• 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.

Partial Feedback EL84PP - Not quite as expected (yet)

Svein_B

Member
2005-10-24 7:34 pm
I have rewired an EL84 PP amp to use "partial feedback", a'la Yves' ECL84PP and Gingertube's Baby Huey.

So far I have only managed to produce a little over 1.5W of clean power (<1%THD) without GFB, and a little over 3W with GFB.

The circuit diagram:
FV-34CIII-rebild-2.gif


I know that it will be difficult to find an optimal operating point with a starved 12AX7 as driver tube. The effect of the anode-anode feedback will change the load on the 12AX7 to a considerably lower impedance than the static load-line. I have however not yet figured out exactly what the load impedance will be, nor the effective Rp of the driver tube.

A rough estimate of the effective load-line is attached. The oscilloscope also shows that the top half of a sine curve start to look flattened with increased volume.

Amplification in driver tube is measured to 19 (withot GFB). The funny thing is that shorting the 15K shunt resistor does not seem to affect the overall gain.

Any ideas of what to try next?
I mean, it should be possible to make it work, since Yves and Gingertube have done it. Yves have also published his measured power and THD.

SveinB
 

Attachments

  • ecc-rh-curve1.gif
    ecc-rh-curve1.gif
    19.3 KB · Views: 1,102
SveinB,
I ran into low power output on one of the Baby Hueys. In that case it turned out to be a wiring fault causing so much feedback that the front end differential amp saturated before generating enough drive to the output tubes. It is VERY suspicious that shorting the 15K shunt resistor did not cause any change in gain. That is also suggestive of a saturating front end. Check that the diff amp anodes have not less than 140V on them. If the voltages are too low you will need to reduce the tail current. Check that the resistor from the EL84 anodes to each end of the 15K are 47K (2 watt). Check that the resistors from the 15K to the anodes of the 12AX7 are 220K AND *** IMPORTANT **** that the PULL side EL84 Anode is feeding (via the resistor network) the PULL side diff amp anode AND that the coupling cap from that side diffamp anode feeds that same side EL84 grid. That is, check that you have not inadvertently wired positive local feedback.

The fact that the power went up when gNFB was applied suggests that maybe you managed to wire POSITIVE local shunt (partial) feedback. If so swap the feeds to the output tube grids.


Can't think of anything else to check.

Cheers,
Ian
 

Svein_B

Member
2005-10-24 7:34 pm
I admit that I sometimes do the occasional wiring error, but I do not think that is the problem here.

It is quite obvious that the rising distortion at higher power is caused by driver stage unlinearity/overload. Traditional output clipping start at around 8W as it should.

I think I will have a closer look at the effect of the 15K shunting resistor. The design idea is that the middle of this one will have zero AC signal (virtual ground). This may hold true as long as the output tubes operate in class A, but I am unsure of what happens when transitioning to class B. I will replace the single 15K with two 6.8K resistors and measure what happens in the middle. Maybe also reduce the values a little to reduce the signal feedback.

Any one know what the effective Rp of the 12AX7 will be as a diff-amp in this circuit?
This value seem to be important to estimate the feedback and signal levels.

SveinB.
 

Svein_B

Member
2005-10-24 7:34 pm
A few measurements show that the circuit works - sort of, but the 12AX7 seem to be under a bit of strain.

My earlier statement that shortening the 15K shunt resistor did not produce any noticable effect in gain is not entirely correct according to my scope. It does give a difference of 4dB, and lowers distortion with the resistor un-shorted. at least at low to moderate volumes. The shunt resistor is now composed of 2 x 5.6K = 11.2K.

Attached is a scope picture of the anodes of the 12AX7, and the midpoint of the shunt resistor. It is a couple of dB below max power with distortion >5%.
It show unlinearity at the 12AX7 anodes. The lowest trace is the midpoint of the shunt resistor, not quite a true virtual zero.

The circuit seems to operate very close to the limits of the 12AX7 capability, and needs very careful tweaking. I wonder if a more conventilal wiring of the driver with anode resistor to B+ and separate feedback ressitors might be easier to work with. Also a different driver tube (12AT7 ?) might be happier under the strain of the anode-anode feedback.

I am still looking for a formula for the effective Rp of the diff pair.

SveinB
 

Attachments

  • driver-2b.gif
    driver-2b.gif
    66.4 KB · Views: 956
This "partial feedback" means a local feedback (any feedback is "partial", so I don't understand why this particular parallel feedback by voltage is called such a way) from anodes of tubes to their grids. As a feedback by voltage it decreases output resistance, but as a parallel feedback (applied counter - phase to an input signal, in parallel) it decreases input resistance of the stage loading the previous stage harder, on a lower resistance. I've always tried to load triodes by more resistance to obtain less distortions from them. In this particular case decrease of distortions in the final stage increases distortions of a previous one. I still don't understand why somebody would like to do that...
 

Shoog

Member
2002-08-15 10:16 pm
Eire
I still don't understand why somebody would like to do that...

