Specifying noise

[NewYorkDave]

OK, I thought I knew how to do this, but I'm coming up with numbers that seem to defy the laws of physics. I must have made a silly oversight somewhere and I'd appreciate it if someone could point it out to me.

Here's the 20Hz-20kHz bandwidth noise plot for my MILA-1 amplifier, measured using RMAA and an interface (Audiophile USB) known to have low residual noise.

The noise level (as distinct from hum, which we see as a bump at 60Hz) is 130dB below the reference level (+20dBM). The hum level is about 112dB below reference. In terms of absolute levels, this works out to -110dBM noise and -92dBM hum.

It's lookin' good, right? But there's trouble when I try to translate these numbers into equivalent input noise (EIN). -110 plus -66 = -176dBM, which seems stellar until you consider the fact that the thermal noise of a 200-ohm resistor at room temperature is about -129dBU. Even if I use the hum figure (the worse of the two figures) as my noise level reference, I still end up with -158dBM.

This doesn't make sense. What am I missing? I get the feeling that I'm supposed to factor the bandwidth into the calculation--and sum the total noise with regard to the bandwidth, instead of just taking the average voltage level across that bandwidth. But I can't remember the particulars and my Googling isn't bringing up the info I seek.

 

[Crusty2]

Scroll down this page:

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

Paul

 

[PRR]

I have not had my third cuppa coffee. Tossing ideas:

Obviously you know to terminate the input in a wire-wound resistor? In a tube preamp, I suppose it makes little difference 200 ohms or short, and open would give huge noise. If you had noise current (i.e. BJT), a short would give wonderful results you'd never get on a real mike.

You are measuring output noise? Are you sure of your gain? If you think it is 60dB, and it is really 55dB, the input noise is 5dB worse than you calculate from the assumed gain.

Have you accounted for 150:600 transformation?

Can you borrow a "very good lo-noise" preamp, trim it to the same gain, and compare output spectra?

The lack of bass-rise is suspicious. Tubes have 1/f noise; you also seem to have a few dB bass-bump in the plot in the other thread; but this shows no rise.

> I'm supposed to factor the bandwidth into the calculation

It's in there. Your "-129" number has "audio band" implied in it. Narrow-band speech assumes a lower noise floor (but we rarely care). 1Hz bands will be about 43dB lower, -172dB? Hmmmm? if RightARC is measuring 1Hz noise bands, "-176dBm" is impossible by 4dB, not 40+dB.

Fornicate the PC-based tools. Get out your Ballantine ACmVM and read the actual output wiggle. If you can dummy-up a 400Hz hi-pass and 15KHz lo-pass, do that. <400Hz noise is usually unimportant, and ~15KHz low-pass approximates a 20KHZ brick-wall, plus it discounts any band-top peaking or supersonic garbage.

 

[Viitalahde]

Is the output cable connected? :green:

 

[NewYorkDave]

Hmmm... So, it seems that the process I'm trying to apply to the data is correct, but the data itself is suspect. That makes me feel better :wink:

Hmmm... You know, when you generate an HTML report in RMAA, for the summary, it spits out a number for noise that's A-weighted. I'm wondering now if that weighting is applied to the plot as well.

The amplifier is at home now and I don't want to drag it back into work, but I'll try to remember to bring my AC millivoltmeter home to do the measurement again the old-fashioned way--the way I always did it before, which always gave results that made sense :thumb:

 

[Boswell]

I suspect there's a scaling problem here. What's the actual output clipping level measured on the same scale? It may be worth doing an independent check. The analyser figure represents about 3uV rms at the output, which is the sort of number I would expect to see at the input.

Try taking the unit back into work and setting up the same trace as you posted. Then apply a 1KHz sinewave from a good generator across the 200 Ohm input resistor using a 10K resistor in each leg (to give about 40dB attenuation). Wind up the sinewave level until you see the 3KHz peak start to rise faster than the 1KHz. Back off a couple of dB and read the height of the 1KHz peak against the Y axis. You could use a wideband ac voltmeter on the output of the unit at this point to get an absolute reading. Then reduce the generator output in 20dB steps and check that the analyser agrees with what you are doing.

The voltmeter gives you the 0dBFS level which you can convert to dBm. The height of the 1KHz peak above the noise floor gives you the dynamic range of the amplifier, which you subtract from the FS output level to get the output noise floor in dBm. If you are certain of the amplifier gain (66dB?), subtract that to give you the (unweighted) noise floor referred to the input.

 

[NewYorkDave]

If you can dummy-up a 400Hz hi-pass and 15KHz lo-pass, do that.

