I've started to post and maintain a list of noise measurements made on a variety of current production (and a few NOS) audio tubes, because I don't find the lists at the tube vendors very helpful. My list shows the actual input-referred noise of each type/brand and how many tubes of each type I've measured. So far the list is modest, with 7 different types or brands (a total of 38 tubes measured). The lowest noise tubes so far include the JJ 5751 and Shuguang 12AX7B. You can see whole list at
Downloads – Tavish Design
Can anyone suggest other (possibly) lower noise types or brands? I can't afford to buy tubes just to test them(!) - but I do want to identify the lowest noise types, so please let me know your experiences.
Scott
Downloads – Tavish Design
Can anyone suggest other (possibly) lower noise types or brands? I can't afford to buy tubes just to test them(!) - but I do want to identify the lowest noise types, so please let me know your experiences.
Scott
I'm a mainly headphones guy, and I've used a number of different tubes. I find that the circuit plays the biggest part by far. The biggest difference I find is how good the heater-cathode insulation is.
It is an interesting question anyway.
It is an interesting question anyway.
Something else to consider, a public list of all the "best" tubes makes it easy for an interested manufacturer to start making a market of those tubes by buying them up. It seems that there is some first hand knowledge of this running around here.
May be best kept as a lower profile item? Just saying it could be an alternative.
May be best kept as a lower profile item? Just saying it could be an alternative.
Nice list! The EF86 is not very good noise-wise. Triode-connected high transconductance pentodes such as the D3a and the Russian 6J52P are much better.
Are you familiar with the work of Merlin Blencowe (Merlinb on this forum)? He wrote a most interesting AES paper on it:
Merlin Blencowe, "Noise in triodes with particular reference to phono preamplifiers", Journal of the Audio Engineering Society, vol. 61, no. 11, November 2013, pages 911...916. See also http://www.diyaudio.com/forums/tube...r-noise-dominates-triode-noise-audio-aes.html
The input-referred noise is quite dependent on the bias point and the optimum can only be found experimentally (or by using Merlin Blencowe's data), because 1/f noise tends to get worse with increasing current while white noise gets better. Did you optimise your bias currents?
The optimum also depends on the weighting. Hence, the optimum for an RIAA amplifier will be different from that of a flat amplifier.
By the way, 390 ohm is not really a good model for the thermal noise of a moving-magnet cartridge. The ESR of a real cartridge increases a lot with frequency.
Are you familiar with the work of Merlin Blencowe (Merlinb on this forum)? He wrote a most interesting AES paper on it:
Merlin Blencowe, "Noise in triodes with particular reference to phono preamplifiers", Journal of the Audio Engineering Society, vol. 61, no. 11, November 2013, pages 911...916. See also http://www.diyaudio.com/forums/tube...r-noise-dominates-triode-noise-audio-aes.html
The input-referred noise is quite dependent on the bias point and the optimum can only be found experimentally (or by using Merlin Blencowe's data), because 1/f noise tends to get worse with increasing current while white noise gets better. Did you optimise your bias currents?
The optimum also depends on the weighting. Hence, the optimum for an RIAA amplifier will be different from that of a flat amplifier.
By the way, 390 ohm is not really a good model for the thermal noise of a moving-magnet cartridge. The ESR of a real cartridge increases a lot with frequency.
Last edited:
I have a spreadsheet for calculation of cartridge noise, taking into account the loading. A quick run-through with real-life cartridge and loading parameters will confirm that a simple small resistor is not at all an accurate model.
Cartridge Noise Calculator
Also... @Tavish, the input capacitance spec on your phono preamp is... unlikely. You might want to actually measure it. I'll bet the minimum C is at least 100pF
Thanks for all the replies, especially the reference to Merlin Blencowe's paper, which I had not seen (but just read).
I have indeed measured the dependence of input-referred noise on bias current, which was what led me to the relatively high (1.5mA) bias point for the 12AX7/ECC83. But I found that the bias current dependence of input noise was swamped by tube-to-tube variation. Figure 2 in Merlin Blencowe's paper clearly shows this big tube-to-tube variation also. So I focused on measuring the noise of various tube types and brands.
I'm aware that the 390 ohm resistor isn't an accurate model for a phono stage, and I did not really intend it to be. I had to choose some value of input resistance to calculate noise figures, and 390 ohms seemed reasonable. The relative noise figures between the tube types would stay about the same with a different source impedance.
I've also heard that the EF86 is a relatively high noise tube, but we'll soon see. So far, the noise of the tubes I've measured seems to depend more on the manufacturer than the tube type (that is, Shuguang vs. Reflector vs JJ). Are tubes currently being manufactured any other places?? I only seem to see tubes from those 3.
I'll update the assembly manual for the phono stage to read "no additional capacitance", or some such thing. The Miller capacitance of the input tube would be on the order of 50-80pF, but I'll measure it and put that number in the spec someplace. Thanks.
