Rigol DS1054Z bandwidth

[TheoB]

To measure the bandwidth I've put a 20MHz square wave signal terminated in 50Ohm to my DS1054Z and measured the rise and falltime. I did this before and after adding the 100MHz bandwidth option.
I determined the bandwidth from the measured 20%/80% rise and falltime as BW=1.39/(6.28*Trf)=0.22/Trf. Trf is the average between rise and fall time.
For completeness I also measured the probe in X1 and X10.

Vert setting	No option	100MHz option	20MHz BW limit	X1 probe (with 100MHz option)	X10 probe (with 100MHz option)
200mv	113MHz	147MHz	24.6MHz	22.9MHz	142MHz
500mv	170MHz	184MHz	25.7MHz	22.7MHz	*

*) Could not measure as I would need 10x bigger signal

I don't know the input rise fall time as I have no other means to measure it. The instrument I use is a pulse generator with max output frequency of 200MHz. What I do find peculiar is that the measured bandwidth is a lot higher than the 50MHz specified (up to three times). And there is quite some difference in bandwidth between the lower gain settings (<=200mV/div) and the higher ones (>=500mV/div). I can hear a relay click when changing from 200mV/div to 500mV so I guess an amplifier is switched. Anyone else noticed this difference in bandwidth as a function of vertical gain?

[saturation]

You only need the rise time.  The standard measurements use 10-90.

The standard way is:
http://literature.cdn.keysight.com/litweb/pdf/5988-8008EN.pdf

For Gaussian roll off its ~0.35/Rise time; for brick wall roll off its ~0.45/Rise time.

You need only the fastest edge, because the basis of the bandwidth calculation is that of a unit impulse or Dirac delta function.

I rarely see anyone use the unit impulse alonse to measure BW because its calculation intensive.  You would differentiate the impulse then apply an FFT to it, which new fast DSO can do easily:

https://community.keysight.com/community/keysight-blogs/oscilloscopes/blog/2016/09/01/how-to-measure-your-oscilloscope-and-probe-s-bandwidth-yourself

In pure math:

http://lpsa.swarthmore.edu/BackGround/ImpulseFunc/ImpFunc.html

To measure Tr accurately you need to spread it out on the screen to confirm the automated readings match the graticule readings.

Hope this helps.  Eyeballing you graphs your data gives the DSO bandwidth as 233-300 MHz, if your measurements are true.

[MrWolf]

Since this test may be vaguely related to my doings... Actually been telling from day 1 that this box is good for speed freaks sitting at lowest timebase pedal to the metal - no surprises here  Go down to 20ns timebase at most to not stall it.

[TheoB]

You only need the rise time.  The standard measurements use 10-90.

I used 20%/80% as the step response is not OK of the rigol below 500mV/div. This means it does not have a Gaussian or flat frequency response in the 200mV range and below. To show I captured the exact same signal in both 200mV and 500mV range. The saved waveform in 500mV range I amplified to match the display in 200mV range so to make comparison easier (now you see the finite resolution also clearly).

You can also see that with 10%/90% the measured rise time starts to become affected by the "knee". 2.4ns  gives a 350/2.4=146MHz.

The standard way is:
http://literature.cdn.keysight.com/litweb/pdf/5988-8008EN.pdf

Thanks I've read it. In the 500mV/div range the rise time is 1.62ns. Because of the ringing it's probably a flat response. So bandwidth would even be 450/1.62=278MHz (instead of 350/1.62=216MHz). Both really impressive for a 50MHz scope 

For Gaussian roll off its ~0.35/Rise time; for brick wall roll off its ~0.45/Rise time.

You need only the fastest edge, because the basis of the bandwidth calculation is that of a unit impulse or Dirac delta function.

