How much noise floor and other things matter in oscilloscope usability

[EEEnthusiast]

For very small signals, use the FFT mode on the scope to find the amplitude. FFT does help to bring out signals from the noise (for periodic signals). For e.g. you could observe a 2mVrms sine wave with even 10mVrms of noise, by using FFT. 

[Fiorenzo]

Ok I have taken other photos but now with BW limit set to 20Mhz and the probe connected but grounded to its tip.

Photos:
1) 1x probe
2) 10x probe

[Fungus]

Ok I have taken other photos but now with BW limit set to 20Mhz and the probe connected but grounded to its tip.

Photos:
1) 1x probe
2) 10x probe

What does it look like at 1v/div? That's what you'll normally be seeing when you're working.

[Fiorenzo]

Fungus you asked me to what I am comparing the noise:
I saw some photos on the web of the siglent sds2000x+ and the measured RMS and peak to peak value, what I see on the Rigol seem to me many times higher....
Because I am no expert I cannot say the Rigol have too much noise and this is a problem for me. I am trying to understand/learn from you if It could be acceptable or a problem in some circumstance because I hope to do not change the oscilloscope for at least 5/10 years.

[Fiorenzo]

Ok I have taken other photos but now with BW limit set to 20Mhz and the probe connected but grounded to its tip.

Photos:
1) 1x probe
2) 10x probe

What does it look like at 1v/div? That's what you'll normally be seeing when you're working.

I didn't change the volt scale only the probe attenuation.

In the afternoon I am going to take the photos you asked, I cannot now.

[Andreas]

In what situation really matter to have an oscilloscope with a "low" noise floor?

The most annoying situation is when you use a FFT to detect small signals within a larger complex signal.
E.g. 50 Hz mains hum within a pre-amplified signal. (AC transformer too near to the setup creating a large error nearly not visible in the time domain).

E.g. actual Chopper frequencies and overlay of different stages (dependant on temperature) at the output of a Chopper stabilized OP-Amp. (had the problem that the measured noise increased suddenly at special temperatures due to interferences)

with best regards

Andreas

[Fiorenzo]

For most uses the noise of the scope is not really an issue. This is because a 1:10 probe is often the mayjor noise source.  Unless the scope is rather high in noise, the noise floor is mainly an issue with the few measurements done with an 1:1 probe or similar signal via coax.
For digital signals itself the scopes noise is usually not an issue, but it may be when looking at the supply ripple.

For comparing the noise, one should take the bandwidth into account. With a higher BW the RMS noise naturally goes up. This effect can make a new higher BW scope look higher noise. So  one usually should compare at the same BW, like the usual 20 MHz BW setting.

2 Gs/s are a bit on the low end for 500MHz BW. This may lead to some aliasing and a few more thoughts about what one is actually seeing at the short time scale end.

Thank you Kleinstein for your reply.

What typology of circuits/signals need measurements with a 1x probe?

Does a higher sample rate give a better visual resolution of square wave signals at high frequency  or its benefit is reduced by the amount of noise of the front end?

[Fiorenzo]

For most uses the noise of the scope is not really an issue.

I strongly disagree with this. An oscilloscope with a lot of noise gives thick traces which make it hard to see the actual level of a signal. Ofcourse you can use high-res or bandwidth limiting but IMHO these should be targeted at cleaning up a signal and not masking a poorly designed analog front end. An oscilloscope with a low noise floor is easier to work with. I used to own an Agilent DSO7104A and it was just horrible to work with for analog stuff due to the massively thick traces it has due its own noise.

Also note that the noise doesn't only apply to the most sensitive V/div setting, it applies similar to all V/div settings. The noise level is usually specified in Volts using the most sensitive V/div setting but it would be more accurate to specify it as a percentage of a division or full range. In the end the V/div setting adjusts an input divider but the actual noise level (what goes into the ADC and what gets added by the ADC) stays the same; it is just scaled differenty.

Thank you, interesting.