Scott
I have indeed measured the dependence of input-referred noise on bias current, which was what led me to the relatively high (1.5mA) bias point for the 12AX7/ECC83. But I found that the bias current dependence of input noise was swamped by tube-to-tube variation. Figure 2 in Merlin Blencowe's paper clearly shows this big tube-to-tube variation also. So I focused on measuring the noise of various tube types and brands.
I'm aware that the 390 ohm resistor isn't an accurate model for a phono stage, and I did not really intend it to be. I had to choose some value of input resistance to calculate noise figures, and 390 ohms seemed reasonable. The relative noise figures between the tube types would stay about the same with a different source impedance.
I've also heard that the EF86 is a relatively high noise tube, but we'll soon see. So far, the noise of the tubes I've measured seems to depend more on the manufacturer than the tube type (that is, Shuguang vs. Reflector vs JJ). Are tubes currently being manufactured any other places?? I only seem to see tubes from those 3.
I'll update the assembly manual for the phono stage to read "no additional capacitance", or some such thing. The Miller capacitance of the input tube would be on the order of 50-80pF, but I'll measure it and put that number in the spec someplace. Thanks.
Scott
Nice piece of work. Can you describe how he rms figure is determined using the pico scope?
Is the 319 ohm resistor in circuit when you measure the noise.
Cheers
Ian
Is the 319 ohm resistor in circuit when you measure the noise.
Cheers
Ian
The Picoscope has a measurements function that computes the AC RMS value of the input. I used a 20-point moving average to stabilize the readings. I checked the answer against an Agilent (Keysight, now, I guess) spectrum analyzer at my place of employment, so I believe the Picoscope is giving the correct answer for white noise.
The 320 ohm resistor is in circuit for the measurement, so 0.227 uV RMS is the minimum input-referred noise that could be measured for the circuit, which would be a 0dB NF.
The 320 ohm resistor is in circuit for the measurement, so 0.227 uV RMS is the minimum input-referred noise that could be measured for the circuit, which would be a 0dB NF.
Thanks for thr info. Can you confirm the 20 point moving average is AFTER the rms measurement?
Not sure how you arrived at the noise voltage for the 390 ohm resistor - what bandwidth have you assumed? In a 20KHz bandwidth it will be much bigger than the values you gave.
The noise voltages in the table of measurements - are they with the noise voltage of the 390 ohms subtracted?
Cheers
Ian
Good stuff!
A good lab dvm could also be used to cross-check rms voltage output measurement, once you know there are no other parasitic signal sources entering.
Can you also indicate the range of NF or IRN measured for each tube type, and not just the average?
Are you able to put a few NOS samples in an operating amp (for example), and then remeasure after a few practical periods (eg. a day, a week) ?
What does your noise floor measure (eg. no tube inserted)?
A good lab dvm could also be used to cross-check rms voltage output measurement, once you know there are no other parasitic signal sources entering.
Can you also indicate the range of NF or IRN measured for each tube type, and not just the average?
Are you able to put a few NOS samples in an operating amp (for example), and then remeasure after a few practical periods (eg. a day, a week) ?
What does your noise floor measure (eg. no tube inserted)?
Some double triode tubes have a quieter triode.
If I remember correctly the 12ax7 lower pin numbered triode is the quieter one.
I always use that one first in my amplification sequence.
I have found most noise comes from bad layout. I get away with AC heaters in everything I have built.
If I remember correctly the 12ax7 lower pin numbered triode is the quieter one.
I always use that one first in my amplification sequence.
I have found most noise comes from bad layout. I get away with AC heaters in everything I have built.
I found that the new production Svetlana AX7's I plugged in to replace EH's were significantly less noisy.
I'd suggest that was hum related only - and not related to intrinsic noise being measured by Scott.
P.S. Just checked 5.5.1960 Philips data sheet and pins 6/7/8 provide the lower hum performance.
Last edited:
All really good questions & comments!! Thanks.
Yes, the Picoscope first computes the AC RMS value of the noise, then does a moving average. The Picoscope software takes care of the math, and I’m fairly certain it is done correctly.
I’ve applied C-weighting to the noise, as I explained on the website & summary page. So the bandwidth is 30 Hz – 8kHz. So En^2 = 4kTR and En = 2.54 nV/rtHz for the 390 ohm resistor. Integrated over an 8kHz bandwidth I get 0.227 uV RMS.
The noise of the 390 ohm resistor is not subtracted from the input-referred noise. Remember that when dealing with noise, you have to add or subtract power, not voltage. So if we did subtract it, we would calculate it as SQRT[(.565)^2 – (.227)^2] = 0.517 uV RMS for the typical Shuguang 12AX7B. But it is customary when dealing with noise calculations not to subtract the noise of the source resistance, but instead to compute a Noise Figure referred to a particular value of source resistance, as I describe in the summary. If all the noise was due to the 390 ohm source resistance, the tube would have a 0 dB noise figure.
