Oscilloscope input noise comparison

[maxwell3e10]

Some time ago I collected shorted input data from several oscilloscopes to compare their noise. More discussion is in this topic https://www.eevblog.com/forum/testgear/owon-xds3062a-input-noise-(good)-and-glitches-(bad)/

Recently I got a Keysight EDUX1002 scope, so I thought I would compare it to the Chinese brands. As one might expect, the noise spectrum of the Keysight is very clean, but the absolute noise level is a little higher for this entry level scope of comparable cost.

Perhaps it would be interesting to compare higher-end scopes as well, but that's above my budget level

[Sylvi]

Hi

That's very interesting

How did you measure the noise?

I recently bought an OWON XDS2102A (100MHz, 12-bit) which is definitely showing me things I could not see with either of my previous CRT scopes - one was 50MHz Philips and the other a 20MHz Instek.

[Hydron]

Perhaps it would be interesting to compare higher-end scopes as well, but that's above my budget level

If you could specify what data is required, then maybe people with higher end instruments could send in the files to analyse and add to the graph? I'd be happy to grab some data from my RTB2004.

[maxwell3e10]

For this comparison I was using the longest time scale that gives 1 GHz sampling rate and the lowest voltage scale that allows full bandwidth. Then just save raw ascii file. To limit file size, use no more than 1M samples.  The input is terminated into 50 Ohm. No averaging or artificial high resolution modes.

[Hydron]

For this comparison I was using the longest time scale that gives 1 GHz sampling rate and the lowest voltage scale that allows full bandwidth. Then just save raw ascii file. To limit file size, use no more than 1M samples.  The input is terminated into 50 Ohm. No averaging or artificial high resolution modes.

I've made a couple of quick measurements at full and 20MHz BW, can't do exactly 1GSa/s but could do 1.25, is that OK? I've saved them as CSV (could do binary as another option if helpful), have sent you a link to the download via PM as it's too big to upload on the forum.
If you need different settings (e.g. due to the 1.25GSa/s thing) please let me know.

[awallin]

Perhaps it would be interesting to compare higher-end scopes as well, but that's above my budget level

If you could specify what data is required, then maybe people with higher end instruments could send in the files to analyse and add to the graph? I'd be happy to grab some data from my RTB2004.

+1 for this.
Put the data-files (zipped?) and analysis scripts (numpy/matplotlib?) into a git repo, with instructions on how to collect data - and just wait for the pull-requests to roll in 

[maxwell3e10]

Thanks, I got  the data for RTB2004! Looks nice, 3-4 nV/Hz^(1/2) noise. I will wait if any other data comes in before making another plot.

[nctnico]

I'll try to pull a file from an R&S RTM3004 later today. A small suggestion: if the new plots are going to contain more scopes then creating 2 graphs (full bandwidth and 20MHz bandwidth) may be a good idea.

[Andreas]

Hello,

interesting project.

For better comparison (and as I am not familiar with all scopes) it would be good to have also the max. bandwidth of the scope used and the max sample rate addditionally to the minimum sensitivity.
And if available the AC-rms (standard deviation) and peak-peak noise for the measurement.

I could part my PicoScope 5444A data but as .csv the amount of data for a 1Ms measurement is huge (even compressed).
With the proprietary internal Format it is below 1 MB so that I could even upload it here.
(You would need the PicoScope Software from PicoTech to export a .csv from here: https://www.picotech.com/downloads )

Also if I export it to Matlab-Format and compress it is below the forum limit.

So the question is if one of these formats would be a alternative.

with best regards

Andreas

[Hydron]

I had the same issue with CSV size - I PM'd maxwell3e10 a download link rather than post it on the forum. A better solution might be to upload it on github or similar, that way nobody has to download proprietary software to convert the file format.

[cruff]

The input is terminated into 50 Ohm.

Do you mean the input connector was shorted and you used the built-in 50 ohm termination mode of the scope?  Or did you use the scope's high value termination mode and terminated with 50 ohms at the input connecotr?

So effectively you are measuring the thermal noise of the termination resistor(s)?  Or am I missing something?

[maxwell3e10]

Some scopes have a build-in 50 Ohm mode and some don't. I meant to add a 50 Ohm resistor outside to short the input. One could also use a shortening BNC. It wouldn't make much difference unless the noise is 2-3 nV/Hz^(1/2).

[nctnico]

Usually the 50 Ohm and 1M Ohm paths are different so the noise is also different.

[maxwell3e10]

Yes, that appears to be the case. I had always assumed in the past that a 50 Ohm input impedance setting on a scope simply switches in a 50 Ohm resistor at the input of the 1 MOhm amplifier.

[rf-loop]

Yes, that appears to be the case. I had always assumed in the past that a 50 Ohm input impedance setting on a scope simply switches in a 50 Ohm resistor at the input of the 1 MOhm amplifier.

It depends oscilloscope. In low price and low frequency oscilloscopes this is "typical" cheap  way to do it. They need only 50 ohm "DC" load resistance. Nothing good in this method for high frequencies - except price. Also noise in these kind of front ends are not at all excellent. Also very small internal noise currents can produce high noise voltages over these 1M pathway after input 50ohm load resistor.

I have seen these many times, connecting external 50ohm feed thru or terminator to 1M scope input it can some times  show even more noise than open or LF-RF shielded input BNC using metal caps.

Time ago one perhaps most worst case  was really amazing effect in some Owon models where lot of internal SMPS circuits was made like wide band radio jammer. "Every kid can design SMPS" so they put cheap unexperienced engineer kids to design these..or for make some modifications to reduce manufacturing costs.  These noise currents flood wild inside scope every part and flow thru different parts and if connect external 50ohm displayed noise level rise a lot of. Because 50 ohm resistor make possible to internal transmitted noise have now pathway to input BNC GND and this generated  noise current produce voltage over 1M.  But if input was open (example if use BNC metal cap without center connection) , less noise. But this was special case.

If scope have this cheap 50 ohm resistor method or if it have real 50ohm input impedance they may be quite different if think analog front end noise.

[David Hess]

Broadband noise of 10 nV/SqrtHz is about right.  But noise at low frequencies increases due to 1/f or flicker noise which can be very high for integrated CMOS designs which are common now in DSOs.  There is also additional low frequency Johnson noise from the parallel RC network in series with the input which provides overload protection.

[maxwell3e10]

Yes, it would be interesting to have another set of data that spans the frequency range 10 Hz to 1MHz to measure 1/f noise. This would require using lower sample rate, about 1 MHz.

[rsjsouza]

Interesting comparison. I can provide Rigol's DS4014 information. I am pretty sure it will be a lot higher than the average, but still worth taking it.

[Performa01]

Yes, it would be interesting to have another set of data that spans the frequency range 10 Hz to 1MHz to measure 1/f noise. This would require using lower sample rate, about 1 MHz.

The problem with a low sample rate like 1MSa/s is that all the noise above Nyquist (500kHz) will fold back and add to the result. This would severely distort the measurements, since a major part of noise energy would come from that folded back frequency range above 500kHz.

Even with 20MHz bandwidth limit, it would still be 20MHz vs. 500kHz noise bandwidth.

So my suggestion would be to use at least 100MSa/s in order to get a reasonable attenuation of the aliased frequency band by means of the 20MHz bandwidth limit.

How low do we need (or want) to go with such a measurement? For 10Hz, we'd need 10ms/div timebase and at least 10Mpts of data, which we could not handle over this forum. You'd need to provide some server/cloud where we could upload it anyway.

[nctnico]

How low do we need (or want) to go with such a measurement? For 10Hz, we'd need 10ms/div timebase and at least 10Mpts of data, which we could not handle over this forum. You'd need to provide some server/cloud where we could upload it anyway.

Wetransfer works excellent for sending big files to people.

[Keysight DanielBogdanoff]

Thanks for doing this! I'm also interested in how you measured noise. We generally use the smallest hardware V/div setting and measure V RMS (NOT Peak-Peak). This makes sure you don't have any issues with the quantity of acquisitions/update rate.

[rsjsouza]

Thanks for doing this! I'm also interested in how you measured noise. We generally use the smallest hardware V/div setting and measure V RMS (NOT Peak-Peak). This makes sure you don't have any issues with the quantity of acquisitions/update rate.

Interesting, Daniel. Have you ever seen any manufacturers skew or butcher RMS measurements that could lead to incorrect results?

