Nice start. This is going to get really deep. I think that you need a lot more samples to accurately test a manufacturer. My guess is that a minimum of 30 tubes to get an accurate sample. Different run dates and factories make a huge difference on some. I know that some companies like Shuguang actually pick the best out at times and sell them as superior tubes.
Also, did you buy matched pairs/quads? That would mean the dealer already took out the garbage.
Finally, instead of giving the high and the low let's get down to real sampling: Use Standard Deviations. It's a much better number to judge by. Then if you get into variances you can compare different kinds of tubes not just different manufacturers...It will let us know things like the best overall consistency from a single manufacturer regardless of tube type.
Hi Scott - I applaud your efforts on the start of a very interesting and relevant project. I thought I'd just bring to the attention of interested readers an AES paper I saw a while ago on this very subject, by Merlin Blencowe of AMS-Neve:http://www.aes.org/e-lib/browse.cfm?elib=17071
This paper is focused more on the influences of tubes' operating conditions on noise performance, but deals tangentially with some of the stochastic variability of different samples. Also some excellent information on specific noise mechanisms, as well as measurement techniques.
Sorry that (like all AES publications) it's not free, but this is an excellent one on a subject that rarely covered anymore. Thorough bibliography, too.
Those are all good comments and suggestions. The current production tubes I tested were not pre-selected in any way. They were purchased in small lots from 3 different retail online tube vendors in the US.
Next time I update the list, I'll include standard deviations (either in addition to, or in place of min/max). I agree that is a better way to judge consistency of a manufacturer. And it is needed to calculate the statistical significance of the measured difference between tubes.
I've read Merlin Blencowe's paper, which is excellent. He also saw a lot of variation between tubes, but did not try to break it down by manufacturer, I believe. His paper has convinced me to test the D3a and similar Russian 6J52P, which I'll add to my list in the future.
We'll see how the list evolves as I test more tubes in the future - I'll post to this forum whenever I update the list.
This is a great start, Scott. Maybe the manufacturers will start paying attention to these results, once they are more statistically robust. I know I will, my active gain MC phono stage uses 12AX7s. Well done.
I don't get this. Noise from tubes is in my experience a variable related to sampling, and of course high gain tubes will tend to have more noise those that afford less gain, when used in a gain stage. Then too, the circuit design is critical and so are the voltage and current chosen to run the tube. These elements and more constitute a calculation of noise that has so many variables as to make the outcome unpredictable in advance. IMO, you have to measure noise in-circuit and experiment to determine whether substitution of one 12AX7 (for random example) with another 12AX7 results in audible or measurable improvement.
"High gain tubes will tend to have more noise than those that afford less gain"
Truer words were never spoken but the real question is how much noise is introduced by different types of the same tube (high or low gain).
Thanks, you have raised some important issues. It is not true that high gain tubes are necessarily noisier than low gain tubes. Tubes, like transistors, are characterized by their input-referred voltage noise (EIN) and input-referred current noise (IIN). The test circuits that I have used have a low input impedance, similar to the impedance of a moving magnet phonograph cartridge, and in such circuits only EIN is important. The high gain 12AX7 has a lower EIN than the low gain 12AU7 and therefore will have a higher signal-to-noise ratio, as an example. Since a lower gain tube amplifies both the signal and the noise less, sometimes people perceive the lower gain tube to have less noise, sort of like turning down the volume control. But the tube with lower EIN will actually produce a better signal-to-noise ratio, regardless of its gain.
The circuit does influence EIN to some extent, but in my test circuits, as in most low noise circuits, the circuit is designed such that the tube EIN dominates the noise performance. The test circuits I’ve used are designed to provide the optimum bias current for lowest tube EIN, and are also designed to be representative of the way the tubes are used in low-noise circuits. Therefore, my measurement results should reflect the way signal-to-noise ratios in audio equipment will be affected by the tubes under test.
Please note that these measurements only characterize the intrinsic noise performance of the tubes (hiss), not their susceptibility to “hum” when AC is used on the tube heaters. In most high-performance audio equipment, tube heaters use filtered (and sometimes regulated DC), so their susceptibility to hum when AC is used on the heaters is not a concern. But since many musical instrument amplifiers do still use AC on tube heaters, I do plan to characterize this “hum susceptibility” in the future – it will be a separate measurement and a separate table in my listings.
