Attempt on a 7591 Se design

[bea]

Thanks, i'll go through this.

BTW: my results seem pretty close to yours.

Using a 6L6GC, i ended up with something around 7.1W clean and some 8-8.5 V clipped. This appears to me as a really nice result - compared with the data sheet of the 6L6GC i should not expect more.

So if i would do the mod in practice i would probably use the 6L6 instead of the 7591: cheaper and significantly more output power. In actual practice tweaking the preamp is more important because of some weaknesses of the present circuitry.

[bea]

First, until you get to some level of refinement, keep it simple. this means lose as many non-essential components as possible and simulate one section at a time. The schematic below will give you the idea.

I actually did this in the static analysis. At some point I wanted to see the complexity of the power supply and the effect of the NFB loop as well.

Lastly, you can estimate the power out and THD when stepping the input using '.four' and '.meas' commands. Note the TRANS statement parameters are crucial to get meaningful results.

I followed especially that suggestion and got results like the following (using a model of the JJ 6L6GC which AFAIK has been derived from µ-tracer-data):

.step in=0.9 (Remark: step 4)
N-Period=1
Fourier components of V(out)
DC component:0.0836239

Harmonic	Frequency	 Fourier 	Normalized	 Phase  	Normalized
 Number 	  [Hz]   	Component	 Component	[degree]	Phase [deg]
    1   	 1.000e+3	 1.099e+1	 1.000e+0	 -161.96°	    0.00°
    2   	 2.000e+3	 5.030e-1	 4.577e-2	    6.81°	  168.76°
    3   	 3.000e+3	 1.060e+0	 9.644e-2	 -118.59°	   43.36°
    4   	 4.000e+3	 1.861e-1	 1.693e-2	  135.32°	  297.28°
    5   	 5.000e+3	 1.267e-1	 1.153e-2	  -28.53°	  133.43°
    6   	 6.000e+3	 9.270e-2	 8.435e-3	 -132.46°	   29.50°
    7   	 7.000e+3	 4.893e-2	 4.452e-3	  101.51°	  263.46°
    8   	 8.000e+3	 3.807e-2	 3.464e-3	  -35.18°	  126.78°
    9   	 9.000e+3	 2.748e-2	 2.500e-3	 -157.58°	    4.38°
Total Harmonic Distortion: 10.920017%(10.923670%)

...

Measurement: power_out
  step	vrms**2/8
     1	4.09692
     2	5.0647
     3	6.19534
     4	7.6529
     5	9.67627

So with a 6L6 i can obtain about 7.5 W at 10%THD.
After these results i concentrated on the EL84 version. I both versions i had to struggle a bit with stability issues, especially with the EL84. I am not sure to which degree these are artefacts by the models and to which degree they are real.

...
.step in=0.55 (Remark: Step 4)
N-Period=1
Fourier components of V(out)
DC component:0.466336

Harmonic	Frequency	 Fourier 	Normalized	 Phase  	Normalized
 Number 	  [Hz]   	Component	 Component	[degree]	Phase [deg]
    1   	 1.000e+3	 9.276e+0	 1.000e+0	 -178.70°	    0.00°
    2   	 2.000e+3	 2.531e-1	 2.729e-2	   21.10°	  199.80°
    3   	 3.000e+3	 7.622e-1	 8.216e-2	 -165.69°	   13.01°
    4   	 4.000e+3	 1.307e-1	 1.409e-2	  105.87°	  284.57°
    5   	 5.000e+3	 1.182e-1	 1.275e-2	  -17.80°	  160.90°
    6   	 6.000e+3	 6.183e-2	 6.666e-3	  175.17°	  353.87°
    7   	 7.000e+3	 5.235e-2	 5.644e-3	   62.64°	  241.34°
    8   	 8.000e+3	 3.023e-2	 3.259e-3	  -64.73°	  113.97°
    9   	 9.000e+3	 2.397e-2	 2.584e-3	  153.30°	  332.00°
Total Harmonic Distortion: 8.916135%(8.923619%)

...