[nctnico]

Thanks for doing this! I'm also interested in how you measured noise. We generally use the smallest hardware V/div setting and measure V RMS (NOT Peak-Peak). This makes sure you don't have any issues with the quantity of acquisitions/update rate.

I'm wondering if that is the best way of doing it on an oscilloscope which uses decimated data for on-screen measurements.  Also the RMS peak measurement may be affected by any DC offset in the signal. Using the actually sampled data and doing a noise spectrum analysis seems like a much better way to me.

[Rich@RohdeScopesUSA]

Thanks for doing this! I'm also interested in how you measured noise. We generally use the smallest hardware V/div setting and measure V RMS (NOT Peak-Peak). This makes sure you don't have any issues with the quantity of acquisitions/update rate.

I'm wondering if that is the best way of doing it on an oscilloscope which uses decimated data for on-screen measurements.  Also the RMS peak measurement may be affected by any DC offset in the signal. Using the actually sampled data and doing a noise spectrum analysis seems like a much better way to me.

We typically recommend the same method Daniel suggested (smallest volt/div setting, AC RMS or std deviation or a vertical histogram) with one addition (which I think Daniel would agree with) - you should really figure it as a percent of full scale as some oscilloscopes have 10 vertical divisions and some only have 8.  Of course the other key thing to keep in mind when measuring noise using this method is that noise is a function of BW.  So you may need to use filters to get an apples to apples compare.

-Rich

[David Hess]

you should really figure it as a percent of full scale as some oscilloscopes have 10 vertical divisions and some only have 8.

It is even more complicated than that.  The number of ADC counts per division varies between DSO designs and has little to do with the number of displayed divisions.  For instance 25 counts per division is (or was) common producing a 10.2 division "virtual" display of which either 10 or 8 vertical divisions might be visible.  My DSOs all use 5:4 displays so 8 vertical divisions with an extra division and then some at the top and bottom so clipping it usually outside the range of the display.

It should always be possible to convert to microvolts RMS within a given bandwidth and nV/SqrtHz versus frequency based on the noise measurement and vertical sensitivity.

[Performa01]

We typically recommend the same method Daniel suggested (smallest volt/div setting, AC RMS or std deviation or a vertical histogram) with one addition (which I think Daniel would agree with) - you should really figure it as a percent of full scale as some oscilloscopes have 10 vertical divisions and some only have 8.

This is why I usually express the noise in LSB instead of µV in my graphs.

Of course, for this you need to know how many LSB/div. a certain DSO has - and keep in mind that this number will be far less for the zoomed (not full resolution) gain settings that some scopes have.

[maxwell3e10]

Here is a compilation of noise spectra for several more oscilloscopes:
R&S RTB2004 thanks to Hydron
Instek GDS2204E thanks to nctnico
R&S RTM3004  thanks to nctnico
Rigol DS6104

I grouped them by the sampling rate, 1 GHz and 5 GHz. It looks like Rohde & Schwarz scopes have the lowest noise among the scopes tested so far.

[Andreas]

Hello,

how many samples are you actually using for calculation for the 1 GS/s sample rate?
the minimum calculated frequency seems to be ~500 kHz with 500 kHz step resolution.
So it seems that you are using 2048 points for spectrum calculation.
(that would be a number which can easily uploaded as CSV here against the 1MS limitation from above).

with best regards

Andreas

[Performa01]

Yes, for these graphs the data can easily be uploaded in this forum and that has been done in that other thread, linked in the opening post. Just have a look there.

The problem only arises when we shall provide data suitable for analysis down to 10Hz.

[nctnico]

It would be interesting to see what the new DSO/MSO7000 from Rigol looks like when it comes to noise.

[maxwell3e10]

how many samples are you actually using for calculation for the 1 GS/s sample rate?

I am using all the samples, about 1M, but I set the linewidth of the FFT to 1 MHz. One can use smaller linewidth, but then the scatter is bigger.

[Keysight DanielBogdanoff]

Thanks for doing this! I'm also interested in how you measured noise. We generally use the smallest hardware V/div setting and measure V RMS (NOT Peak-Peak). This makes sure you don't have any issues with the quantity of acquisitions/update rate.

I'm wondering if that is the best way of doing it on an oscilloscope which uses decimated data for on-screen measurements.  Also the RMS peak measurement may be affected by any DC offset in the signal. Using the actually sampled data and doing a noise spectrum analysis seems like a much better way to me.

We typically recommend the same method Daniel suggested (smallest volt/div setting, AC RMS or std deviation or a vertical histogram) with one addition (which I think Daniel would agree with) - you should really figure it as a percent of full scale as some oscilloscopes have 10 vertical divisions and some only have 8.  Of course the other key thing to keep in mind when measuring noise using this method is that noise is a function of BW.  So you may need to use filters to get an apples to apples compare.

-Rich

Everything I learned about measuring scope noise I learned from Rich, back before he went to the dark side 

I second this comment.

Interesting, Daniel. Have you ever seen any manufacturers skew or butcher RMS measurements that could lead to incorrect results?

What we've seen some manufacturers do is measure pk-pk and not have the same number of captures. Generally as your quantity of captures go up, so will your max pk-pk measurement.

[Rich@RohdeScopesUSA]

Everything I learned about measuring scope noise I learned from Rich, back before he went to the dark side 

  LOL

-Rich

[Andreas]

how many samples are you actually using for calculation for the 1 GS/s sample rate?

I am using all the samples, about 1M, but I set the linewidth of the FFT to 1 MHz. One can use smaller linewidth, but then the scatter is bigger.

Ok,
now I think I found a way: if I remove the time-stamps from the data (only the voltages) it compresses down below 1 MB.
Per default my output is in mV: is this a problem or should I scale it to Volts.

with best regards

Andreas

[maxwell3e10]

Instek scopes make the best csv files from space point of view. They save only the waveform data in bits, so its typically only a single integer number for shorted input. It compresses down to 147 kB for 1 M data points. The only problem is that the preamble doesn't specify the volts/bit conversion in a very clear way. Its easy enough to scale the y axis and make the x axis.

[nctnico]

Instek scopes make the best csv files from space point of view. They save only the waveform data in bits, so its typically only a single integer number for shorted input. It compresses down to 147 kB for 1 M data points. The only problem is that the preamble doesn't specify the volts/bit conversion in a very clear way. Its easy enough to scale the y axis and make the x axis.

Some clarification here: I choose the 'fast CSV file' method when creating the files on the GDS2204E. There is also a slower version which likely has the data in Volts but is also more bulky and probably takes more time to create.

[Andreas]

Hello,

first measurements of PicoScope 5444A
8-Bit Mode with 1 GS/s and 1 Ms of data.
first with 200 MHz (full bandwidth)
and 2nd with 20 MHz (hardware filter).

Best y-resolution is 2mV/div (+/-10mV)
so I should vote for the comparison relative to the range.
Due to the low y-resoution it will not compare to devices with 0.5mV/div.

20 fold Y-Zoom in the diagram shows that the step width is 110 uV/step
so for the +/-10 mV range only 70% are used for the display (28.3 mV total range.)

Measurements are done with 50 Ohms terminator on one input.

the 200 MHz measurement shows 125.7uV AC rms (std dev).
with 200 MHz bandwidth this computes to 8.9 nV/sqrt(Hz) in average.
the 20 MHz measurement is around 11.7 nV/sqrt(Hz). (mostly quantisation noise)

from the .CSV data the timestamps have been removed so that 1Ms of data could be packed within the forum upload limits.
Also the mV output of the scope has been converted to Volts.

Next will be 16 Bit measurements in 16 Bit mode of the scope.

with best regards

Andreas

[lordvader88]

can someone tell me and others that don't know, the basics here, whats the graph mean ? whats good/bad ?

[nctnico]

The easiest way is to send it by Wetransfer or another file sharing service. There is no need to jump through all kinds of hoops to attach a file to a forum message.

[Carrington]

The easiest way is to send it by Wetransfer or another file sharing service. There is no need to jump through all kinds of hoops to attach a file to a forum message.

Absolutely, edited deleted. I was bored, sorry. 

[rsjsouza]

can someone tell me and others that don't know, the basics here, whats the graph mean ? whats good/bad ?

lordvader88, there are many references around on the web.

A training:
https://training.ti.com/ti-precision-labs-adcs-high-speed-SNR-NSD

An article:
http://www.ni.com/white-paper/3304/en/

[Hagrid]

Hello,

first measurements of PicoScope 5444A
[...]