I rarely see anyone use the unit impulse alonse to measure BW because its calculation intensive.  You would differentiate the impulse then apply an FFT to it, which new fast DSO can do easily:

https://community.keysight.com/community/keysight-blogs/oscilloscopes/blog/2016/09/01/how-to-measure-your-oscilloscope-and-probe-s-bandwidth-yourself

In pure math:

http://lpsa.swarthmore.edu/BackGround/ImpulseFunc/ImpFunc.html

I tried but that's too much for the little rigol. It cannot do math over math.

To measure Tr accurately you need to spread it out on the screen to confirm the automated readings match the graticule readings.

Hope this helps.  Eyeballing you graphs your data gives the DSO bandwidth as 233-300 MHz, if your measurements are true.

What is truth. Any of these measurements must be compensated for the rise time of my E-H research Model-122 pulse generator. So bandwidth of scope must even be higher.

[MrWolf]

Good to see some practical stuff for once. Care to sample same wfm on same timebase, but with more channels enabled to drop sampling rate to 500 and 250 MSa/s? Lots of pages have written if this is a problem or not... Experiment could decide.

[saturation]

That is improved but its still suboptimal.

You simply need to measure the rise time, not the whole pulse to insure most accurate placement of the cursors and insure most accurate horizontal resolution.



You can confirm the auto measurement thus more easily with manual cursors as well the visual graticule readings as you choose. 

Enjoy.

Also be aware of problems Mr Wolf's thread has shown were the timebase affects the automated measurements.

https://www.eevblog.com/forum/testgear/testing-dso-auto-measurements-accuracy-across-timebases/msg1093842/#msg1093842

What is truth. Any of these measurements must be compensated for the rise time of my E-H research Model-122 pulse generator. So bandwidth of scope must even be higher.

Yes, but you must eliminate sources of removable error before accepting the limits of real world equipment.   FWIW what the Instek shows is a Heaviside step function, but limited by real world equipment, and its derivative is the Dirac delta as discussed earlier.  They are all 'related' and thus the step alone is valid to calculate the frequency response.

https://en.wikipedia.org/wiki/Heaviside_step_function

[TheoB]

I was not doubting the auto measurement of the scope. I can already see myself that the measured value is about what is shown on the screen. Timebase cannot be further zoomed in (5ns is the fastest) so the measurement resolution is not that great. I did notice that the timebase itself affects the measurements. All measurements are done from the screen buffer not from the bigger capture memory. I've seen that with time and voltage measurements (up to FFT). The FTT they improved a bit by using 2400 samples from the internal buffer. Vertical resolution could be improved by averaging. That not done. Instead you get truncated vertical values, not higher resolution.

Just to make you happy, I manually placed the cursors at the 10%/90% points and removed the falling edge. I then measure 1.5ns, but it could also be 1.6ns as that's the resolution you get. The raw data shows points every ns (of course). My message is only that the bandwidth depends (a lot) on the vertical settings that is chosen. And it should not matter if you measure rise or fall time and if you show one or both (I checked and it did not affect the measurement having both on the screen). I showed both as my risetime is not the same as the falltime. Could be my generator, could be the scope. I have no means to tell.

[Fungus]

Just to make you happy, I manually placed the cursors at the 10%/90% points and removed the falling edge. I then measure 1.5ns, but it could also be 1.6ns as that's the resolution you get.

And that's not bad for a "100MHz" oscilloscope. 

[saturation]

Theo, that is a good clean result.  I can reference the graticule against your measurements and the slope remains quite sharp.  That would mean fairly certain that your bandwidth is ~> 200 MHz - 280 MHz depending on the type of filter it has on input amps.   Congratulations!

[janoc]

AFAIK, the ~200MHz bandwidth on an unlocked DS1054 seems to match the measurements done previously of this scope, if I recall right. If you search the forum you will find the thread. So your values are likely correct. This scope is quite a value for the money.

[Fungus]

AFAIK, the ~200MHz bandwidth on an unlocked DS1054 seems to match the measurements done previously of this scope. This scope is quite a value for the money.

Yep. That's why I put the "100MHz" in quotes. Most people get over 150MHz after unlocking it.

