6SL7 driver stage - a sanity check, please

[thorpej]

Ok, on my stereo SE 6L6 project, I think I have the power supply and output stage all squared away. Now I'm on to the driver stage. My initial plan is to use a single 6SL7, and drive each 6L6 with one half of the 6SL7. There will be a volume pot between the 6SL7 and the input jacks, which will be fed with garden variety consumer line-level signals. From my output stage load line, it looks like I need ~8.5Vrms to drive the output to full power.

On paper, the power supply node feeding the driver stage is at 251V.

Because I want to experiment with NFB, I want to maximize the gain (because it's eventually going to be reduced by the feedback loop), and I obviously want to minimize the distortion generated by the driver stage. To minimize the distortion, I want to keep the plate curves as evenly spaced as possible along my load line. It looks like I can do that by using a fairly small plate load and center-biasing, but then the gain of the stage is significantly reduced. But it also looks like I can do this with a pretty large plate load and biasing a little hotter than center.

The load line I came up with for the latter scenario is based on a 510K plate load, which gives me 0.5mA for the 0V end of the load line, and 0.3mA idle current for a -1.5V bias point. On paper, the gain is 59 (35dB) for a fully-bypassed cathode. I went back to the 6SL7 data sheet and the example circuits table does give an example for a 510K plate load, so the idea doesn't seem absurd on it's face, but I've never personally seen a triode gain stage with a plate load any larger than 220K (then again, I haven't been at this very long).

So, my question for you all -- am I off the rails here, or does my design seem reasonable?

 

[kward]

What is the bias of the output stage? Meaning, at quiescent conditions, what will the voltage be at the grid of the 6L6 with respect to it's cathode? You can use that as a poor man's estimate of drive swing needed in the driver stage.

Global feedback then factors in based on the amount of feedback added.

 

[thorpej]

What is the bias of the output stage? Meaning, at quiescent conditions, what will the voltage be at the grid of the 6L6 with respect to it's cathode? You can use that as a poor man's estimate of drive swing needed in the driver stage.

Global feedback then factors in based on the amount of feedback added.

Bias point is -18V, 300V to plates, 50mA idle current.

 

[thorpej]

Global feedback then factors in based on the amount of feedback added.

Yah, so about that... I've been reading about how the closed-loop gain is calculated (and thus how to determine the desired ratio of the feedback divider). Aiken Amps has a nice article that explains the basics. But it's missing a couple of details for this newbie.

http://www.aikenamps.com/index.php/designing-for-global-negative-feedback

When calculating the open loop gain, am I also concerned about the voltage gain in the output stage itself? That seems a little odd since the output stage voltage is stepped down in the OT. I guess I should start a thread specifically about this topic, and focus on getting the rest of the design correct first.

 

[kward]

At the quiescent conditions you quoted for the 6SL7, with a 510K plate resistor and let's say a 470K grid leak resistor on the 6L6 (can go that high if you are doing cathode bias), then I calculate an open loop stage gain of about 70 with cathode fully bypassed. Let's also assume you want about 0.5V input sensitivity. That means you can apply somewhere around 8 dB of feedback max. If you could live with 0.9V input sensitivity, then you can apply about 12 dB feedback. With a 470K grid leak resistor on the 6L6, you can go pretty low on the coupling capacitor, probably down to 0.033 uF. With a 1.5K cathode resistor on the 6SL7, you can use a 47uF bypass cap and be fine.

So I think this could work, especially if you are just experimenting and you are going to tear it down when the experiment is done. But if you are trying to make something more permanent, I would really like to see quadruple the current on the 6SL7 at quiescent conditions to be able to provide more drive capability. That would mean quite a bit more rail voltage to start with.

 

[kward]

When calculating the open loop gain, am I also concerned about the voltage gain in the output stage itself? That seems a little odd since the output stage voltage is stepped down in the OT. I guess I should start a thread specifically about this topic, and focus on getting the rest of the design correct first.

I find it easier for this type of analysis to consider the output stage as a "black box," where the black box internally contains the output tube and the transformer. Input to the black box is how much voltage will drive the stage to full power. Output from the black box is the speaker terminals. Thus the whole analysis pivots around the drive requirements for the output stage.

 

[thorpej]

At the quiescent conditions you quoted for the 6SL7, with a 510K plate resistor and let's say a 470K grid leak resistor on the 6L6 (can go that high if you are doing cathode bias), then I calculate an open loop stage gain of about 70 with cathode fully bypassed. Let's also assume you want about 0.5V input sensitivity. That means you can apply somewhere around 8 dB of feedback max. If you could live with 0.9V input sensitivity, then you can apply about 12 dB feedback. With a 470K grid leak resistor on the 6L6, you can go pretty low on the coupling capacitor, probably down to 0.033 uF. With a 1.5K cathode resistor on the 6SL7, you can use a 47uF bypass cap and be fine.