Thanks a lot! I was looking for exactly this data. Please someone correct me if I am wrong, but is the noise of the 5444 about five to ten times higher than the noise from the cheap Chinese scopes from the first page?
This makes little sense to me, since it has modes up to 16 bit resolution.

I would really appreciate it if somebody could explain this to me, thanks .

Greetings, Hagrid

[MrW0lf]

I would really appreciate it if somebody could explain this to me, thanks .

Picos just do not have low input ranges.

2000 ±20mv to ±20v
4262 ±10 mV to ±20 V
5000 ±10 mV to ±20 V

If think in divs then 2mV/div is lowest for 5000.

Is it a problem? Depends on application. When can average, do FFT etc then noise drops quite good:
https://www.eevblog.com/forum/testgear/fft-spectrum-analysis-reviewed/
https://www.eevblog.com/forum/testgear/cheap-chinese-ad584-voltage-reference-legit-cal-data-let_s-find-out!/msg1154930/#msg1154930
https://www.eevblog.com/forum/testgear/picoscope-2000/msg1155735/#msg1155735

[egonotto]

Hallo,

@Hagrid:
"I would really appreciate it if somebody could explain this to me, thanks"

You are right, the noise of the 5000 serie is not that good.
https://www.picotech.com/support/topic16031.html

But in some cases you can lower the noise.
https://www.eevblog.com/forum/testgear/looking-for-a-dynamic-signal-analyzer-with-extended-bandwidth/?all

You can limit the bandwith and lower the noise. To limit the bandwith with the 5000 series you have different ways.

From PicoScope 6 help:
"Resolution enhancement is a technique for increasing the effective vertical resolution of the scope at the expense of high-frequency detail. In some scope operating modes, PicoScope may reduce the number of samples available to maintain display performance.

For this technique to work, the signal must contain a very small amount of Gaussian noise, but for many practical applications this is generally supplied by the scope itself and the noise inherent in normal signals.

The resolution enhancement feature uses a flat moving-average filter. This acts as a lowpass filter with good step response characteristics and a very slow roll-off from the pass-band to the stop-band."

You can limit the bandwith to 20 MHz

With math you can average to lower the noise.

Best regards
egonotto

[Andreas]

With math you can average to lower the noise.

Hello,

you can also directly configure the low pass frequency of a software filter in the menu.

 Please someone correct me if I am wrong, but is the noise of the 5444 about five to ten times higher than the noise from the cheap Chinese scopes from the first page?
This makes little sense to me, since it has modes up to 16 bit resolution.

I guess with a 0.5mV/Div range and only 8 Div instead of 10  the factor of 5 can be explained.

At least for that what I measure I do not need high bandwidth and high resolution at the same time.
With a software filter of 1 MHz or below (besides the hardware 20 MHz filter) the noise decreases significantly.

with best regards

Andreas

[Andreas]

Hello,

and a 16 Bit noise measurement with .csv data (had to split it into 2 parts)
Again 2mV/Div 62.5 MS/s maximum sample rate in 16 Bit mode and with 20 MHz hardware bandwidth limiter.

Further Pictures:
Software low pass with 1 MHz, 50kHz, 1kHz additional to the 20 MHz.
and 20 Bit enhanced resolution out of 20 MHz.

with best regards

Andreas

[Andreas]

Hello,

and again 16 Bit values, this time without bandwidth limit.
the 16 Bit values are only 625000 samples because of upload limit.

with best regards

Andreas

[David Hess]

Thanks a lot! I was looking for exactly this data. Please someone correct me if I am wrong, but is the noise of the 5444 about five to ten times higher than the noise from the cheap Chinese scopes from the first page?
This makes little sense to me, since it has modes up to 16 bit resolution.

I would really appreciate it if somebody could explain this to me, thanks .

The Picoscope details I have seen show an integrated CMOS transimpedance buffer intended for DSOs from TI that by itself has 100 times the noise of a discrete front end at low frequencies.  I mean literally 1000nV/SqrtHz where a discrete design could be 10nV/SqrtHz.

So a broadband RMS noise of 125.7uV over 200MHz does not surprise me at all and that is about 5 times worse than my 40+ year old 200MHz analog oscilloscopes.

[Hagrid]

Thank you all for sharing this Information.

Thanks a lot! I was looking for exactly this data. Please someone correct me if I am wrong, but is the noise of the 5444 about five to ten times higher than the noise from the cheap Chinese scopes from the first page?
This makes little sense to me, since it has modes up to 16 bit resolution.

I would really appreciate it if somebody could explain this to me, thanks .

The Picoscope details I have seen show an integrated CMOS transimpedance buffer intended for DSOs from TI that by itself has 100 times the noise of a discrete front end at low frequencies.  I mean literally 1000nV/SqrtHz where a discrete design could be 10nV/SqrtHz.

So a broadband RMS noise of 125.7uV over 200MHz does not surprise me at all and that is about 5 times worse than my 40+ year old 200MHz analog oscilloscopes.

Very interesting. I really wonder why they made this design decision. In my opinion it looks like this contradics the whole purpose of a high resolution mode.

[MrW0lf]

Very interesting. I really wonder why they made this design decision. In my opinion it looks like this contradics the whole purpose of a high resolution mode.

Maybe it's related to their main business background - automotive. High resolution is not only for looking at very low voltages. Would be useful with x100 probe etc.

[nctnico]

Very interesting. I really wonder why they made this design decision. In my opinion it looks like this contradics the whole purpose of a high resolution mode.

Maybe it's related to their main business background - automotive. High resolution is not only for looking at very low voltages. Would be useful with x100 probe etc.

The problem is that the input attenuator will attenuate the signals before they hit the amplifier so you'll get the noise also at lower sensitivities. A high resolution ADC doesn't make sense with that much noise. You also have to read the datasheet for the Picoscopes very carefully. On some models the high resolution mode is nothing more than oversampling an 8 bit ADC while other models do have an ADC with more bits.

[MrW0lf]

The problem is that the input attenuator will attenuate the signals before they hit the amplifier so you'll get the noise also at lower sensitivities. A high resolution ADC doesn't make sense with that much noise.

Cannot agree, it does makes very much sense, just as stated, when poking LV & HV:

Reference, x100 setting, 100V/div, input shorted with 50Ω:

PicoScope 2205, 195.3kSa/s
8bit AC RMS = 1100mV
12bit (software) AC RMS = 80mV

PicoScope 2408B, 125MSa/s
8bit AC RMS = 2400mV
12bit (software)  AC RMS = 250mV

Analog Discovery 2, 320kSa/s
14bit decimate AC RMS = 990mV
14bit average (hardware*) AC RMS = 250mV

*ADC is sampling at full speed (100MSa/s) in background

Pico TA041 probe at x100 setting, 100V/div, 10x H/V zoom windows, observing mains:

PicoScope 2205:



8bit trace fairly stepped, 12bit (software) trace deforms shape due to bw hit. Instrument not very suitable for the task.

PicoScope 2408B:



Much higher sample rate & analog bw, 8bit trace drowned in noise. 12bit (software) processing works because of much larger recordset hiding bw hit. However response becomes a bit sluggish because PC has to crunch the numbers.

Analog Discovery 2:



Here can see why 14bit hw is superior despite very poor AC RMS noise performance (see reference). Darker green trace is decimated one. Noise is visible but UI response is very good because of native 14bit and no software processing. When turn on hardware averaging (bright green) trace clears up w/o noticeable performance hit.

So can conclude that native high bit scopes are very useful, even if their frontend is not low noise.

[Performa01]

Please someone correct me if I am wrong, but is the noise of the 5444 about five to ten times higher than the noise from the cheap Chinese scopes from the first page?
This makes little sense to me, since it has modes up to 16 bit resolution.
 

Well, these tests have been with just 8 bits resolution.

We need to be cautious to not compare apples with oranges. I have used the best of the “cheap Chinese scopes” and tried hard to resemble the test scenario for the PicoScope 5444 as close as possible, i.e. 1GSa/s, 1.4Mpts, 100µs/div and 2mV/div:


Siglent SDS1104X-E_100µs_2mV_100MHz_T50

Yes we get lower noise (63µV), but bandwidth is only 110MHz for this scope. The average noise density over the full bandwidth is about 6nV/sqrt(Hz) and I cannot see how this could be 5-10 times better than the ~9nV/sqrt(Hz) of the Pico 5444.