Measurement: power_out
  step	vrms**2/8
     1	2.86495
     2	3.71463
     3	4.65839
     4	5.58837
     5	6.25489
     6	6.67498
     7	6.96952

5.6 W at 9% THD (Step 4) ... that's pretty close to the predictions of the data sheet for an anode voltage of 250 V (Valvo, Telefunken). Even the values for k2 and k3 do match reasonably well.

Overdrive:
when the EL84 is overdriven, the output power can reach 8 W. To my understanding that means a heavy thermal overload of the anode - and indeed, my amp does not like to be overdriven for a long time.
With the 6L6GC the fully clipped output corresponds to about 12W which should be on the safe side

[nickb]

Sounds like you are getting on well.

You might want to check the models you have for inter-electrode capacitances. The thing that will have the biggest impact on stability is the output transformer. At a minimum I would add DC resistance, primary self-capacitance and inter-winding capacitance. You'll just have to guess at values (unless you have access to a network analyzer).

[bea]

Sounds like you are getting on well.

You might want to check the models you have for inter-electrode capacitances. The thing that will have the biggest impact on stability is the output transformer. At a minimum I would add DC resistance, primary self-capacitance and inter-winding capacitance. You'll just have to guess at values (unless you have access to a network analyzer).

I looked at Daten_AÜ and found nothing even there. Well i could estimate from the Ls and the resonance frequencies. But is that really worth the effort?
When i modded the amps i had a lot less experience - and i did neither own an oscilloscope and a signal generator and could not look what i thought would happen (in terms of oscillation). The simulations give me strong hints what to look for and how to fix it - but reality ids different anyway.

[bea]

Preliminary final model, 6L6 version

Maybe i'd show important results as replies to my first posting....

So here it goes - Schematics and output form intermediate level into saturation. As there is no output voltage of the bias voltage i'll have to derive it from the main HT. If i decide to do the mod i will work on only one of my two amplifiers. Simple reason:

Amp A has one volume control and one tone control, and it has the switch for the NFB on the front.
Amp B has an additional "master" volume and the NFB switch on the back.

So Amp B looks like the ideal candidate for a circuit like the following, and Amp A will keep the EL84 unless i find something better. Maybe the russian military variant of the EL84 with a Pa_max of 14W instead of 12 (my results for the JJ-7591 are not better than what would be expected from that tube, and the effort is smaller.)

Because of the small value of the grid leak resistor the 6L6 requires with static biasing a master volume is a bit difficult. Therefore i omitted it in the following plans. You might notice the unusually large anode resistors of the ECC83. This is well and safely within the limits, the headroom remains sufficient to fully overdrive the 6L6GC, and the valve has less distortion. Disadvantage is the larger output impedance - but in connection with the small load impedance this is rather an advantage than a disadvantage - the effective (dynamic) load will be roughly as usual.





And the log:

.step in=0.85
N-Period=1
Fourier components of V(out)
DC component:0.157466

Harmonic	Frequency	 Fourier 	Normalized	 Phase  	Normalized
 Number 	  [Hz]   	Component	 Component	[degree]	Phase [deg]
    1   	 1.000e+3	 1.002e+1	 1.000e+0	 -161.36°	    0.00°
    2   	 2.000e+3	 2.870e-1	 2.864e-2	   48.65°	  210.01°
    3   	 3.000e+3	 5.369e-1	 5.357e-2	 -101.84°	   59.52°
    4   	 4.000e+3	 1.652e-1	 1.648e-2	 -176.60°	  -15.23°
    5   	 5.000e+3	 8.401e-2	 8.383e-3	   74.13°	  235.49°
    6   	 6.000e+3	 5.443e-2	 5.431e-3	  -47.40°	  113.96°
    7   	 7.000e+3	 4.279e-2	 4.270e-3	 -157.69°	    3.68°
    8   	 8.000e+3	 3.093e-2	 3.086e-3	   89.65°	  251.01°
    9   	 9.000e+3	 2.360e-2	 2.354e-3	  -26.24°	  135.12°
Total Harmonic Distortion: 6.398771%(6.408589%)

