6AN8 Triode Section, Phase Splitter

There appear to be various names for the phase splitter topology I used; It is variously known, depending on which text you're reading at the time, as;

  • the Concertina phase splitter
  • the Cathodyne phase splitter
  • the Split Load Inverter

This type of phase splitter (and there are several alternative topologies), is cheep, easy to implement, and capable of good signal voltage balance between the two anti-phase outputs.

Amongst it's disadvantages are the absence of voltage gain, and limitations to signal voltage swing, and relatively high output impedance, with the potential to significantly misbehave should the subsequent output valve(s) be (over)driven into grid current.

For these reasons the stage is typically driven at relatively high current, and, if possible, as high a B+ as the particular valve used will tolerate.

6AN8 pinout and internal arrangement



An intermediary stage, between each of the two phase splitter anti-phase outputs, and the grids of the following output stage valves, is regularly seen in the literature; the Williamson circuit is a classic example. Such a stage is added to reduce the loading on the phase splitter itself, confer some additional voltage gain, and isolate the phase splitter from possible variations between anode and cathode circuit loads. I elected not to add such an additional stage, but to try to drive the phase splitter reasonably hard (that is, with a high quiescent current), instead.

I have read discussion on the apparent difference in source impedance between anode and cathode outputs. Semantics, definitions of what is actually being considered and meant, and considerable mathematics, all seem to be to the fore in such discussion (and, at times, not a little passion and vitriol too!).

There does appear to be an apparent paradox. The cathode output, having lower apparent source impedance than that at the anode, may intuitively be expected to have better high frequency response, for a given amount of capacitive loading (by the subsequent stage). The implication is that this would give dissimilar phase and frequency performances between the two phase splitter outputs, and hence distortion.

However, with the critical proviso that the loads are at all times equal, the output voltages and frequency responses are at all times equal. The apparent paradox is resolved by considering both the source impedances and the effect within the valve of each load impedance on the voltage of the opposite terminal. For the special and particular case of both anode and cathode loads being equal, these two effects are self compensating for each other

A notable article on this is Notes on the Cathodyne Phase-Splitter, Albert Preisman, Audio, April 1960, pp22~23. At the time of writing my link to the on-line source of this article had evaporated, but I'm subsequently (3/03) helpfully advised by fellow enthusiast, Mr. Ray Moth, that it is again available here.

From time to time I see circuits where the anode and cathode resistors have been deliberately made slightly unequal. I presume this is a supposed design countermeasure to the supposed dissimilar performance characteristics at anode and cathode. If so, then said countermeasure is in fact the problem rather than the cure!

To maximise signal voltage swing at each of the two anti-phase outputs, the circuit DC condition is designed to set 1/4 of the B+ at the phase splitter's cathode, and 3/4 of the B+ at the anode. This, and the desirability of running as high a quiescent current as continuously tolerable, pretty well fix the design. The cathode resistor is sometimes seen split, in capacitivly coupled circuits, to give a cathode resistor portion (to which the grid resistor is returned), of suitable value for self-bias. In direct-coupled application, as here, neither grid nor cathode self-bias resistor is required.

At near or partial overload, there exists readily apparent asymmetry in the circuit. Significant distortion is likely, predominantly even-order in nature. The clip is approached slowly and progressively, so the overload distortion characteristic is low-order rather than high-order, harmonic content dominated. This is perhaps one of the reasons for favouring other types of phase splitter(!), but I feel that there is an alternate view. I suggest that overload characteristic is amongst the factors that give an amplifier its sonic character. I suggest that even during listening levels well within an amplifier's capabilities for all intents and purposes, brief overloads or near-overloads do, and regularly, occur, without is being outright aware of them as singular events. I hold that such, albeit brief, events contribute to the sound-clues that yield overall "sonic character" or sonic signature".

Probably the simplest type of RC phase inverter is made by simply putting half of the plate load in the cathode circuit .. . . It does not matter where the resistance in series with a tube is, so long as it has the same effect in restricting the plate-to-cathode voltage according to the current the tube draws. A load-line for this method of operation can be drawn in exactly the same manner as for a normal voltage-amplifier stage, using a resistance value made up of the two resistances in series.

Designing Your Own Amplifier, Part 3: Phase Inverters, Norman H. Crowhurst, originally appearing in Audiocraft, May 1956, and reproduced in Glass Audio 2/00 pp1~5

I further suggest that it is preferable for that sonic character to be dominated by even-order in preference to odd-order harmonic content, although both may well be present in the sonic signature. Now in practice it is likely that one stage of an amplifier will reach its limits before all others; rather than all "hitting the rail" simultaneously at some point, in response to a musical peak. That stage that overloads first is likely to dominate the overall sonic character. It is therefore better that that stage be single ended, rather than push-pull or differential. In conclusion, I suggest that it ain't necessarily a bad thing that the concertina phase splitter is B+ hungry and therefore likely to run into overload problems first, and possesses an asymmetric overload characteristic. I suspect that this stage may well confer the dominant sonic signature to my push-pull amp.

6AN8 triode Va vs. Ia curves with loadlines plotted.  Click here for full size image.

Triode Curves

Complete amplifier schematic; click here for full size image.

Amplifier Schematic

Originally I simply used 22 kOhm resistors in cathode and anode circuits - all of the published designs I had seen stuck to these values. I hadn't seen any examples of heavier, higher current, loading than this. Later inspection of the triode characteristic curves suggested that the triode could be run quite a bit harder, and I changed the cathode and anode resistors to 10 kOhms accordingly. That places the loadline on a more linear region of the curves.

As in the explanation above, the treatment of this circuit is by considering the two resistors as one, being the sum of the two, for DC loadline purposes - here 44 kOhm reduced down to 20 kOhm.  Now this is a "DC" loadline - for the AC loadline the grid resistors of the subsequent stage are also first summed, and then the sum is considered as being in parallel with the DC loadline resistance.