• 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.

6336-SE modification

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Joined 2003
Paid Member

As reported in last spring in another thread I’ve finished a 6336-SE.
I’ve attached the design of Grover Gardner, he encouraged me to do so. It has a classic power supply (no electrolitics in the last filter stage) and the heater voltage is lifted some to prevent hum.
I’ve been listening for a while now and I’m still really pleased with it. Thanks again Grover for sharing this.

Curiosity – as always - drives me to using the solvering iron to try a few options.
There are some modifications I’d like to try. Some thoughts are:
- incorporate a grid choke in the third stage. Some say it will make the 6336 really sing.
- change the second stage of the EL34 to a screen drive mode to make it more linear (inspired by an article in Sound Practices). Would this be wise/feasable?
- a constant current source (ccs) instead of a cathode resistor in the third stage. However, somewhere the datasheet that says the 6336 needs at least 100 ohm cathode resistor. How does this relate to installing a ccs?
- Replace the 6336 by a 6528. I think the 6528 will drive on a lower B+ voltage with a lot more current.

I don’t have a EE-competence to design such options. I’d like to ask members of the diyaudio-community to comment on such options. And may be one would be even willing to contribute to a development effort. With concrete suggestions for change of components/values I could experiment some and report it back.

Any thoughts?

Best regards, Arjen


  • 6336_baby845_grover.jpg
    93.9 KB · Views: 2,302
hi Arjen,

i have not too much tube knowledge yet, but...about your third option...

about adding a ccs instead of a cathode resistor....
As far as i know one of the things about at a ccs is that it has a virtually infinite resistance...(somewhere about 1Mohm-2Mohm)

It tends to make the load line more horizontal then with a normal resistor....thus linearising output....

if it is an anode ccs or cathode ccs both have this mechanisme because tube current flows there

have a look at:



Joined 2003
Paid Member
6336 modification

Hello Kathodyne,

Thanks for your kind reply (we could talk Dutch off-list could we?).
Yes the theory on a ccs looks easy enough.
But to put it in practice in the case of the schematic that I uploaded; that's a quest I cannot accomplish on my own.

I hope this forum offers some input to progess in the next months.

Best regards, Arjen.


Joined 2003
Paid Member
> a constant current source (ccs) instead of a cathode resistor in the third stage.

If the cathode bypass capacitor is big enough, this should make no difference to sound except possibly in more sonsistent bias conditions.

That opens a problem. The cathode resistance of 6336 at 330V 80mA is about 167Ω. 47uFd is barely big enough; some would say not big enough at all. You might try a 200uFd 150V cap and see what it does to the bass.

> However, somewhere the datasheet that says the 6336 needs at least 100 ohm cathode resistor. How does this relate to installing a ccs?

They didn't have CCSes in those days. For what they are saying, a CCS is better than a resistor. What they do not want you to do is bias it with a fixed voltage: tube-to-tube variation is high, and these tubes can go into a run-away mode. Some cathode (or plate) resistance makes them self-bias and find their happy operating point. A CCS will FIX the current and eliminate this type of runaway.

> Replace the 6336 by a 6528. I think the 6528 will drive on a lower B+ voltage with a lot more current.

Rather the other way round. Compare curves for 6336 and 6528.

plate _50V_ _70V_
6336 250mA 450mA
6528 120mA 200mA

> 6528 will drive on a lower B+ voltage with a lot more current.

Looking at this like an engineer, it seems mismatched (with either tube). Say the 6336 idles at 330V 80mA. For maximum power we can swing 80mA +/-80mA, or from zero mA to 160mA. At 160mA the 6336 has 40V plate-cathode drop, so we can swing 330V-40V= 290V. 290V/80mA suggests 3.6K load, not too far from the 2.5K specified. BUT we can't swing any tube, and especially these, to zero mA. For low distortion we might only swing to 40mA, maybe 30mA; or 80mA +/-50mA. Taking 290V/50mA suggests 5.8K for good power and efficiency with low distortion.

Alternatively, given 2.5K 80mA iron, you could reduce the voltages without reduction in power. Too late now that you already built it: the cost of new power iron exceeds the cost of power wasted unless you will run it 24/7.

What the high-Mu 6528 does is waste less voltage in the cathode bias resistor. If we stay at 330V plate-cathode, about 40V instead of 120V, so a 370V supply will do. Or we can take it up to 400V P-K, 50V K-G, holding to 75mA to avoid melting. Saying 75mA +/-45mA, 50V drop at the peak 120mA, 350V/45mA suggests 7.8K loading.