[Sighound36]

This was one fo ther main reasons for purchasing Lecroy scopes, (we also have Rigol 8000's) which are great general scopes and have some cracking features, but for low noise rail voltage measurments (and 2 channel FFT's) that are consistant, accurate (gain accuracy of 0.5%) and reliable coupled with the superb selection of probes then Wavepro HD is the R&D go to. We alosmuse powerananlysers (both DC and AC which are again very accurate) along with a 6705C.

nctnico is on the money, scope accuracy is very important imho

[Fiorenzo]

For example the noise floor:
In what situation really matter to have an oscilloscope with a "low" noise floor?

When you're measuring very small signals.

I am going to use It for many different things, I am a "begginer" but I do digital stuff with embedded electronics and I also aim to learn more things as possible about analog electronics starting from working on an old valve radio that i would like to repair and experiment with.

It won't make any difference at all on your digital stuff.

For the radio? If it turns out to be a problem you can easily add a preamplifier and make it even better than a lower-noise oscilloscope.

So how much this matter in electronics? How this could preclude its usability?

It's not a showstopper. You can still do everything, just maybe not as easily for a few specific things.

The real questions are: How often do you do those things? How much would you have to spend to get a lower-noise 'scope with the same abilities as your Rigol? Is the extra money well-spent?

Yes I understand what you mean, but I don't have the knowledge to understand ​all the "abilities" of my scope.
My budget is <2000€ but if it is possible to spend less of 500€ It would be (offcourse) better because I could buy other equipments.
It Is a difficult decision because there are so many models of oscilloscopes and no one is perfect.
I would like to have the right instrument that will let me expand my knowledge and build circuits without particular constrains, maybe I am exagerating because I don't have a complete understanding of all the fields of electronics, but at the same time I do not expect to work with Ghz exotic electronics.

[Kleinstein]

For the DSO there are 2 types of noise: noise relative to the input (e.g. input amplifier noise)  and noise relative to the output (ADC noise). For the noise relative to the input the amplifier a 10:1 probe can be a major noise source. The 1 M resistor in the divider has a natural noise of some 130 nV/sqrt(Hz), which is higher than a reasonable JFET based input stage (more like 10 nV/sqrt(hz) range). A low noise for the input is nice, but not relevant when using a 10:1 probe. It can be when using a 1:1 probe.
An 1:1 probe is mainly used when one rellay needs low noise for very small signals (e.g. noise at some point on a circuit, supply ripple). It comes with the price of reduced bandwidth (except with a special, expensive active probe), but with significant lower input noise (e.g. factor 10 from the divider ratio and an additional factor of some 3-10 from not having the resistor noise).

The lower gain settings usually use an internal input divider and if not designed good this may add some noise to 1 or 2 of the higher ranges, which can be a bit annoying as it is avoidable. So ideally a full noise testing would test all ranges, at least with 1 BW setting.

The output related noise, e.g. from the limited resolution  ADC noise, of just ADC noise because of a limited effective ADC resolution that may be lower.
For the ADC noise the sampling rate can make a difference. So the same DSO may look noisy at the higherst sampling rate but looks much better at a lower sampling rate when more samples are averaged. The noise relevant bandwidth is different from the 3 dB bandwith and can be quite a bit higher if the sampling rate is high, or closer to the -3dB BW when the sampling rate is barely sufficient. So it needs some case for the comparison to get comparable condictions (e.g. same sampling rate, relatively close to the maximum, like some 1 Gs/s for the scopes in question).
With the high speed scopes the ADC noise can be a factor.

This is mainly relevant with non repetitve (single trigger) relatively fast signals (so one needs a high sampling rate).
Spectrumanalysis (FFT) is also an example where the noise can matter.  Here also the way the math is done (e.g. does it support averaging, if so the unused time for the math) can make a difference.

Besides the noise of the raw data the methods available for noise reduction can also make a difference. So how good is the averaging over multiple triggers or bandwidth limiting working.

