Noise Floor of Oscilloscope and How to Specify DUT Noise.

[Mechatrommer]

related thread are perused... https://www.eevblog.com/forum/general-chat/rigol-1052e-noise-at-100mhz-3mhz-etc/msg4474/#msg4474 ...but this time its different, i dont believe its "outside radio interference" since i'm using 0ohm shorting BNC terminator (maintaining fully shielded input), but maybe something penetrating through/from inside the DSO body/circuit. Johnson Noise is also studied, 50ohm should give 9uV Johnson noise at 100MHz (4kTRB) thats why i think using 0ohm short terminator is wise to rule out Johnson noise at the input (but i maybe wrong that something else is not within my insight). opamp noise formula is also studied, but i prefer more practical way instead of relying on formula alone. from my calc, opamp noise is lower than i can measure, but i need a real life prove on it.

1st picture: apparently cleaner 0.8mVpp noise of Rigol DS1052E (hacked to DS1102E) is just deceiving from my observation, when not connected to anything.
2nd picture: connecting a simple such as T-connector increases the noise to 1.44mVpp.
3rd picture: putting a 0ohm (50ohm, 1Mohm terminator tried to the same result) shorting terminator, boom! 4.32mVpp noise!
4th picture: removing T, just direct terminator doesnt help much (3.6mVpp jumping up and down around the same 4mVpp)
5th-7th pictures: closer look at the offending 100MHz intermittent noise.
last picture: DS1102E frontend circuit as drawn by A-Hallene FWIW. thanks to him i hope he wouldnt mind i reproduce his work here.

now, this got to my attention when i tried to specify my circuit noise output (around 2.8-4mVpp read by the DSO). i thought its my circuit noise floor. but when doing the test above, now i cannot say about it. i simply cannot specify my DUT with this noisy system (DS1102E). So i'm opening for discussion, such as do you find this similar to your oscilloscope? what about the Agilent Infinity X000X? are they exhibit the same behaviour? any other DS1052-1102E can confirm this behaviour?

Is there anyway i can do to be able to specify my DUT with this noisy scope? is there any trick? or other more suitable device (preferably at hobby level budget? or pro device just for our knowledge) i'm aiming at DUT noise around 20µV (or at least at the DSO theoritical noise floor ie 800µV). so i see, from experience i'm going through, things get messy when we/i dealing with this micro level stuff. i also would like to hear comment from our admin Dave about this matter. how did he specify his µCurrent before he got his Agilent DSO. ISTR his µCurrent is using 100X gain amp. so 4mVpp back to 100X opamp can simply deceives us into believing that the opamp input is at 40µV noise level (4mVpp / 100). so any idea? friends and dave?

edit: i may find this thread also usefull... https://www.eevblog.com/forum/beginners/question-about-preamp-for-measuring-noise/ FWIW

[jahonen]

Note that Johnson noise calculation gives RMS voltage (or standard deviation), not p-p. P-p of noise voltage is a tricky business to determine (although an estimate may be calculated from RMS value if the noise distribution can be assumed to be gaussian), RMS is much simpler. Also, noise bandwidth is generally not equal to -3 dB bandwidth, and is a topic of another discussion.

Regards,
Janne

[Mechatrommer]

thanks janne for the reminder. yes i forgot it should be rms value. i tried 10x postamplifier as your suggestion in another thread on my circuit. and it showed 30-40mVpp, so i guess my circuit is indeed that... 3-4mVpp noise or 1.41mVrms (assuming Vrms = (Vpp/2)/sqrt(2).

[free_electron]

Scopes dont have a large enough s/n or dynamic range for these kinds of measurements.
The 8 bit converter has a 10 millivolt resolution and for most low end scopes you only get 6 effective bits.
You need a spectrum analyser. those things have a much lower noise floor.

[Mechatrommer]

The 8 bit converter has a 10 millivolt resolution and for most low end scopes you only get 6 effective bits

i dont know how to calculate it, but how a dso cheat to show 0.8mVpp while not connected to anything, and then 4mVpp when shorted? as shown above?

