DSO Bandwidth test SDS1104X-E DSOX1102G TO1104 GDS1054B

[David Hess]

With manufacturers like Rigol producing DSOs with unspecified and variable full power bandwidth, these tests are close to useless without specifying the input sensitivity.  In the past this was not a consideration because the full power bandwidth was always higher than the bandwidth at any input sensitivity which was a result of operating the input stages over a much smaller signal range.

If I measure freq response using (example) 5, 10, 20, 50, 100, 200, 500mV, etc/div and so on and show all these results. Or if I measure freq response for 100mV/div and show result. What is then wrong there. Can you say this response is not true with used setup.    Do you mean all this is bullshit.

What is wrong. If it is wrong, please give explanation and instructions how to measure.
Or is it just only wrong because these are made in China.

That would be one way and in practice because of the common designs now, it may be the only way without knowing more about the vertical signal chain.

The problem comes about when a gain stage has a limited full power bandwidth (slew rate).  In the past, all of the attenuation stages were very early in the signal chain so every stage operated with the same signal level.  With the input always attenuated down to say 5mV/div, each stage saw the same signal level.  (1)  Later designs changed this a little and replaced some of the high impedance attenuators at the input with low impedance attenuators immediately after the high impedance buffer but the result was the same as long as the buffer always operated within its full power bandwidth.

Now it is common to have those low impedance attenuators after the early gain stages which results in the early gain stages operating over a much wider signal range requiring a correspondingly higher full power bandwidth.  This is reflected in measurements of instruments like the Rigol DS1000Z series where bandwidth varies markedly with sensitivity, over more than a 2:1 range.

I have repaired and calibrated several Tektronix old and also made frequency response measurements as told in Tek service manuals. Are these also useless.

Those old Tektronix oscilloscopes had full power bandwidths in excess of their bandwidth so it was never a problem.

Measured frequency response is just as result is in told situation. Or do you claim this result is wrong. If it is wrong please come here and show how it need measure so that also you are satisfied.

I am claiming that with a modern oscilloscope, the measurement cannot be replicated without knowing the input sensitivity because many oscilloscopes now have wildly varying bandwidth with different input sensitives.  This has shown up repeated on this forum with users making rise time and bandwidth measurements of Rigol DS1000Z DSOs which vary over more than a 2:1 range depending on the test conditions.

(1) Often old oscilloscopes had a small number of switched gain stages which were used at the highest sensitivities but at least the manufacturers specified the bandwidth if it differed.  For instance an old 200 MHz Tektronix TDS410A is 200 MHz from 5mV/div to 10V/div, 150 MHz at 2mV/div, and 95 MHz at 1mV/div.

However this is the opposite of what is going on now where bandwidth decreases when an input attenuator is removed because of pull power bandwidth (slew rate) limits.  Slew rate limiting causes harmonic distortion however it is very difficult to discern from the shape of the waveform with the harmonics all above the -3dB bandwidth.

[tautech]

A picture update..

Wrong, there is err...afaik.

Look tautech 1104X-E result. Where from you get 150MHz data point?
This around -18dB point is 200MHz!

Yes counter error when 1104X-E is pushed past its -3dB point.......but it might also be some trigger level hysteresis. 
Will add some more scopes tomorrow when I do their checks before they need to be delivered.
Now 

[iMo]

A picture update v3.
PS: not enough colors, sorry

[rf-loop]

A picture update v3.
PS: not enough colors, sorry

And still same error.

[iMo]

A picture update v4.
Fixed the TAU 1104 100M.

PS: plz provide the numbers only.. 

[nctnico]

I have two remarks: it would be better not to use the polynomal fit because you don't have enough points. Also the frequency axis would be better if it where logarithmic.

[rf-loop]

I am claiming that with a modern oscilloscope, the measurement cannot be replicated without knowing the input sensitivity because many oscilloscopes now have wildly varying bandwidth with different input sensitives.  This has shown up repeated on this forum with users making rise time and bandwidth measurements of Rigol DS1000Z DSOs which vary over more than a 2:1 range depending on the test conditions.

