| andrei |
Class AB is often touted as having a higher efficiency than Class A (which is true), but it often overlooked that Class AB distortion characteristics are essentially the same as Class B (i.e. really bad). Here is an example:
You decide to build a super-duper 300B amplifier, and settle on a push-pull output stage. You know a single 300B is happy to drive 3k load, so for push-pull you get 6k:8 output transformer. You decide to save a little bit on power, and run the stage at 30mA idle current per tube. It's Class AB (Class A would idle at 70-80mA), so you should get 15W power output no problem, right?
Wrong. Here is what happens instead: |
|
|
| andrei |
Distortion at 7W output is about 10%, and the waveform looks really bad. In comparison, if I was running the tubes in Class A at 60mA idle current per tube, I would get my 15W at 0.5% THD.
The moral is that for a given tube, Class AB operation is not going to give me more output power than Class A at a given distortion level (say, 5%). Of course, if I use feedback, I can make Class AB stage respectably linear, but I can do the same to Class A, so there is no real advantage to using AB... |
|
|
| SHiFTY |
That is why negative feedback is necessary in a class AB amp. I would imagine a small amount of negative feedback would fully clear up this crossover distortion in the schematic pictured.
If anyone was making a PP amp with no feedback they would be well advised to make it operate fully in class A, even if it shortens the life of the output tubes. |
|
|
| Geek |
| Another often overlooked item is the P-P load difference in class-ABx. Two steps to fixing the problem, 1) Add NFB and 2) increase the P-P load :) |
|
|
| SY |
That's odd, I've had plenty of p-p amps around here with no loop feedback, and the distortion is no higher than SE, in fact, often much lower. For example, in an amp in front of me at the moment, a pair of 7027A in AB1 triode, 8K plate-to-plate give me 10 watts at under 1% THD before feedback. 1W distortion is under 0.05%, totally dominated by third.
So, if what you're saying is that one can come up with a hack circuit that doesn't work well, OK, that's true. But if you're trying to claim any universality to that observation, I'm sorry, it's just not correct.
edit: to do your comparison properly, you need to recognize that keeping the load constant requires the plate to plate load to be 4x higher than a single tube. |
|
|
| Tubes4e4 |
Hi Gregg,
| quote: | Originally posted by Geek
Another often overlooked item is the P-P load difference in class-ABx. |
Yeah, and because of this the spice simulation done by Andrei proves nothing at all. Or since when does spice adjust the load conditions from Raa/2 to Raa/4 (as seen by a single tube in a PP pair) automagically and smoothly, when doing the slow transition from class A1 to B1?
Damn simulations ;)
Tom |
|
|
| tubetvr |
Hi,
I am assuming that the distortion figures comes from a simulator? If so then the result would be very dependant on how good the model is.
Even so the result is not that surprising for a tube like 300B and is well known. In this paper http://www.clarisonus.com/Archives/...y/BeamPower.pdf
on pages 38 and forward this phenomena is described for theoretical triodes and also make a comparison with beam power tubes showing that these can give lower distortion in AB than triodes.
I think it is important to note that distortion level in AB with real triodes depend on the triode curve itself, and the chosen operating point.
Please also read what is written in the paper about the 0 distortion class B1 operation point that gives lower distortion than AB.
In some other books like Terman it is described about using projected cutoff operating point for tubes operating in close to B, the theory here is that there is a operating point for some triodes which give a minimum of distortion, this is usually at very low idle current.
So in contradiction to what many believe at least for triodes lower idle current and operating close to class B can give better results than operation closer to A.
Some regulator triodes like 6AS7 and 6C33C have more remote cutoff than e.g 300B, and they operate with a clear distortion minima at a quite low idle current in class AB.
Regards Hans |
|
|
| Geek |
Hi Tom,
I have found simulations only useful for finding DC operating points and quite useless for analog AC parameters ;) |
|
|
| Tubes4e4 |
| quote: | Originally posted by tubetvr
on pages 38 and forward this phenomena is described for theoretical triodes and also make a comparison with beam power tubes showing that these can give lower distortion in AB than triodes.
|
Yes, theoretically they can, but the ideal load range is very small. For many well known tetrodes/pentodes the manufacturers do supply distortion nomograms in the spec sheets. You can see there, that the least distortion point is very distinguished and distortion rises steep sideways from that ideal load point.
In real world circuit with real world loads (like speakers changing their impedance over frequency - thus giving a reflected load to the tubes with an error multiplied by OPT impedance ratio!) that means, that triodes usually are much more forgiving to load mismatching by real world loads like speakers.
Tom |
|
|
| tubelab.com |
The impedance ratio in an OPT is the square of the turns ratio. To present each tube with a 3K ohm load you need a 12K P-P transformer.
