, but different amounts of nonlinearity in that resistance.
Cheers,
Bob
Hello Bob,
Afew years ago I did something similar with the special speaker relays from Amplimo. These are special-purpose relais that have two contacts. On closure, a 100A-capable tungsten contact closes first followed by a gold parallel contact; the reverse takes place on opening, of course.
I tested 2 relais (same batch) and could not detect any difference in THD with or without the relais.
These relais, for about 15 $ apiece, are not well known which is a pity as they are excellent.
If you want to try one out, send me a mailing address via PM and I'll ship you one.
(And no, Mr Moderator, I have no relation to Amplimo except that of a happy customer)
Jan Didden
Afew years ago I did something similar with the special speaker relays from Amplimo. These are special-purpose relais that have two contacts. On closure, a 100A-capable tungsten contact closes first followed by a gold parallel contact; the reverse takes place on opening, of course.
I tested 2 relais (same batch) and could not detect any difference in THD with or without the relais.
These relais, for about 15 $ apiece, are not well known which is a pity as they are excellent.
If you want to try one out, send me a mailing address via PM and I'll ship you one.
(And no, Mr Moderator, I have no relation to Amplimo except that of a happy customer)
Jan Didden
janneman said:
Hi Jan,
In the test that you did, did you just test THD of the total signal with and without the relay in series, or did you look solely at the signal across the relay?
The reason I ask is that these observed distortion levels are fairly low, and the former test, using a conventional distortion analyzer, would have many of the relay distortion components I measured lying under the analyzer noise floor.
Thanks,
Bob
I *think* I used a current through the relay equivalent to 50W in 8 ohms, but I am not sure anymore of the exact topology. I'll try to find it back.
Jan Didden
Jan Didden