Done! (But I decided to go for a wider bandpass...)

It's just a simple first-order job:

 

[Samuel Groner]

What am I missing?

The problem is that there is no easy relationship between a spectral analysis and a time domain RMS measurement for non-stationary signals. The only easy thing I could think of right know is that they are proportional--if you increase the level in the time domain, the same rise can be noted in the frequency domain. Conclusion: the spectrum analysis doesn't do what you need.

BTW, I remember some FFT plots in an Apogee leaflet where they tried to proof that their converters have "true 144 dB dynamic range" with the same trick that fooled you.

Samuel

 

[bcarso]

Also, when you use your passive filter be sure to account for noise bandwidth being different than "3 dB" bandwidth, except when the filter order is very high.

When I used to do a lot of noise measurements the most frequent sources of error were 1) improper assumptions about gain 2) improper assumptions about bandwidth.

It does look like something is cutting off lows on your spectral plot.

 

[NewYorkDave]

Hmmmmm....!

http://www.osha.gov/dts/osta/otm/noise/images/a_weighting_network.gif

Kinda makes ya wonder if RMAA's finger is on the scale... if ya know what I mean.

I'll measure using the trad method tonight and will report back.

 

[NewYorkDave]

Well, well... As we expected, I ended up with results that seem much more realistic.

With a 200-ohm source (carbon-film, I didn't have wirewound), 620-ohm load, gain control at max and my passive 15Hz-30kHz bandpass filter interposed between the load and the AC millivoltmeter, I get the following results:

Range 1:
Gain: +54dB
Hum + Noise level: -59dBM
EIN: -113dBU
S/N: 63dB (ref. +4dBM)
Dynamic range: 79dB (ref. +20dBM)

Range 2:
Gain: +60dB
Hum + Noise level: -60dBM
EIN: -120dBU
S/N: 64dB (ref. +4dBM)
Dynamic range: 80dB (ref. +20dBM)

Range 3:
Gain: +65dB
Hum + Noise level: -55dBM
EIN: -120dBU
S/N: 59dB (ref. +4dBM)
Dynamic range: 75dB (ref. +20dBM).

I haven't confirmed this, but I suspect that the reason why Range 1 is not ~6dB better than Range 2 (in terms of actual noise level) is because the input stage cathode is unbypassed in Range 1, and the heaters are powered by AC. The cathode is bypassed in Ranges 2 and 3.

The noise levels are about 2dB higher across the board with a 10K load instead of 600. But the circuit can also furnish a greater maximum output voltage into a bridging load (+24dBU instead of +20dBM).

Without the bandpass filter, noise levels are a couple or three dB higher (in a 1MHz bandwidth, according to the spec of my millivoltmeter. But that's beside the point since I've never seen audio equipment specified that way).

RMAA had summarized the noise level as -78dBU A-weighted. And obviously it applied weighting to the spectrum plot as well. So... RMAA users should take note: RMAA uses a weighted noise measurement that can mislead you when comparing to equipment that was measured without weighting.

Thanks for the replies, everybody!

 

[PRR]

> RMAA had summarized the noise level as -78dBU A-weighted

The digi-boys always A-weight.... not to lose the lows, but to discount the way they shift residual digi-noise around the spectrum.

IAC: -120dBU isn't that good. You should be doing better. I repeat: limit your spectrum to 20KHz, which for a simple 1-pole filter means a corner at something like 15KHz (not 30KHz!). And cut the lows at 400Hz: your random noise is "the same" whether bandwidth is 15Hz-20KHz (19,985Hz) or 400Hz (19,600Hz) (0.08dB difference), but 400Hz discounts your hum and buzz, which isn't random noise and is not totally unavoidable. Your goal would be "the same reading" (-0.08dB) with or without the 400Hz low-cut; but for initial bench testing it may be expedient to just filter-away the hum/buzz and see if there is any hope of getting the random noise reading "good".

Not that -120dBU sucks. But I think the random noise should be 5 or 6dB better, over 20KHz bandwidth, with a selected tube, and discounting the hum/buzz which you will fix (or not) as a separate study.

 

[pstamler]

Well, let's look at some theoretical numbers. Second stage first.

You're running it at about 0.8mA, which means a transconductance of about 1 millimho. So your equivalent noise resistance is about 2.5k; add 2.15k for the cathode resistor paralleled with the feedback resistor, ignore the plate resistor because divided by the gain^2 of the stage it's too small to matter, and your total noise resistance is about 4650. In a 20k bandwidth this should produce an equivalent grid noise of 1.2uV. The gain of the stage is about 46x, so output noise is about 58uV, or about -87dBu, at the output of the active circuits, or about -99dBu after the output transformer. So most of your noise is not coming from the output stage.