.step in=0.9
N-Period=1
Fourier components of V(out)
DC component:0.264288

Harmonic	Frequency	 Fourier 	Normalized	 Phase  	Normalized
 Number 	  [Hz]   	Component	 Component	[degree]	Phase [deg]
    1   	 1.000e+3	 1.098e+1	 1.000e+0	 -161.47°	    0.00°
    2   	 2.000e+3	 3.623e-1	 3.299e-2	   18.50°	  179.97°
    3   	 3.000e+3	 8.716e-1	 7.938e-2	 -108.56°	   52.91°
    4   	 4.000e+3	 2.035e-1	 1.853e-2	  164.12°	  325.59°
    5   	 5.000e+3	 1.213e-1	 1.104e-2	   33.53°	  195.00°
    6   	 6.000e+3	 9.677e-2	 8.813e-3	  -89.59°	   71.87°
    7   	 7.000e+3	 7.245e-2	 6.598e-3	  158.27°	  319.74°
    8   	 8.000e+3	 5.153e-2	 4.693e-3	   34.78°	  196.25°
    9   	 9.000e+3	 4.332e-2	 3.945e-3	  -86.17°	   75.30°
Total Harmonic Distortion: 8.951782%(8.978948%)

.step in=0.95
N-Period=1
Fourier components of V(out)
DC component:0.327321

Harmonic	Frequency	 Fourier 	Normalized	 Phase  	Normalized
 Number 	  [Hz]   	Component	 Component	[degree]	Phase [deg]
    1   	 1.000e+3	 1.173e+1	 1.000e+0	 -161.18°	    0.00°
    2   	 2.000e+3	 5.127e-1	 4.370e-2	   -3.86°	  157.31°
    3   	 3.000e+3	 1.301e+0	 1.109e-1	 -114.23°	   46.95°
    4   	 4.000e+3	 1.984e-1	 1.691e-2	  136.24°	  297.41°
    5   	 5.000e+3	 1.857e-1	 1.583e-2	   -4.12°	  157.05°
    6   	 6.000e+3	 1.378e-1	 1.175e-2	 -120.85°	   40.33°
    7   	 7.000e+3	 9.415e-2	 8.025e-3	  118.47°	  279.65°
    8   	 8.000e+3	 7.524e-2	 6.413e-3	  -17.86°	  143.31°
    9   	 9.000e+3	 6.267e-2	 5.341e-3	 -139.23°	   21.95°
Total Harmonic Distortion: 12.256270%(12.306296%)

.step in=1
N-Period=1
Fourier components of V(out)
DC component:0.297474

Harmonic	Frequency	 Fourier 	Normalized	 Phase  	Normalized
 Number 	  [Hz]   	Component	 Component	[degree]	Phase [deg]
    1   	 1.000e+3	 1.220e+1	 1.000e+0	 -160.12°	    0.00°
    2   	 2.000e+3	 7.063e-1	 5.792e-2	  -11.61°	  148.51°
    3   	 3.000e+3	 1.820e+0	 1.492e-1	 -117.08°	   43.04°
    4   	 4.000e+3	 1.895e-1	 1.554e-2	   72.07°	  232.19°
    5   	 5.000e+3	 2.616e-1	 2.145e-2	  -29.51°	  130.61°
    6   	 6.000e+3	 1.683e-1	 1.380e-2	 -149.09°	   11.03°
    7   	 7.000e+3	 1.089e-1	 8.929e-3	   76.29°	  236.41°
    8   	 8.000e+3	 9.909e-2	 8.125e-3	  -57.29°	  102.83°
    9   	 9.000e+3	 6.333e-2	 5.193e-3	 -178.03°	  -17.91°
Total Harmonic Distortion: 16.338583%(16.386446%)