For the true high resolution models, like the 4262, we get much lower noise (4.33µV) because of the limited bandwidth, but also a lower noise density of <2nV/sqrt(Hz) at 2mV/div (zoomed to 100µV/div):


Pico 4262_Noise_10ms_2mV_5MHz_T50_Z100µV

The noise spectrum of the 4262 from 5Hz – 5MHz captured at a RBW of 14Hz looks like this:


Pico 4262_Noise_10ms_2mV_5MHz_T50_FFT

1/f noise is not very pronounced down to 50kHz and noise floor is below -156dBV (16nV) at 50kHz and even lower above. The strongest spurious signal measures -149.5dBV (32nV).

A closer look at the low frequency spectrum finally shows 1/f noise quite clearly:


Pico 4262_Noise_10ms_2mV_5MHz_T50_FFT_Z50kHz

We can see -156.5dBv (15nV) at 50kHz and -135.7dBV (164nV) at 100Hz. Given the FFT bin width of 4.768Hz and the flat-top window, we can expect a noise bandwidth of some 14Hz for the measurement.

This would result in 4nV/sqrt(Hz) at 50kHz and 44nV/ sqrt(Hz) at 100Hz, yet all these figures have to be taken with a grain of salt, because even at 5MHz, where we get -160dBV (10nV) the calculated noise density would still be ~2. 4nV/sqrt(Hz), which does not conform to the total noise density of <2nV/sqrt(Hz) as calculated earlier. I would tend to rather trust the FFT and believe that the RMS measurement in y-t mode might have given a slightly low reading.

[maxwell3e10]

Here is the comparison of PicoScope modes from Andreas as well as Rigol DS4014 data I got from rsjsouza.

As one might expect, at low mV/div setting the noise does not depend on the bit resolution, it is simply limited by the input amplifier noise.  The advantage of high resolution modes is that they would give the same noise even at higher mV/div setting (for example Owon 12 bit scope has similar noise on 5mv/div setting).

Also, small details, like how many divisions are in the full screen, do not matter when noise is limited by the input amplifier.

The 16 bit 200 MHz BW noise is higher due to aliasing with 62.5 MHz sample rate. I am surprised it is even allowed by the firmware.

[_Wim_]

Hi Maxwell3e10 and Performa01. Thanks for the plots above. Very interesting!

I am surprised it is even allowed by the firmware.

Yes, I am surprised about this to, I always thought that for reduced sample rates a corresponding hardware aliasing filter would have been activated. If only a single 20MHz filter is available, then only the 12-bit mode makes sense for the 200Mhz model.
Makes me wonder how they do the bandwidth limiting on the lower bandwidth models like mine? I always thought it was done in hardware, but seeing this that makes no sense. If it done in software, I normally should be able to create aliasing effect on the higher resolution modes. I will give that a try later today.


I wonder now if this is common on other scopes also. If four example you enable all channels, samples rate drops below nyquist for the lower end scope (ex. "upgraded" Rigol ds1054z has only 125MS/s for 100Mhz bandwidth"

[_Wim_]

Ok, done quick a few tests. Setup: CMU200 signal generator connected to Pico 5442B with 50 ohm input adaptor.

First: 8 bit mode (1GS/s), my scope has 60Mhz bandwidth

* reference signal at 5Mhz (500mV RMS)
* -3db found at 87Mhz
* -6db at 200Mhz
* -10db at 300Mhz
* -10db also at 400Mhz
* when switching to 200mV/div sensitivity, signal drops completely (this path has much lower bandwidth)

Second: 16-bit mode (62.5MS/s)

* reference signal at 5Mhz (500mV RMS)
* at 57.5 Mhz we see a 5 Mhz alias only 3db down (no bandwidth limit)
* at 57.5 Mhz with 20Mhz bandwidth limit we see a 5Mhz alias 10db down => 20Mhz bandwidth limit only 6db/oct or less filter slope?

So indeed much higher bandwidth noise can come back into the spectrum under test, and the software should at least active the 20Mhz bandwidth limit automatically when running at 62.5MS/s only.

[MrW0lf]

So indeed much higher bandwidth noise can come back into the spectrum under test, and the software should at least active the 20Mhz bandwidth limit automatically when running at 62.5MS/s only.

And what if you want to actually use alias?
The problem cannot be handled properly with 20MHz limit anyway so better leave it alone.
AFAIK they disabled FFT plot > official analog bw not to confuse dumbusers. So now you cannot look 101MHz peak on -3dB@~140MHz scope etc

[_Wim_]

So now you cannot look 101MHz peak on -3dB@~140MHz scope etc

Why not, you know were it will fold back to, so you still can but just need to calculate a bit...

[MrW0lf]

Why not, you know were it will fold back to, so you still can but just need to calculate a bit...

...with scope running at 1GSa/s? I would sort of get it if they handicapped dedicated FFT only which has auto control over sampling rate, but no. It's also imposed on FFT viewport. Also FFT is completely disabled for ETS So sorry but these boxes already have enough of dumb (or perhaps marketing derived?) limitations.

[_Wim_]

...with scope running at 1GSa/s? I would sort of get it if they handicapped dedicated FFT only which has auto control over sampling rate, but no. It's also imposed on FFT viewport. Also FFT is completely disabled for ETS So sorry but these boxes already have enough of dumb (or perhaps marketing derived?) limitations.

Ok, but now you are suggesting they deliver you a 500Mhz product for the price of a 60Mhz product, I am off course also a fan of that (but would be bad marketing).

I think almost all base scopes  are just a software cripled version of the high bandwidth model of the same series, this is not only so for pico, but is the case for all brands.

[_Wim_]

Here is the frequency response of the 20 Mhz BW Limit filter. Setup: Pico FFT with peak hold, CMU200 sweep frequencies between 10Mhz and 60Mhz.

[MrW0lf]

Ok, but now you are suggesting they deliver you a 500Mhz product for the price of a 60Mhz product, I am off course also a fan of that (but would be bad marketing).

But it is not the case. Scope still has unhackable analog bandwidth and FFT plot would drop accordingly. I know of no other (low end) scopes where FFT is artificially crippled like that. AD2 will honestly go to SR/2 so all the others I know of.
When using ref compensation one can flatten the plot nicely, but it would be still limited to reduced resolution according to the actual analog bw. So it's not like you suddenly getting more than you paid for. In Pico & FFT case you get less what you paid for if compare to other scopes. But thats OT - it's just a little soft spot suppose - I've hassled a bit too much to overcome various artificial limitations is PS6. To bad it does not have in-GUI scripting like AD2 which is real limitation-killer. Going SDK for each little hack is cumbersome.

Edit: So sorry if it did seem I jumped on you, it just "do not give them any more ideas what to limit" reaction

[srce]

Here are some results for a Keysight MSOS204A.

Vertical: 1mV/div - 8 divisions - 10bit
Horizontal: 5uS/div - 20GSa/s - 1Mpts

Yellow is 20MHz B/W (s/w filter) - 50 Ohm - 44uV AC RMS - 350uVpp
Blue is 2.1GHz B/W - 50 Ohm - 115uV AC RMS - 1.1mVpp
50 Ohm terminators on BNC inputs.

2GHz data 50 Ohm

20MHz data 50 Ohm

[maxwell3e10]

Thanks, srce, I was looking for data from a high-end Keysight scope. Are these with 50 Ohm impedance, as the screen shot says?

Could you also take data for 1 MOhm setting to compare.

[srce]

Yellow is 20MHz B/W (s/w filter) - 1MOhm - 140uV AC RMS - 1mVpp  - 1mV/div

Blue is 500MHz B/W (s/w filter) - 1MOhm - 355uV AC RMS - 3.3mVpp - 5mV/div (note b/w is limited to 200MHz at 1mV/div)

20MHz 1MOhm data

500MHz 1MOhm data

[rsjsouza]

maxwell3e10, thanks for adding the DS4014 to the graph. It seems a bit noisier than some of the other models (although the R&S seems unbeatable).

I forgot that you were measuring other modes as well. I can add five more measurements to the mix:

BW limit	Impedance
20MHz	50Ω
200MHz	1MΩ and external 50Ω terminator
200MHz	50Ω
500MHz (no limit)	1MΩ and external 50Ω terminator
500MHz (no limit)	50Ω

[egonotto]

Hallo,

@_Wim_: "Makes me wonder how they do the bandwidth limiting on the lower bandwidth models like mine?"

