Specifying noise

[NewYorkDave]

More noise measurements tonight... But I'm gonna have to call it good after this, at least for a while. Between being sick for 2+ weeks--and before that, a month's worth of "special events" at my job--I have about six weeks' worth of work backed up.

Here are the numbers using the modified filters:

Gain.............No Filter......40Hz-15kHz BP......400Hz-15kHz BP

+54dB..........-57dBM.......-59dBM.................-67dBM
+60dB..........-58dBM.......-62.5dBM..............-65.5dBM
+65dB..........-53dBM.......-57dBM.................-60.5dBM

This translates to EIN of.:

+54dB..........-111dBU.....-113dBU...............-121dBU
+60dB..........-118dBU.....-122.5dBU............-125.5dBU
+65dB..........-118dBU.....-122dBU...............-125.5dBU

 

[pstamler]

That's more like it. The EIN's for 60dB and 65dB gains are within 3dB of theoretical performance, so you're within the ballpark. I'm a little concerned that the performance at 54dB gain improves so drastically when you go from a 40Hz to a 400Hz bottom cutoff; that means you're picking up a good deal of hum, presumably from those AC filaments. And there's more hum than is desirable even in the higher-gain settings.

First try might be to hoist the filaments up some 60V or so and see if the hum goes away or at least gets quieter. Otherwise, I hate to say it, but it might be DC filament time.

The random-noise figure might get closer to theoretical with different transformers; I don't know what the DC resistance of the Beyers is. (I have a few down in the basement...somewhere......)

Peace,
Paul

 

[clintrubber]

Right ! No data without conditions. :thumb:

It remembered me of this nice file:

http://www.rane.com/pdf/note145.pdf
(be sure to check the latest page as well)

http://www.rane.com/note145.html

Regards,

Peter

 

[NewYorkDave]

Grrrr... Why must you fill me with doubt like this?

:wink:

presumably from those AC filaments.

More than likely. In the +54dB range, the cathode is unbypassed.

Biasing the heater supply is a possibility, but a DC heater supply is not unless I build a new PSU. The heater winding on that xfmr is rated 1A, and I'm already drawing 900mA.

Since I really don't want to put more time into this project if I don't have to, this could be a good time for me to actually listen to the thing and find out whether or not the noise is actually perceptible... or if I'm just engaged in a numbers quest with diminishing returns. But in any case, it's been an educational experience.

BTW, Beyer's catalog spec for the primary DCR is 85 ohms, and the secondary is 4.75K. I have not measured it myself.

Oh, another BTW: Brad made reference earlier to the noise bandwidth being different from the -3dB bandwidth of my filter. (The actual measured -3dB points are at 37Hz and 16kHz, incidentally). So, if I'm quoting a noise spec based on measurements made with that filter, what would be the best concise way to state the measurement bandwidth?

 

[pstamler]

[quote author="NewYorkDave"]

presumably from those AC filaments.

More than likely. In the +54dB range, the cathode is unbypassed.

Biasing the heater supply is a possibility, but a DC heater supply is not unless I build a new PSU. The heater winding on that xfmr is rated 1A, and I'm already drawing 900mA.

Since I really don't want to put more time into this project if I don't have to, this could be a good time for me to actually listen to the thing and find out whether or not the noise is actually perceptible... or if I'm just engaged in a numbers quest with diminishing returns. But in any case, it's been an educational experience.[/quote]

Indeed. Well, see if elevating the heaters does the job.

BTW, Beyer's catalog spec for the primary DCR is 85 ohms, and the secondary is 4.75K. I have not measured it myself.

That's a *lot* higher than the Jensen. It brings the total resistive noise up to 13.5k or so, making it only a bit less than the source noise (still the main noise source, theoretically). With the Beyer, random noise becomes -127dBu rather than -128.7dBu, with a 200 ohm source. (-128.7dBu? I said -128dBu yesterday, but I'd forgotten to take into account the loading resistor on the Jensen.) So you lose about 1.7dB of S/N.

With a 150 ohm source, the EIN's are -129.5dBu (Jensen) and -127.7dBu (Beyer), so you again take a hit of about 1.8dB.

Oh, another BTW: Brad made reference earlier to the noise bandwidth being different from the -3dB bandwidth of my filter. (The actual measured -3dB points are at 37Hz and 16kHz, incidentally). So, if I'm quoting a noise spec based on measurements made with that filter, what would be the best concise way to state the measurement bandwidth?

