I’ve made no secret of my high regard for the qualities of the 6SN7 family of twin triodes when properly used as voltage amplifiers. My experience was affirmed by an interesting set of measurements presented by Morgan Jones in “Valve Amplifiers.” Using a basic pentode-loaded mu follower topology biased by LEDs in the cathode circuit, Jones slogged his way through hundreds of distortion measurements of various species within the 6SN7 genus. A really useful piece of work except for one thing… he did not include the 5692, a tank-like construction presumably designed for use in the gadget that verified the ratio of Tang to water in NASA’s Gemini capsules. And that’s a serious omission for American tube enthusiasts (and some non-American ones, too, most notably Frank De Grove). I would have remedied that lack myself except that my own capabilities to do spectral analysis were limited by the HP wave analyzer on hand (reliable to perhaps -75dB or so), whereas Jones presented data down to -97dB, and third harmonic and higher products for 6SN7s tended to fall below -80dB.
Well, “diy” does mean “do-it-yourself,” so I managed to cobble together a nice PC-based system capable of reaching -100dB reliably. A bit of crude jig construction and I can duplicate Jones’s work. Except that I’m really quite a bit lazier than Jones, so there are some modifications to his procedure which, as will be seen, don’t totally obviate comparisons between his results and mine.
The test circuit used is shown here:
Note that I have retained Jones’s use of LEDs to provide cathode bias. Note also that I have chosen not to use his mu follower plate load, opting instead for a solid-state cascode current source adjustable to the 8.0mA current used in the previous work. The CCS also makes use of LEDs, this time as a series-paired reference voltage source of 3.5V (2 * 1.75V). The circuit is repeated for both triode sections. With a total of 8 bright red LEDs cheerfully glowing away during the test, my garage lab takes on a festive tone appropriate for the holiday season. The use of those horrid bits of bipolar silicon to form the CCS was rationalized by considering the need for separate heater supplies for the DUT and pentode, and the inconvenience and size associated with building the pentode loads. After all, this is a TEST jig, confined to my lab, and with no need to be allowed to foul the carpet of my living room.
The parts used for the current source were carefully optimized through a resort to first principles: what parts are rattling around in the coffee cans I use as a junk box? With the engineering priorities thus set, the output resistance of the CCS is quite reasonably high: to a good approximation, Rout = hfe1*hfe2*Re. In this case, Re = (3.5 – 0.6)V/8.0mA, which is about 350ohm, the product of the hfes is about 6000, so Rout is roughly 2Mohm, safely three orders of magnitude greater that the plate resistance of the DUT. The current sources were built, then 20Kohm dummy loads were attached to verify operation and facilitate adjustment to the 8.0mA test current before attaching the CCS to the tube. For the chosen current, the plate voltages settled in at about 155-160V for all tubes tested.
Instrumentation is the essence of simplicity and is based on an Audigy 2Zs 24/96 sound card. The one hardware tweak involves levels and loading. Frustratingly, Audigy does not specify an input voltage range nor do they bother disclosing any information regarding input impedance. Nonetheless, it is safe to assume that the input impedance is too low to allow direct connection to the output of the test jig without severely perturbing the test. Likewise, it is safe to assume that the output level used by Jones (+28dBU or 19.8VRMS) will well exceed the input range of the Audigy. In order to circumvent these two issues, an interface box was constructed comprising a BUF-03 high speed buffer (the bandwidth, input impedance, and distortion safely orders of magnitude better than needed for this measurement) and a BNC input with a 1Mohm resistor and 20pF capacitor strapped across it. Attached to this input is a 10x probe borrowed from my oscilloscope- the input network for the buffer simulates the ‘scope input, allowing the probe’s compensation to work properly. Though the probe does alter the loading a bit (10Mohm, 10pF), such loading is pretty negligible compared with the much lower source resistance of the tube.
There are quite a few nice software options, and my major criterion, as always, was minimal cost. Preferably free. In that impecunious spirit, I tapped the generous resources of Dazyweb Labs. Their SG2102a signal generator and SA3000 spectrum analyzer teamed with the inexpensive ($US95) sound card to provide a startlingly good instrument. I still haven’t figured out how to manipulate some display functions, but the functionality is irreproachable. When the interface box was inserted into the loop, the baseline was below -130 dB using optimized level settings and signal averaging (16 samples per measurement).
Though probably simple to more computer-literate folks, I spent quite a time diddling with various faders and gains to optimize the performance to this level. A calibrated analog oscilloscope was used to verify analog waveforms and levels. Putting the whole test setup in the loop and using a 1kHz sine wave as excitation, the baseline 2nd harmonic was at -116dB, 3rd harmonic at -109dB. Not too shabby performance by the test gear.
All right, then, let’s get to measuring. I had on hand nine 5692s, consisting of 4 red base RCA, two red base GE, and three brown base CBS. Also at hand was a single 6SN7GTB of rather late manufacture, branded RCA. This version had a shortened octal base, reminiscent of the old 7591s.
The average distortion results are as follows, presented in dB below the fundamental (standard deviation in parentheses):
Tube: sections tested 2nd 3rd 4th 5th (dB below +28dBU)
RCA red base 8 -55(1.3) -89(1.8) -99(2.4) -110(3.3)
GE red base 4 -52 -86.4 -93.4 -111
6SN7GTB 2 -52/-56 -93/-51 -91/-69 -105/-53
CBS brown base 6 -54(1.7) -87(2.6) -103(1.7) -105(4)
The RCA red base 5692 sections were quite consistent. The GE versions both showed one better-than-average section and one worse-than-average section (in one case, the worst 2nd HD of any section on test here), but with only two tubes, it is impossible to draw any conclusion. And they still weren’t THAT far off the median. By contrast, the sections of the 6SN7GTB were so wildly different, I present both of them rather than averaging. Section 1 was quite good; section 2 showed slightly higher THD, but a horrific distribution of harmonics, bad enough to warrant showing here:
As a comparison, Jones showed an average for Russian 6J5s (his choice for a driver tube for the Crystal Palace amplifier) of -52dB for 2nd, -90dB for 3rd, and -96dB for 4th. His distortion champion was a carbonized-envelope version of 6SN7GTB, which boasted -54dB 2nd, -94dB 3rd, and 4th also below -100dB.
So, it’s gratifying to see that these measurements are fairly close to published data and that, indeed, the 5692 stacks up quite well to the best of what English scroungers can dig up. Add the heroic construction, low level of microphonics, and long lifetime, and it’s easy to see why Frank chose it for his Ultimate Line Stage.
My thanks to the Dazyweb people for providing such useful software, Ace_3000 for some great advice on soundcards and measurement, and Morgan Jones, who was kind enough to respond to my inquiries and provide lots of insight into the practicalities of the measurements. Any mistakes are mine, not theirs.