My PicoScope 100MHz 5243A has no bandwidth limit at all.

It has full 200MHz.

I ask this in the picotech forum:

https://www.picotech.com/support/topic33731.html

They don't explain what is matter.

Best regards
egonotto

[_Wim_]

But it is not the case. Scope still has unhackable analog bandwidth and FFT plot would drop accordingly. I know of no other (low end) scopes where FFT is artificially crippled like that. AD2 will honestly go to SR/2 so all the others I know of.
When using ref compensation one can flatten the plot nicely, but it would be still limited to reduced resolution according to the actual analog bw. So it's not like you suddenly getting more than you paid for. In Pico & FFT case you get less what you paid for if compare to other scopes. But thats OT - it's just a little soft spot suppose - I've hassled a bit too much to overcome various artificial limitations is PS6. To bad it does not have in-GUI scripting like AD2 which is real limitation-killer. Going SDK for each little hack is cumbersome.

Edit: So sorry if it did seem I jumped on you, it just "do not give them any more ideas what to limit" reaction

Ok, this makes indeed more sense. I also agree the software could use an update. I still like the pico overall, and have used it quite a bit. So far I did not use the SDK at all, but it is still nice to know this is an option.

[_Wim_]

My PicoScope 100MHz 5243A has no bandwidth limit at all.
It has full 200MHz.

That's nice!. I am also happy with only -6db @200Mhz for my 60Mhz scope. I am still wondering if there is actually a hardware difference between the scopes of the 5000 series (with same channel count).

[maxwell3e10]

Here is the plot for the Keysight MSOS204A. It is actually not that great compared with R&S scopes. At low frequencies, its even worse than low-cost scopes! I am surprised the noise is so poor on their "high definition" 10 bit scope.

Does anyone have any other high-end Keysight scopes to compare?

[Andreas]

Does anyone have any other high-end Keysight scopes to compare?

Unfortunately not; only a Tek 4104 MSO (from work).

The device has 1 GHz bandwidth and 5 GS/s maximum sample rate.
so this gives 20us/div * 10 div = 200 us for 1 MS on the x-axis.
the most sensitive range is 1mV/Div * 10 Div so +/-5mV. @ 8 Bit giving 40uV/LSB.

First set of measurements is done with internal 50 Ohms input.
with full 1 GHz BW and  250 MHz /20 MHz limited band width.

Interestingly the noise is much smaller with the 50 Ohms input (int) than with a external 50 Ohms terminator (ext).
(see next post).
So it seems that there are different signal paths for the 1 Meg and the 50 Ohms input resistance.

with best regards

Andreas

[Andreas]

And now the measurements with external 50 Ohms terminator. (End cap).

with best regards

Andreas

by the way: the RMS-value on the scope measurement seems to be including DC-Offset
If I calculate standard deviation the values are much lesser.

internal 50 Ohms:
1 GHz bandwidth      SD = 90 uV
250 MHz bandwidth  SD = 68 uV
20 MHz bandwidth    SD = 29 uV

external 50 Ohms: (1Meg input).
1 GHz bandwidth      SD = 136 uV
250 MHz bandwidth  SD = 108 uV
20 MHz bandwidth    SD =  46 uV

[maxwell3e10]

Here is a comparison of the 1 GHz+ scopes that I have so far. For full bandwidth I used 50 Ohm input, for 20 MHz bandwidth - 1 Mohm input. So far Rohde&Schwarz still looks the best.

[_Wim_]

Hi, could it be that in the above graph for the RTM3004 the 20Mhz curve is for 1M input, and the 1Ghz one is for 50 ohm? If not, why is the low frequency noise higher with the 20Mhz limit enabled? 

I should read also, and not only look at the picture... 

[JPortici]

is it me or is the 20 MHz trace for the keysight showing a digital filter?

[nctnico]

is it me or is the 20 MHz trace for the keysight showing a digital filter?

Well spotted and I agree that it looks like a simple digital filter.  y[n]= (y[n-1] * (a-1) + x[n]) / a.

[EEVblog]

New Rigol 5000 & 7000, 20MHz and full bandwidth

https://www.dropbox.com/s/hxzjzidrofd9eli/Rigol5000-Noise.zip?dl=0

[BravoV]

To OP, suggesting to update the 1st post with all accumulated result, rather than piling up at new post that is hard to keep up. Just add change list as a note.

[maxwell3e10]

Dave, thanks for posting. I looked at the files, they are taken with 200 MHz sampling rate, so the noise is pretty high due to aliasing. Could you take data at the maximum sampling rate of the scope, as the other ones are.

[rf-loop]

Dave, thanks for posting. I looked at the files, they are taken with 200 MHz sampling rate, so the noise is pretty high due to aliasing. Could you take data at the maximum sampling rate of the scope, as the other ones are.

Is it possible that your suspected aliasing rise this noise to this enormous level. Also there is 20MHz front end BW on.
(!) 163.9 uV stdev (AC-rms) and over 1.6mVp-p. (Dave's 5000-1-20.csv)


Is it made using 4mV/div?  (most low true full resolution)

Vertical resolution in file is  135.4uV

Rigol: Note[3]: 1 mV/div and 2 mV/div are a magnification of 4 mV/div setting. For vertical accuracy calculations, use full scale of 32 mV for 1 mV/div and 2 mV/div sensitivity setting.

[srce]

is it me or is the 20 MHz trace for the keysight showing a digital filter?

Well spotted and I agree that it looks like a simple digital filter.  y[n]= (y[n-1] * (a-1) + x[n]) / a.

You can choose between 20MHz or 200MHz h/w filter or s/w with any bandwidth. That was indeed with s/w filter.

One other point to bear in mind is that the scope h/w does actually have >8GHz of b/w. Mine is just s/w limited to 2GHz.

[2N3055]

Keysight MSOX3104T (1GHz 5/GSa/sec) and Picoscope 3406D (200MHz 1/GSa/sec)

50 Ohm terminator on input.

https://drive.google.com/open?id=1kZEO_hKSXGY_-ScaOQO3ydRLsjV2okpD

Regards,
Sinisa

[maxwell3e10]

It looks like the noise on 50 Ohm setting and 1 MOhm setting is identical for Keysight MSOX3104T. Just want to make sure its not a mix-up of files.

[maxwell3e10]

One other point to bear in mind is that the scope h/w does actually have >8GHz of b/w. Mine is just s/w limited to 2GHz.

This brings an interesting general point with the recent trend of software-upgradable scopes. When a scope is software limited in the maximum sampling rate and bandwidth, does it typically run the ADC at the maximum rate and then averages samples? Does it have a switchable analog front-end  filter?  It needs some type of anti-aliasing filter, both for noise purposes and to avoid signal aliasing.

[2N3055]

I checked, Std Dev on those files shows lower RMS noise on 50 Ohm.  Cca. 326uV RMS vs 343uV for 1GHz bandwidth (50 vs 1M) and 112uV vs 176uV for 20MHz bandlimited. I believe they should be right.

[srce]

One other point to bear in mind is that the scope h/w does actually have >8GHz of b/w. Mine is just s/w limited to 2GHz.

This brings an interesting general point with the recent trend of software-upgradable scopes. When a scope is software limited in the maximum sampling rate and bandwidth, does it typically run the ADC at the maximum rate and then averages samples? Does it have a switchable analog front-end  filter?  It needs some type of anti-aliasing filter, both for noise purposes and to avoid signal aliasing.

For the S series the ADCs still run at 20GSa/s (with downsampling being selectable) - I think it's just a s/w low pass filter (similar to what a user can select for any lower frequency).

[srce]

MSOX2024A - 200MHz - 2GSa/s - 1mV/div - 1MOhm

200MHz - 1.4mVpp - 174uV AC RMS

20MHz - 800uVpp - 119uV AC RMS

200MHz data

20MHz data

[rf-loop]

MSOX2024A - 200MHz - 2GSa/s - 1mV/div - 1MOhm

200MHz - 1.4mVpp - 174uV AC RMS

20MHz - 800uVpp - 119uV AC RMS

200MHz data

20MHz data

My calculator give from your 200MHz data  168 Sdev (AC RMS) and 1.45mVpp from full data
But as we know Keysight (eta: some X models) use somehow decimated data for measurements when modern scopes use full acq mem data.