A higher sampling rate can reduce the artifacts from signal parts beyound the Nyquist limit (f_s / 2). With a relatively low sampling rate for a given BW it would need either some extra anti aliasing filtering that can add phase shifts or one has to accept some artifacts if the signal actually contains very fast components. At the same BW, the higher sampling rate scope would usually give a slightly more accurate response.
Noise whise the highest sampling rate often comes with more noise, but after averaging of consecutive samples (the usual way to reduce the data rate - though it can be done worse) the noise improves and thus no panelty from a faster ADC.


edit:  I just remembered: the 10:1 probe is resistive only for low frequencies (e.g. < 10 kHz) but capacitive at higher frequenices. So there is not that much extra noise from the divider and input amplifier noise can still be a factor with the 10:1 probe (at least in the lower ranges).

[gf]

Also note that the noise doesn't only apply to the most sensitive V/div setting, it applies similar to all V/div settings. The noise level is usually specified in Volts using the most sensitive V/div setting but it would be more accurate to specify it as a percentage of a division or full range. In the end the V/div setting adjusts an input divider but the actual noise level (what goes into the ADC and what gets added by the ADC) stays the same; it is just scaled differenty.

Eventually it depends on the individual attenuator and front end design. For instance, I'm aware of a low-cost scope model where the relative noise levels are best at 100mV/div, 1V/div and 10V/div; already a bit worse at 50mV/div, 500mV/div and 5V/div; even worse at 20mV/div, 200mV/div and 2V/div; and finally getting more and more worse at 10mV/div, 5mV/div and 2mV/div

[Fiorenzo]

Which kind of signal need an oscilloscope with a low noise frontend?

A 1mV signal.

(for example)
[/quote]

What kind of electronics works with such low intensity signals?

[Fiorenzo]

What circuits works with such a low signal?  This is my question from the beginning.
I understand that noise sucks but it is the practical application of a low noise oscilloscope that i cannot figure out.
For example, if I work with on an old valve radio am I going to encounter such kind of low signal? It Is only an example....

On an old valve radio you might want to use a 100X probe, in which case a 'small signal' that would merit using the lowest range of the scope might be hundreds of millivolts.  With a 10X probe, which is what you will almost always use, you might start caring about front-end noise at 50mV.  Low front end noise gives you better FFT performance as well.  There are all sorts of cases where noise is an issue and if you are just starting out,  I can't predict which ones you will run into.

Thank you for all your explanations. Very usefull. I am trying to understand: what kind of electronics work with such kind of low signals?

[tautech]

What circuits works with such a low signal?  This is my question from the beginning.
I understand that noise sucks but it is the practical application of a low noise oscilloscope that i cannot figure out.
For example, if I work with on an old valve radio am I going to encounter such kind of low signal? It Is only an example....

On an old valve radio you might want to use a 100X probe, in which case a 'small signal' that would merit using the lowest range of the scope might be hundreds of millivolts.  With a 10X probe, which is what you will almost always use, you might start caring about front-end noise at 50mV.  Low front end noise gives you better FFT performance as well.  There are all sorts of cases where noise is an issue and if you are just starting out,  I can't predict which ones you will run into.

Thank you for all your explanations. Very usefull. I am trying to understand: what kind of electronics work with such kind of low signals?

Few but if ever needing to measure low values indeed noise can get in the way.
When probing high impedance circuits and connection can effect the circuit operation and as 10x probes are most commonly used a 1mV/div scope setting need be used for a 10mV signal however it will only be displayed ~1div high.
To take this to extremes a popular tablet DSO has just 50mv/div max sensitivity which when coupled with a 10x probe permits only 500mV/div sensitivity which is useless for anything but the simplest of tasks.

When we need higher sensitivity 1x is used but at the expense of higher probe capacitance loading on the circuit so such use is often restricted to low impedance measurements like the ripple on a DC rail or that of a PSU.

When most move to a DSO forgetting to set the input attenuation to match the probe is a common newbie error instead of letting the scope display the correct measured value. This is where probe autosense like in the other DSO you are looking at is of substantial value.

[David Hess]

What typology of circuits/signals need measurements with a 1x probe?

20 MHz AC coupled power supply noise and ripple measurements work well with a 1x probe.  This is actually specified in the ATX power supply standard.  A 50 ohm cable can be used in place of a 1x probe but it will deliver worse performance.

Audio measurements are another area where 1x probes are useful.

Does a higher sample rate give a better visual resolution of square wave signals at high frequency  or its benefit is reduced by the amount of noise of the front end?