Measurement: ig2
  step	RMS(ix(u5:screen))	FROM	TO
 
     5	0.00184526	0	0.01
     6	0.0024257	0	0.01
     7	0.00289037	0	0.01
     8	0.00317757	0	0.01

Measurement: power_out
  step	vrms**2/8

     5	6.22658
     6	7.57143
     7	8.81957
     8	9.76553

[jazbo8]

You might notice the unusually large anode resistors of the ECC83. This is well and safely within the limits, the headroom remains sufficient to fully overdrive the 6L6GC, and the valve has less distortion. Disadvantage is the larger output impedance - but in connection with the small load impedance this is rather an advantage than a disadvantage - the effective (dynamic) load will be roughly as usual.

I still don't see the point of using such large Ra's, what you get in the extra gain is just thrown away by the voltage divider prior to the 6L6, not to mention making the Miller effect more prominent...

[bea]

I still don't see the point of using such large Ra's, what you get in the extra gain is just thrown away by the voltage divider prior to the 6L6, not to mention making the Miller effect more prominent...

A large Ra gives You more gain with less distortion. The price is less headroom and a larger output resistance. Soundwise it is audibly cleaner and more open (i played with this in my current amp). I checked other possibilities and that one appears to me the most reasonable for a cleanly sounding amp.

The voltage divider in front of the 6L6 has the purpose of

- presenting a reasonable load to the ECC83 that will not reduce the gain too much (we are at 230 kOhms here, important even if i reduced the anode resistance - load in this range has a pretty strong impact)

- being a measure against blocking distortion.

The Miller effect has no practical impact - the frequency response is linear in the range of interest

Gain reduction: there is still enough gain do slightly overdrive the 6L6. So why bother?

[nickb]

A large Ra gives You more gain with less distortion. The price is less headroom and a larger output resistance. Soundwise it is audibly cleaner and more open (i played with this in my current amp). I checked other possibilities and that one appears to me the most reasonable for a cleanly sounding amp.

The voltage divider in front of the 6L6 has the purpose of

- presenting a reasonable load to the ECC83 that will not reduce the gain too much (we are at 230 kOhms here, important even if i reduced the anode resistance - load in this range has a pretty strong impact)

- being a measure against blocking distortion.

The Miller effect has no practical impact - the frequency response is linear in the range of interest

Gain reduction: there is still enough gain do slightly overdrive the 6L6. So why bother?

Since we seem to have swung back to this topic, looking at the second stage which has a lower output load and a bigger swing, I see the gain is lower and the distortion is higher with the 390K plate resistor. The lower gain of the 390k lowers the Miller capacitance and improves the high frequency response (a little).

For the first stage and using your figure of 230k for the output load, the gain of the 390K stage is only 0.25dB bigger. The 390K case distortion remains higher.



Gains:

Ra_gains.pdf

Distortion:



EDIT: I just looked at the new version you posted. This has Ra=430K and a much higher load impedance but the point remains. You get a tiny bit of gain at the expense of an increase in distortion.

[bea]

back to the 7591.

Under no way i could reproduce the output of the data sheet. If i simulate a 3 kOhm load i end up with slightly more output power than for 5.2 k load - about 0.5 W ore maybe even less. Clipping will occur if the output increases above about 7.5 W and about 7.2 W in the 5.2 k case. This is in sharp contrast to the 6L6 where my results are immediately plausible from the examples in the data sheet.

Under similar voltages at anode and G2 and an anode current of a bit less than 60 mA both tubes will deliver nearly the same power, the 6L6 systematically a bit more. IMO eeasy to explain: Power pentodes and tetrodes act as current sources on a load impedance which is small compared to the internal impedance of the tube itself. The closer the load impedance will come to equality with the internal resistance of the source, the larger the output will be.