What about the level control? Worst case, its output resistance is 25k, for an equivalent noise level of about 2.9uV. Amplify that by 46x, divide by 4 and you get about 33uV, or -87dBu at the output of the transformer. Not that either.

Now the input stage. I'll pretend for the moment you're using a Jensen transformer instead of the Beyer. You're running that stage at about 1mA, so your transconductance is about 1.2 mmho and the equivalent noise resistance is about 2080 ohms. Your cathode resistor is bypassed in ranges 2 and 3, so it won't contribute any noise. Mu is about 27.5, Rp is around 20k and Rload is 50k, so the gain should be about 20x. Divide Rload by the gain squared and you get 125 ohms. So the total noise resistance of the tube is close to 2200 ohms.

The reflected impedance of a 200 ohm source is of course 20k. If we were using the Jensen 115K-E transformer, the secondary DC resistance would be 2465 ohms, and the primary resistance reflected to the secondary would be 1970 ohms. Add 270 ohms for the resistor in series with the grid, and the total equivalent DC resistance would be 4705 ohms.

So our three noise sources are, in resistance terms:

DC resistance: 4,705 ohms
Source impedance: 20,000 ohms
Tube noise: 2,200 ohms

The total grid-referred noise would be almost exactly 3uV, or -108dBu. That's the equivalent, at the transformer's primary, of -128dBu in random noise. So this should be a much quieter preamp than you're measuring.

Of course, what you're seeing on the meter may not be random noise, but hum. As PRR notes, you should set your lower bandlimit at 400Hz to dump hum components. And he's right about setting the upper bandlimit to 15kHz; that's the equivalent of measuring a sharp 20kHz bandwidth.

The other possibility is that your Beyer transformer has higher DC resistance in its primary and secondary. But even if the DC resistance was doubled, the noise would only increase 0.7dB.

So my guess is you've got hum. And the RMAA graphs suggest the same thing. I suspect that if you get rid of the hum, and measure with a 400Hz-15kHz bandwidth, you'll find a really quiet preamp waiting.

Peace,
Paul

 

[pstamler]

Well, let's look at some theoretical numbers. Second stage first.

You're running it at about 0.8mA, which means a transconductance of about 1 millimho. So your equivalent noise resistance is about 2.5k; add 2.15k for the cathode resistor paralleled with the feedback resistor, ignore the plate resistor because divided by the gain^2 of the stage it's too small to matter, and your total noise resistance is about 4650. In a 20k bandwidth this should produce an equivalent grid noise of 1.2uV. The gain of the stage is about 46x, so output noise is about 58uV, or about -87dBu, at the output of the active circuits, or about -99dBu after the output transformer. So most of your noise is not coming from the output stage.

What about the level control? Worst case, its output resistance is 25k, for an equivalent noise level of about 2.9uV. Amplify that by 46x, divide by 4 and you get about 33uV, or -87dBu at the output of the transformer. Not that either.

Now the input stage. I'll pretend for the moment you're using a Jensen transformer instead of the Beyer. You're running that stage at about 1mA, so your transconductance is about 1.2 mmho and the equivalent noise resistance is about 2080 ohms. Your cathode resistor is bypassed in ranges 2 and 3, so it won't contribute any noise. Mu is about 27.5, Rp is around 20k and Rload is 50k, so the gain should be about 20x. Divide Rload by the gain squared and you get 125 ohms. So the total noise resistance of the tube is close to 2200 ohms.

The reflected impedance of a 200 ohm source is of course 20k. If we were using the Jensen 115K-E transformer, the secondary DC resistance would be 2465 ohms, and the primary resistance reflected to the secondary would be 1970 ohms. Add 270 ohms for the resistor in series with the grid, and the total equivalent DC resistance would be 4705 ohms.

So our three noise sources are, in resistance terms:

DC resistance: 4,705 ohms
Source impedance: 20,000 ohms
Tube noise: 2,200 ohms

The total grid-referred noise would be almost exactly 3uV, or -108dBu. That's the equivalent, at the transformer's primary, of -128dBu in random noise. So this should be a much quieter preamp than you're measuring.

Of course, what you're seeing on the meter may not be random noise, but hum. As PRR notes, you should set your lower bandlimit at 400Hz to dump hum components. And he's right about setting the upper bandlimit to 15kHz; that's the equivalent of measuring a sharp 20kHz bandwidth.