Because experience shows that it works very well, especially if the "partial feedback" resistor can be designed to have a high value and so maintain a reasonable load on the triode. It is a hard balancing act, as I am discovering with my latest design.

Listen to one of these amps and you will understand why people recommend them.

Shoog
 
Svein_B said:
A few measurements show that the circuit works - sort of, but the 12AX7 seem to be under a bit of strain.

My earlier statement that shortening the 15K shunt resistor did not produce any noticable effect in gain is not entirely correct according to my scope. It does give a difference of 4dB, and lowers distortion with the resistor un-shorted. at least at low to moderate volumes. The shunt resistor is now composed of 2 x 5.6K = 11.2K.
The effect of this "inner" feedback loop is somewhat hidden when the "outer" (from opt seconadry to the free LTP grid) one is active.
Indeed, it apears that the outer alone define the final gain.
Attached is a scope picture of the anodes of the 12AX7, and the midpoint of the shunt resistor. It is a couple of dB below max power with distortion >5%.
It show unlinearity at the 12AX7 anodes. The lowest trace is the midpoint of the shunt resistor, not quite a true virtual zero.
This means that the voltage at output stage anodes are not equal.
One goal of this configuration is to improve symetry.
Could you indicate the scale of the scope ?
Perhaps what you see is not a distortion introduced by a wrong operating point of the LTP, but the result of the feedback action, I mean a predistorded signal to compensate the one from the "in the loop" output stage which looks at the limit of the overload ?
The circuit seems to operate very close to the limits of the 12AX7 capability, and needs very careful tweaking. I wonder if a more conventilal wiring of the driver with anode resistor to B+ and separate feedback ressitors might be easier to work with. Also a different driver tube (12AT7 ?) might be happier under the strain of the anode-anode feedback.
Strain ? Mmmh ? no over heat nor over voltage !
Perhaps do you dislike such lo current ? It's the key !
Just use an higher value for plate resistors an don't try to pass more current.
You will have difficulties running 12AT7 in such a way
The 6BQ5 are very sensitive and do not need more than 6V rms drive level. The AX7 can do that easily even with a somewhat reduced load.
I am still looking for a formula for the effective Rp of the diff pair.
The higher, the better the inner feed back loop will act.

Yves.
 

Shoog

Member
2002-08-15 10:16 pm
Eire
Is there any particular reason for so much suffering trying to use AX7 as a driver? May be it is easier to change a tube and use something like 5687 etc.

You have to understand the circuit before you can make this type of substitution. One of the requirements of the driver is a very high rp, which precludes the 5687 and its excellent type. The best type of drivers for this circuit are pentodes.

Go and find the Tubecad article, read it about four times and then design the circuit.



Shoog, there are different ways to skin the same cat, more elegant, without torture of driving triodes.

Try it....

Shoog
 

Svein_B

Member
2005-10-24 7:34 pm
Thank you Yves for joining in. I believe you are the original inventor of this implementation of balancing the shunt feedback.

My measurements are done with the outer global feedback disconnected in order to isolate the inner loop.

My design goal (wishful thinking) was to achieve around 6dB of local feedback and maybe get the output impedance down to around 5 ohm before applying global feedback. So far I have about 10 ohm output impedance. The reduction in output with feedback I believe is a combination of feedback, and the reduced output of the AX7 due to load impedance.

The scope pictures above are 10V/div for the anodes. I am attaching another trace with driver signal at the onset of clipping, and the speaker output. This shows that the signal at the midpoint of the feedback shunt coincides with some "shoulders" of the driver (and output) signals. I am not sure which one is cause and effect.

It is clearly important to keep the load impedance as high as possible. My use of 180K anode resistors instead of your 220K might just be tipping the balance.

I will try to rewire one channel to use separate anode and feedback resistors and see if I can compare the behaviour.

SveinB
 

Attachments

  • shunt-scope.gif
    shunt-scope.gif
    94.7 KB · Views: 757
Svein_B said:
. . .
The scope pictures above are 10V/div for the anodes. I am attaching another trace with driver signal at the onset of clipping, and the speaker output. This shows that the signal at the midpoint of the feedback shunt coincides with some "shoulders" of the driver (and output) signals. I am not sure which one is cause and
effect
.

Neither me, but at this level (near 40V pp) the 6BQ5 are clearly over driven.
The shoulder/flat topping is probably the result of some current in the 6BQ5 grids that the driver is unable to deliver.
When this occurs, the link cap charges itself in such a way that the 6BQ5 grids become more and more negative.
You could confirm/infirm that by scoping at those grids in DC mode.

The fact that so much drive is needed to obtain a bit less than 8W suggests that something may be wrong in the output stage.

Thel signal at FB shunt midpoint contains plenty of H2 that could indicate severe unbalance in the PP ?


Yves.
 

Svein_B

Member
2005-10-24 7:34 pm
SveinB[/i] ... the signal at the midpoint of the feedback shunt coincides with some "shoulders" of the driver (and output) signals.I am not sure which one is cause and effect[/QUOTE] [QUOTE][i]Originally posted by Yvesm said:

Neither me, but at this level (near 40V pp) the 6BQ5 are clearly over driven. The shoulder/flat topping is probably the result of some current in the 6BQ5 grids that the driver is unable to deliver.
...
Yves.