Oh these. Wellll...   

I have laughed many times looking many of these. Some times it looks that some peoples do not even know what they are doing with things what they have get to hand or table for playing. But if it is fun and educating, perhaps it is cheap and health way to keep fun and learn. Good and healthy value for spent money.

[rf-loop]

I have two remarks: it would be better not to use the polynomal fit because you don't have enough points. Also the frequency axis would be better if it where logarithmic.

 

And same for a straight lines.

[iMo]

Guys, no rocket science here, just depicting a trend based on a few measurement points..

[rf-loop]

Guys, no rocket science here, just depicting a trend based on a few measurement points..

Yes free for artists.

[tautech]

OK so if TK can bring an 2GSa/s entry level DSO into this comparison then so shall I. 

Signal source SSG3021X and setup as before, 0dBm into 50 ohm Tek termination and 100mV/div unless shown otherwise.

SDS2022X-E stock 200 MHz $619 model.
Completely missed 900 MHz. 
Pushed to 1 GHz !  At the start of aliasing.

[Rerouter]

Tautech, can I ask why you did not increase the VDIV for most of the screens, the measure command is more accurate when the wave is at least 65% of the screen height,

[tautech]

Tautech, can I ask why you did not increase the VDIV for most of the screens, the measure command is more accurate when the wave is at least 65% of the screen height,

Look at the OP, all measurements were done @ 100mV/div.

If we think about it, what David Hess bought to the discussion is valid in that displayed/measured accuracy could be compromised if different vertical attenuations are selected. A valid point IMO.
It certainly would affect the BW roll off curve however probably just to a small degree.

So we just compare apples with apples @ 100mV and call it quits at that and let the viewers decide which are the lemons. 

[rf-loop]

With manufacturers like Rigol producing DSOs with unspecified and variable full power bandwidth, these tests are close to useless without specifying the input sensitivity.  In the past this was not a consideration because the full power bandwidth was always higher than the bandwidth at any input sensitivity which was a result of operating the input stages over a much smaller signal range.

If I measure freq response using (example) 5, 10, 20, 50, 100, 200, 500mV, etc/div and so on and show all these results. Or if I measure freq response for 100mV/div and show result. What is then wrong there. Can you say this response is not true with used setup.    Do you mean all this is bullshit.

What is wrong. If it is wrong, please give explanation and instructions how to measure.
Or is it just only wrong because these are made in China.

That would be one way and in practice because of the common designs now, it may be the only way without knowing more about the vertical signal chain.

The problem comes about when a gain stage has a limited full power bandwidth (slew rate).  In the past, all of the attenuation stages were very early in the signal chain so every stage operated with the same signal level.  With the input always attenuated down to say 5mV/div, each stage saw the same signal level.  (1)  Later designs changed this a little and replaced some of the high impedance attenuators at the input with low impedance attenuators immediately after the high impedance buffer but the result was the same as long as the buffer always operated within its full power bandwidth.

Now it is common to have those low impedance attenuators after the early gain stages which results in the early gain stages operating over a much wider signal range requiring a correspondingly higher full power bandwidth.  This is reflected in measurements of instruments like the Rigol DS1000Z series where bandwidth varies markedly with sensitivity, over more than a 2:1 range.

I have repaired and calibrated several Tektronix old and also made frequency response measurements as told in Tek service manuals. Are these also useless.

Those old Tektronix oscilloscopes had full power bandwidths in excess of their bandwidth so it was never a problem.

Measured frequency response is just as result is in told situation. Or do you claim this result is wrong. If it is wrong please come here and show how it need measure so that also you are satisfied.

I am claiming that with a modern oscilloscope, the measurement cannot be replicated without knowing the input sensitivity because many oscilloscopes now have wildly varying bandwidth with different input sensitives.  This has shown up repeated on this forum with users making rise time and bandwidth measurements of Rigol DS1000Z DSOs which vary over more than a 2:1 range depending on the test conditions.