It does give better fidelity but how it affects noise depends on the implementation.  High end instruments now use an ADC which can trade sample rate with noise and resolution, but for most, sample rate has no effect.  The reason for this is that if the ADC is not doing noise shaping, then it operates at a constant (maximum) sample rate and different sample rates are produced by discarding samples during decimation, which has no effect on noise within the input bandwidth of the ADC.

Also note that the noise doesn't only apply to the most sensitive V/div setting, it applies similar to all V/div settings. The noise level is usually specified in Volts using the most sensitive V/div setting but it would be more accurate to specify it as a percentage of a division or full range. In the end the V/div setting adjusts an input divider but the actual noise level (what goes into the ADC and what gets added by the ADC) stays the same; it is just scaled differenty.

Most DSOs these days only have a single input divider.  Older DSOs have at least two which allows the input buffer to operate over 1/10th of input range so the input full power bandwidth does not limit performance.

Separately there is a low impedance output divider, usually now in the form of a PGA (programmable gain amplifier), with its own noise characteristics.  At high sensitivity noise is dominated by the input buffer, and at low sensitivity noise is dominated by the preamplifier and ADC.

For the noise relative to the input the amplifier a 10:1 probe can be a major noise source. The 1 M resistor in the divider has a natural noise of some 130 nV/sqrt(Hz), which is higher than a reasonable JFET based input stage (more like 10 nV/sqrt(hz) range). A low noise for the input is nice, but not relevant when using a 10:1 probe.

For a typical tip capacitance of a 10x probe, the 1 megohm shunt resistance is in parallel with about 100 picofarads of compensation capacitance producing a noise bandwidth of only 2.5 kHz, so its noise contribution is only about 6.5 microvolts RMS over a wide bandwidth which is close to insignificant.

For the same reason, the noise contribution from the roughly 500 kilohm resistance in series with the gate of the input transistor for protection adds basically no noise.  It is bypassed with about 1000 picofarads reducing its noise bandwidth to an insignificant level.

The lower gain settings usually use an internal input divider and if not designed good this may add some noise to 1 or 2 of the higher ranges, which can be a bit annoying as it is avoidable. So ideally a full noise testing would test all ranges, at least with 1 BW setting.

The internal high impedance input dividers are also capacitively compensated limiting their high frequency noise.  The output dividers are low impedance so require no compensation, but have low noise anyway.  In a modern DSO, these are part of the PGA.

For the ADC noise the sampling rate can make a difference. So the same DSO may look noisy at the higherst sampling rate but looks much better at a lower sampling rate when more samples are averaged. The noise relevant bandwidth is different from the 3 dB bandwith and can be quite a bit higher if the sampling rate is high, or closer to the -3dB BW when the sampling rate is barely sufficient. So it needs some case for the comparison to get comparable condictions (e.g. same sampling rate, relatively close to the maximum, like some 1 Gs/s for the scopes in question).

With the high speed scopes the ADC noise can be a factor.

I have only seen high end DSOs take advantage of noise shaping in the ADC.  There is probably some effect on noise for DSOs which use an interleaved ADC for multiple channels.

Some old DSOs with relatively low real time sample rates, like 100s of MSamples/second, have less ADC noise (and preamplifier noise) than the quantization noise of their ADC.  When I first saw this on my 2232, I thought something was broken or misconfigured.  This might actually be considered a disadvantage when averaging where added noise would produce a better result and I have actually seen this happen with the averaged signal producing a stair-step from the ADC's quantization noise.

[tautech]

Here's a simple example where the noise increases at the higher input sensitivities required to properly scale the higher attenuation probes.
Done some years back and grabbed from an old post.
SDS1104X-E with 1x, 10x, 100x and 1000x probes all connected to the probe Cal output.

[Fiorenzo]

What circuits works with such a low signal?  This is my question from the beginning.
I understand that noise sucks but it is the practical application of a low noise oscilloscope that i cannot figure out.
For example, if I work with on an old valve radio am I going to encounter such kind of low signal? It Is only an example....

