L4 EL34 Power Amp Design and Circuit Description Support Information from Andy Grove

First off a brief overview of the circuitry: Actually it's quite a simple schematic and simplicity usually means good sound as long as you don't have to make too many compromises to achieve it.

Circuit Description

The input and driver stages use the triode/pentode type ECF80 (6BL8), I've used this valve a fair few times over the years and it's always given me good results. Its electrical characteristics make it very suitable for this application, it's not too scarce or sought after to push NOS prices through the roof and, of course very importantly, it sounds great. The pentode section is configured as a voltage amp and all the driver gain comes from this one stage. The setup is fairly straightforward with R1 as the grid resistor, R2 as a grid stopper to help prevent parasitic oscillation (the ECF80 pentode is a fairly high transconductance RF type) and R3 is the screen dropper resistor. The anode load is made up from two resistors R4a and R4b, this is done to reduce the voltage across the resistors. This is particularly important if you intend to try carbon film or (especially) composition resistors by the way. R5 and R6 are the cathode resistors, most of the bias voltage is formed across R5 and this is bypassed with the capacitor C2 and C1 is the screen bypass capacitor which is returned to the cathode rather than to ground (sometimes seen elsewhere) to isolate the recirculating screen current. Feedback from the transformer secondary is fed, in the usual manner, to the cathode of the input valve via R16 and the compensation component C7. There is also a small amount of HF feedback taken from the transformer primary via R14 and the two 47pF capacitors in series (in series for voltage rating) to get 23.5pF.

The next stage is a "concertina" phase splitter. This type uses a single triode and has equal anode and cathode resistors R6 and R7. Basically the approximate gain from grid to cathode is +1 and from the grid to the anode is -1 or, it's in phase at the cathode and 180 degrees out of phase at the anode and with a magnitude of gain of approximately 1. The magnitude of gain being around one (actually slightly less) means that the input impedance of the stage is very high - even at high frequencies due to minimal Miller effect. That, in turn, means the HF bandwidth is also pretty good and remember we've only got one stage before - the voltage amp stage - so the overall phase shift at HF is also controlled and therefore so is stability more easily controlled. From the PS grid to ground is a step network R9 and C6. This reduces, or rather shelves, gain in a controlled manner at HF and forms part of the amp's HF compensation scheme.

The push pull output of the phase splitter is fed via C4 and C5 and the grid resistors R10 and R11 and 10k grid stoppers to two EL34 power pentodes. These things are great because they can kick out a lot of power at sane voltages and, if set up right, sound totally groovy. Their operating conditions are more or less according to the Mullard* datasheet, I've not implemented anything "clever" just for the sake of it here.

Experimentation

I have tried various setups for the EL34 with different voltages and transformer impedances, split supplies for the screen and anodes, ultra linear operation and various combinations and permutations of those and more but this setup really is just about the best of the bunch without going on a trip into the land of techno la la - and that applies to both HiFi and guitar amps. It's one of those things where to make the thing 10% better you need to go to great lengths and often in doing so something is lost in the process.

Dynaco Ultra Linear Mode vs L4 Pentode

The schematic of this amp bears a resemblance of the old Dyna which was a great amp. The input and phase splitter follow the same pattern, and for much the same reasons but I chose a pure pentode output stage over the Ultra Linear used in the Dyna. Apart from the fact that a pair of EL34s set up as I've done here possesses a pure groove factor which you don't get with the Ultra Linear output stage there are technical reasons to consider too which I'll outline:

Ultra Linear Issues

The Ultra Linear connection attempts to make a pentode behave more like a triode and, to a certain extent, it does. But it brings with it some of the "problems" triodes have too, such as requiring a lot more voltage to drive them and now, because the screen grid has signal on it, we suddenly have some nonlinear input capacitance to deal with.

The nice, simple concertina phase splitter works well when presented by constant, linear loads, when the output stage (or whatever stage it's driving) also remains pretty well constant over a full signal cycle. In addition the concertina doesn't like to supply a lot of signal voltage - it's like two stacked stages and as such it can't supply a great deal of voltage compared to its HT supply. Or to look at it another way, the signal it supplies is split between two outputs so you can only get half the swing at each.

Class AB1 – EL34

To get power from the output valves they are run in Class AB1. That means that at or near full power one of the output valves is turned completely off. When the output valve is a pentode not much happens to it's grid at that point as it's screened from the anode by the screen grid (hence its name) and the main components of the load the driver stage sees remain fairly constant. If the valve is a triode or partial triode then a fair part (it could be the majority of) the load the driver stage sees is created by the Miller effect. Of course when the valve cuts off its anode continues to swing because the transformer is being driven by the other valve of the pair but as only one valve is now conducting the output stage gain changes and so does the driver's load - in an abrupt and nonlinear fashion. The unloaded section of the output transformer will also most likely ring and this too will be coupled into the output valve's grid.

