I seem to recall seeing some proposed circuits where a long tail pair had 2 (equal) current sources, one in each anode.
I can't quite understand how/why this would perform better than a single current source in the (common) cathode. Can someone explain it to me?
I can't quite understand how/why this would perform better than a single current source in the (common) cathode. Can someone explain it to me?
Because the currents would be the same through each valve.
If the valves have slightly different mu's then a single current source would allow different currents through each valve.
If the valves have slightly different mu's then a single current source would allow different currents through each valve.
Using two separate CCS's in the plate circuits (only) would no longer allow any differential effect between the two tubes. The constant current thru the tail resistor would give a fixed voltage on the tail, just like for grounded cathode stages. This prevents any swing of the tail voltage to re-distribute current between the two tubes.
If the tail also has a CCS, then differential effect is re-enabled, but then no stable plate voltages are possible unless the plate CCS's are changed to gyrator mode.
If the tail also has a CCS, then differential effect is re-enabled, but then no stable plate voltages are possible unless the plate CCS's are changed to gyrator mode.
OK, so the part I'm not grasping is this: it's no longer a differential amp. So there isn't any point to tying the cathodes together. Is there? It's no different than 2 separate amps. You lose the benefits of diff operation. Correct?
No. You still have a differential amp with CCS in both anodes. You get exactly mu/2 gain from either grid to either anode, if the tail resistor is high enough or a CCS too.
If you have CCS in all three places then there is a risk that the anode voltage may settle to a poor quiescent value, unless the CCS are adjustable. This is because you are, in effect, trying to set one current through two CCS in series and this means the voltage at the junction is poorly defined as it depends on their impedances and current difference.
If you have CCS in all three places then there is a risk that the anode voltage may settle to a poor quiescent value, unless the CCS are adjustable. This is because you are, in effect, trying to set one current through two CCS in series and this means the voltage at the junction is poorly defined as it depends on their impedances and current difference.
Hmmm, last time I checked Ohm's Law, a constant current (sum of the two plate CCS's) thru a resistor was still a constant voltage. Maybe if the imperfect impedances of the plate CCS's (MegOhms?) are comparable to the tail resistance some differential effect would occur. A CCS tail with gyrator plate loads will work in differential.
I don't see how. Let's say each source is 2mA. The tail current is 4mA fixed. If the cathode resistor is 1k, both cathodes sit at a fixed 4V. Anything going on in one tube has no influence on the other.
OK, I see what you are saying. The solution to this puzzle is that the anode CCS set the quiescent current, but there is also the signal current in the applied load. The signal current, although nearly balanced, can generate a cathode signal voltage.
Next problem: what if there is no load (apart from a voltmeter/scope)? Then it all depends on the ratio between the CCS dynamic impedances. No CCS is perfect.
Next problem: what if there is no load (apart from a voltmeter/scope)? Then it all depends on the ratio between the CCS dynamic impedances. No CCS is perfect.
Morgan Jones has suggested in "Designing Valve Amplifiers" that LTP set up with 3 CCS should have a much higher performance CCS in the tail than in the anode loads, and seems also to be generally recommending not using CCS anode loads in LTP applications.
Precisely. Ohm's law applies in its differential form to real CCS, which have a finite (although large) impedance. As I said, it all depends on the ratio of the CCS impedances. A useful way to think about a LTP is to consider common-mode and differential-mode inputs. This still works if CCS appear everywhere.
If the cathode resistor is only 1K, then you have an STP (short tail pair) not LTP! I did say the tail resistor has to be high enough. To get a differential effect Rtail must be much bigger than 1/gm, and also significantly bigger than Ranode/mu. Note that here Ranode includes the load (e.g. the next stage). Of course, if you have a perfect CCS in the anode and no load then Rtail must be infinite too. I don't know where to buy a perfect CCS.
If the cathode resistor is only 1K, then you have an STP (short tail pair) not LTP! I did say the tail resistor has to be high enough. To get a differential effect Rtail must be much bigger than 1/gm, and also significantly bigger than Ranode/mu. Note that here Ranode includes the load (e.g. the next stage). Of course, if you have a perfect CCS in the anode and no load then Rtail must be infinite too. I don't know where to buy a perfect CCS.
The big dumb blonde one decided that Ohm didn't really apply here, so I set up an LTP with 3 CCS's. After much experimentation I came to the realization that I had reinvented the flip flop. One plate voltage will rail while the other saturates. The cure as mentioned is to water down the CCS in the tail or both CCS's in the plate. Putting resistors across the plate CCS works better than putting one across the tail CCS.
Not totally true. Even with "perfect" CCS's, applying enough signal to one grid in the positive direction will saturate that tube. Continuing the application of signal, the grid will attempt to go positive causing current flow and bringing the cathode voltage with it. It the grid of the other tube is fixed, the rising cathode voltage will cut that tube off. Hence the circuit behaves like an LTP with infinite gain.