Either 6336 at 80mA 5.8K or 6528 at 75mA 7.8K gives over 7 Watts, and will give essentially the same current to the load. Either will give more load current than 6336 at 330V 80mA and 2.5K load, which is about 3 Watts.

> change the second stage of the EL34 to a screen drive mode to make it more linear

Beware. That second stage can't be linear trying to slap 120V peak with just 450V supply. However the 6336 stage is nonlinear too. The nonlinearities are opposite and will cancel. Making the second stage more linear (I'm not sure G2-drive will do it) or using the higher-gain 6528 as third stage will reduce the cancellation and actually increase numeric distortion.

While I critique it on engineering grounds, I do believe someone optimized the sound of it. The audible difference between 3 watt and 7 watt is almost nil, while the audible effect of nonlinearity cancellation is probably what makes the very curved 6336 pleasing. I think if you mess with it, many things will be no-improvement or even make it worse. It might be interesting to use the general plan with completely different parts: 2A3 or 6F6 driving 6528 with 8K load, or 6336 with 2.5K load working at 250V P-K, 120mA idle, -75V bias (325V total supply), and EL34 with choke for the plate load (so it can kick far above B+ and make those 75V swings on 6336 grid). Or even take the 6336 grid resistor to its 500K max, use 6SN7 driver with 50K plate resistor. But this would be a very different amp, and a LOT of re-thinking.
Joined 2003
Paid Member
Hello PRR,

Thank you for your clear analysis.
You point out that tweaking the design as it is would mess up thing probably more than better it.
Also you suggest some openings for a better design from an engineering point of view.
This might be worthwhile, although you conclude it would become a different amplifier and would require a lot of rethinking. Exactly this however is my weak point.
May be it would be an option to start with some small steps that you’ve mentioned, like:
- replace the cathode resistor of the 6336 with a CCS. I know the principle of a CCS and used it in preamps. But I’ve not seen yet any readily available designs for an output tube.
- replace the anode resistor of the EL34 with a choke as you suggest. What would be the property of a suitable choke? And would it require a change in value of the cathode resistor?

As next steps it might be interesting to experiment with:
- different setting for the 6336 to minimise for non-lineairity. The output transformer (air gapped Edison 60 OPT) would allow for different impedances (2,5k to 6k)
- use a completely different driver as the 6336 is hard to drive. You suggest the 6SN7 with 50K plate resistor. An alternative would be in interstage iron or even a powerdrive option as mentioned by tubelab in the 833-thread. Somehow a cathode follower as a driver (like in the early version of the ongaku-211 or other designs) is not popular. Although SY in the thread on the unity gain stage offers an appealing approach for a pre-amp environment.

Could anyone comment on these options?

Best regards, Arjen.
Hi Arjen,

As requested, I add my opinions.... Even though I haven't posted in 6 months...:cannotbe:

I can agree with PRR that some "distortion cancellation" is taking place in this design. The resultant sound may or may not please you.
By improving the linearity of the driver stage the THD will probably increase. But I think it may sound better. My reasoning is as follows:
The output stage has a transfer curve that is mainly 2nd order. The driver stage has a similar but inverse effect. But they don't match over their length. Therefore the composite curve will need higher orders to describe it. As such the distortion spectrum will be more complex. Furthermore the higher order components are likely to to be sporadic. I've seen these kind of distortion components rising and falling as the temperature and mains varies over time.
In other words it cancels some of the 2H and adds some 3H & 5H.

My best idea for linearizing the driver stage is to replace the anode load resistors with an audio choke. This will almost double the drive headroom, and so the audio can use the straighter, (middle) part of the curve.
If you don't have a choke to hand, you could try using the primary of a small SE output transformer, such as can be salvaged from an old radio or tape recorder.
Watch out for sparks if you overdrive the stage. There will be a lot of volts on the EL34 anode!

No doubt any mistakes on my part will be speedily corrected ;)

Your idea of using a grid choke on the output tube might be worth investigating. I've had nice results when using them in my 12e1 amp.
Choke itself is a special item typically 1500H and 3.5k dc resistance. Made of very thin wire, it won't tolerate much current so be careful when wiring up.

more info here

and here
Joined 2003
Paid Member
Hello martinab2,

Thanks for your encouraging words on a grid choke.
I've looked around for further examples but I don't get much of a clue on how to calculate it. What I notice is that the 'bottom' of the grid choke in most cases is connected to the middle node of a split cathode resistor.

Can any of you give a hint on how to calculate the value of the grid choke and how to connect it?