[guenthert]

[..]
All measurements are done from the screen buffer not from the bigger capture memory. I've seen that with time and voltage measurements (up to FFT). The FTT they improved a bit by using 2400 samples from the internal buffer. Vertical resolution could be improved by averaging. That not done. Instead you get truncated vertical values, not higher resolution.
[..]

Do you happen to have means (and inclination  ) to download the measurements and perform the calculations on a PC?  I'd wished I had a well-defined fast enough signal source, then I'd do it myself.

[ebastler]

Many thanks, TheoB! Good to see posts from someone who (a) knows what he is doing, and (b) gives a balanced view of the DS1054Z, putting things in perspective, mentioning strengths and limitations without much fanfare.

[Fungus]

Do you happen to have means (and inclination  ) to download the measurements and perform the calculations on a PC?  I'd wished I had a well-defined fast enough signal source, then I'd do it myself.

All you need to measure rise times is a mechanical switch.

(...in the 100-200Mhz range)

[guenthert]

Do you happen to have means (and inclination  ) to download the measurements and perform the calculations on a PC?  I'd wished I had a well-defined fast enough signal source, then I'd do it myself.

All you need to measure rise times is a mechanical switch.

(...in the 100-200Mhz range)

Do you happen to have some mercury-wetted relays for me to borrow?

[tggzzz]

My message is only that the bandwidth depends (a lot) on the vertical settings that is chosen.

Do you have a rough indication of how it varies and by how much?

Such gain variations aren't uncommon. Sometimes it is due to a differing number series amplifiers being switched in/out, or because an analogue multipliers bandwidth is level dependent.

[TheoB]

With two channels selected, the sample rate drops to 500Msa/s and the 10%/90% rise fall time becomes 2.095ns. With four channels on (250Msa/s) bandwidth cannot be higher than 125MHz. Now it becomes interesting as I have some options. In normal display the edge becomes very wobbly (as there are not many points remaining for the 1.5ns edge). So I can average. But the risetime also depends on the sin(x)/x option being on or off. Only with four channels enabled the sin(x)/x option can be disabled. With one or two channels it's always enabled. The sin)x)/x option seems to change the filtering from flat to Gaussian. So less ringing but also less bandwidth if you disable the option.

CH #	Sinc	Average	Risetime
1	On	Normal	1.5ns
2	On	Normal	2.11ns
2	On	Averaged	2.09ns
4	Off	Normal	4.61ns
4	On	Normal	2.83ns
4	Off	Averaged	4.53ns
4	On	Averaged	3.34ns

I also captured the single channel data and plotted it with octave. That gives a feeling of the raw data the scope captures. You can image what happens for lower sample rate:

[nctnico]

I'd use an (RF) generator and use that to determine the -3dB point. No need to derive anything from the risetime which may not be measured accurate enough.

[TheSteve]

I'd use an (RF) generator and use that to determine the -3dB point. No need to derive anything from the risetime which may not be measured accurate enough.

Agreed - and people will find the 3dB point is not 150 MHz(or 200 lol), it is generally between 115 and 135 MHz.

[TheoB]

Do you have a rough indication of how it varies and by how much?

Such gain variations aren't uncommon. Sometimes it is due to a differing number series amplifiers being switched in/out, or because an analogue multipliers bandwidth is level dependent.

That's the information I posted when I started the thread. For a not upgraded scope the difference I find huge (+50% more bandwidth at 500mV and higher). I think indeed it has to do with amplifiers being switched in. Since they try to limit a faster intrinsic scope I would have expected a much more accurate bandwidth.
And the scope only need to do 50MHz, so there is no issue at all. It's just something to mention if you are considered about the bandwidth. I use my scope as spectrum analyzer. It has a dynamic range of 100dB (yes, that's not what you get on your screen, that's with postprocessing in Matlab/octave). From my experiments I can use this scope up to about 150MHz as an SA. It has some shortcomings, but I don't want to spend a lot of money on the real thing 

[TheoB]

I'd use an (RF) generator and use that to determine the -3dB point. No need to derive anything from the risetime which may not be measured accurate enough.