This situation is further complicated by the fact that there is a coupling mechanism between the lower (in phase) output of the concertina and the upper output. If you look you can see a grounded grid amplifier in the making; put a signal in at the cathode and you'll see a version of it at the anode. This means that the valve connected to the lower output can talk to the upper one....and now, with UL, we have a coupling system via the OPTX and the screen grids and we've created a very nice opportunity for incredibly annoying, and sometimes intractable, parasitic oscillation. This kind of thing might not present itself with a pair of 300Bs but it will with a pair of high transconductance wide bandwidth EL34s.

Williamson used a differential type stage between his concertina and the output valves and this works to both isolate the concertina and to provide more voltage swing to drive the output valves but but then you bump into (especially) LF stability problems. There are more complex schemes which will work too but...

Custom Design Output transformer

Finally, one of the big decision makers was the output transformer. There are certain, simple, transformer topologies which just work. A good friend of mine once said to me, when he saw a half completed transformer drawing, "You've designed that same transformer over and again for the last twenty years!" and to a certain extent he's right! The reason why is because that layout works and sounds good. Ultra Linear operation disallows the best of these simple topologies because you have to consider the screen taps and the coupling to them and very quickly things start to get messy and/or complex. On the subject of the L4 output transformer it's designed for full output (35W) at 20Hz.

Cathode Biased vs Fixed Biased

There are a couple more things about the EL34 output stage "datasheet" setup. Firstly the L4 is cathode biased. There's a fair amount of blurb around about this and fixed bias. Of course fixed bias can give more power and lower distortion at high powers but, depending upon the conditions, that can vary from hardly anything to quite considerable. If the valves are set somewhere near Class A then the difference is small and something which isn't often considered is that the difference in power output and distortion disappears for dynamic signals. Fixed bias only really rules when it comes to full, continuous full power output in Class B. Cathode bias wins when it comes to not having to monitor your amp every day to check the output valve bias (especially to avoid offset currents from saturating the OPTX) and for not totally cutting the output valves off due to grid rectification when the amp clips. I once tested a well known valve amp and just could not believe what I saw when it was driven to clip: Err guys, the waveform totally collapses as soon as you go one zillionth of a dB past clip and because you used enormous coupling caps it takes a minute and a half to recover. Finally, of course, you don't need an extra negative supply with cathode bias.

Cathode Bias Scheme

There are two versions of the cathode bias scheme in the datasheets. One, Mullard, specifies two 260R resistors, and Philips and Telefunken specify one 130R resistor. Now, I've tried both and I have to say that the single resistor, I think, very slightly beats the two resistor approach for sound. The problem is that now you're back to square one with offset currents and you will lose a bit of power. Also if one of the EL34s croaks then the anode of the other will glow red hot (been there and have the T shirt).

Common Screen Resistor

Something else which is a little different with this "datasheet" output stage setup is the common screen resistor. This is pretty neat and gives you a more or less free lunch, or maybe it's just a very cheap lunch. Essentially what happens is that when one valve is drawing more current (over a signal cycle) it pulls more from the screen which in turn causes a drop in the screen voltage due to the presence of the screen resistor. This creates a feedforward effect by reducing the screen voltage to the partner valve and helps it to cut off. There is a slight penalty in overall power output but with experimentation an ideal resistor value can be found and the Mullard* engineers did just that.

Power Supply

The power supply is quite straightforward with a CLC sequence after a set of silicon diodes.

What is different is that the anodes are fed from the raw supply straight after the rectifiers and the screens are fed from a point after the choke.

The anodes of a pentode are pretty high impedance and so aren't so affected by supply ripple when transformer loaded, especially in push pull. The screen grids and the more sensitive stages instead are filtered more heavily. This requires a smaller choke but also means the hungry power section is connected right to the reservoir cap and the rest of the circuitry is isolated from that section which makes for a more solid bass without resorting to complex voltage stabilisation schemes.

A Final Note

Valves sound like they do not just because they are valves, but also because they are, or rather can be, used in a certain way; a vague analogue would be nature and nurture. Select valves which look good on paper and then put them together in "complex and interesting ways" and pretty soon you're on the road to hell.

Brian and Co. have done a great job in turning my mad prototype into a buildable kit, I hope that that you enjoy building and listening to it.

Andy Grove - Audio Note design engineer