Best regards, Arjen.
As far as I know the only current manufacturers of grid chokes are Magnequest and Lundahl. It probably doesn't matter which one you choose as in this application the choke behaves as a 'smart' grid resistor with low dc resistance and high ac resistance. The actual inductance is probably irrelevant, just as long as it's high enough which both of these ones are.

Connecting it up couldn't be simpler - just replace the 180k resistor from the 6336 grid to ground with the grid choke.

I have read somewhere about adding a resistor in parallel with the choke to remove any peaks in LF response, however I have no idea how to calculate this and have not needed to do this in my amp.

There are various theories as to why the use of a grid choke would improve the sound (or not). In my application I suspect that the low dc resistance grid circuit improves bias stability.
I have experimented with a few series regulator type tubes (low voltage high current) in fixed bias and noticed that some current was being drawn from the bias supply. That's not meant to happen according to the textboks but might explain why some circuits sound better with cathode bias.
If some (tiny) current does flow in the grid circuit of certain tubes, the dc resistance of the circuit could be significant
Joined 2003
Paid Member

Your suggestions made me reflect on how to proceed. Herewith an update of the options. I guess PRR is right when this would turn out to be a complete different design in the end.
Before I proceed to implement the options I would like to invite you to comment on this update. There are too many choices where EE-knowledge is presupposed that I do not cover enough yet.

First step: Improve the driver stage
Some advised to use a plate choke and some an interstage.
I have bought an ISO-Tango NC-14 interstage (5k, 30mA,1+1: 1+1) during my recent visit in Japan. This might also be wired as a plate choke. With the primaries in parallel it can handle even 60 mA. I would prefer to use it as an interstage in the first place.
Only, how to connect it? I have looked at many schematics with interstages in the past months I did hardly see a schematic where the primaries and the secundaries are both in series.
Most connect the primaries in series and the secundaries in parallel. Which I guess means the drive is focussed on current. The EL34 driver in the design is now set at some 25 mA. With the interstage in series it can only be set to a maximum of 15 mA. Is it correct that this lower dc-current will be ‘compensated’ by the fact that the choke will allow for a larger voltage swing? Moreover the coupling capacitor will not be needed anymore.
Another way of wiring is to parallel the primaries and have the secundaries in series. I did not see this option apart from one example, and it used precisely the 6336 (see below the Blue Magoo amp of John Day). Do you have an advise, then please let me know.

Another step that can easily be taken is to provide the driver stage with it’s own power supply. Now it is derived from the output stage via 100 ohm/2W.

Step 2: DC-coupling between the first stage and the driverstage
In the school of Shishido and others it appeared that DC-coupling between the first stage (here 2 x 6SN7par) and the driverstage (here EL34) offer a considerable audible improvement.
I do not know yet how to calculate this but I have some examples of the Japanese magazine Mussen to Jikken that used the EL34 in the driver stage. If anyone is interested I can send the schematics to evaluate if this can be a drop-in in this 6336-amplifier. I do not know if it is allowed to post the schematics in this specific case. I would suggest it poses no problem as long as is referred to the original source. Also the difference in language/culture can be bridged by discussing elements of such an article in an open way. May be the moderator can shed his light here?

An interesting experiment for the longer term is the so called ‘direct-coupling’moderated-bias circuit as described by Ari Polisois in the GA-special 2002 page 51. This requires an extra supply with separate winding but that is no problem in my case. The supplies of the first and second stage are kind of ‘build on each other’ so a coupling cap nor an external bias voltage supply is needed (my intuitive way of understanding).

Step 3: the output stage
With installation of the interstage I could simply leave the 180k bias resistor and the 1500/50W cathode resistor in place to obtain the –90 bias voltage. Is this correct?

To experiment further the choice is to maintain the autobias with the 180k or to apply fixed bias. With fixed bias there still is a small cathode resistor needed of 100 to 150 ohm for the 6336 (as indicated by Groover Gardner and others). However, PRR seems not in favour of fixed bias with low-Mu triodes working at such ratings: small shifts in the tube can mean large shifts in bias and thus a melt-down.
Replacing the 180k autobias resistor by a grid choke also is surrounded by concern. First of all, I have never seen such an arrangement in other schematics. Is this feasable or sensible at all? And as Martinab2 said: ‘I have experimented with a few series regulator type tubes (low voltage high current) in fixed bias and noticed that some current was being drawn from the bias supply. It won't tolerate much current so be careful when wiring up.’ And PRR said in another thread: ‘Also note that the 6336 grids are leaky, you need a fairly low grid resistor.’
As grid chokes do not like dc-current may be I should measure first if any current is passing the 180k resistor.
Furthermore as a measure of prevention on leakage I might add a diode from grid to cathode on the 6336.