Interesting findings. I'm glad experimental science still lives on.
When you say one goes up while the other comes down, do you mean infinite gain like a comparator or just a CCS mismatch so the poor valve anode impedance has to try and take up the slack betwween warring CCS's?
When you say one goes up while the other comes down, do you mean infinite gain like a comparator or just a CCS mismatch so the poor valve anode impedance has to try and take up the slack betwween warring CCS's?
Ah, the load current should give another degree of freedom, but only if the two outputs aren't exactly complementary. So it could work if the amp doesn't work, like catch 22......
In any case I think it could work with gyrators up top and a CCS on the bottom. But until the gyrators adjust their current to sum -exactly- to the bottom CCS, it would be a flip-flop like George says. It's a circuit that lives dangerously. Any distortion in the output signals could cause the gyrators to have to re-adjust their steady currents, with recurrent instability rampant until they do re-adjust each time. This could be the service technicians nightmare. You should put this circuit in if you want the amp always returned for service.
Hmm,, I wonder what would happen if we put gyrators in for ALL the CCS's. Then they would all be trying to adjust their current outputs slowly to match the others. With all those time constants and instabilities in there, it would be almost certain to motor boat and flip-flop endlessly. A Chaos generator?
In any case I think it could work with gyrators up top and a CCS on the bottom. But until the gyrators adjust their current to sum -exactly- to the bottom CCS, it would be a flip-flop like George says. It's a circuit that lives dangerously. Any distortion in the output signals could cause the gyrators to have to re-adjust their steady currents, with recurrent instability rampant until they do re-adjust each time. This could be the service technicians nightmare. You should put this circuit in if you want the amp always returned for service.
Hmm,, I wonder what would happen if we put gyrators in for ALL the CCS's. Then they would all be trying to adjust their current outputs slowly to match the others. With all those time constants and instabilities in there, it would be almost certain to motor boat and flip-flop endlessly. A Chaos generator?
Last edited:
The LTP used a 6SN7 and all CCS's were IXYS 10M45's with 10 turn trimmer pots. When the CCS's were tweaked exactly to achive plate voltages near each other and in the linear region the LTP would output square waves with just a few mV of input. After just a few minutes of operation the circuit would drift out of the linear region and latch with one tube saturated and the other cut off. Applying a bunch of drive would provoke square wave output again, but the output did not always latch up in the same direction when the drive was removed.
These experiments were part of the dual LTP driver developed during this thread. There are over 500 posts and I don't know if I posted the schematic or not somewhere in the early part of the thread:
http://www.diyaudio.com/forums/tubes-valves/133034-6l6gc-ab2-amp.html?highlight=6L6GC+AB2
OK, this 3 CCS issue is just a sideline. My original question is why, or to what benefit, would you put 2 CCSs in the plates as opposed, NOT IN ADDITION, to one in the (common) cathode. This could be for either a LTP or a short tail (plain diff amp).
The only suggestion was that it will compensate for different mu in the 2 halves of the tube. Fair enough. But my thinking is that you would largely turn it into 2 simple and not-so-differential amplifiers. I don't see how that's better than a diff amp with slightly different currents. Am I wrong?
The only suggestion was that it will compensate for different mu in the 2 halves of the tube. Fair enough. But my thinking is that you would largely turn it into 2 simple and not-so-differential amplifiers. I don't see how that's better than a diff amp with slightly different currents. Am I wrong?
I can see where the two plate CCS's would keep the DC idle currents matched.
Seems to me if the two actual output load currents from complementary voltages into equal load resistances sum to zero, then the tail resistor still has to be constant current and hence constant voltage. So no differential action. If the two input signals had some common mode swing upwards, then the tail would try to draw more current and the plate voltages would drop in common mode until the actual loads drew the additional current. Where as a CCS tail would avoid that. So common mode rejection has dissappeared. Seems to be just two separate tubes.
Seems to me if the two actual output load currents from complementary voltages into equal load resistances sum to zero, then the tail resistor still has to be constant current and hence constant voltage. So no differential action. If the two input signals had some common mode swing upwards, then the tail would try to draw more current and the plate voltages would drop in common mode until the actual loads drew the additional current. Where as a CCS tail would avoid that. So common mode rejection has dissappeared. Seems to be just two separate tubes.
During the development of the driver board used in the AB2 amp discussed in the above mentioned thread I tried an LTP with 1, 2 and 3 CCS's. In each case it was the second LTP in the driver chain. There were two designs. The first used a 12AT7, 5751, or a 6FQ7 for the first tube and the second tube was always a 6FQ7. THe second design used a 6SN7 or 6SL7 for the first tube and a 6SN7 for the second tube. There were mosfet followers to drive the output tubes so the LTP's were very lightly loaded. The 3 CCS design was deemed unconditionally unstable and abandoned early. The 1 and 2 CCS circuits were tested in working P-P amps using 6L6GC, KT88, and several more unconventional output tubes all wired in triode mode. A fully differential design with a CCS in the tail of the output tubes was also tested. No negative feedback was used in any test.