So I think this could work, especially if you are just experimenting and you are going to tear it down when the experiment is done. But if you are trying to make something more permanent, I would really like to see quadruple the current on the 6SL7 at quiescent conditions to be able to provide more drive capability. That would mean quite a bit more rail voltage to start with.

Huh, what's the formula you're using to calculate the open loop gain? Your results are different from my calculation, but readily acknowledge that mine is probably wrong

I do actually have a fair bit of rail voltage available; I'm dropping quite a bit to get down to 250V for the 6SL7. I'm pretty heavily filtering the power supply before the 6L6 plates, so I can afford to drop quite a bit less without any power supply noise penalty for the 6SL7.

Note that I'm not wedded to the idea of using a 6SL7 there, either. I have plenty of Russian ECC83-alikes in my tube stash, but was hoping to use an octal for looks. I do intend to let this thing live on after I finish playing in the sandbox, so I'd like to get it as "right" as possible.

I'm attaching a PDF of the current state of the schematic (that also includes a few random notes documenting my thought process on the power supply).

 

[kward]

Stage gain of a common cathode gain stage: mu*RL/(rp +RL+Rk(mu+1)). When cathode is bypassed Rk goes to zero for that calculation. But you can also just use the load line method. I did both and got only a 5% error.

• mu is the gain factor of the tube assuming infinite plate load (I used 80 in the example, as eyeballed from the plate curves at the quiescent point chosen.
• rp is the dynamic plate resistance at the quiescent point chosen (which you determine from the plate curves). I used 26.67K in the example.
• RL is the load the plate sees. It's the plate resistor in parallel with the input impedance of the stage it's connected to. So I used RL as 510K || 470K = 244K in the example.

The situation here is if you raise the quiescent current while capping the rail voltage, then stage gain must go down, because to support those conditions RL must go down. You would need a lot of rail voltage to keep quiescent current up, while also keeping stage gain up. In fact you'd need a rail voltage of about 570V to support 1mA quiescent current with a 470K plate resistor. So you have competing requirements that cannot all be met simultaneously.

No biggie, you can compromise. Let's say you can live with 1 mA idle current and you're limited to 325V rail. Furthermore, let's assume you can live with 0.9V input sensitivity. Here's what you could do:

• Set the quiescent point at 1V cathode, 1 mA idle current, and 108V across the tube (plate to cathode). That supports a load line from 1.5mA to 325V. That gives you stage gain of about 56 before any feedback is applied (assuming bypassed cathode).
• To obtain no worse than 0.9V input sensitivity, apply no more than about 11 dB feedback.

 

[thorpej]

Stage gain of a common cathode gain stage: mu*RL/(rp +RL+Rk(mu+1)). When cathode is bypassed Rk goes to zero for that calculation. But you can also just use the load line method. I did both and got only a 5% error.

• mu is the gain factor of the tube assuming infinite plate load (I used 80 in the example, as eyeballed from the plate curves at the quiescent point chosen.
• rp is the dynamic plate resistance at the quiescent point chosen (which you determine from the plate curves). I used 26.67K in the example.
• RL is the load the plate sees. It's the plate resistor in parallel with the input impedance of the stage it's connected to. So I used RL as 510K || 470K = 244K in the example.

Ok, I'm using the wrong rp ... I'm just plucking it from the data sheet, which I figured was close, but it's obviously getting me the wrong result. Can you explain how to figure this out from the plate curve? It's not obvious to me.

The situation here is if you raise the quiescent current while capping the rail voltage, then stage gain must go down, because to support those conditions RL must go down. You would need a lot of rail voltage to keep quiescent current up, while also keeping stage gain up. In fact you'd need a rail voltage of about 570V to support 1mA quiescent current with a 470K plate resistor. So you have competing requirements that cannot all be met simultaneously.

No biggie, you can compromise. Let's say you can live with 1 mA idle current and you're limited to 325V rail. Furthermore, let's assume you can live with 0.9V input sensitivity. Here's what you could do:

• Set the quiescent point at 1V cathode, 1 mA idle current, and 108V across the tube (plate to cathode). That supports a load line from 1.5mA to 325V. That gives you stage gain of about 56 before any feedback is applied (assuming bypassed cathode).
• To obtain no worse than 0.9V input sensitivity, apply no more than about 11 dB feedback.

If you look at the schematic I posted, I could very easily bump the last 2 filter caps to 100uF (I was planning on using a JJ 2-section can for it anyway), run the screen and driver supplies in parallel rather than series, and use a 1.5K resistor for the driver supply. 1.5K | 100uF still has a roll-off frequency of < 2Hz, and since I'm filtering the whole rail with a choke in the first place, I wouldn't expect to have any supply noise. On paper, that gives me 340V at the rail to play with.

Using an ECC83 and a 100K plate resistor and 340V rail voltage at the driver stage would be 1mA idle current at a -2V bias point. The gain is > 70 in that case (I'll calculate it more accurately later). Perhaps that's the route I'll go.