But then, also 1mV/div is not its most sensitive full resolution sensitivity. It is digitally magnified from 4mV/div what also is its most sensitive full resolution V/div. Perhaps it show near same noise with 4mV/div setting.
(it have only FullScale/64 resolution using 1mV/div aka 6 bit resolution)
Example if I set Siglent for 5mV/div and other ways comparable setting it have 134uV Sdev. But if use 1mV/div (also full resolution) it show 79uV Sdev
This is why I suspect that it (Keyshit) can not compare using 1mV/div to scope what have full resolution 1mV/div, least it is questionable.

[srce]

My calculator give from your 200MHz data  168 Sdev (AC RMS) and 1.45mVpp from full data
But as we know Keysight use somehow decimated data for measurements when modern scopes use full acq mem data.

It depends on the scope, I think. The X uses what is on screen - the S can use all data.

Perhaps it show near same noise with 4mV/div setting.

Yes, it does.

[maxwell3e10]

Here is a comparison of Keysight MSOX2024 and MSOX3104. They have very similar noise performance. Another peculiar thing is that MSOX3104 has virtually identical noise on the 1 MOhm and 50 Ohm setting. I guess it means that the scope doesn't have a different front end amplifier, just an internal 50 Ohm resistor switched across the input.

[TheSteve]

On the Keysight 3104 the 50 ohm path is a different input on the front end ASIC.

[2N3055]

On the Keysight 3104 the 50 ohm path is a different input on the front end ASIC.

As I stated above, there is visble change in RMS noise level when changing from 50 to 1M , but it is not great...

I will be traveling for few days. When I get back I will repeat measurements.
I would like to see somebody's else 3104T noise figures...
Maybe my 3104 is noisy.

[TheSteve]

I can post mine but it will need a big star next to it as it's a MSOX3024t modified to 1+ GHz.

[ci11]

Teledyne LeCroy HDO4032 12-bit 350MHz, no averaging, no ERES.

DC50 Ohm

 20MHz                   62,90 µVrms   312,44 µVpp
 350MHz               124,37 µVrms   847,66 µVpp

DC1M Ohm

 20MHz                   83,95 µVrms   313,00 µVpp
 350MHz               165,33 µVrms   899,47 µVpp

DC50 Ohm data:
 
 20MHz BwL DC50 Ohm
 350MHz DC50 Ohm

DC1M Ohm data:

 20MHz BwL DC1MOhm
 350MHz DC1MOhm

A couple of interesting points from the FFTs: the 20MHz BwL seems very gentle but still effective as can be seen in the numbers, and for a 350MHz scope, it appears to have BW well beyond 1GHz. Perhaps the folks at LeCroy do not believe in brickwall filters.

[rf-loop]

Here is a comparison of Keysight MSOX2024 and MSOX3104. They have very similar noise performance. Another peculiar thing is that MSOX3104 has virtually identical noise on the 1 MOhm and 50 Ohm setting. I guess it means that the scope doesn't have a different front end amplifier, just an internal 50 Ohm resistor switched across the input.

Of course these graphs and tables and curves are nice and useful.
Btv, is it possible to keep all graphs vertical scale and position in scale same for more quick and easy compare without thinking. 

But then also, what we do IRL. We look oscilloscope screens and it also matters what we really see (and measure).

Like these:
first @srce previously published  MSOX2024A - 200MHz - 2GSa/s - 1mV/div - 1MOhm
then same settings with SDS1204X-E. same BW, 1M input, 5us/div  but samplerate 1GSa/s (and memory of course due to samplerate and time scale "equal")

Not very easy imagine what is visible on the screen if only look V/(Hz^1/2) graphs.

[maxwell3e10]

Here is a comparison of the LeCroy HDO scope with the R&S RTM scope. On the LeCroy scope the noise is identical for 1 MOhm and 50 Ohm paths.

[ci11]

@maxwell3e10

Thanks for the comparison - would you mind adding the RTM 20MHz BW plots to the LeCroy HDO4032 vs.RTM 3004 chart?

[JPortici]

what's that peak? at.... 625MHz? only with 1Meg?

[ci11]

what's that peak? at.... 625MHz? only with 1Meg?

Sorry - cannot tell you. Same peak at 625MHz shows up in both 50R and 1M FFTs. Probably something LeCroy put in to deter overusing the BW actually paid for, but keep the roll-off gentle to reduce outband noise.

[JPortici]

My first guess is that it's correlated with the sampling clock (2.5GS/s -> 4 interleaved ADCs at 625MS/s)
that wouldn't be so strange but geez.. i would have expected better.
Shall i power up my ancient boat anchor and try to get some data, just for kicks ?

[ci11]

My first guess is that it's correlated with the sampling clock (2.5GS/s -> 4 interleaved ADCs at 625MS/s)
that wouldn't be so strange but geez.. i would have expected better.
Shall i power up my ancient boat anchor and try to get some data, just for kicks ?

That thought did not come to mind since this HDO is supposed to use a single TI ADC12D1600 per channel, for dual 1.6 GS/s or interleaved 3.2 GS/s. 625MHz is not an easy guess. But perhaps you're right - only LeCroy knows.

[JPortici]

mine is just a guess a peak at a hint right of 600MHz suggest 625 which is a divider of 1250 and 2500, the latter being a very common sampling frequency for oscilloscopes.
I don't know about the HDO internals unfortunately

EDIT: I looked at the HDO4032 datasheet: Sample Rate (Single-shot) 2.5 GS/s

[_Wim_]

Today I was measuring a high frequency spectrum with my Picoscope 5442B, and started wondering the following:

According to the datasheet the 5443B and 5444B have a better time base accuracy than my 5442B (50ppm for the 5442B, and only 2ppm for the higher end models). When I run a high bandwidth FFT, I always see some spurs around 33Mhz (see attached screenshots with shorted input). These are the worst in 12bit mode, and get better in 15 and 16 bit mode (due to paralleling of the ADC's?). I was wondering if these are simular on a 5443/5444 model or if these are related to the lower quality clock?

[Performa01]

FFT spurs are not related to time base accuracy.
The close-in phase noise is related to time base stability, but certainly not visible in a wideband spectrum plot without a signal.

The reason for spurs with a signal applied are non-linearities of the ADC. But without any signal, the non-linearities are pretty much irrelevant and all spurs can be attributed to internal interference signals.

Since the Pico 5000 series use some tricky configurations of the ADC cores in order to get the various resolutions, the 12-bit mode obviously generates more interference than other modes.

By the way, here you can see what a true 16 bit ADC in a Pico 4262 (which is optimized for low noise and low distortion) can do with regard to noise floor and spur level (with the same 50ppm time base tolerance!):


Pico_4262_FFT_Noise_50Ohm_term

[_Wim_]

FFT spurs are not related to time base accuracy.
The close-in phase noise is related to time base stability, but certainly not visible in a wideband spectrum plot without a signal.

The reason for spurs with a signal applied are non-linearities of the ADC. But without any signal, the non-linearities are pretty much irrelevant and all spurs can be attributed to internal interference signals.

Since the Pico 5000 series use some tricky configurations of the ADC cores in order to get the various resolutions, the 12-bit mode obviously generates more interference than other modes.

By the way, here you can see what a true 16 bit ADC in a Pico 4262 (which is oiptimized for low noise and low distortion) can do with regard to noise floor and spur level (with the same 50ppm time base tolerance!):

Thanks for this plot and the explanation. If only the 4262 would have had more bandwidth! Sometimes I do which I had bought that one, but the additional bandwidth made me choose the 5442B.
I agree this is related to the trickery, but do not understand why the 12bit mode is worse, because that sample rate is supported directly by the ADC without trickery (unless the ADC does some internal trickery off course).

[Performa01]

I agree this is related to the trickery, but do not understand why the 12bit mode is worse, because that sample rate is supported directly by the ADC without trickery (unless the ADC does some internal trickery off course).

It could be different signal paths for clock and input signals (even on the chip) with slightly more crosstalk in 12-bit mode?

-110 dBu isn't a very high level after all (<2.5µVrms or <7µVpp). In the +/-10mV range, this is equivalent to some -80dBFS. PicoTech specifies only 60dB SFDR (but that is irrelevant without a signal) and 110µVrms noise in 12-bit mode. So there is no specification for spurs without a signal, but I guess the -80dBFS is simply the performance limit of a 5000 series PicoScope.

[thanasisk]

Would be very very interesting to plot the low frequency input noise density  of the scopes, i.e. dc to 500 kHz. Is this possible  already with the same measurement files?