I came to the same conclusion each time. The single CCS in the tail has the lowest distortion and sounds better. The twin CCS's in the plate with a resistor in the tail has more gain and a higher output swing, but higher distortion. In some cases the differences were small, but became more noticible where the drive requirements were higher. As the tail voltage increases the resistor behaves more like a CCS anyway. All of these experiments were direct coupled between the two stages making the tail voltage about 100 volts.
I decided that for reasonable drive requirements I would always use the single CCS in the tail. For screen drive or cathode follower output stages where hundreds of volts of drive is needed, the resistor tail, CCS in the plate will provide more drive voltage if enough B+ is available.
Consider the "Western Electric Parallel Feed Amplifier" and the "Differential Parallel Feed Amplifier" in Lynn Olson's article.
http://www.nutshellhifi.com/library/ETF.html
Dave
http://www.nutshellhifi.com/library/ETF.html
Dave
"Consider the "Western Electric Parallel Feed Amplifier" and the "Differential Parallel Feed Amplifier" in Lynn Olson's article."
Interesting circuits. Here is my take on them:
Looking at the differential parallel feed 1st, any difference in voltage gain between the two tubes shows up as a shift in the null AC voltage point in the output transformer primary. Effectively the transformer alters its two primary turns count complementarily so as to match the tube gains together. Clearly no signal current can pass thru the cathode resistor, so it operates at constant voltage to provide biasing only. (the series OT primary connection insures that any current drop in one tube is matched by the same current increase in the other tube.) Each tube runs at constant (and presumeably matched) DC idle current, but has varying AC current, so it can deliver power to the load. (the difference in tube AC currents is shuttled thru the OT primary to keep the CCS's happy) Since the cathodes operate at constant voltage, there is no differential cathode effect to remove common mode signal or match gains there like the usual diffl. stage, but the output series OT circuit performs the same function there instead. Very interesting re-shuffle of the diffl. stage.
I initially suspected that the ordinary diffl. stage with a CCS tail still had a slight distortion advantage (for the matched tubes case anyway) since the tubes are prevented from mismatched gains when the input signal has shifted the AC current balance significantly (so gm versus current have become un-matched), while the diffl. parallel case fixes the problem after the fact. But on closer analysis, they are the same, I think. In both cases the tube currents are forced to sum to a constant, so the tube gains are similar. So net advantage of the parallel case is simply equalized DC idle currents at the cost of an extra CCS.
The Western Electric Parallel Feed almost does the same thing, except the connection from OT center to the cathodes causes some separate tube output loading and also some positive feedback. The output loading effect deteriorates the gain balancing effect in the OT, and the positive feedback to the cathodes causes the tube AC gain imbalance to get worse. What this does to the distortion would require Spice simulation to decipher, but it can't be good. The positive feedback also appears to make the common mode gain worse, but the sorta series OT connection mostly blocks that. I would call this one a defective circuit design.
Interesting circuits. Here is my take on them:
Looking at the differential parallel feed 1st, any difference in voltage gain between the two tubes shows up as a shift in the null AC voltage point in the output transformer primary. Effectively the transformer alters its two primary turns count complementarily so as to match the tube gains together. Clearly no signal current can pass thru the cathode resistor, so it operates at constant voltage to provide biasing only. (the series OT primary connection insures that any current drop in one tube is matched by the same current increase in the other tube.) Each tube runs at constant (and presumeably matched) DC idle current, but has varying AC current, so it can deliver power to the load. (the difference in tube AC currents is shuttled thru the OT primary to keep the CCS's happy) Since the cathodes operate at constant voltage, there is no differential cathode effect to remove common mode signal or match gains there like the usual diffl. stage, but the output series OT circuit performs the same function there instead. Very interesting re-shuffle of the diffl. stage.
I initially suspected that the ordinary diffl. stage with a CCS tail still had a slight distortion advantage (for the matched tubes case anyway) since the tubes are prevented from mismatched gains when the input signal has shifted the AC current balance significantly (so gm versus current have become un-matched), while the diffl. parallel case fixes the problem after the fact. But on closer analysis, they are the same, I think. In both cases the tube currents are forced to sum to a constant, so the tube gains are similar. So net advantage of the parallel case is simply equalized DC idle currents at the cost of an extra CCS.
The Western Electric Parallel Feed almost does the same thing, except the connection from OT center to the cathodes causes some separate tube output loading and also some positive feedback. The output loading effect deteriorates the gain balancing effect in the OT, and the positive feedback to the cathodes causes the tube AC gain imbalance to get worse. What this does to the distortion would require Spice simulation to decipher, but it can't be good. The positive feedback also appears to make the common mode gain worse, but the sorta series OT connection mostly blocks that. I would call this one a defective circuit design.
Last edited:
- Status
- Not open for further replies.