On an old valve radio you might want to use a 100X probe, in which case a 'small signal' that would merit using the lowest range of the scope might be hundreds of millivolts.  With a 10X probe, which is what you will almost always use, you might start caring about front-end noise at 50mV.  Low front end noise gives you better FFT performance as well.  There are all sorts of cases where noise is an issue and if you are just starting out,  I can't predict which ones you will run into.

Thank you for all your explanations. Very usefull. I am trying to understand: what kind of electronics work with such kind of low signals?

Few but if ever needing to measure low values indeed noise can get in the way.
When probing high impedance circuits and connection can effect the circuit operation and as 10x probes are most commonly used a 1mV/div scope setting need be used for a 10mV signal however it will only be displayed ~1div high.
To take this to extremes a popular tablet DSO has just 50mv/div max sensitivity which when coupled with a 10x probe permits only 500mV/div sensitivity which is useless for anything but the simplest of tasks.

When we need higher sensitivity 1x is used but at the expense of higher probe capacitance loading on the circuit so such use is often restricted to low impedance measurements like the ripple on a DC rail or that of a PSU.

When most move to a DSO forgetting to set the input attenuation to match the probe is a common newbie error instead of letting the scope display the correct measured value. This is where probe autosense like in the other DSO you are looking at is of substantial value.

Thank you tautech and everyone again for the time spent replying. I find everything you wrote very usefull and interesting

[Fiorenzo]

What typology of circuits/signals need measurements with a 1x probe?

20 MHz AC coupled power supply noise and ripple measurements work well with a 1x probe.  This is actually specified in the ATX power supply standard.  A 50 ohm cable can be used in place of a 1x probe but it will deliver worse performance.

Audio measurements are another area where 1x probes are useful.

Does a higher sample rate give a better visual resolution of square wave signals at high frequency  or its benefit is reduced by the amount of noise of the front end?

It does give better fidelity but how it affects noise depends on the implementation.  High end instruments now use an ADC which can trade sample rate with noise and resolution, but for most, sample rate has no effect.  The reason for this is that if the ADC is not doing noise shaping, then it operates at a constant (maximum) sample rate and different sample rates are produced by discarding samples during decimation, which has no effect on noise within the input bandwidth of the ADC.

Also note that the noise doesn't only apply to the most sensitive V/div setting, it applies similar to all V/div settings. The noise level is usually specified in Volts using the most sensitive V/div setting but it would be more accurate to specify it as a percentage of a division or full range. In the end the V/div setting adjusts an input divider but the actual noise level (what goes into the ADC and what gets added by the ADC) stays the same; it is just scaled differenty.

Most DSOs these days only have a single input divider.  Older DSOs have at least two which allows the input buffer to operate over 1/10th of input range so the input full power bandwidth does not limit performance.

Separately there is a low impedance output divider, usually now in the form of a PGA (programmable gain amplifier), with its own noise characteristics.  At high sensitivity noise is dominated by the input buffer, and at low sensitivity noise is dominated by the preamplifier and ADC.

For the noise relative to the input the amplifier a 10:1 probe can be a major noise source. The 1 M resistor in the divider has a natural noise of some 130 nV/sqrt(Hz), which is higher than a reasonable JFET based input stage (more like 10 nV/sqrt(hz) range). A low noise for the input is nice, but not relevant when using a 10:1 probe.

For a typical tip capacitance of a 10x probe, the 1 megohm shunt resistance is in parallel with about 100 picofarads of compensation capacitance producing a noise bandwidth of only 2.5 kHz, so its noise contribution is only about 6.5 microvolts RMS over a wide bandwidth which is close to insignificant.

For the same reason, the noise contribution from the roughly 500 kilohm resistance in series with the gate of the input transistor for protection adds basically no noise.  It is bypassed with about 1000 picofarads reducing its noise bandwidth to an insignificant level.

The lower gain settings usually use an internal input divider and if not designed good this may add some noise to 1 or 2 of the higher ranges, which can be a bit annoying as it is avoidable. So ideally a full noise testing would test all ranges, at least with 1 BW setting.

The internal high impedance input dividers are also capacitively compensated limiting their high frequency noise.  The output dividers are low impedance so require no compensation, but have low noise anyway.  In a modern DSO, these are part of the PGA.