"Equivalent noise bandwidth 20kHz".

Incidentally, per Jung, the bandwidth to measure 20kHz equivalent is 12,738Hz, assuming a 6dB/octave filter. So a 16kHz filter is actually measuring about 25kHz bandwidth, and your real-world figures are about 2dB high. They're also about 1dB low if you're not using a true-RMS meter, again per Jung. So if you're measuring -60.5dBm it's really -61.5dBm, and your EIN's are -126.5dBu on the two high ranges and (maybe) -122dBu on the low range (maybe because of the hum factor). On the high ranges, that's within half a dB of the theoretical value, a nice match. So all that really remains is the hum.

[Edit:] And maybe a quieter transformer.

Peace,
Paul

 

[Guest]

maybe compute little:
FFT analyzer is bank of filters. You must respect bandwith of this:
-10*log10(NFFT)
and equivallent noise bandwidth of window.

By the way, if you measure via comparative method, you need not
to know anything. Only to connect 200 Ohm resistor and short to input of preamp (if preamp have sufficiently high impedance)
Othervise you need this:
http://www.rudolfinea.cz/flaska.jpg
pure tungsten direct filament diode.

xvlk

 

[NewYorkDave]

Incidentally, per Jung, the bandwidth to measure 20kHz equivalent is 12,738Hz, assuming a 6dB/octave filter. So a 16kHz filter is actually measuring about 25kHz bandwidth, and your real-world figures are about 2dB high. They're also about 1dB low if you're not using a true-RMS meter, again per Jung. So if you're measuring -60.5dBm it's really -61.5dBm, and your EIN's are -126.5dBu on the two high ranges and (maybe) -122dBu on the low range (maybe because of the hum factor). On the high ranges, that's within half a dB of the theoretical value, a nice match. So all that really remains is the hum. And maybe a quieter transformer.

Thanks, Paul. It's your typical average-responding AC millivoltmeter, and the filter is 6dB/octave, so it seems all of the above apply. Is that Jung material available online, by the way?

So, if I'm understanding all this correctly, a snappy but reasonably accurate summary could read:

"Equivalent input noise (20kHz bandwidth, 200-ohm source): -122dBU (+54dB gain), -126.5dBU (+60 or +65dB gain)."

Or, in less technical terms, "decent, could probably be made better with some tweaking." :wink:

EDIT: Ah... I found the following handy summary of noise bandwidth of nonideal filters on Analog's site.

The noise bandwidth of a non-ideal brick wall filter is defined as the bandwidth of an ideal brick wall filter which will pass the same noise power as the non-ideal filter. Therefore, the noise bandwidth of a filter is always greater than the 3-dB bandwidth of the filter by a factor which depends upon the sharpness of the cutoff region of the filter. Figure 2 shows the relationship between the noise bandwidth and the 3-dB bandwidth for Butterworth filters up to 5 poles. Note that for two poles, the noise bandwidth and 3-dB bandwidth are within 11% of each other, and beyond that the two quantities are essentially equal.

Bien voila, 20kHz divided by 1.57 is 12.738kHz. 15.96kHz (37Hz-16kHz first-order filter) times 1.57 is 25.057kHz. And 15.6kHz (400Hz-16kHz first-order filter) times 1.57 is 24.49kHz.

It's always fun to discover these little gaps or lapses in my knowledge of the fundamentals, and then to work on filling them in.

 

[pstamler]

[quote author="NewYorkDave"]Thanks, Paul. It's your typical average-responding AC millivoltmeter, and the filter is 6dB/octave, so it seems all of the above apply. Is that Jung material available online, by the way?[/quote]

Dunno; I found it in Audio IC Op-Amp Applications.

So, if I'm understanding all this correctly, a snappy but reasonably accurate summary could read:

"Equivalent input noise (20kHz bandwidth, 200-ohm source): -122dBU (+54dB gain), -126.5dBU (+60 or +65dB gain)."

Or, in less technical terms, "decent, could probably be made better with some tweaking." :wink:

Yup.

It's always fun to discover these little gaps or lapses in my knowledge of the fundamentals, and then to work on filling them in.

For me too, and I'm always finding new stuff I didn't know.

