Owon XDS3062A input noise (good) and glitches (bad)

[maxwell3e10]

I took a closer look at  the 12-bit Owon scope to compare the input noise levels to other scopes. Here are the RMS noise for 50 Ohm shorted inputs. It does not depend very much on the time scale of the scope.

Shorted RMS noise   1mV/div        1 V/div
TEK TBS1064          250 uV          16 mV
Instek GDS-1054B  95 uV            14 mV
Micsig TO1104        85 uV            10 mV
Owon XDS3062A     45 uV            3.3 mV

So on high sensitivity settings Owon input amplifier has about a factor of 2 lower noise. On higher voltage settings the 12-bit ADC has about 1 LSB RMS noise, while the 8 bit ADCs are limited by 1-bit flickering.

I also tested some of the scopes for time jitter with 10 MHz 1.3 Vrms sine wave from a Rb frequency standard.

RMS deviation from sine fit for   1V/div     500 mV/div
Micsig TO1104                          15 mV     9mV
Owon XDS3062A  (8 bit mode)   13 mV
Owon XDS3062A  (12 bit mode)  4.9 mV   3.8 mV

One can see that time jitter contributes somewhat to noise, but is not a dominant noise source even for relatively fast signals.

While looking at Owon signals I also found some peculiar glitches which are summarized in plots below. For some settings every 2^16th point is equal to zero. Also, there are occasionally repeated points, but they disappear periodically for 2^11 points. Its probably some digital interference. Hopefully Owon guys can look into it.

[radhaz]

What hardware version is your scope? Mine is 1528006, FW 2.4.0.0.

[maxwell3e10]

Owon is 1539xxx, the firmware they have listed  for it is 1.5.0.0.

[radhaz]

I forgot to mention, mine is a 3102A.

[ZomBiE80]

Looks a LOT of quantization noise of ADC.

[rf-loop]

I took a closer look at  the 12-bit Owon scope to compare the input noise levels to other scopes. Here are the RMS noise for 50 Ohm shorted inputs. It does not depend very much on the time scale of the scope.

Shorted RMS noise   1mV/div        1 V/div
TEK TBS1064          250 uV          16 mV
Instek GDS-1054B  95 uV            14 mV
Micsig TO1104        85 uV            10 mV
Owon XDS3062A     45 uV            3.3 mV

So on high sensitivity settings Owon input amplifier has about a factor of 2 lower noise. On higher voltage settings the 12-bit ADC has about 1 LSB RMS noise, while the 8 bit ADCs are limited by 1-bit flickering.

Is it good?  Not very bad at all but also not very good. (I have here also This Owon 100MHz 3kXA and these some tests can find around forum older messages)



Here for compare Siglent.
500uV/div  20M BW  RMSnoise 27uV
1mV/div 20M BW RMSnoise 30uV

RMSnoise equals SDEV (in this case. It removes mean so it show this noise RMS without possible DC offset. This can use because we measure noise and we are interested only amount of noise itself)
(random noise mean (DC - offset) removed)

Here comparable image for Owon 1mV/div with forced 20M BW. (Because Owon do not have full bandwidth available)

[egonotto]

Hallo,

I am interested in inherent noise in the most sensitive mV/div range.

Can someone please post a file with about 100 kSamples in the most sensitive range with a 50Ohm termination without input.

Nice if I get the data from different scopes

This gives me the possibility to calculate some values.




Best regards
egonotto

[Performa01]

I am interested in inherent noise in the most sensitive mV/div range.
Can someone please post a file with about 100 kSamples in the most sensitive range with a 50Ohm termination without input.
Nice if I get the data from different scopes

What bandwidth?
What sample rate?

[egonotto]

Hi

Sample rate: best the maximal possible.
Bandwidth: Full bandwidth and 20MHz if possible.

I have already data from R&S. But is good to have data from other scopes.

Best regards
egonotto

[rf-loop]

Hallo,

I am interested in inherent noise in the most sensitive mV/div range.

Can someone please post a file with about 100 kSamples in the most sensitive range with a 50Ohm termination without input.

Nice if I get the data from different scopes

This gives me the possibility to calculate some values.




