• WARNING: Tube/Valve amplifiers use potentially LETHAL HIGH VOLTAGES.
    Building, troubleshooting and testing of these amplifiers should only be
    performed by someone who is thoroughly familiar with
    the safety precautions around high voltages.

Constant Current Sources in Tube Amps

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
The regulator thread generated this post. Once again, I realize that most of the folks on this list understand what CCSs do for tube circuits (at least some of them). So, I thought that if you all would indulge me once again, we might talk about it some. :)

Grounded Cathode

To start we will use the basic grounded cathode amplifier. This diagram shows two identical gain stages except that the one on the right is loaded with a perfect constant current source. I'm going to look at the gain, distortion and frequency response of both of these and compare them. Once again, I will use PSpice for the simulator and the trusty old 6DJ8 for the tube.

An externally hosted image should be here but it was not working when we last tested it.


First, we can look at the gain. The tubes are driven by a 0.1v 1KHz sine wave and Ca=Cb=10u. Here is the output from points A (green) and B (red):

An externally hosted image should be here but it was not working when we last tested it.


Notice that the CCS loaded device has more gain, about 25 for A and 30 for B. This is helpful, but not that important.

To look at the distortion, we'll take the Fourier transform and look at the second harmonic at 2KHz. Here it is, blown up:

An externally hosted image should be here but it was not working when we last tested it.


Notice that the 2KHz amplitude for the non-CCS amplifier is about 5mV, but for the CCS loaded amp it is 2.5mV. This reduction in distortion can be of great benefit. When compared with the increase in gain the relative distortions are: 0.2% for A and 0.08% for B. Small numbers to be sure, but in the world of audio perfection, well . . . . . :goodbad:

Why is this so? Without dragging out the plate curves, this change simply comes from the fact that in the CCS loaded amp the load line is completely horizontal (minimizing distorion), whereas in the non-CCS stage, the load line has a slope. I know that you all know this anyway.

Perhaps, more important, is the frequency response curves. Here we will sweep the input from 10Hz to 1MHz. Again, Ca=Cb=10u. Plots are in DB of gain. Here is the comparison:

An externally hosted image should be here but it was not working when we last tested it.


Notice that the CCS loaded amp is very flat all the way to 10Hz. Why is this so? Because, as all the better circuit guys are screaming to me, if the current is constant through the tube, then the bypass capacitor is essentially out of the circuit and, therefore, does not limit the low frequency response. In fact, if the CCS device really were perfect, we could eliminate Ck altogether.

The only way to approach this response for the non-CCS stage is to put C=1000u. Still, you won't get there and you've got another big ugly electrolytic in the circuit.

However, the goodness of this result also depends on the response curves of the actual CCS device. Since it will be made of real components, it won't have a perfect response and, hence, any real circuit will deviate from this result. How much, you may ask? Well, that's the subject of the next post. :wrench:
 
Tube CCS

Will a tube CCS really be fast (or slow) enough? Here's a fairly typical CCS tube load for our previous grounded cathode amp.

An externally hosted image should be here but it was not working when we last tested it.


This circuit draws the same idle current as above. I am setting Ck=10u and Cc=470n. The gain of this configuration is nearly identical to the perfect CCS above. Here is the three-way distortion comparson:

An externally hosted image should be here but it was not working when we last tested it.


The new blue curve is the Tube CCS stage. Its distorion is identical (at 2KHz) to the perfect CCS stage.

What about the response? Here is the three way comparison:

An externally hosted image should be here but it was not working when we last tested it.


Notice that the Tube CCS blue curve is very close to the perfect response curve. This graph shows just the perfect and Tube CCS response comparison:

An externally hosted image should be here but it was not working when we last tested it.


The Tube CCS stage is only 0.2db down at 10Hz (with a 10u Ck). I would hope that this would be unoticeable to all but the most discriminating ears. :)

So, if you want to go tubes all the way, you can load your grounded cathode amp with its mating tube ccs and you should get very good results. Of course, the price for this is an additional 100V or so on the PS (from 200 to 300 volts).