As for the output transformer I bought the improved version FT-20S of Andix (Akihabara in Tokyo) of the ISO-Tango XE-20S that can be installed with some minor changes in lay-out of the chassis. However, in the "Blue Magoo" 6C45PE-6336-Amp of John Day
(see: http://audiokarma.org/forums/showthread.php?t=15587 )
Quote: ‘The transformer is a Hammond 2.5K SE design with minor surgery performed to separate the primary windings 60:40 at the ultralinear tap. The 60% winding loads the plate of the 6336, and the 40% winding loads the cathode, providing part of the cathode bias resistance along the way. This provides a higher damping factor for the output stage and gives a strong, sure, competent bass sound...’ This is an interesting option that might be feasable with my current Edison 60 airgapped OPT’s. Hmm. Interesting option?

A next experiment will be to use the 6528 output tube of which I have some in the drawer.
Here the question will be as PRR analysed before so clearly whether the EL34-driver and the 6528 will like each other. Then I probably can experiment with a low impedance driver of my stock like the 6BX7 or the A2293.

Well, all in all next winter cannot be cold enough to experiment with such options. Please feel free to comment as I need all resources to stay on track.

Best regards, Arjen.
Joined 2003
Paid Member

Time to start with the first step to improve the driver stage as indicated in my last mail.
I found an example of how to connect an EL34 for the Tango interstage NC-14 interstage (5k, 30mA,1+1: 1+1). It is a schematic of the Japanese MJ-magazine of 12/2003. I attached the schematic. The plate of the EL34 is connected through the primary in series at 408V/19mA. The feedback part I will leave out for the moment.
The secudairy is wired in parallel for an 845. So this way I should have enough drive for the 6336A. Contrary to the direct heated 845 however in my case I will need a grid resistor to bias at some 90V. I will leave the original grid resistor and the cathode resistor of the 6336A for now.

One more interesting thing in this schematic is that the first stage with 12R-LL3 and the driver stage are direct coupled. This can be my next step in the experiment. Would there be any risk of direct coupling in my case? I wonder if I can substitute the first tube with something more familiar like a 6BX7. Have to look at the datasheet first.

Any comments before I start soldering?

Best, Arjen


  • mj-12-2003_12r-ll3-el34-845.jpg
    36 KB · Views: 930
Joined 2003
Paid Member

I did find a description of Alan Kimmel's choke loaded mu stage at the audioasylum site:

In a next post of this thread by twystd (A) on January 05, 2004 the modification of the original version is described as: ‘First buy a pair of Lundahl 1668 chokes, I mounted mine on the sides of the wooden basses away from the signal part of the amp for EMI reasons. Then jumper pin 2 and 4 together on the choke (pin one being furthest from the mounting base, pin 4 closest to the mounting base). Remove R4 and replace with the choke, pin 1 on the choke going to the cathode of the 6AC7 and pin 3 going to the plates of the 6SL7. Remove C2 and replace with a jumper. Remove R6 entirely, then replace R7 with a jumper. That's all there is to it. You said you tried other mods, out of curiosity, what were they, and did they work for you? ‘

Allthough it is not a schematic yet one could work out one.

Best , Arjen
Joined 2003
Paid Member

While progressing slowly in implementing the first modification step it can be interesting to explore the suggestion of a mu-stage as a driver. Apart from the choke loaded mu-stage there also is the inexhaustible source on CCS of Gary Pimm at www.pacifier.com/~gpimm

Under the heading presentations in the document ‘CCS's and signal current control’
Is described how to use a CCS as a combined CCS and mu follower. Near the end some is said on how to drive a difficult output stage. Quote:
‘Output stages are a different case. The CCS is used as a high performance replacement for the plate choke. The CCS offers performance that is simply not achievable with inductors.
The object is not to operate the triode at constant current but to provide the output stage with a constant bias current that does not change over the operating frequency range of the amplifier.’
End of quote.

The object to provide the oputput stage with a constant bias current looks interesting. If I read well this also is at stake in the so called power drive of Tubelab. I have read a lot about driver stages now, and there is much copy and paste. The approach of Gary Pimm offers a perspective that can be relevant for the difficult to drive 6336 tube. Would there be anyone out there that can shed a light on what this means for redesign the driver stage of the 6336-SE of my first post?