Best regards
egonotto

100k on the display width. (many scopes what have more memory do not map all to display width. Many oscilloscopes also do not calculate mesurements from full memory or undecimated data. This may make big difference.
Noise is random distribution. Peak value depends amount of data and here we go very easu to area where we do not compare scopes as apples to apples. We have seen this over years also here. Also what you see on the screen depends not only how oscilloscope decimate and/or map acquistion tata to screen buit also it depends how fast wfm/s it do (again, amount of data)

What I did in previous image from Siglent here. Noise RMS. How we get reliable noise RMS so that it have really noise itself. This is why I use Sdev measurement instead of RMS. Then, also, as can see it is calculated from quite lot of data as can see this statistics.

And you see memory length is 700pts.
But this image screen have stacked several axquisition. Amount of sample data is over 500000 samples in this image one trace. And all is mapped in this image. This is one reason why different scopes are bit difficult to directly compare.
But if we calculate RMSnoise it do not so much matter how fast scope is. 


Here in this
https://www.eevblog.com/forum/testgear/siglent-sds1204x-e-released-for-domestic-markets-in-china/msg1290409/#msg1290409
thread first message is attached noise images and also 1M inputs but covered with RF shield cap for eliminate external noise.
Full bw and 20M BW, 500µV/div and 1mV/div.
Sorry they have also 700pts data length but, as told they have on the screen around 500000 samples  every trace. (so much over 100k you want.) 

Peak value rough loose estimate for random noise you can get when you calculate  6 * Sdev. (or if want be very tight then 8*Sdev)



with automatic mesurements Siglent use full acquisition length and resolution (but here it do not matter because one sweep length is 700 (roughly 20000 times/s)

I have also checked it calculate Sdev right.

[maxwell3e10]

It would be useful to have raw data so one can also calculate other things, such as spectral noise density. Siglent input noise is actually very impressive, almost on par with R&S and other much more expensive scopes. Roughly 27 uV/20 MHz gives 6 nV/Hz^(1/2), while the best scopes might have 100uV/1 GHz= 3 nV/Hz^1/2 (ignoring NEB effects).

It would be interesting to compare the noise spectra, often scopes have a bunch of spurious peaks above 100 MHz and its hard to tell how much it contributes to RMS. I don't trust build-in FFT routines, its much better to do fft of raw data so it can be properly normalized to give noise spectral density in nV/Hz^(1/2) independent of number of points, etc.

[egonotto]

Hi,

why not post raw data?

Best regards
egonotto

[Performa01]

Okay, here it is for the Siglent SDS1104X-E, channel 1, input terminated with 50 ohms, acquisition mode normal.

Input sensitivity = 500µV/div;
Timebase = 10µs/div;
Total captured time window = 140µs;
Sample rate = 1GSa/s;
Record length = 140kpts;

The ZIP archive contains two .csv files, one for 20MHz bandwidth limit, the other one for the full bandwidth of 110MHz (-3dB).

Have fun!

EDIT: Input sensitivity added.
EDIT2: Corrected the sample rate in the file names. The files are still the same.

[Performa01]

Since not all scopes have a true (full resolution) 500µV/div sensitivity, I've also created the corresponding files for 1mV/div.

The difference should be pretty small anyway, i.e. the higher sensitivity does not significantly improve the noise figure.

EDIT: Corrected the sample rate in the file names. The files are still the same.

[maxwell3e10]

Thanks! So if I calculate the standard deviation for the 20 MHz files, I am getting higher noise:
1 mV, 110 MHz:  69 uV
1 mV, 20 MHz:  44 uV
500 uV, 110 MHz: 61 uV
500 uV, 20 MHz: 43 uV

For reference, all the numbers I reported at the start of the post came from downloading the data and calculating standard deviation.

[Performa01]

Well, I have only briefly checked the 500µV/110MHz data myself and found the standard deviation to be about 61µV, which is in accordance with both your findings as well as the readings from the automatic measurements on the scope.

The 69µV for 1mV/div sensitivity sound also very plausible.

Your 20MHz calculation is not plausible though:

SQRT(20/110) = 0.4264;
0.4264 x 61µV = 26µV;

So the theory conforms much more with the automatic measurements of the scope (which were about 27µV) than your calculations.

EDIT: The time base was 10µs/div for the files, so the noise may be a tad higher than the 1µs/div demonstration in the other thread, but only a few percent and not x1.65!

[egonotto]

Hi,

thanks Performa01

Now my results. Sorry I am an novice in OCTAVE.

I calculate history and FFT and got the same result as maxwell3e10.

Please maxwell3e10 can you send data from 12Bit OWEN.