[Performa01]

Would be very very interesting to plot the low frequency input noise density  of the scopes, i.e. dc to 500 kHz. Is this possible  already with the same measurement files?

No it isn't. We need limit the record length - hence the sample rate - or the files would get huge.

Because of the the split-path design of the input buffer (and the FET amplifiers required for this), the noise in the DC/LF path is very high. So the noise plot below some 10kHz will most likely be ugly for just about any reasonably modern standard scope.

[nctnico]

Would be very very interesting to plot the low frequency input noise density  of the scopes, i.e. dc to 500 kHz. Is this possible  already with the same measurement files?

No it isn't. We need limit the record length - hence the sample rate - or the files would get huge.

Because of the the split-path design of the input buffer (and the FET amplifiers required for this), the noise in the DC/LF path is very high. So the noise plot below some 10kHz will most likely be ugly for just about any reasonably modern standard scope.

But this would need to be measured with the bandwidth limit on and the samplerate needs to be 100Ms/s or more otherwise the HF noise will alias into the lower frequencies.

[egonotto]

Hello,

I have on my 5243A and 5444B strong spur at 31.25 MHz.


Best regards
egonotto

[_Wim_]

I have on my 5243A and 5444B strong spur at 31.25 MHz.

Just zoomed in, exactly 31.25Mhz here also

[egonotto]

Hello,

31.25MHz * 4 = 125MHz
Perhaps this has to do with the clock frequency of the ADC



Best regards
egonotto

[_Wim_]

Hello,

31.25MHz * 4 = 125MHz
Perhaps this has to do with the clock frequency of the ADC



Best regards
egonotto

If that is the case, these spurs should move to a different frequency when you run an FFT with 200MHz bandwidth, because I expect the sample rate will then be 400MS/s, which is not dividable by 31.25. Or does is sample at 500MS/s and throws away a bit of the FFT?

Could you post a plot to 200MHz to check this with your 5444B?

[Performa01]

If that is the case, these spurs should move to a different frequency when you run an FFT with 200MHz bandwidth, because I expect the sample rate will then be 400MS/s, which is not dividable by 31.25. Or does is sample at 500MS/s and throws away a bit of the FFT?

Could you post a plot to 200MHz to check this with your 5444B?

PicoTech have pioneered the fully digital trigger system, which requires the ADC to run at full sample rate all the time. This sample data stream is fed into the digital trigger system. Only after that, the sample data get decimated if required to fit shorter record buffer lengths, so we get the effective sample rate at this point.

In other DSOs, where the FFT is not a completely independent operating mode like it is for PicoScopes, one additional decimation step is required whenever the max. FFT length is less than the record length.

Apart from that, the old 3206B samples at 500MSa/s and the FFT shows quite a few strong spurs, starting at 31.25MHz again. That’s probably just the frequency of the clock generator signal, which is then multiplied by a PLL in order to get the required ADC sample clock.

See the noise plot from 1kHz to 200MHz:


Pico 3206B FFT 50Ohm 1kHz-200MHz

EDIT: Please keep in mind that this plot is not directly comparable to the 4000 and 5000 series, because the most sensitive input range is +/-50mV (10mV/div) for the 3206B.

[_Wim_]

I sniffed a bit around internally with a DIY near field probe, but could not locate any strong 31.25MHz interference signal. The closest I could find was the 1.5Vdc switchmode regulated which had a spur at 31.20MHz, but also at around 29, 27... that were as strong, and we see only the one at 31.25MHz.

I also enabled all four channels (so only 1 ADC per channel running at 125MS/s, the spur at 31.2MHz remained. So it seems the ADC must do some interleaving internally for the 12 bit mode, but the datasheet is a bit confusing about this topic:

The high speed modes all utilize interleaving to achieve high sampling speed. Quad channel mode interleaves 2 ADC branches, dual channel mode interleaves
4 ADC branches, while single channel mode interleave all 8 ADC branches. In precision mode interleaving is not
required and each ADC channel uses one ADC branch only.

So it seems ADC1 consists of 2 ADCs, and only one of them is used is precision mode (14 bit). This also corresponds with the reduction in spurs in 14 bit.

[jpb]

I am interested in these Picoscope results as I'm thinking of getting the 4262 for ADEV and phase noise type measurements but would like more than 2 channels if possible.

I have a couple of questions.

Looking at the data sheet for the 5000 series, it looks as if 16bit mode only works with 1 channel - is this correct? (i.e. if you're in 16 bit mode you can only have one out of four channels operational) If it is correct, then for 16 bit (below 5MHz) you can't get more than 2 channels (4262) anyway.

Secondly, the relatively noisy plots you posted _Wim_ are labelled "shorted" while the plots you posted Performa01 are labelled as being terminated in 50 ohms. The input impedance of the scopes is 10M not 50 ohms but I wondered if this made a difference or perhaps "shorted" just means a 50 ohm terminator anyway.

[2N3055]

3406D MSO 500MSa/s from 1kHz to 200MHz 50 Ohm terminator on input (I didn't have a short handy)

[Performa01]

Looking at the data sheet for the 5000 series, it looks as if 16bit mode only works with 1 channel - is this correct? (i.e. if you're in 16 bit mode you can only have one out of four channels operational) If it is correct, then for 16 bit (below 5MHz) you can't get more than 2 channels (4262) anyway.

Yes. You can get more than 2 channels only in 14 bit mode. But then, the 15 and 16 bit modes appear to be extreme trickery and I don’t see a real advantage using them.

I think that the 4262 is rather unique and cannot be compared to the 5000 series scopes in any way.

Just look at the dynamic performance specifications in the data sheet to get the idea.

                     5000 series         4262
Crosstalk            >400:1              >50000:1
Harmonic Distortion  -70 dB (≥12 bit)    -95 dB
SFDR                  70 dB (≥14 bit)     96 dB
Noise                 70 µV RMS (16 bit)  8.5 µV RMS

Secondly, the relatively noisy plots you posted _Wim_ are labelled "shorted" while the plots you posted Performa01 are labelled as being terminated in 50 ohms. The input impedance of the scopes is 10M not 50 ohms but I wondered if this made a difference or perhaps "shorted" just means a 50 ohm terminator anyway.

Input short or 50ohm terminator don’t make a difference at that level.

3406D MSO 500MSa/s from 1kHz to 200MHz 50 Ohm terminator on input (I didn't have a short handy)

Yes, this is much improved over the older 3000 series. Higher input sensitivity, switchable 20MHz bandwidth limit, twice the ADC sample rate…

Interesting to see the much reduced spurious signals and the spur at 31.25MHz is missing altogether. Probably because of the doubled ADC speed, whereas the various physical sampling speeds in the 5000 series might have different clocking requirements.

[_Wim_]

Looking at the data sheet for the 5000 series, it looks as if 16bit mode only works with 1 channel - is this correct?

Yes, but 16 bit mode makes no real sense for this scope. Above 14 bit there is no performance increase noise floor wise, and give you already 84 db theoretical, with an SFDR or only 70dB, this is more than sufficient. In 14 bit you can use all 4 channels up to 60Mhz. But as stated above, the 4262 has much higher performance than the 5000, but lacks the bandwidth. Depending on your needs, this can be important or not.

Secondly, the relatively noisy plots you posted _Wim_ are labelled "shorted" while the plots you posted Performa01 are labelled as being terminated in 50 ohms. The input impedance of the scopes is 10M not 50 ohms but I wondered if this made a difference or perhaps "shorted" just means a 50 ohm terminator anyway.

I was using a short, not 50 ohm terminator. I have use a 50 ohm terminator in the past also, but result were indentical.

[_Wim_]

[Yes, this is much improved over the older 3000 series. Higher input sensitivity, switchable 20MHz bandwidth limit, twice the ADC sample rate…

Interesting to see the much reduced spurious signals and the spur at 31.25MHz is missing altogether. Probably because of the doubled ADC speed, whereas the various physical sampling speeds in the 5000 series might have different clocking requirements.

Indeed, the 3406D perform quite wel, I am impressed. Still dreaming of an 4262 with the bandwidth of the 5000 series (that is also affordable    )

[jpb]

Still dreaming of an 4262 with the bandwidth of the 5000 series (that is also affordable    )

I'd like a 4 channel 4262 with an external reference input - this is unlikely to happen.