For the ADC noise the sampling rate can make a difference. So the same DSO may look noisy at the higherst sampling rate but looks much better at a lower sampling rate when more samples are averaged. The noise relevant bandwidth is different from the 3 dB bandwith and can be quite a bit higher if the sampling rate is high, or closer to the -3dB BW when the sampling rate is barely sufficient. So it needs some case for the comparison to get comparable condictions (e.g. same sampling rate, relatively close to the maximum, like some 1 Gs/s for the scopes in question).

With the high speed scopes the ADC noise can be a factor.

I have only seen high end DSOs take advantage of noise shaping in the ADC.  There is probably some effect on noise for DSOs which use an interleaved ADC for multiple channels.

Some old DSOs with relatively low real time sample rates, like 100s of MSamples/second, have less ADC noise (and preamplifier noise) than the quantization noise of their ADC.  When I first saw this on my 2232, I thought something was broken or misconfigured.  This might actually be considered a disadvantage when averaging where added noise would produce a better result and I have actually seen this happen with the averaged signal producing a stair-step from the ADC's quantization noise.

Mr. David you gave a lot of explanations, this is very kind and usefull.

Trying to do a recap: at this point It seem to me that a "low noise" oscilloscope is important when working with FFT analysis, power supply ripple, audio signal, and high impedance circuits?

In regard of the mso5000 with its high sample rate of 8GSa/s I am not sure if in the balance It is an advantage due to its apparently noisy front end.
As an ignorant, at the beginning I thought: the Rigol is better because It has better specs, so I bought it. Could you suggest me a different model if you think It could be better?

[Fungus]

In regard of the mso5000 with its high sample rate of 8GSa/s I am not sure if in the balance It is an advantage due to its apparently noisy front end.

It definitely is. It will help reduce the Gibbs Phenomenon on your digital circuits.

(Ever wonder how "ringing" can occur before a signal starts to rise? Undersampling combined with sin(x)/x reconstruction...)

[Fungus]

Could you suggest me a different model if you think It could be better?

At the same price? Not a chance.

You'll have to go up quite a lot in price to get something better. Think: The difference would buy a good multimeter, a soldering iron and a bench power supply.

(And the oscilloscope still wouldn't be perfect, you'd just have other things to worry about)

OTOH you could even go down to a 500 Euro oscilloscope like the Siglent SDS1204X-E. It's probably all you need and you'd have enough money left over to create an awesome lab.

[bdunham7]

It definitely is. It will help reduce the Gibbs Phenomenon on your digital circuits.

(Ever wonder how "ringing" can occur before a signal starts to rise? Undersampling combined with sin(x)/x reconstruction...)

Meh, ringing on step responses isn't usually Gibbs unless something has been designed or set wrong.  You can have ringing (and even pre-ringing) on an analog scope. 

[Fungus]

Meh, ringing on step responses isn't usually Gibbs unless something has been designed or set wrong.

It can be a mixture of both.

You can have ringing (and even pre-ringing) on an analog scope.

But you can't have Gibbs.   

[Fiorenzo]

Could you suggest me a different model if you think It could be better?

At the same price? Not a chance.

You'll have to go up quite a lot in price to get something better. Think: The difference would buy a good multimeter, a soldering iron and a bench power supply.

(And the oscilloscope still wouldn't be perfect, you'd just have other things to worry about)

OTOH you could even go down to a 500 Euro oscilloscope like the Siglent SDS1204X-E. It's probably all you need and you'd have enough money left over to create an awesome lab.

Yes I know, this is one of my doubts. I will check some review of that scope but in the mean time what do you think about its bigger brother sds2000x-plus?

[Fiorenzo]

Can you try again only set the vertical scale to 1mV/div and measure with no probe and with the shorted probe?  Then do the same at 100mV/div.

Here I am, I did the test you asked for, please let me know whath do you think about It.
Thank you very kind!

Photos:
1) 1mv/div, no probe, 20Mhz BW limit, 1x
2) 1mv/div, probe, 20 MHz BW limit, 1x
3) 100mv/div, no probe, all the same
4) 100mv/div, probe, all the same