Best Arjen.
Joined 2003
Paid Member

There is little yield on my searches for a ccs in the cathode of a tube. For instance in post 29 of the ‘12B4 line stage’ thread Brian Beck stated that he often use a LM317 for a ccs in the cathode.
For the 6336-SE I consider to replace the cathode resistor of the driver stage, EL34 at 28 mA/410V, with a ccs. The cathode resistor of 750 ohm measures a voltage of
Also in the output stage with the 6336 I would like to replace the cathode resistor with a ccs. Here the cathode resistor of 750 ohm/80mA measures a voltage of –90V. I intend to reduce the cathode voltage to – 10V (10% of the bias voltage) and use fixed bias for the rest.
As said I am now installing an interstage transformer between the EL34 and the 6336, and might consider a ccs as a plateload of the driver later on, may be in conjunction with the power drive of Tubelab.

Questions I now have are:
- Can the LM317 be used in both the driver stage and the output stage? From my knowledge with the ccs in the driver stage of the superb PP-2C of Allen Wright the active device in the ccs must be able to withstand the total anode voltage (here 410/450 V). The LM317 would not be capable to handle such a voltage.
- With the LM317 however I will need a bypass cap and that is exactly what I want to get rid off. What are other options to install a ccs without the need of a bypass cap? For instance, this might be done in a ccs-design with a transistor or mosfet arrangement as in the SS-design ‘Power Follower 99C’ as designed by Andrea Ciuffoli.
This also would allow to pass AC, I think.

Could someone please comment on my questions?

Best, Arjen

Email: vrhff04_@_xs4all.nl (remove the _ signs)
I am using a LM317 as a CCS in the cathode of a 6080 output stage. I am running it at 100mA and -30V bias. Works fantastically. However by its very definition it has to be bypassed with a cap or it wouldn't be doing its job. If thats a problem then forget it.
In PP stages the output tubes can only be biased into class A.

Sonically you can expect a tightening up of the sound with strong definition in the bass. Definately worth it as far as i'am concerned and I wouldn't go back.

Joined 2003
Paid Member
Hello Shoog, Others,

Thanks for your experience and encouragement.
I might do an experiment with the LM317 and learn if the bypass with a ccs gives less impact than with a resistor.

Apart from that I still wonder what is the option for a ccs that does not need a bypass by nature.

Best, Arjen.
av-trouvaille said:

Time to start with the first step to improve the driver stage as indicated in my last mail.

Any comments before I start soldering?

Best, Arjen

I would use an easier to find tube in the first stage. If much gain is needed, a triode wired C3g would be nice. Any other high gm pentode with 3-4 watt Pa will do.
The direct coupling loftin-white-like is not safe. If the second stage comes before the first has reached it's operating point, the grid sees the full PS voltage. Nice firework and possible damage of the second tube.

I use a seperate negative supply for the first stage with an anode choke as a load. The anode choke is grounded, so in the worst case, the following grid has about 10 volts higher than it schould be. It adds cost, but i built only 1 amp and 40-50 euros is not an issue, compared to the total costs. Of course, if it's a commercial product, things change.

You don't need fancy tango or whatsoever anode chokes. Go to any transformer manufacturer and order about 2x4500 turns of 0,15mm wire on a dual EI84a bobbin (this is essential). Gap with 0,05mm per leg - enough for 15mA DC.
Even with the worst quality iron (max 1T) this will give you about 140 H with low exitation - 0,2-0,3Vrms and >160 for 5Vrms. Self capacitance is about 100pF (dual bobbin), that's 80Kohm / 20kHz, negligible for a valve with 3-4Kohm Ri.

I pay for a pair of these about 20-25 euros - depending on the mood of the winder. Tell the winder to put isolation every 4-5 layers, otherwise it will "sing". The small local business usually do not impregnate transformers :)

A nice side effect is that you now need to drop only about 25-30volts with your cath0de resistor of the EL34- 2watt less heat.

By the way: EL34 is overkill. A 5687-6N6 is enough and has more gain. The only problem i see, is that you need at least 60-70H pri. inductance in the IT, but with ca 8-10mA DC.


PS. The direct coupling with a choke as an anode load was described in Glass Audio. I can't recall the issue, but the title was something like "safe loftin-white".
The anode choke i described is exactly what i use with 5687 and the like valves (Ri about 3k, current 10-15mA) and the values of inductane and capacitance are measured, not estimated from the formulas.
I forgot to write that with the mod i proposed, you have to use a 300v supply instead of 400 for your EL34 to stay with the operating point of the schematic .
This doesn't mean that i think 270v/20mA is a good operating point for the tube- quite the opposite. But the design is compromised. A better op. point would be not practical - higher PS voltage, more heat dissipated from the cathode resistor, recalculated op. point of the first stage - possibly a worse one, etc.

Of cource, wenn i say that the EL34 is an overkill, i mean for your output tube, not for the 845 of the japanese circuit.

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.