I've looked at CCS loaded cathode follwers (at least for a few configurations) and there seems to be no significant effect from the CCS. But, real circuits may be different.

A further question is, does a CCS help a circuit that has no bypass cap? For example, does it help a common cathode amplifier? Next post.
 
Hi Runeight,

Love your postings!!!

From an amateur's point of view I am somewhat suprised however by the performance of the triode CCS...

Alan Kimmel says that a triode cannot be considered a true CCS since "because it hasn't enough gain to respond adequately to the small voltage changes which occur across Rk2"

http://home.zonnet.nl/horneman/ml/akimmel.htm

Regards,
Bas
 

Attachments

  • figure2.gif
    figure2.gif
    2.3 KB · Views: 1,820
Another thing I would be interested in is how could the shunt capacitance be lowered without a cascoded CCS? Is that something you can easily measure/simulate.

From Gary Pimm's pages...
http://home.pacifier.com/~gpimm/Active_loads_and_signal_current_control.html
Another poor performance aspect of single mosfet CCS’s is high shunt capacitance. Mosfets look like great parts until you take into account their high capacitance issues. I suspect that the shunt capacitance is the source of most of the sound coloration’s that make CCS’s sound different.

The use of a cascode circuit will greatly reduce the shunt capacitance and improve the DC performance of a CCS.

As a diy'er I was thinking...how about just using a 1k resistor or so? Would that "hide" the shunt capacitance of the mosfet CCS??? In the same way that you can hide capacitance from a tube rectifier with a choke or resistor???

Cheers,
Bas
 
Ex-Moderator
Joined 2003
dBs please!

Hello Runeight, audio runs on dBs. It produces slopes of 6dB/octave etc, and humans respond to sound logarithmically. We just can't get enough dBs...

Although constant input voltage is fine when looking for changes to gain, if you want to look at how a CCS changes distortion, it's best to hold the output voltage constant.
 
Hello Runeight, audio runs on dBs. It produces slopes of 6dB/octave etc, and humans respond to sound logarithmically. We just can't get enough dBs...

Thanks EC. I guess this was in response to my line about discriminating listeners? I think my hyperbole was a little more effective than I meant it to be. :) I do know how the ear hears loudness, although my wife might argue with that.

Although constant input voltage is fine when looking for changes to gain, if you want to look at how a CCS changes distortion, it's best to hold the output voltage constant.

Could you explain this another way. I'm not sure I understand it. Do you mean that to get an accurate picture of distorion vs. frequency, we should keep the output signal the same amplitude? Please help.

Thanks for the review. Also try load capability drawing output from the top tube's cathode, and using a pentode instead.

Ah yes. I was trying take the very simplest case for comparison, but if we're looking at Zo then, of course. I've added point B to the picture above at the cathode of the CCS device. Just for completeness, here is the simulated Zo for:

No CCS = 3200R

Tube CCS A = 3700R

Tube CCS B = 135R

Taking advantage of the follower is the natural thing to do if you're driving a load or even if you're not.

OK, so I'll try some pentodes, but at least according to this simple model, the triode will do quite well. I'll have to read Mr. Kimmel's paper in detail.

Dear God, for once we agree on something!!!

If I have performed no other service to this forum, this happy agreement may be enough. :clown:
 
Ex-Moderator
Joined 2003
What I meant was that graphs expressing distortion in terms of dB down on the findamental are clearer than absolute distortion voltages.

As for my other point, when you bypass a cathode resistor with a capacitor, you change the gain as well as the distortion. Because distortion is proportional to output level, if you want to compare the effect on distortion, you need to adjust the input level so that both circuits produce the same output level.

I'm going to have to say that I don't think any of the PSPICE triode models are sufficiently good to accurately predict the distortion of a real valve. I have seen people trying to show distortion cancellation/reduction schemes down to -80dB based on PSPICE. All it "proves" is that the model cancels. No more.
 