 

[kward]

Yup, if a 12AX7 is an option, I think I would go that route, since for this application you need every ounce of gain and drive current you can get.

To determine the dynamic plate resistance from the plate curves, first find the quiescent point and mark it. Then using a ruler, draw a line tangent to the plate curve at the quiescent point, where the tangent line just barely touches the plate curve at one spot at the quiescent point. You may need to extrapolate the plate curve at quiescent point from two adjoining plate curve lines. Find two points on that tangent line and mark them. I usually put the first point down low on the tangent line at zero plate current, and the second point up on the line where it intersects one of the cross hairs on the graph. Dynamic plate resistance is the change in voltage divided by the change in current (i.e., Ohm's law). Find the change in voltage between the two points and then find the change in current between those same two points. Divide voltage by current and that becomes the dynamic plate resistance at the quiescent operating point. And by the way, if you invert that number you have the dynamic transconductance of the tube at that operating point.

In mathematical terms the dynamic plate resistance is simply the slope of that tangent line, inverted.

 

[thorpej]

Ok, here is where I'm at right now. @kward, curious why current matters at the driver stage... (isn't it just providing voltage swing to the output stage?) I could increase gain and lower driver stage distortion by increasing that stage's plate load, but at the cost of current.

 

[jazbo8]

Current matters for all stages, you can't have voltge without current nor vice versa. Try this handy program which allows you to play with the load line and estimate the distortion, you should also include the grid leak resistor of the power tube for the AC load line, i.e., 100k//470k.

 

[thorpej]

Current matters for all stages, you can't have voltge without current nor vice versa. Try this handy program which allows you to play with the load line and estimate the distortion, you should also include the grid leak resistor of the power tube for the AC load line, i.e., 100k//470k.

I've always used the mental model of "voltage is the potential that allows current to flow, but does not require it to do so". But I take your point; electricity is very Heisenberg-y ... to measure the potential, you have to flow some current

I guess what I meant with that question was this: Why does a higher current flow matter for the driver stage, if you're getting the voltage swing that you want? After all, current isn't flowing into the grid of the output stage.

Thanks for pointing out that simulator! Looks pretty cool!

 

[jazbo8]

Because higher plate current usually lowers the distortion (within limit of course), so even if the power tube's grid does not draw current, you still want to provide optimal operating conditions for the driver.

 

[thorpej]

Yup, if a 12AX7 is an option, I think I would go that route, since for this application you need every ounce of gain and drive current you can get.

FWIW, I do also have some 6SJ7s in my tube stash. I use them in guitar amps mostly, but if maximum gain is what we're after, small signal pentodes certainly deliver. But I'd rather not use 2 tubes where one would suffice.

 

[Pio1980]

FWIW, I do also have some 6SJ7s in my tube stash. I use them in guitar amps mostly, but if maximum gain is what we're after, small signal pentodes certainly deliver. But I'd rather not use 2 tubes where one would suffice.

Could a 6SJ7 drive a class 'A' 6L6? A 6K6 or type 41 might have the gain and drive if not if it must be a pentode. Otherwise, if a 6SQ7 could do it, half a 6SL7 should.

 

[kward]

Why does a higher current flow matter for the driver stage, if you're getting the voltage swing that you want? After all, current isn't flowing into the grid of the output stage.

Ah but current is flowing--at AC. It flows through the coupling cap and to ground (and vice versa) through the resistance the low side of the cap sees to ground (i.e, the following stage's grid leak resistor). Capacitor charge time is proportional to the RC coupling's time constant, which is the product of capacitance and resistance that the low side of the cap sees to ground. Audibly, amps with not enough drive current driving a RC coupled stage sound anemic, slow, lack dynamic punch, etc.

 

[kward]

With your latest schematic (post #11), I show a stage gain of 55. This will support the application of up to 11 dB feedback while still delivering 0.9V input sensitivity. This is not bad and I think if you built it as is, it would work pretty well.

However, even in worst case where you allow input sensitivity to be as high as 0.9V, I think you can squeak out a bit more performance from the first stage by moving the quiescent point to the left a bit on that same load line--let's say to 1.5mA quiescent current and 1.2V cathode bias. This will increase quiescent current, raise stage gain, and thus support just a bit more global feedback. The difference in distortion will be inconsequential, especially with 10 to 12 dB feedback applied, but you will have gained another 0.5 mA drive current.

 

[for_p1]

6SL7 is a poor choice for driver. This tube (as well as 12ax7) is designed for high gain, low signal stage. If you need good driver tube, use 6SN7 instead.

 

[kward]

But note the input impedance of the following stage is 470K. Under those conditions the 12AX7 will drive it fine, given the other constraints the OP has placed on this project, such as single tube frontend, a reasonable amount of feedback being applied, and being able to use standard line source to drive the amp to full power. Can't think of a better triode to use than the 12AX7 in this situation given the requirements the OP has stated.