At least that is what the theory says. My amplifier (P-P 300B) runs a 6.6 K transformer that was made for 6L6's. I set the bias at about 45 mA per tube which allows class A operation up to about 10 Watts, at that point the amp crosses into AB1. I get 26 to 28 watts per channel from this amp. It has no local or global feedback. There is no visible crossover distortion, and the measured performence is quite good also. |
|
|
| Yvesm |
| quote: | Originally posted by Geek
Hi Tom,
I have found simulations only useful for finding DC operating points and quite useless for analog AC parameters ;) |
Yeah !
It's very difficuly to fully "switch off" a "real" 300B !
At very hi anode voltages where this should occur, expect a huge negative grid voltage swing to remove the least milliamp !
Yves. |
|
|
| kevinkr |
Some of you are no doubt familiar with the 300B PP amplifier I designed for an article in VTV about 7 years ago, that amplifier normally runs class A at 70-80mA per tube and will produce 25Wrms out at <1%, 30W @ 2% thd no feedback, reduce the plate current to about 40mA per tube and the distortion increases to less than 2% at 25W.
Note that at power levels of 10W and below I was not able to measure any significant difference at all. This design used fixed bias incidentally which makes it very simple to do these sorts of experiments.
Plate to plate impedance of the output transformer used is 3.8K and VP is 400V.
The article on this design is now available on my site if you are interested. |
|
|
| andrei |
P-P load difference in class-ABx is indeed often overlooked. Especially the fact that the load seen by the tube changes depending on wether the other tube is conducting or not. If the other tube is in cut-off, the load is 1/4 of primary, while if the other tube is conducting, it's 1/2 of primary impedance.
| quote: | | Since when does spice adjust the load conditions from Raa/2 to Raa/4 (as seen by a single tube in a PP pair) automagically and smoothly, when doing the slow transition from class A1 to B1? |
Since always. Transformers are correctly modelled as coupled inductors in Spice, dynamical effects and all. (n1/n2)^2 impedance reflection formula is a simplification.
| quote: | | I am assuming that the distortion figures comes from a simulator? If so then the result would be very dependant on how good the model is. |
Simulation of course is not exact, and depends on the quality of the models used (the ones I used are fairly accurate and are made from Sophia-traced 300B data), but it illustrates the problems you might have. So no, I do not believe THD number of Class AB, but yes, I am pretty sure it will be higher than in Class A if I prototype.
The example I chose is intentionally extreme to make a point, but is not that atypical (nor is it a hack circuit). You can do various things to improve Class AB, like play with the load or use a remote cutoff triode. But IMO, it is better to simply run it in Class A, unless there are special reasons that prevent you from doing so. And BTW, I don't buy "the feedback will take care of it" argument, as the same feedback loop will make Class A stage more linear still.
| quote: | | Even so the result is not that surprising for a tube like 300B and is well known. |
It is. But AB amplifiers still abound and have their proponents, as this thread is the evidence of :). |
|
|
| Geek |
Hi Yves,
| quote: | Originally posted by Yvesm
It's very difficuly to fully "switch off" a "real" 300B !
At very hi anode voltages where this should occur, expect a huge negative grid voltage swing to remove the least milliamp ! |
Indeed!
At high Va, it is likely we'd get a nasty g-k or g-a arc in the tube before we cutoff the last mA in real life :bigeyes: |
|
|
| tubelab.com |
Some new production 300B's have some filament wire exposed above the grid wires. This creates a path for current to flow directly from the filament to plate. Even though the path is small and long, if the plate voltage is high enough no amount of negative grid voltage can cut the tube off.
I have seen this on some Sovteks at voltages over 400 volts. |
|
|
| Miles Prower |
| quote: | Originally posted by andrei
Class AB is often touted as having a higher efficiency than Class A (which is true), but it often overlooked that Class AB distortion characteristics are essentially the same as Class B (i.e. really bad). Here is an example:
You decide to build a super-duper 300B amplifier, and settle on a push-pull output stage. You know a single 300B is happy to drive 3k load, so for push-pull you get 6k:8 output transformer. You decide to save a little bit on power, and run the stage at 30mA idle current per tube. It's Class AB (Class A would idle at 70-80mA), so you should get 15W power output no problem, right?
Wrong. Here is what happens instead: |
"You decide to build a super-duper 300B amplifier, and settle on a push-pull output stage. You know a single 300B is happy to drive 3k load, so for push-pull you get 6k:8 output transformer."
That's not right. Give it the appropriate load and see what happens. Secondly, take a second LQQK at the results. That cross over ain't all that bad, as compared to what you routinely see with transistors. Add some NFB and it won't be that bad at all. Select a proper op-point for the 300Bs and it'll probably be pretty good.
