Flea-Power 6N1P - 6P14P SET Amplifier

1. The Story 2. The Details 3. Conclusion 4. References

1.   The Story

2.   The Details

From the angle of the plate curves, you can tell that this tube has a fairly high plate resistance (1700 - 2000 ohm) and therefore likes loads above 5k, ideally 6..10k. Obviously, higher load translates into lesser output power, and the trick is to optimize the biasing for 0...1 W RMS output, since this is my target power range for this project. Because the amp has a 8 ohm and a 16 ohm outputs, and since my line array speakers can be configured in various impedances, I was able to test the effects of various loads. The transformer results in ~ 4k load if speakers are connected as specified. If I connect a 12ohm speaker into the 8 ohm tap, I effectively get a 6k load, 16ohm gives 8k, and so on.

This data is for the input in a range 10mV...500mV, step is 50mV. Somewhere close to 1.5W the amp begins to clip due to increasing grid current on the output tube, but that's outside of our target range.

I use a mild form of harmonic cancellation here achieved by lowering of the plate voltage on the driver tube. This can be adjusted by R7. First, let's use a stronger form of cancellation by adjusting R7 to ~120K, resulting in 76V on C2. The effect is evident from the following two plots: the first one is voltage on the grid of the output tube and the second one is voltage on the load.

You can see on the first plot that the positive halves of Sin waves are slightly compressed (curves 7,8,9). The shape of this compression is as produced by a second harmonic. Further with amplitude increased, the upper points of Curve 10 are visibly clipped. This happens because the output tube grid starts eating current (the amplitude swing approaches grid-cathode bias) and is "destructive", whereas the compression introduced by the driver is "constructive". This "constructive" compression cancels some of the second harmonic distortion produced by the output tube. The negative halves on the second plot would have been visibly extended without driver compression, and among other effects, this would lower the WRMS at which clipping kicks in - the curve 9 would be clipped. The third harmonic slightly raises because of that but higher products stay low. This raise in higher products, and specifically in bad sounding odd harmonics is the reason anyone should be very careful with THD cancellation techniques, and is the reason I put the word "constructive" in double quotes. THD cancellation can be misleading if only THD is calculated and THD reduction is the target. Cancellation without understanding of the spectrum can lead to bad sound. I could have achieved greater cancellation by lowering the driver tube voltage even more - the "optimal" value for the schematic presented here is around 50V - but that would change the spectrum by trading lower second harmonic for an increase in higher products, and I would not want that.

Also, as evident from the next section, working in a vicinity of "optimized" region is nothing but an illusion. Not only achieving that is hard, but also it's likely that musical material might be colored by that. I feel I do not want higher voltage, either. For comparison purpose, on the right is a distortion plot for the first stage receiving 140V.

Another modification of the circuit I'm currently auditioning - a further simplification, in a sense - is removal of bypass caps in both stages. I know, this smells like a negative feedback. At the time of this experiment (2001) I'm somewhat reserved about this. Of course, it is only negative current feedback, and a feedback that is essentially instantaneous (not global, not local - inner to the tube - the kind of feedback that is very similar to the emission process itself. Nevertheless, I'm a believer that even this variant of feedback may damage clarity and magic of SET. I'm convinced that this feedback at least aggravates clipping behavior. And raises output impedance of each stage. And may cause ugly transient distortions. The latter is something I plan to study. For now, one comment regards yet one more frequently implied property of unbypassed cathode resistors - tradeoff between lower and higher harmonics. I did not find that it is the case, at least with the PSPICE models I use. Whatever simplifications and/or imperfection these models contain and which might worsen the sound in real listening tests should be offset by imperfections and sonic distortions of bypass capacitors. I'm currently auditioning a unbypassed version of my amp and so far the results are favorable. I'll post more detailed findings later. For now I present distortion data of the same schematic with cathode capacitors removed. The optimal B+ voltage for the driver stage in this configuration appears to be ~ 92V. This is for the value of R7 equal 100k as shown on the schematic. Of course, this will vary depending on the individual characteristics of the driver triode. Instead of R7C2, the first stage can be fed from a resistive divider and then C2. The divider can be 33k/33k. The latter has an advantage in that changes in the driver tube will not change the driver stage B+ too much:

Here is the data, for the input in a range 200mV...1200mV, step is 50mV.

Surprising. At one watt *both* the second and the third harmonic are lower. Higher order harmonics are not visibly higher. Therefore the mechanism of cathode-resistor feedback and THD cancellation give different spectral tradeoffs. The only evidence of something bad I can derive from the data is more aggressive clipping. Same is visible from Sine wave plots. This is "what is expected" but on the other hand knowing that this is grid-current-induced clipping, and knowing that the bias did not change, what'd we expect? The tube will run into the grid0leaking regime no matter what and in this scenario it happens slightly more aggressively. Here is more reveling plot of the unbypassed configuration, with distortion, in %, presented in Log scale. For comparison, I first give the 76V, bypassed data in the same Log-Y format, and then unbypassed 92V data:

3.   Conclusion

For a friend living in Ukraine, I designed a variant of 6n1p/6p14p topology which is closer in spirit to Decware Zen yet without its somewhat artificial soundstage widening caused by cathode resistor sharing, based entirely on ex-USSR parts and containing no electrolytic. This variant is described in [3].

4.   References

[2] High-efficient open baffle line arrays

[3] 6P14P amp using PIO caps