Peace,
Paul

 

[NewYorkDave]

I threw the preamp in my car as I left for work this morning, just in case I should happen to find the time to mess around with it during the day. As luck would have it, a meeting ended early today and I found myself with about 30 mins. to spare, so...

Well, see if elevating the heaters does the job.

It did. I haven't used that trick in a long time--I'd forgotten how effective it can be.

Here's a new set of measurements, with the heater centertap biased to about +30VDC.

The differentials between the columns of noise levels are consistent (~3dB between columns 1 and 2, ~4dB between columns 2 and 3), at all gain settings, which is encouraging. The +54dB gain setting is no longer degraded by hum to a greater degree than the other settings.

Paul, I owe ya a beer. :thumb:

So, crunching the numbers in column 3 using the +1dB +(-2dB) = -1dB correction factor mention earlier:

EIN (200-ohm source, 20kHz bandwidth): -126dBU (+54.5dB gain), -126.5dBU (+60dB gain), -127.5dBU (65.5dB gain).

Not too shabby.

:sam: :sam:

 

[pstamler]

[quote author="NewYorkDave"]Paul, I owe ya a beer. :thumb:[/quote]

Le chaim!

So, crunching the numbers in column 3 using the +1dB +(-2dB) = -1dB correction factor mention earlier:

EIN (200-ohm source, 20kHz bandwidth): -126dBU (+54.5dB gain), -126.5dBU (+60dB gain), -127.5dBU (65.5dB gain).

Not too shabby.

:sam: :sam:

Not shabby. :grin: At the highest gain setting you're actually half a dB better than theory predicts. There's one more correction: your 400Hz cutoff. If that weren't there, and you had a cutoff of 20Hz instead, then you'd have about 0.2dB more noise. Still 0.3dB better than theory says is possible; probably you've got a tube there with higher-than-normal transconductance.

No, not shabby at all. And a good deal quieter than several commercial tubed preamps I've measured.

Peace,
Paul

 

[NewYorkDave]

Thanks, Paul. I'll apply the 0.2dB correction, but could you tell me how you derived it?

 

[bcarso]

Probably just the net 380 Hz bandwidth difference.

 

[pstamler]

Got it in one. :sam:

Peace,
Paul

 

[NewYorkDave]

Hmmm... Is that based on equal-power-per-bandwidth (a 3dB increase per doubling of bandwidth?). I keep crunchin' it on my calculator and I don't arrive at the same result. Maybe I've gotten lost in the thicket of equivalent noise bandwidth and other correction factors we've been applying along the way. Or maybe my math just sucks. I'd really like to understand this before we call it a day.

 

[pstamler]

[quote author="NewYorkDave"]Hmmm... Is that based on equal-power-per-bandwidth (a 3dB increase per doubling of bandwidth?). I keep crunchin' it on my calculator and I don't arrive at the same result. Maybe I've gotten lost in the thicket of equivalent noise bandwidth and other correction factors we've been applying along the way. Or maybe my math just sucks. I'd really like to understand this before we call it a day.[/quote]

The reason you don't arrive at the same result is simple. Mine was wrong.

I got mine by the simplistic method of dividing the assumed 20kHz bandwidth by 19,600 (the bandwidth when you subtract the 400Hz on the bottom. That's not really exact, but it's pretty close. So I did 20 x the log of that ratio and got .175dB, which rounds off to .2dB.

The reason it's wrong is that I forgot to take the square root of the ratio (as you said, equal power...). The real answer should have been .0877...dB, which rounds off to 0.1dB. Oh, and if you do the same calculation with the 12.7kHz bandwidth you get a correction factor of about .14dB which also rounds to 0.1dB. Which is the real correction factor.

You've still got a damned quiet preamp.

Peace,
Paul

 

[bcarso]

With most amplifying devices the noise spectral density at 400 Hz and below is rising too, so the real change in total noise will likely be a good deal more highpassing at 400.

 

[NewYorkDave]

Oh, brother :?. Well, I rounded off to the nearest whole dB and specified the hum+noise EIN (20kHz bandwidth, 200-ohm source) as -127dBU for +66dB gain and -126dBU for +54 and +60dB gain. I hope that's not too far off the mark. I already pulled the PDF for editing once.

 

[pstamler]

Just add "+/-1dB" as a fudge factor and you'll not only be cool, but look more professional than the commercial companies.

Peace,
Paul