There's no real reason to say "No" to AB2 in every single case. :whazzat: Like anything else, it all depends... |
|
|
| kevinkr |
Something else to be aware of in LTspice, the transformer K coupling coefficient does not always work as expected, in very recent earlier releases it did not work at all for most typical audio transformer models, this has been recently fixed. Setting K to anything other than 1 with large inductors may result in erratic convergence behavior and inaccurate results - if you are attempting to model leakage inductance it is a lot safer to just add the leakage terms as small external (to K) inductors, this incidentally is also true of the dcr.
Note that 300B models (and other tubes) captured using Andrei's curve capture program do not model the non-linearity accurately near cutoff. (Capture is only 2 points on the curve.) There may be a few better models out there, I use a mix models and many of Duncan Monroe's are reasonably accurate even at the extremes. 3F4 models will work in LTspice without problems.
With good models I do get reasonably good correlation between simulated results and what I actually measure. Just changing tube brands sometimes has a significant effect on the measured linearity. |
|
|
| cerrem |
I don't think you can make such generalizations....
Each tube, each topology, Plate voltage used, Screen voltage used..ect.ect...the selected plate loading all play a role in the outcome of the amp and not really comparable...
For example the Macintosh circuit breaks the rules about breaking into AB without the cross-over notch...since the Class A drivers/ followers also drive the cathode/plate windings..they keep the OPT's small amount of leakage from seeing any cut-off of audio signal current...
You can get acurate simulations...the simulator is only as good as the model you put in there and if you rely on the crappy models that currently exist, then your wasting your time... Write your own models but don't use the standard equations...
Chris |
|
|
| andrei |
| quote: | | Note that 300B models (and other tubes) captured using Andrei's curve capture program do not model the non-linearity accurately near cutoff. (Capture is only 2 points on the curve.) |
That is false. Here is the 300B curve tracer capture: |
|
|
| andrei |
| And here is the model: |
|
|
| andrei |
| As you can see, the cutoff is modelled correctly (and it is easy to do for 300B, which is a hard-cutoff tube). The average model error is ~1.4mA. |
|
|
| SY |
| OK, so if you can't blame the model, then it really IS a poorly designed circuit. Try modelling again, but with a 4x increase in plate to plate load compared to SE, and with the idle current chosen to minimize distortion instead of maximize it. Then your simulation might give you something closer to how competently designed real-world push-pull AB amps measure. |
|
|
| andrei |
| Leaving SY's implications of my competence aside, the answer to his question is this: with 12KCT:8 load and the same bias as in original example you obviously get less distortion - about 6.5% at 12W in Class AB2 - but much more than what is achievable in Class A. More instructive is the answer to the second suggestion - vary the idle current to minimize the distortion. Distortion is more or less monotonic function of idle current, with distortion figure always best at largest available idle current - i.e. optimizing Class AB stage for distortion you are driven to Class A operation. There is no "sweet spot" in idle bias. This has been generally true for all triode push-pull stages I've modelled so far. |
|
|
| SY |
A couple of questions:
1. What's the B+?
2. Why is the source impedance of the driver set at 27k? |
|
|
| andrei |
| quote: | Originally posted by SY
1. What's the B+?
|
425V.
| quote: | | 2. Why is the source impedance of the driver set at 27k? |
No particular reason. A convenient value to represent grid-stopper resistor + previous stage impedance. It is irrelevant for the point I'm trying to make, as the grid current draw is not modelled, and Miller capacitance is too small to cause trouble at 100Hz. If I were designing a Class A2 stage, then I would worry about it :). |
|
|
| SY |
Thanks. In a real AB2 amp, you'd want a source impedance well lower than that.
If, for this tube and this model, your sim shows distortion decreasing monotonically with increasing idle current, you should set that at something more like 85mA. Why use a 40W tube and not run it at 40W? I'd be curious to see what your model predicts under those conditions.
I haven't used 300Bs in p-p AB (there are better tubes for this application), but in one published circuit I have at hand (C. Schafer, "30 Watt High Fidelity Audio Amplifier" Audio Engineering, July 1947), with a 425V B+, 80mA idle, 4000 ohms plate-to-plate load, the measured distortion of the entire amp (no loop feedback!) was under 2% IMD (40Hz/12kHz) below 22 watts. Not too terrible, but I'll bet some playing with load lines could come up with something better. |
|
|
| andrei |
At 425V B+ and running at 85mA idle per tube into 4kCT:8 load, the stage models really well. You get 15W out at 0.1% distortion. I would not believe that number too much, but distortion is definitely low. (In the real amp, some second order distortion will creep in from the tube mismatch.) Getting comparable distortion from a driver is going to be tough...
| quote: | | If, for this tube and this model, your sim shows distortion decreasing monotonically with increasing idle current, you should set that at something more like 85mA. Why use a 40W tube and not run it at 40W? |
That's what I've been trying to say :). |
|
|
| kevinkr |
Hi Andrei,
I quite obviously owe you an apology, I had heard mistakenly that the software did not do multi-point curve fitting when it is obvious that it does, and the curves look like a really close match in the model.