I'm going to have to say that I don't think any of the PSPICE triode models are sufficiently good to accurately predict the distortion of a real valve. I have seen people trying to show distortion cancellation/reduction schemes down to -80dB based on PSPICE. All it "proves" is that the model cancels. No more.

I mostly agree with this. I don't think that the spice models can predict absolute levels of distortion with detailed accuracy. But, I do think that they can give you order-of-magnitude results and I think that they give very good relative results. That is, they can tell you if one topology is better than another. For example, I have simulated transcendent's grounded grid preamp and the distortion figure that I get is very close to the published value (on the assumption, of course, that the published value is accurate).

What I meant was that graphs expressing distortion in terms of dB down on the findamental are clearer than absolute distortion voltages.

Yes, I just don't have that capability in the PSpice that I have. And, also, we are accustomed to seeing distortion figures in % values too.

if you want to compare the effect on distortion, you need to adjust the input level so that both circuits produce the same output level.

Yes, I should have done this. Although, in this case, I think the non-CCS device will only get worse in comparison.
 
diyAudio Senior Member
Joined 2002
Hi,

What a shame that your version insists on giving distortion as an absolute voltage. We'll just have to live with it...

While I quite understand the objections uttered, it doesn't bother me all that much.
All it takes is a little intellectual effort to extrapolate these figures and see how they correlate.

All in all I think runeight is doing us a great service investigating/demystifiying the CCS and its' impact on the circuit.

Cheers,;)
 
I'll keep plugging along Frank. :)

To adjust for EC's correct assertion about distorion, here is the comparison for the non-CCS gain stage against the tube CCS gain stage where the input to the non-CCS stage is set to give the same output level (.118V in). Non-ccs distortion is green, tube ccs is red. Ck=10u.

An externally hosted image should be here but it was not working when we last tested it.


As everyone expected the second harmonic increased to 7.5mV which is about .25%. Worse than before.

We agreed before that these absolute numbers cannot be counted on to be precise because of other circuit affects. But, I believe that the trend is correct, even if it is not necessarily important.

In case you're wondering, this is a graph of the current draw from the PS for the tube CCS stage. I have placed a 100K load through a 1u cap to provide something for the stage to drive. Output is taken from the cathode of the ccs device (point B), so that this is a mu follower (Ck=10u):

An externally hosted image should be here but it was not working when we last tested it.


So that when the device is loaded, it draws about 20uA peak from the PS. I don't know if the PS designers would consider this to be constant current enough.

This little bit of work led me to look at a constant current draw pair. Here is the diagram:

An externally hosted image should be here but it was not working when we last tested it.


The idea here is that since both tubes are at the same operating point, but their currents are 180d out of phase, their demands on the PS will cancel. And, therefore, they can share a common Rk which won't need much, if any, bypass. Since the output is a follower you still get low Zo.

First, the distortion comparison to the tube ccs device:

An externally hosted image should be here but it was not working when we last tested it.


This is as expected, because, although the pair of tubes is constant current, each tube still sees a load resistor and doesn't get the benefit of a CCS load. So the distortion is higher than for the pure CCS device.

Now we may ask, does this device's PS current requirements differ from the mu follower for the same load? Here it is at 100K load. I've taken the fourier transform and am showing the current amplitude at 1KHz. Ck=10u.

An externally hosted image should be here but it was not working when we last tested it.


So, from the PS point of view the current draw should be nearly identical. However, the mu follower needs 100 extra volts on the PS while the constant current pair needs twice the idle current. Also, the mu follower doesn't depend on the tubes being in balance (which they usually aren't).

What about the frequency response? If we set Ck=10u for both stages and if we normalize the output, we get:

An externally hosted image should be here but it was not working when we last tested it.


If we set Ck=0 we get:

An externally hosted image should be here but it was not working when we last tested it.


For Ck=0 the curves are nearly identical. Even for Ck=10u they are very close. (I realize that I haven't normalized the outputs precisely, but they're close enough).