I'm dithering between a 4262 and something like an RME UX usb interface. If I want to measure 3 oscillators with a reference/calibration channel then this needs 4 channels.
The 4262 is the easiest solution but I can only ever measure 1 oscillator against a reference oscillator.
Using an 8 channel audio USB interface the frequency range is limited (1Hz to 80kHz basically), only 2 channels have pre-amps and setting levels is more of a pain. The noise figures are good though and it is a cheaper option.

If I want to measure 3 oscillators in a 3 cornered hat approach then I'd need 3 two channel 4262s as I'd have to use one channel on each for reference but it could be done on a single 4 channel 4262 if such a thing existed.

[egonotto]

Hello,

@_Wim_: "Could you post a plot to 200MHz to check this with your 5444B?"

Perhaps we should ask this problem in the Picotech forum.

Best regards
egonotto

[_Wim_]

Perhaps we should ask this problem in the Picotech forum.

That is a possibility, but as the spurs are "within spec", I do not expect much of an answer. I have been into contact with them a few times (using their feedback functionality build into the software), and although they are quite responsive, the answer is typically not very enlightening.

[Andreas]

Hello,

at these sensitivities also the exact measurement setup is important.
The below 10 MHz peaks on Egonottos plot are shurely from a switchmode supply or from some other radiating cables nearby.
I usually separate carefully the power supply cable and the USB cables on my laptop.
(sometimes I also operate the laptop on battery only).

Attached 2 plots from my PS5444A with 50 Ohms terminator on the input.
One is with PS5444A outside my cookies box.
the other with PS5444A inside my cookies box (USB-cable hanging out)
So already there we have some (small) differences.
I guess that also a ferrrite on the USB-cable gives some improvement.

with best regards

Andreas

[_Wim_]

I guess that also a ferrrite on the USB-cable gives some improvement.

I have a ferrite on both USB and power cable, but when I installed it, I did not see a whole lot of difference. I also looked at the common mode noise "delivered" via the power cable, but did not find any correlation between the spurs and the common mode noise. Should do the same for the USB also. I will try to post some results of this later on.

[2N3055]

I guess that also a ferrrite on the USB-cable gives some improvement.

I have a ferrite on both USB and power cable, but when I installed it, I did not see a whole lot of difference. I also looked at the common mode noise "delivered" via the power cable, but did not find any correlation between the spurs and the common mode noise. Should do the same for the USB also. I will try to post some results of this later on.

I did see a bit of improvement on 4262 by making two loops of USB cable trough ferite... But not much..

[ADT123]

The PicoScope 5000 contains 8 x ADCs each of which can sample at 125MS/s.  In 12 bit mode time interleaving is used to increase the sampling rate - matching between ADCs is not quite perfect hence the interleaving spurs.  These are small / within spec so this is normal.

In 14 bit mode each channel has 2 ADCs in parallel (to increase resolution / reduce noise) so there are no interleaving spurs.  In spectrum mode 14 bits is great unless you need > 62.5MHz. 

More generally for all scopes using FFTs its best to increase the number of points to spread the noise between as many bins as possible.  Also its best to measure to a higher bandwidth than you need.  Eg if measuring just audio to 20kHz then measure to say 10MHz and zoom in on the 0 to 20kHz range - this ensures higher frequency noise does not get folded back into the range you are measuring / displaying.  Add in some averaging and perhaps also the hardware bandwidth limit and its amazing what you can detect.

[_Wim_]

These are the common mode profiles for both USB and power cable. These look "bad", but this is due to the sensitivity of the used current probe (>40V/A from 10Mhz and up). When I compare this with other result I have got in the past, the peaks are quite low and also did not match with the 31.25MHz. 
The probe used is a Tegam 95242-1 (some model number is also sold by ETS-lindgren, I suspect they are identical).
I also made a comparison plot (with shorted input) when the Picoscope power adaptor was plugged into a very clean power outlet (isolation transformer output) and from a normal "noisy" power outlet. Some very small differences were seen. The wider trace is due to shorter averaging time. For me this means that the spurs are mainly internally generated by the ADC (because they practically disappear when the "precision 14 bit mode is used".
P.S: the above current probe is normally intended to inject common mode current, but so far I have no suitable high bandwidth power amp for this.

[_Wim_]

The PicoScope 5000 contains 8 x ADCs each of which can sample at 125MS/s.  In 12 bit mode time interleaving is used to increase the sampling rate - matching between ADCs is not quite perfect hence the interleaving spurs.  These are small / within spec so this is normal.

In 14 bit mode each channel has 2 ADCs in parallel (to increase resolution / reduce noise) so there are no interleaving spurs.  In spectrum mode 14 bits is great unless you need > 62.5MHz. 

More generally for all scopes using FFTs its best to increase the number of points to spread the noise between as many bins as possible.  Also its best to measure to a higher bandwidth than you need.  Eg if measuring just audio to 20kHz then measure to say 10MHz and zoom in on the 0 to 20kHz range - this ensures higher frequency noise does not get folded back into the range you are measuring / displaying.  Add in some averaging and perhaps also the hardware bandwidth limit and its amazing what you can detect.

Hi, thanks for your explanation, I didn't notice before you had posted also (got no notification some-one posted while I was typing). Great to have somebody (previously) related to Pico here on the forum.

Can you provide some more info on the differences in clock specs between the 5442B and the 5444B? Is it only because of the higher ETS sample rate?

[ADT123]

_Wim_ the crystal oscillator in the higher end model is simply a more expensive one that is calibrated at final test.  Other than the accuracy of frequency measurements there should be no other differences.  Other specs such as jitter are the same.

[_Wim_]

_Wim_ the crystal oscillator in the higher end model is simply a more expensive one that is calibrated at final test.  Other than the accuracy of frequency measurements there should be no other differences.  Other specs such as jitter are the same.

Thanks for this info, if it is only about absolute frequency accuracy, that less of interest to me.

[maxwell3e10]

I got a more recent model of Owon XDS3062A scope (due to a sale at Arrow.com). The input noise spectrum of it looks somewhat better than an earlier version of the same scope. White noise level is quite low at about 3.5 nV/sqrt(Hz). The rms noise is 46 uV and it remains the same up to full scale of 100mV, thanks to 12 bit ADC. But there is significant 1/f noise and a number of prominent noise  peaks, in particular one near 1.3 MHz. 

The scope also came with a built-in function generator and it has a basic frequency response analysis function (gain and phase) that seems to work OK. The function generator has the same clock as the scope, which can also have some advantages. Not bad for less than $200!

[srce]

Here are some results for a Keysight MSOS204A.

Vertical: 1mV/div - 8 divisions - 10bit
Horizontal: 5uS/div - 20GSa/s - 1Mpts

Yellow is 20MHz B/W (s/w filter) - 50 Ohm - 44uV AC RMS - 350uVpp
Blue is 2.1GHz B/W - 50 Ohm - 115uV AC RMS - 1.1mVpp
50 Ohm terminators on BNC inputs.

2GHz data 50 Ohm

20MHz data 50 Ohm

And here's another one with the same settings:

8.4GHz - 50 Ohm - 248uV AC RMS - 2.4mVpp

8.4GHz data 50 Ohm

[Nanoman]

The Picoscope details I have seen show an integrated CMOS transimpedance buffer intended for DSOs from TI that by itself has 100 times the noise of a discrete front end at low frequencies.  I mean literally 1000nV/SqrtHz where a discrete design could be 10nV/SqrtHz.

So a broadband RMS noise of 125.7uV over 200MHz does not surprise me at all and that is about 5 times worse than my 40+ year old 200MHz analog oscilloscopes.

3000/5000 series ADA4817-1 (4nV), recent series of scopes. Obsolete versions (early 2000's) employed discrete jfet front ends.
2000 series          ADA4891-1 (9nV) or AD8065 (7nv) (25MHz versions)
4000 series          ADA4891-1 (8 channel scope) and AD8065 on older 20MHz scopes and 4817 on 16bit scope
6000 series          BF998 mosfet (~6nV, obsolete device), 6404 is ~12nv due to attenuator configuration

Either Analog Devices parts or discrete jfet/mosfets. The ADA4891 measures closer to 12nV due to 1/f contribution.

I used to design scopes at Pico and wanted to clear up this misinformation, noise is a critical scope spec and we'd use the best parts whenever possible.

[ILIJA]

Hi
Can you share script you used  to compute nV/sqrt(Hz)?
Thank you