Best regards
egonotto

[Performa01]

I've just had a look myself, and the standard deviation is some 41µV for the 20MHz file indeed.
Very curious, I have no explanation for that. Theory and scope measurements agree, but the data do not...


EDIT: Btw. Egonotto, thanks for the analysis data, I just don't know how to read the FFT. What's on the x and y axes?
x: 70000 Hz? y:10e-1 volts? 

[rf-loop]

Here is zip. It include SDS1104X-E
one 70k acquisition full CSV
oscilloscope screen image about exactly same data.
(20M BW, 500µV/div, 70k data samples (one 70k acquisition using 500MSa/s (1), 10µs/div, 14 div, input terminated 50ohm)
(1) when scope was running there was CH3 and 4 on so this is reason for 500MS/s and when I save .csv I just turn other channels off for get just one channel data for avoid any confusions.

oscilloscope automatic mesurement Sdev  result is 29.30 µV
oscilloscoppe automatic mesurement Mean result is 40.16 µV

Also I check  these scope own calculations using Apache OpenOffice.
(and have done it many times with different data and signal and t/div and V/div. Allways right. So  no reason to compute from  .CSV delivered to PC. Scope can do it much more fast and just reliable. (but I do not know other scopes what do many trics like calculation from decimated data or even only from "screen" depending model and manufacturer.

STDEV = 29.303195489906 µV  equals RMSnoise 29.30 µV
AVERAGE = 40.162573751045 µV  (DC offset)

also   @maxwell3e10  can check calculations. If your calculations give same result, then I, scope and you have same result what is perhaps also right.

[egonotto]

Hi,

sorry for the incompleteness.

The scale in the FFT is in x about 7142 Hz and in y about 0.0000143V.
The x-axis reach to 500MHz. (70000*7142).

1 in the y-axis means 0.0000143V
0.1 in the y-axis means 0.00000143V

Best regards
egonotto

[Performa01]

Thanks Egonotto for the explanation!

Yet I'd need to know the RBW (resolution bandwidth) of the analysis in order to make sense of the amplitude level.

What frequency step (or: what number of FFT points)?
What window function?

[egonotto]

Hi,

I use the fft-function from OCTAVE. The number of points are 140000
a bin is 7142.8 Hz there are 70000 bins.
I suppose the fft-funktion dont use a window.

Enclosed the example from rf_loop

The bin is also 7142.8 Hz the y-axis factor is 0.0000286V.

Best regard
egonotto

[Performa01]

I have tried a lot to find out the discrepancy, but nothing seemed to help. The scope itself displayed the correct value for the rms noise, but the data didn’t match – at least not with 20MHz bandwidth limit on.

I eventually figured out that I should have stopped the scope during the data export instead of doing it while the scope was still running.

I have attached new data – only for 500µV/div this time, because everyone pulls out their strongest weapon when it comes to some competition don’t we?

Anyway, at 1mV/div it would be about 31µVrms for 20MHz and 58µVrms for 110MHz.

The FFT thingy got me curious and so I tried it directly on the scope. And I think it displays quite reasonable data. Here is the FFT analysis from DC to 500MHz for both records. The first one for 110MHz, the 2nd one for 20MHz. What a difference!


SDS1104X-E_Noise_500uV_140us_1GSa_BW110MHz


SDS1104X-E_Noise_500uV_140us_1GSa_BW20MHz

EDIT: Just for clarification: The FFT shows the amplitude in dBm for the 50 ohm input termination. -100dBm = 2.24µV.

[egonotto]

Hi,

looks good. Very low noise.

The scale in the FFT is in x about 7142 Hz and in y about 0.0000143V.


Best regards
egonotto

[maxwell3e10]

Here is an updated summary of input noise for various scopes:

                       
Micsig TO1104
0.5 mV/div Full BW: 75uV, 20M BW: 37 uV
1 mV/div   Full BW: 85 uV

Owon XDS3062A
1 mV/div 20M BW: 44 uV
5 mV/div Full BW: 78 uV, 20 M BW: 47 uV

Instek GDS1054B
1 mV/div FullBW 102 uV, 20 M BW: 95 uV

Siglent SDS1104X-E
0.5 mV/div FullBW 60 uV, 20M BW 28 uV

So Siglent is a clear winner here, at least on the high gain settings.
Is anyone willing to post data for Rigol1054Z? I guess its not that great, but would be good to know for comparison.