The other possibility is that your Beyer transformer has higher DC resistance in its primary and secondary. But even if the DC resistance was doubled, the noise would only increase 0.7dB.

So my guess is you've got hum. And the RMAA graphs suggest the same thing. I suspect that if you get rid of the hum, and measure with a 400Hz-15kHz bandwidth, you'll find a really quiet preamp waiting.

Peace,
Paul

 

[CJ]

Incredible posts you guys!

 

[NewYorkDave]

Absolutely! I'd much rather read analyses from the Two Pauls than another "what capz should i use in my gssl?" post. This is exactly the sort of discussion that sets the Drawing Board apart, IMHO.

Well, I was actually pretty pleased to come up with -120dBU EIN (that's the spec of the old Langevin tube preamps, so it can't be all that bad, right?). But youse guys had to go and piss on my parade and now I have to drag that ACmVM home with me again :wink:. But the detailed analysis of the noise sources in my circuit is more than adequate compensation for the trouble :thumb:

I've altered the bandpass filter I built yesterday. I changed C1 to .001uF and C2 to .0047uF. I also built a single-pole highpass (.033uF and 12K) in a separate box, which can be added-on in front of the bandpass filter. Here's a response plot.

The idea here was to use the bandpass for a total hum + noise measurement, and add-on the 400Hz highpass for "noise only" measurements.

I was using my ACmVM as a hi-Z buffer in front of the audio interface, and I had to hit its front-end pretty hard to get enough level into the interface to reach full-scale. That might account for some of the jaggedness in the response.

 

[clintrubber]

[quote author="PRR"]> I'm supposed to factor the bandwidth into the calculation

It's in there. Your "-129" number has "audio band" implied in it. Narrow-band speech assumes a lower noise floor (but we rarely care). 1Hz bands will be about 43dB lower, -172dB? Hmmmm? if RightARC is measuring 1Hz noise bands, "-176dBm" is impossible by 4dB, not 40+dB. [/quote]
It'd be nice indeed if RMAA would state its noise-bin-BW. And a noise-density integrating-function with user-definable f_start & f_stop would be nice as well.
But I'm overasking, it's free & already a nice program as it is now.

Regards,

Peter

 

[soundguy]

[quote author="CJ"]Incredible posts you guys![/quote]


seriously!

seems like making measurments is really a very subjective thing, kinda adds more creedence to the idea that quoting specs doesnt account for too much without the lenghty details of the test procedure. If I was a spec quoting type of fellow I think I would be pretty detailed about the test procedure along with the measurement just so people can see exactly what the deal was.

dave

 

[bcarso]

[quote author="soundguy"]seems like making measurments is really a very subjective thing, kinda adds more creedence to the idea that quoting specs doesnt account for too much without the lenghty details of the test procedure. If I was a spec quoting type of fellow I think I would be pretty detailed about the test procedure along with the measurement just so people can see exactly what the deal was.

dave[/quote]

I wouldn't so much say subjective, but definitely subtle and prone to all manner of errors. I quite agree that specifying the setup and procedure would be a good idea.

There have been some pretty egregious errors made by many who should know better. I remember a Pass amp that claimed a signal-to-noise of 155dB below max output. Details and justification were skimpy :roll:

 

[NewYorkDave]

"I think I would be pretty detailed about the test procedure along with the measurement just so people can see exactly what the deal was."

Check out this page from the '61 Langevin catalog:
PDF

Even though somebody goofed on the "Distortion and Output Power" diagram, you've got to applaud Langevin for having a high enough opinion of their customers to go into that kind of detail.

Here's the page for the 5116B preamp, from the same catalog. Note the lack of ambiguity in the specs:
PDF

Now, in the present day, cheap "consumer" and "prosumer" crap usually comes with pitifully vague and meaningless specs. But that syndrome isn't limited to the low end (dollar-wise) of the market. Here's an example of pretty poorly-presented specs from a company that caters to big spenders:
Link

"Flat frequency response from 10Hz-60KHz"

At what level? What source and load impedances? Define "flat."

"<0.05% THD"

Level? Load impedance? Harmonics only or noise +harmonics? Band-limited or weighted?

"S/N ratio: 80dB"

Reference? Bandwidth? Weighting?

"EIN: -125dB"

Relative to what?

"Max. input (with input pad fully open): 450mV"

Peak or RMS? Frequency?

"Max. output: +32dBu"

Into a 10K load? 100K load? 1 Meg scope input? Define "maximum": 1% THD point, onset of clipping, other?

Well, you get the idea...

Here's a couple of examples of somewhat more relevant specs:

Link1
Link2