Accordingly, in case of a load of 32 Ohm on the 8 Ohm tap the 6L6 would deliver full 15 W or more (in the simulation!!!), and the 7591 with its larger internal resistance some 13 W. Mhmm...

[nickb]

Note the data sheet THD is 11%. Also, the model may be some way off. Here I've overlaid the Koren model over the data sheet using the same -14 to 0V G1 range. Sorry - it is a bit hard to read as the model curve lines are very faint, but hopefully you can see that it doesn't model the low anode voltage region very well and the grid 1 voltage is about 1.5V out. You might want to try the same experiment with your model.

[bea]

I was not aware that the difference would be THAT large - that should really affect the modeling results. But anyway, my main goal seems to be reached: it looks as if my idea for the Valve Junior is a realistic option promising me roughly 50% more power. And both tubes in question seem to be usable.

Before i pick up the soldering iron, i might do a similar case study for a PP amp which needs to be fully rebuilt anyway. Again a design for the EL84:
I am currently trying to collect data on the power transformer (Wüstens ATRA 0201, 2x100 mA, 230V, usable in parallel) and the OT (Welter Ü3, Raa 8kOhm, 50mA).

(If i like to take the soldering iron, i'll first restore the power stage of a Dynacord HiFi Favorit - one of those beasts with 80W from 2 EL34)

[bea]

EDIT: I just looked at the new version you posted. This has Ra=430K and a much higher load impedance but the point remains. You get a tiny bit of gain at the expense of an increase in distortion.

Ok, i'll start with the first changes - in the amp which one day might be run with the 7591. This amp is my main practice and backup amp, therefore tiny steps. That amp sounds a bit dark, and that has two reasons: a) the large output resistance of the 1st stage lead to loss in treble if the anode resistor was too large (320 kOhms). b) the tone stack i used was not the AMZ stack but a simpler variant lacking one resistor. My workaround was to use an ECC81. Due to its smaller output resistance, a was a lot less pronounced.

But unfortunately ECC81s tend to be very microphonic.

What i did: reduce the anode resistor of the 1st stage to 100k and to leave the anode resistor of the 2nd stage on 47k in the hope that it will be still easier to overdrive it a bit. I also added 330 pF in parallel to the NFB resistor in the hope to increase the stability margin.

In spice - small signal analysis - everything looks good now. Let's see how it sounds (tomorrow, not deep in the night).

EDIT: it does what it is suspected to do. Fine.
It is pretty hard to drive the 2nd stage into distortion. Need to think about that.



The next step will be looking at the 2nd amp. That one has no master volume installed and is indeed my candidate for the 6L6 mod. Plan: first do the preamp, notably the tone stack, and then install the 8-pin socket. I will not touch the 1st amp unless i am sure the 2nd amp runs reliably with the 6L6.

[bea]

Please let me pick this up again. As already mentioned i would like to modifiy one of the two amps, and i want to do the 6L6 variant first.

There is still an open question: biasing. As i want to squeeze as much power as safely possible from this little amp, cathode biasing is not applicable any more, and every volt on the rail voltage counts... The 6L6 needs a fairly small grid leak resistor if static biasing is applied. This will impose some difficulties to the design of the coupling of the ECC83 to the 6L6 - the permissible 100k is a pretty small load for that tube.

The reason of my question is which method to obtain a biasing voltage might be preferrable. I have a strong tendency to use a capacitive voltage divider similar to the Ampeg V4B, which also uses power tubes from the 6L6 family.

My question is to learn about different approaches and their pros and cons, and of course if that approach is the preferrable one.

Thanks



PS: I'll probably use the preamp i did in another project, where the ECC83 also sees a load of 100k: i will use 3.3k on the cathodes and 220 k at the anodes. From the previous project i learned that this will sound fine with such a small load resistance.