(1) Often old oscilloscopes had a small number of switched gain stages which were used at the highest sensitivities but at least the manufacturers specified the bandwidth if it differed.  For instance an old 200 MHz Tektronix TDS410A is 200 MHz from 5mV/div to 10V/div, 150 MHz at 2mV/div, and 95 MHz at 1mV/div.

However this is the opposite of what is going on now where bandwidth decreases when an input attenuator is removed because of pull power bandwidth (slew rate) limits.  Slew rate limiting causes harmonic distortion however it is very difficult to discern from the shape of the waveform with the harmonics all above the -3dB bandwidth.

Here is small simple form test what show that least in Siglent whole range starting from 500uV/div is roughly same when compare 10MHz and 200MHz using all steps from 500uV/div up to 2V/div  (can not higher level due to lack of trusted high level signal source with enough flatness.) But test include 2 lower bands all steps and band 3 first step. THis do not proof anything and test was with simplest method but it give some image that least not any dramatical Band Width differences between input sensitivity (V/div) settings. (like example ion older Tek analog scopes where are some times significant differences specially in highest sensitivities, depending model)

https://www.eevblog.com/forum/testgear/siglent-sds1204x-e-released-for-domestic-markets-in-china/msg2390184/#msg2390184

[TK]

Tautech, can I ask why you did not increase the VDIV for most of the screens, the measure command is more accurate when the wave is at least 65% of the screen height,

Look at the OP, all measurements were done @ 100mV/div.

If we think about it, what David Hess bought to the discussion is valid in that displayed/measured accuracy could be compromised if different vertical attenuations are selected. A valid point IMO.
It certainly would affect the BW roll off curve however probably just to a small degree.

So we just compare apples with apples @ 100mV and call it quits at that and let the viewers decide which are the lemons. 

I am sorry to have caused a confusion... I started at 100-200mV/div and went down all the way to 2-10mV/div for the higher frequencies.

The objective of the test was to get the highest possible frequency that the instrument can measure

[tautech]

For shits and giggles let's run the 200 MHz 1 GSa/s 2ch SDS1202X-E, the first of the X-E scopes but without some of the nice features added into the 4ch X-E range and 2kX-E models.

Source as before; SSG3021X, 0dBm into Tek 1x 50 ohm termination.

[rf-loop]

Nice alias signals in 600, 700 and 800MHz images
(
600 ---> 400MHz alias
700 ---> 300MHz alias
800 ---> 200MHz alias
)

500 can see because sampling and signal is not perfectly synch and same with fN.

This is why real best modification is reduce analog BW  (aka example make 200MHz LPF more steep after corner freq, With 500MSa/s this situation is lot of more "terrible", real trap to peoples who have lack of enough experience and knowledge. As can see even with 1GSa/s it is not good except that many times probing make some freq rejection but better solution is if front end is bit more designed.  Personally I'm not satisfied with these too widely open analog front ends with very lazy slope down after corner. Looks like it is not designed at all for this matter.  These alias productions smudge also many other signals where is markable amount of example harmonics. Oscilloscope need design for analyze unknown signals. Example inputting extremely fast changing signal may produce "what ever" miracles and phenomena what are then wondered here and there.

[tinhead]

Personally I'm not satisfied with these too widely open analog front ends with very lazy slope down after corner. Looks like it is not designed at all for this matter.  These alias productions smudge also many other signals where is markable amount of example harmonics.

this is what of the reason why i use, when unsure or for RF, my DIY active probes with LPF from minicircuits with my SDS1204X-E
I have only 550 and 600MHz models here, these here are for sure better -> https://www.minicircuits.com/WebStore/dashboard.html?model=BLP-450%2B

[nctnico]

It is much less of a problem then you might think. To avoid aliasing the front end should attenuate around 40dB to 50dB at fNyquist * 1.2. Sure you can lower the V/div setting to see a signal but that has no real value. If you are looking at a signal within the oscilloscope's bandwidth you'll use 8 bits of ADC resolution at most and with multiple traces on screen likely 6 or 7 bits. If the higher harmonics above the Nyquist frequency are attenuated by 40dB to 50dB they won't trouble you at all.