I do admit that i cranked the drive up a bit for the left scope picture to better visualize the effects of the drive circuit. The symptoms are there also at lower volumes. The drive required for 8W is less than 20V p-p as it should be, and there is no serious imbalance anywhere.

The overall result is that the THD start to rise gradually at fairly modest levels, with a high ratio of higher harmonics. Below 1w there is only 0.2% of H2 and H3.

Svein.
 
Svein_B said:
My earlier statement that shortening the 15K shunt resistor did not produce any noticable effect in gain is not entirely correct according to my scope. It does give a difference of 4dB

Let's take a rough estimate of the feedback situation:
What you need to know is the a) gain in the output stage, b) the feedback fraction and c) the plate resistance of the driver tube.

Gain of the output stage is approximate gm x Rl = 11mA/V x 8k/4 = ~20x. That indicates that you will need ~10Vp for 200Vp output, I guess that's not far away from the realities.
The feedback fraction depends of the 47k and 15k shunt resistor and the plate resistor and the plate resistance. First there is a voltage divider of 47k and 15k/2, the feedback fraction after this stage is something like 7.5/(47+7.5) = 0.14, if the plate resistance is 60k the signal fed back to the plate will be about 0.14 x 60/(180+60) = 0.035.
Ok, then we know the gain (A) and the feedback fraction (B), then we can calculate the feedback: A’ = A / (1+BxA) = 12, and log10(20/12)*20 = 4.4dB not far away from the 4dB that you got. ;)

Jan E Veiset
 
Svein_B said:
I have rewired an EL84 PP amp to use "partial feedback"...

So far I have only managed to produce a little over 1.5W of clean power (<1%THD) without GFB, and a little over 3W with GFB.

SveinB

I am not surprised at this. By your schemo, you have 1.1mA in the tail, for 0.55mA of plate current. There is no way that a small signal triode pulling sub-milliamp currents is going to be able to drive the grids of those finals. The 12AX7 can handle high impedance, low capacitance loads, and that does not describe the grid circuit of an EL84, or any other power amp pentode.

I would rewire the thing again, this time losing the funky partial feedback scheme and add cathode follower grid drivers (6SN7, 6CG7, 12AU7, etc. medium gain triodes). (If you don't want to make another hole, then use a MOSFET source follower, and get a MOSFET with the smallest reverse transfer capacitance you can find.) Configured as cathode followers, the input impedance could be bootstrapped to a very high value, and the topology minimizes input capacitance since Miller effect doesn't come into play. Your input capacitance is mainly the triode's reverse transfer capacitance since the Cgk is "reverse bootstrapped" to a very low value. This is the type of load that a 12AX7 can handle.

For an 807 project, this is exactly how I did the design: 6SL7 LTP splitter drawing sub-milliamp currents into 6SN7 cathode followers. I didn't have the problem described here.

After you get the grid drivers installed, then you can start thinking about local feedback. Of course, I would first listen to the thing without any feedback to see what sonic defects you can identify. I haven't done an EL84 project, so I can't say. However, I did an 807-based project. For this one, based on the sonics running open loop, local feedback was necessary. For a 6BQ6GTB-based project, I determined that local feedback was not necessary.

Maybe you don't even need local feedback here?

"It is VERY suspicious that shorting the 15K shunt resistor did not cause any change in gain. That is also suggestive of a saturating front end. Check that the diff amp anodes have not less than 140V on them. If the voltages are too low you will need to reduce the tail current.

(Note: Emphasis mine)"

Really bad advise here: the current is too low already.

And another thing: change how that gNFB in/out switch works. Use it to short out the 1.0K resistor, not open the entire loop. That's just asking for instability and/or noise troubles.
 

ray_moth

diyAudio Moderator Emeritus
2004-01-27 8:55 am
Jakarta
I've experimented with modeling 'partial' feedback using LTSpice, having been enthused by reading Gary Pimm and John Broskie on the subject.

I've tried with output stage using EL34 both triode and pentode connected, and with both 6SN7 and 6AU6 as the first stage. I must admit I didn't try with 6DJ8. I wouldn't even think of trying with a 12AX7 or 6SL7, because the first stage is a voltage-to-current converter and needs high Gm.

The problem I find is getting enough voltage and current for the first stage, even if I use a negative supply as the common (ground) for the first stage. The problem seems to be that if I set the NFB resistor value high, I need less current but I get too much voltage drop across it. If I set it lower, I need more current in the first stage so the voltage drop is still high.

The only way I can get it to work is if I use the first stage as a and LTP splitter, insert an extra (driver) stage in the middle and use cross-coupled NFB from OP tube plates to first stage plates. Now that works really well!

Frankly, I've given up trying to get the simple partial FB arrangement to work. There must be some reason why it never caught on. Morgan Jones (Valve Amplifiers, 3rd Ed.) doesn't seem to be interested in it either.