The Zo of the constant current pair is about 80R. The tube mu follower has about 135R.

So, the only distinguishing feature of a real CCS loaded device compared to what could be done with a constant current draw pair is the level of distortion. Assuming that a real triode performs the way the models do, then this difference might be enough to choose the CCS loaded amplifier. ANd, of course, the need for some balance in the tubes for the constant current pair.

I guess the PS trade-off depends on what you like to do, supply current or supply voltage.

Some caveats again. This is simulation and these are perfect Rs and Cs. Also, I've only explored a few possibilities. There are many others, but I think these get the ideas across. At least, this helped me to see what's going on.

Pentodes next. But these are a real pain in the :censored: compared to triodes (model-wise).
 
Pentodes

Ah yes. The claim is that pentodes are much better CCSs than triodes. Perhaps they are, but are we straining out gnats while swallowing camels?

I loved pentodes when I was a geek kid. They were the best tubes to break open to see the internal construction. I used to buy tons of them for $0.10 apiece at surplus stores. In case you want to try this, put the tube into a paper (plastic) bag and whack the glass with a hammer. Just hard enough to shatter the envelope. The bag will collect the junk from the implosion and you can retrieve the tube to pick apart the internals to see how this stuff is really made (warning: be very careful with the glass and tubes have heavy metals in them). I just wanted you guys to know that the stuff you pay big money for and view as new age crystals that bring cosmic goodness to your gear, I used to smash regularly. :devily:

Bas, I think this begins to answer your post. I have read Alan Kimmel's article.

This post will compare triode and pentode CCSs. Mr. Kimmel's article recomends high S tubes. I didn't have a really high S pentode, so I created a 6JC6 (S~=16000 umohs) model. It's not perfect, but it is close enough for this analysis. Here is the circuit:

An externally hosted image should be here but it was not working when we last tested it.


The left side is the triode CCS from before. The right side is a topologically equivalent pentode CCS using the 6JC6. The biasing is set so that the idle current is the same (within a few uA) for both circuits.

The cap that locks the screen to the cathode is 10u. Mr. Kimmel recommends that Xc for this cap be no more than 1/10 the value of Rsc (100K). I believe that 10u does the trick. Note that this is a high Rsc, but is what seems to be necessary to make this configuration work. Besides, Isc hardly changes at all.

First, let's look at the CCS characteristics of both configurations. This graph shows the current variation in the bottom triodes at 1KHz input (green=triode, red=pentode):

An externally hosted image should be here but it was not working when we last tested it.


Indeed, the pentode current variation is smaller. To see this better, here is the fourier transform at 1KHz:

An externally hosted image should be here but it was not working when we last tested it.


So the triode CCS keeps the current constant to within 4.5uA and the pentode CCS keeps it constant to within 0.5uA. The pentode is a remarkably good CCS, but does this difference really matter? I'd like to hear from you guys about this.

How does this effect the distortion? Well, both bottom triodes are now operating close to maximum possible gain, with very flat load lines. And so, there is very little change to the distortion figures (looking at second harmonic):

An externally hosted image should be here but it was not working when we last tested it.


And finally, does the more perfect pentode lead to better frequency response? Well, here's what the simulation says:

An externally hosted image should be here but it was not working when we last tested it.


Gain is practially identical with the pentode approaching mu a little better, but still they are within 1db. The triode rolls off a little more quickly above 200KHz, but I can't imagine that this would be noticeable since it's way less than 1db at 300KHz.

Mr. Kimmel shows another, more sophisticated CCS for pentodes. I can't imagine that could help much. But, he does indicate that he built all of his designs and compared them.

So, my take on this, recognizing the limits on the simulations, is that the pentode CCS may not be worth the extra trouble or parts. Go ahead, fire away . . . . :)

Next, I think I'll look at a 6BM8 for the pentode as these may be more available today. 6BM8 has about half the transconductance of the 6JC6.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.