[w2aew]

If your signal is repetitive (not a single-shot acquisition), you can employ averaging.  This will improve your SNR if the noise is uncorrelated to your signal.

Some scopes can give you "high-res" sampling when your sweep speed (horizontal scale) and memory depth aren't being fully utilized.  In these cases, excess samples that are normally decimated away, are instead averaged together such that each waveform point is actually a boxcar average of n-points.   I don't know if the Rigol supports this mode, but it would most certainly support ordinary waveform averaging.

But, as was already mentioned, you can only get so much out of an 8-bit converter.  Averaging may be able to give you an effective 11-12 bits at best.

[muvideo]

Scopes dont have a large enough s/n or dynamic range for these kinds of measurements.
The 8 bit converter has a 10 millivolt resolution and for most low end scopes you only get 6 effective bits.
You need a spectrum analyser. those things have a much lower noise floor.

Shouldn't have a 62.5uV of resolution at 2mV per div?

Fabio.

[Mechatrommer]

@w2aew: rigol has averaging, and yes it helps with reading repetitive signal buried in noise. but in my case here, i want to specify the noise, so thats not repetetive and averaging (and/or even 25GSps eqv sampling rate) will worsen the result (noise spec) or... to make it "too good to be true" 0V noise.

[free_electron]

Scopes dont have a large enough s/n or dynamic range for these kinds of measurements.
The 8 bit converter has a 10 millivolt resolution and for most low end scopes you only get 6 effective bits.
You need a spectrum analyser. those things have a much lower noise floor.

Shouldn't have a 62.5uV of resolution at 2mV per div?

Fabio.

you wish.... and how much noise is injected becasue of the amplifier stage ? lots of cheap digital scope simply 'pump up' the signal digitally ... ( mathematical scaling )

if that ads runs off a 2.5 volt reference and has 8 bits this is 10mV / step ....
to measure in the mv range you would have a gainstage in front that does *10 or *20

it depends on the input stage construction... do you divide first , then amplify ? do you amplify 'mathematically' ? do you use a 1:1 probe ( it does no good tyring to measure with a 1:10 probe and then amplyfing that again .... )
You need to know the particularities of your scope.

[muvideo]

you wish.... and how much noise is injected becasue of the amplifier stage ? lots of cheap digital scope simply 'pump up' the signal digitally ... ( mathematical scaling )

if that ads runs off a 2.5 volt reference and has 8 bits this is 10mV / step ....
to measure in the mv range you would have a gainstage in front that does *10 or *20

it depends on the input stage construction... do you divide first , then amplify ? do you amplify 'mathematically' ? do you use a 1:1 probe ( it does no good tyring to measure with a 1:10 probe and then amplyfing that again .... )
You need to know the particularities of your scope.

Ok now I understand what you mean. I was thinking to that Rigol model in particular.
By the way,part of the input stage schematic is in the last photo of the original message.

Fabio.

[Mechatrommer]

skip the detailed schematics above. my riddle has not been answered, i repeat here... how the 8bits 6bits effective scope can sense between 800uV - 4mVpp and values in between? (from above pictures) if theoritically, its only 10mV resolution? software trick?

the way i see it (again skipping the details, just by eye only), the dso is capable to sense 800uV difference (or say 1mV difference) at the input. the problem is... something going on (noise) by even just connecting short terminator, albeit full shield. the bottleneck here i believe is inside noise, not resolution. it will not be very useful to specify dut noise less than or equal or near this 4mVpp noise floor, let alone the unshielded diy circuit like what i have here. so my suspicion its something noise from inside, but the input stage is able to see that (1mV difference).

edit: and the input stage i mean, all the way from BNC connector to attenuator up to adc and SW post processing if you may. the output?... is what i see on the screen, the rest are input. so the schematics above is just a fraction of it.