Your observations about the monotonic variance of the distortion in a 300B PP output stage jibe well with my previous experience in two 300B class A PP amps I designed, this incidentally appears to be true of 300B SE output stages as well, thd continues to decrease linearly as the quiescent current is increased to the point where the tube would actually self destruct. (I stopped just short of that point.) There may be no measured "sweet spot" but sonically there usually is and imho it is not usually at the minimum distortion point, but perhaps at a point where 2nd harmonic is slightly favored over third.. (Have not investigated this phenomena in PP amplifiers, have clearly seen this in SE amplifiers however.)
My last 300B PP amplifier runs in class A at 80mA/400V per tube and manages >30W @ 2% thd, and as I recall is predominantly 2 & 3rd harmonic. (Obviously most 2nd harmonic should be cancelled in the transformer.) Transformer is 3800 ohms plate to plate. Driving a pair of them to this power level requires differential drive of about 320Vpk-pk which is getting close to onerous with non transformer coupled driver stages. |
|
|
| andrei |
No problem. That just goes to show that I need to document the damn thing properly :). Well, maybe one of these days...
| quote: | | My last 300B PP amplifier runs in class A at 80mA/400V per tube and manages >30W @ 2% thd, and as I recall is predominantly 2 & 3rd harmonic. (Obviously most 2nd harmonic should be cancelled in the transformer.) Transformer is 3800 ohms plate to plate. Driving a pair of them to this power level requires differential drive of about 320Vpk-pk which is getting close to onerous with non transformer coupled driver stages. |
That's probably what I would do too if I were to build 300B PP amp... OTOH, even my SE 300B is over-powered :), so I am currently thinking about small (1-2W/channel) amps. I am working on a small (and cheap) 6BX7 Class A push-pull amp now... |
|
|
| tubelab.com |
I have recently done several experiments for the cheap transformers thread where I varied the current from low to just below the melting point, while watching the distortion spectra on an FFT analyzer. These tests were all done on an SE amplifier using 45's, 2A3's and 300B's. Several different low cost OPT's were tested. There is a point where there is an obvious null in the third harmonic distortion that is highly transformer dependent. It is nearly independent of the tube being used. The dip can be as much as 30 db. The second harmonic is usually inversely related to current for most DHT's that I have tested. I have found thet the "sweet spot" is coincident with the null in the third harmonic, which is largely transformer dependent.
I have not run similar tests on a P-P amplifier yet, but I have a screen driven 6AV5 amplifier being breadboarded. I will compare it to my 300Beast when it is done, and repeat these tests.
http://www.diyaudio.com/forums/show...2654&highlight= |
|
|
| kevinkr |
Hi Tubelab,
That is interesting... At the time I was doing the experiments I was unable to observe the spectral content of the distortion but was able to determine that there was a definite operating point sweet spot in all of my se amplifiers where the sound was markedly better.. I experimented extensively with the 45 fixed bias design and determined that the plate voltage and current that resulted in maximum power output and the minimum measured distortion at all power levels was not the point that sounded best. Later on I was able to determine that my preferred operating point resulted in very low levels of 3rd harmonic, but not the lowest overall distortion...
I will say I am impressed at your thoroughness and dedication. :) |
|
|
| tubelab.com |
Actually it's pretty easy with a high end sound card and some FFT software. I use WINaudioMLS, but there are freeware programs that do almost as much. You can watch the spectrum in real time and turn the current knob (fixed bias). I just discovered the null in the third harmonic while testing the Edcor transformers. The effect was similar, but at different current levels, with all of the other low cost transformers. I noticed that it was independent of the output tube used, but all testing was done in the same amp. Now I need to go back and try this with some different amps, including the 300B P-P.
I have since noticed the same effect with triode strapped 6AV5's in SE. The null occurred at the same current as the other tubes. 80 mA on the Edcor unit. |
|
|
| andrei |
| That is an interesting observation. It might be transformer-induced distortion. Have you poked at FFT spectra upstream on transformer primary? |
|
|
| tubelab.com |
No I haven't had much time to investigate, since I just discovered the effect. I suspect that it has to do with the gap in the SE transformer, so I think I will get an extra transformer next time I order some Edcors and offer it up to science as an $18 experiment. I would like to use these transformers with 45's but they have a lot of H3 at 30 mA.
The order will have to wait until I determine the optimum load impedance for the class B push pull amp that I am prototyping currently. Maybe this weekend, depending upon the weather. |
|
|
|