I hope this is helpful.
I updated the measurement list on my website.
1. I included a few more samples of Shuguang 12AX7B and Shuguang 12AT7.
2. I included new measurements made on the JJ EF86 (they call it EF806S). I've read in many places that pentodes are noisier than triodes, but I've never measured it before and was very curious to see how much. The EF86 measurements make it one of the higher noise tubes I measured, but not the highest. I also measured it as a triode, because I've read that a triode-connected pentode should produce less noise. At least in this case, it did not seem to be true. Within the accuracy of my measurements, it was the same. Hmmm......
3. Finally, I have begun to compute sample standard deviation (sigma) instead of posting the min & max for each type. Standard deviation is a better way to compare data sets of different sizes.
Since my last post here, I’ve updated the list on my website to include more tube types (such as the European D3a), and many more samples of the original tubes. The list at this point includes 95 samples and 15 different brands and types, along with some comments on the measurements and on specific tubes. And it includes the average noise, standard deviation, and number of samples of each type, so statistical significance of the measurements can be computed (if you are so inclined).
You can see the list on my “downloads” page at:
I’d be interested in hearing about other tubes or brands you think should be included.
Tavishdad, Thanks for pointing out the logical inconsistency in my generalization to the effect that "High gain tubes will tend to have more noise than those that afford less gain". After posting, I kind of wished I hadn't said that, for reasons you state; it's a matter of semantics, and I got the semantics wrong. But how about the other variable factors that to my mind also determine noise, at the practical level, which is to say when listening to music?
Thanks for posting your results, and keep at it, please. Have you tested ECC99?
Is the tube noise data significantly effected by the equipment used to measure it? Will your results pretty well match those of Roger Modjeski (RAM Tube Works) and Kevin Deal (Upscale Audio) if the same tube was to be tested by all three of you? Or is it more a matter of the relative noise levels of a group of tubes obtained by each tester?
Thanks, another good question. If I understand correctly, you are asking if the intrinsic noise (EIN) of a tube really makes any practical difference in your audio equipment. It will only matter if the tube is the first active device in a signal chain, that is, the first tube in a phono stage, the first tube in the analog section of a hybrid DAC, or the first tube in a line amplifier or power amp. In that case, the tube EIN should determine the signal-to-noise ratio of the equipment (unless the equipment is poorly designed and has, for instance, a lot of hum).
Whether the signal-to-noise ratio matters depends on your signal source. Let’s take a phono stage as an example. The best signal-to-noise ratio achievable on an LP is around 70dB, I believe (it is often much worse). A tube phono stage using one of the lowest noise tubes on my list should have an 80dB signal-to-noise ratio, the way such things are usually measured (A-weighted and referred to an input level of 5mV at 1kHz). But one of the noisier tubes on my list may give a signal-to-noise ratio 8 – 10 dB worse, comparable to the surface noise on a very quiet LP. You wouldn’t want that. So it could make a practical difference, yes.
I haven’t measured the ECC99, but I have some on hand and could do so. I had not considered the ECC99 as a preamp tube, but it might be worth measuring. Thanks.
Good question. The intrinsic, input-referred voltage noise of a tube (EIN) should be the same regardless of how it is measured. On my “downloads” page, I’ve provided a lot of technical details about how I made the measurements, so that anyone who wants to repeat them can do so. I’ve also compared my results to other tube noise measurements in the technical literature, which have shown comparable numbers. So I believe my measurements are correct and reproducible, and anyone who measures the same thing should get the same answer.
I visited the webpages for RAM Tube Works and Upscale Audio, and although I may have missed it, I did not see any technical details about how their noise measurements are made. I also did not see that they are posting the actual EIN or noise figure of the tubes they sell (again, I may have missed it). What they seem to be doing is selecting and grading the noise levels of tubes of a given brand and type, which is fine. Other tube vendors do the same. But since they do not specify the actual noise level in a technically precise way, you can’t make comparisons between the various tube types and brands.
It is an odd situation – no one would buy a low noise transistor unless the manufacturer or vendor specified the EIN or noise figure. But no manufacturer or vendor that I know of provides this data for tubes.
