'A'weighting tends to improve noise measurements which is why it is so loved by marketing men. ITU weighting is specifically designed to represent the ear's response to noise. It never was intended for acoustic measurements - that's what A and C were intended for.
Cheers
Ian
Yes, sorry, A weighting makes the figures marginally better. But either way they have a more or less uniform scaling effect on white/pink noise figures, which makes them pointless. Only when there is significant HF or burst noise do you get any useful reflection of the way noise sounds to the ear. That sort of background noise doesn't often occur in electronic circuits, it's more likely in the acoustic world around us (drills, cars, factories etc).
This doesn't make any sense to me. A-weighting and ITU-R 468 weighting both suppress the low frequencies compared to the midrange frequencies and therefore 1/f noise has much less impact in a weighted measurement than in a flat measurement. By the way, RIAA correction has the opposite effect.
I just updated the measurements table on my website to include the minimum and maximum measured values for each tube type, in addition to the average value. I also made some edits to clarify the text, based on some of the questions I've gotten in this thread.
Downloads – Tavish Design
And yes, the 3 - 4 uV RMS noise floor at the output of the test circuit is with the B+ DCV turned on. The power supply regulator is very noisy, of course, but there is a two section RC low-pass filter between the HV regulator and the test circuit. I guess I should have included that in the test circuit schematic, but you can see the second section of it in the photo of the test fixture. The test circuit "sees" two 33uF caps in parallel.
I'll update the table when I've measured the EF86 and 6SL7 tubes. And I may try to find some of the less common tubes suggested in this thread. Thanks!
Scott
ITU considerably boosts mid frequencies; it doesn't just have unity gain like A weighting does. ITU also requires quasi-peak detection which further increases the measured result. But the LF cut is quite gentle in both cases, so their weighting effects are relatively uniform for any mixture of white/pink Gaussian noise (esp for A weighting). Differences mainly occur when you have some blue noise in there, or highly peaky (non Gaussian) noise like bursts.
Yes, RIAA gives a noise benefit for white noise but a noise disadvantage for pink noise.
I tend to agree that for comparison purposes a flat response is preferred. You can then add any weighting you fancy thereafter. At least everyone starts with a set of comparable measurements. Equally important is to use a true rms measurement rather than the typical average reading calibrated for sine wave rms meter often used but this is easier to achieve these days with FFT type scopes than with analogue meters.
Cheers
Ian
Hi Ian,
I disagree entirely. If you only give one figure rather than some sort of spectrum plot, you have no idea what part is white and what part is 1/f noise and therefore you can't calculate what the number would have been with weighting. Hence, you either have to give some indication of the spectral distribution of the noise, or measure with all relevant weightings right from the start.
Regards,
Marcel
Hi Marcel,
LOL. I disagree entirely. 'Some sort' of spectrum plot is very vague and there is a huge range or parameters that can be set in the capture of such a plot. Then there is the assumed ability of the person looking at the plot to interpret it relative to plots from other tubes and from that decide which tube actually sounds quieter. Not going to happen. A single measurement of integrated power over the audio bandwidth is much more useful. The ITU weighting is specifically designed to mimic the response of the ear to noise, so it is my personal preferred measurement but it does tend to give worse answers than A or C weighting which is probably why it is not used much.
Cheers
Ian
I think what Marcel was saying is that you can't apply weighting after the fact to an unweighted (single number) measurement unless you know the noise spectrum (e.g. the noise corner frequency for Gaussian white/pink noise. Not that this makes a huge difference where A- and ITU- weighting is concerned).
Well, in the case of these tube noise measurements, the noise is mostly white. I see a rise due to 1/f only at low frequencies, below 200 Hz or so. It varies a bit from tube to tube, of course, but for calculation or comparison purposes, there would be very little error from assuming it was all white noise, given the 8kHz BW. I'll post a spectrum next time I update the website.
Scott
Scott
You need to be very careful about how you extract the spectrum. Things like moving average filters and windowing can change the apparent corner frequency. It is quite unusual for the corner frequency to be lower than 1kHz where tube noise is concerned, so I would be suspicious of any numbers consistently lower than 200Hz.
I agree, but from the point of view of the original purpose of this thread, i.e. the comparison of perceived noise from different tubes, the spectrum is even less useful.
Cheers
Ian
Hi Ian,
Apparently we don't disagree as much as I thought. A single number measured with a sensible weighting is indeed very useful.
Best regards,
Marcel
I do not think there is any serious disagreement. We all want the same thing, a repeatable, representative easily implemented measurement of noise.
Cheers
Ian
I just updated the summary and webpage to include some captured noise waveforms from the Picoscope. You can see them at Downloads – Tavish Design
There is a time domain waveform, and frequency domain spectra with 1kHz and 20kHz bandwidths, all taken at the output of the C-weighting filter. They happen to be taken with a Tung Sol Reissue 12AX7, but are typical of the waveforms seen. I don't think I've seen a vacuum tube noise spectrum before (except for my own circuits), and few (if any) rigorously taken noise measurements on current productions tubes, which is why I wrote up this work.
I'm still working on the 6SL7 & EF86 test circuits, and also the heater-cathode leakage (hum) measurements.
Best,
Scott
There is a time domain waveform, and frequency domain spectra with 1kHz and 20kHz bandwidths, all taken at the output of the C-weighting filter. They happen to be taken with a Tung Sol Reissue 12AX7, but are typical of the waveforms seen. I don't think I've seen a vacuum tube noise spectrum before (except for my own circuits), and few (if any) rigorously taken noise measurements on current productions tubes, which is why I wrote up this work.
I'm still working on the 6SL7 & EF86 test circuits, and also the heater-cathode leakage (hum) measurements.
Best,
Scott
Tim,
I'm using the same test setup, but with AC on the heaters instead of regulated DC. Then measuring the powerline harmonics on the output and computing an input-referred "hum" analogous to the input-referred noise. As you might expect, that number is much higher than input-referred intrinsic noise, and it also varies form tube to tube.
Scott
I'm using the same test setup, but with AC on the heaters instead of regulated DC. Then measuring the powerline harmonics on the output and computing an input-referred "hum" analogous to the input-referred noise. As you might expect, that number is much higher than input-referred intrinsic noise, and it also varies form tube to tube.
Scott
For ac heater powering, were you going to modify the cabling to attempt to minimise capacitive and electromagnetic field coupling?
For capacitive coupling, the 12AX7 spec is <0.15pF, but the EF86 is way way lower, so external parasitics can dominate the intrinsic tube level. Tuning a humdinger is then sort of the practical implementation performance, but not the intrinsic tube performance per se.
The cathode-heater resistance is very tube sample dependant, but given you may have a very low noise floor, and are nulling the capacitive contribution, may be discernible if the tube sample is poor and you don't dc bias to a higher resistance region.
For capacitive coupling, the 12AX7 spec is <0.15pF, but the EF86 is way way lower, so external parasitics can dominate the intrinsic tube level. Tuning a humdinger is then sort of the practical implementation performance, but not the intrinsic tube performance per se.
The cathode-heater resistance is very tube sample dependant, but given you may have a very low noise floor, and are nulling the capacitive contribution, may be discernible if the tube sample is poor and you don't dc bias to a higher resistance region.