The noise floor of my setup is approximately 3 – 4 uV RMS at the output of the test circuit (with the tube removed), which is mainly due to the tube load resistor (either 100k or 41k ohms) and the input noise of the 100X noise measurement amplifier. I just double checked that in the lab. Referred back to the input of the tube, that is below the noise of the 390 resistor (although if I had a really low noise & low gain tube, it might start to affect the measurement).
Adding the high and low noise measurement (or maybe standard deviation) for each tube type is a good idea. I’ll revise the table to include those, maybe this weekend, and post to the forum when I’ve updated my website. And I’ll also try burning in 1 or 2 of the NOS tubes to see if they improve.
Finally, I agree with several folks who point out that HUM is usually a bigger practical problem in tube circuits than intrinsic noise (or HISS). But HUM can be eliminated with good circuit design and layout (which is often a lot of work). But after you’ve worked so hard to eliminate the HUM, you’re left with the HISS, and it is definitely worth the effort to select the tube with the lowest HISS.
Scott
Yes, the Picoscope first computes the AC RMS value of the noise, then does a moving average. The Picoscope software takes care of the math, and I’m fairly certain it is done correctly.
I’ve applied C-weighting to the noise, as I explained on the website & summary page. So the bandwidth is 30 Hz – 8kHz. So En^2 = 4kTR and En = 2.54 nV/rtHz for the 390 ohm resistor. Integrated over an 8kHz bandwidth I get 0.227 uV RMS.
The noise of the 390 ohm resistor is not subtracted from the input-referred noise. Remember that when dealing with noise, you have to add or subtract power, not voltage. So if we did subtract it, we would calculate it as SQRT[(.565)^2 – (.227)^2] = 0.517 uV RMS for the typical Shuguang 12AX7B. But it is customary when dealing with noise calculations not to subtract the noise of the source resistance, but instead to compute a Noise Figure referred to a particular value of source resistance, as I describe in the summary. If all the noise was due to the 390 ohm source resistance, the tube would have a 0 dB noise figure.
The noise floor of my setup is approximately 3 – 4 uV RMS at the output of the test circuit (with the tube removed), which is mainly due to the tube load resistor (either 100k or 41k ohms) and the input noise of the 100X noise measurement amplifier. I just double checked that in the lab. Referred back to the input of the tube, that is below the noise of the 390 resistor (although if I had a really low noise & low gain tube, it might start to affect the measurement).
Adding the high and low noise measurement (or maybe standard deviation) for each tube type is a good idea. I’ll revise the table to include those, maybe this weekend, and post to the forum when I’ve updated my website. And I’ll also try burning in 1 or 2 of the NOS tubes to see if they improve.
Finally, I agree with several folks who point out that HUM is usually a bigger practical problem in tube circuits than intrinsic noise (or HISS). But HUM can be eliminated with good circuit design and layout (which is often a lot of work). But after you’ve worked so hard to eliminate the HUM, you’re left with the HISS, and it is definitely worth the effort to select the tube with the lowest HISS.
Scott
As the basic idea is to make measurements of noise, why not use the ITU-R 468 noise weighting which is specifically designed for this purpose?
Cheers
Ian
Cheers
Ian
That is true mainly for radio electronics. For audio it is much more useful to know the true EIN of the amplifying device alone, since this does not tie you to any specific source resistance or configuration (input current noise is negligible under these circumstances). Pure unweighted (flat, defined bandwidth) noise measurements are by far the most useful as they do not hide or second-guess anything about how the device will be used. ITU and A-weighting should certainly be avoided as they do practically nothing other than make the numbers worse; they are better suited to measurements of acoustic noise, especially when there is significant high frequency content, rather than white/pink noise.
Although I measured both A-weighted noise and C-weighted noise, I only posted the flat C-weighted results. The relative tube rankings were the same either way. I posted the flat C-weighted results because I thought they were more generally useful, if someone wanted to try to repeat my measurements or use them in a calculation or comparison.
I confess to being a radio engineer.......... But the input-referred noise I measured is very close to the true EIN of the tube, integrated over the 8kHz BW. And anyone who wants to subtract the contribution of the 390 ohm resistor can do it. The NF calculation gives an idea of how much the SNR of your amp will change with the different tubes, which I thought was an interesting comparison.
Scott
I confess to being a radio engineer.......... But the input-referred noise I measured is very close to the true EIN of the tube, integrated over the 8kHz BW. And anyone who wants to subtract the contribution of the 390 ohm resistor can do it. The NF calculation gives an idea of how much the SNR of your amp will change with the different tubes, which I thought was an interesting comparison.
Scott