[rf-loop]

Personally I'm not satisfied with these too widely open analog front ends with very lazy slope down after corner. Looks like it is not designed at all for this matter.  These alias productions smudge also many other signals where is markable amount of example harmonics.

this is what of the reason why i use, when unsure or for RF, my DIY active probes with LPF from minicircuits with my SDS1204X-E
I have only 550 and 600MHz models here, these here are for sure better -> https://www.minicircuits.com/WebStore/dashboard.html?model=BLP-450%2B

This is produced using freq hopping sinewave (yes there is also rf generator harmonics what also produce aliasing but they are buried behind others in display) sweep so that it leave enough spacing for also get alias peaks visible. So, it is not beautiful image but it is ok for this purpose. So it is swept using constant level signal from 10MHz to 1000MHz (not true constant level at scope BNC input point but enough for this purpose.)

In this tiny example fNyquist is 250MHz (500MHz sampling frequency)
(maximum if all 4 channels are in use)

What other things these aliases do. They can example distort square/pulse etc waveforms of course specially if signal have fast edges where higher level harmonics goes over fN.



For peoples who do not know alias things at all, clarify: These arrows mean freq sweeping up direction
(example in image (blue arrow) if signal input is 300MHz then FFT show alias 200MHz. and then when inout freq rise alias goes down.. until 0, after then when signal still rise then alias freq rise etc..

[iMo]

Where to place a few pF to make the BW shape better (1204X-E)?

[rf-loop]

Where to place a few pF to make the BW shape better (1204X-E)?

What is better shape?

If want better shape then filter need follow ADC sampling frequency and so on.

Here same experimental with 1GHs ADC sampling frequency.

Then if have BW limited more steep with this 500MSa/s then can ask is it ok anymore for this next.


As previous, also this is produced using freq hopping sinewave (yes there is also rf generator harmonics what also produce aliasing but they are mostly buried behind others in display) sweep so that it leave enough spacing for also get alias peaks visible when freq rise and aliases start fold back from fNyquist.


Of course often we use oscilloscope with usual passive probes. This is also filter.

Then, example, if we look, say example square waves. Harmonics levels are not at all same as fundamental freq. (3th level 1/3, 5th: 1/5, 7th 1/7, 9th 1/9  etc... IF square is ideal square with 0 risetime, what is is never). 
So in practice this is not at all as bad situation what it first may feel example if look these images where input freq is constant level over whole sweep freq range. But important is that users undestand these basics in theory and also his scope and his probes etc. Enough knowledge and scope do not fool user so easy.

Also when look these images, there vertical scale is LOG. Normally when we use time domain, vertical scale is linear.  Why I draw there 26dB range.  If example normal time domain display we have 1Vrms sinewave on the screen.  When it drop 26dB there is 50mV on display. (5%) 
If we have 10M square and think its harmonics example 11th. It is 110MHz but it is dropped say example over 20dB... if it have very fast edges it may easy have exaple 31th harmonic 310MHz. But it is perhaps -30dB down from fundamental level. If Sampling freq is 500M (as previous image)  it is aliasing,  its alias is 190MHz but if we are in time domain display with linear vertical and our fundamental is example 6div height, how is level of this harmonics. Can you see its effect. If front end attenuate 310MHz example 5dB then this harmonic alone is around 0.1div or so. In practice, no visible effect in this square wave shape.

But, with FFT we really need care about alias frequencies. There these can mess lot of.

Just example.  Some RF generator 3rd  harmonic can be example -30dBc.
If some one is looking  124.9MHz carrier on FFT and example case he is using 500MSa/s.
Say example he look 0dBm carrier. 3rd harmonic -30dBc (374.7MHz)
Now he can see 124.9MHz 0dBm peak and he can also see 125.3MHz peak level -36dBm. (If scope attenuate 6dB 375MHz.) And then user wonder why there is fundamental carrier 124.9MHz  and after then 400kHz higher also signal. Think these with different frequencies and different waveforms with different harmonics. In some cases this situation with many aliases can be really confusing.

Keep care and not input over fN frequencies (including harmonics and nonharmonics)