If these sites (or other sites) are actually providing measured tube noise data, please point me to it, I haven’t found it yet.
Tavishdad, I find the ECC99 to be very useful in a variety of applications, partly or mostly because it has the same pin-out as do the 12AX7/AT7/AU7, etc, family, whilst having a much lower Rp and higher Gm than any of those do. I use it as a CCS and as a follower in some cases in gear that was previously wired for a 12AT7 or 12AU7. However, in my phono stage, I use ECC99 as the top tube in a hybrid dual-differential cascode topology at the input to my phono stage, a very critical application, as you point out. Because of the very very high Gm of the bottom transistor, the gain at the input is enormous and the S/N is very good, enabling me to run just about any very low output MC cartridge with room to spare.
I'll put it on my to-do list. The only current-production ECC99 is JJ - is that what you use? What bias current do you run the ECC99 at?
JJ invented the tube and is the only maker, so far as I know. So, yes, that's the brand I use. I typically run them at 5-6mA per section, but I also feel that is not necessarily optimal; a little more current probably would be better. I have been constrained from using more current by the limitations of the rest of the circuit, in the 2-3 devices where I've implemented them.
I updated the measurement list on my website "Downloads" page to include additional samples of the JJ 5751, and also an additional tube type, the JJ ECC99.
The JJ 5751 continues to be one of the most consistent and low-noise tubes in my study. But I’ve noticed that as more tubes are measured, the average input-referred noise tends to rise. This is true for all tube types, and it is because as more tubes are measured, it is more likely to get a high noise tube that raises the average. The noise distribution is actually not Gaussian – while there seems to be a lower limit for the noise of a given type, there is no upper limit, so there is a tighter distribution on the low side of the average and a wider distribution on the high side of the average. In the future, I may begin reporting the median as well as the average for each type.
The ECC99 is a high transconductance tube that I believe was originally intended for RF applications. I’ve experimented with it as a driver tube for my hybrid amplifier, but had not considered it as a low noise tube until Lewm on this forum suggested it to me. The ECC99 has transconductance comparable to an ECC88, but without the extremely tight electrode spacing or wide bandwidth.
I’ve only measured 4 samples of the ECC99 so far, but it seems promising. The average input-referred noise of 0.711 µV RMS puts it in the middle of my list, and I intend to measure more. One of the 4 samples had “popcorn” noise and was discarded.
Scott, Re the ECC99, If one uses it as a driver, i.e., a cathode follower, would the noise be so important? Thanks for taking the trouble to both purchase and measure several of them.
I am using them in my very hi-gain phono stage as gain tubes; no obtrusive issues with noise, although I am sure there is noise at some level.
Intrinsic noise is not important if a tube is used at high signal levels. It is only important in the first stage of an amplification chain, where it can set the signal to noise ratio of the system.
You mentioned that you use the ECC99 in the first stage of a phono preamp, as the upper device in a hybrid cascode. In that location, intrinsic noise would definitely matter, although not as much as the lower device in the cascode. The ECC99 is a good performer, within a couple of dB of the best current production tubes I've measured.
I've updated the project page on my website again, to include the actual noise measurement histograms for the JJ 5751, JJ E88CC, and Tung Sol Reissue (Reflector) 6SL7GT. It is interesting to compare the three different plots - they tend to suggest that JJ is having some trouble getting consistent noise performance for their E88CC. The JJ 5751, on the other hand, shows fairly tight distribution and pretty consistent low noise performance.
The list of tube noise measurements on our “Downloads” page has had several updates since my last post here. It now includes measurements on the Toshiba 2SK209 JFET, which is an interesting comparison to the tubes and provides a sanity check on my measurement technique.
The list also now includes the Sovtek 6C45, which has the lowest measured noise of tubes on the list, but the samples tested were also microphonic.
Samples of the Sovtek 6H9C / 6SL7GT have shown very low intrinsic noise, comparable to the best 12AX7s tested, and 3dB lower than the Tung Sol Reissue 6SL7GT. I’ve only tested 3 samples of the 6H9C but intend to test more.
I’ve also recently tested the RU-6H6, a NOS Russian tube comparable to the JJ ECC99. And there are additional samples of many previously tested types, so the statistics are becoming more meaningful.