I don't have one  at home
A step response says more about the quality of the scope frontend than a single figure of merit like bandwidth or risetime. At the end you might be using the scope to measure a fast changing signal and wonder if the signal is not properly terminated. It is, it's your scope not performing so great. I traded my 30kg Philips 200MHz sampling scope for this little baby. And I'm very happy about it, but there are things that are just not as good anymore. You know what I also missed today? The scope has no variable horizontally  timebase (vernier)  .
After I discovered I saw others had already reported it before. That's a much bigger problem of this scope for me than any bandwidth above 100MHz.....

[nctnico]

A step response says more about the quality of the scope frontend than a single figure of merit like bandwidth or risetime.

That is a different discussion. You can also choose to make a graph of the amplitude versus frequency using a levelled (or somewhat accurate) generator. Still you have to take into account that getting a signal into a 1M Ohm with (ball park) 15pf paralllel typical oscilloscope input is somewhat challenging if you want to get it absolutely right. A 50 Ohm terminator doesn't really do the trick here because the 15pf capacitor has an impedance around 100 Ohms at 100MHz.

[TurboTom]

Coudn't resist to test the DS1054Z (liberated) bandwidth the "classic" way. I'm not too sure how accurate my measurement is but the ballpark should be correct. I used the following setup: Signal source is an SSA3021X TG in zero span mode. The TG output is routed to the DS1054Z CH1 input where it is connected via a BNC T. The other side of the BNC T is routed back to the SA input by an identical BNC cable (so everything should be well terminated). I manually selected several individual frequencies and adjusted the TG output level to provide a constant reading on the SA input (within a range of +- 0.5dB). The input sensitivity on the o'scope was selected at 100mV/div. With this configuration, I found the measured Vpp to drop to 0.707 of the value at 50MHz not before I reached 299MHz!

Can someone try to confirm this -- possibly cable reflections might have affected my measurements?? A similar test without TG level compensation and a 50 Ohm terminator at the "free end" of the BNC T already showed soemthing like 263MHz 3dB single channel bandwidth on my DS1054Z, so the result may also well be accurate.

Cheers,
Thomas

[TheoB]

Coudn't resist to test the DS1054Z (liberated) bandwidth the "classic" way. I'm not too sure how accurate my measurement is but the ballpark should be correct. I used the following setup: Signal source is an SSA3021X TG in zero span mode. The TG output is routed to the DS1054Z CH1 input where it is connected via a BNC T. The other side of the BNC T is routed back to the SA input by an identical BNC cable (so everything should be well terminated). I manually selected several individual frequencies and adjusted the TG output level to provide a constant reading on the SA input (within a range of +- 0.5dB). The input sensitivity on the o'scope was selected at 100mV/div. With this configuration, I found the measured Vpp to drop to 0.707 of the value at 50MHz not before I reached 299MHz!

Can someone try to confirm this -- possibly cable reflections might have affected my measurements?? A similar test without TG level compensation and a 50 Ohm terminator at the "free end" of the BNC T already showed soemthing like 263MHz 3dB single channel bandwidth on my DS1054Z, so the result may also well be accurate.

You still have a discontinuity at the scope. It's 20pF or so. So the signal from the TG reaches the scope input and sees 20pF in parallel with 50Ohm going to your SA input. That might give reflections that are frequency dependent. It will then depend on the length of your cables and the Instruments termination accuracy.
Note that I terminated the input at 50Ohm andere have the same issue. Any reflections should be covered by my 50Ohm generator (so not be reflected again). It would have been better to add an attenuator I guess. You better use a power splitter/isolater if you have those.

[TurboTom]

Thanks TheoB -- Good idea! I'm a litttle bit short on SMA / BNC adapters but I managed to route a MiniCircuits 15dB directional coupler into the RF path which now looks like this:

SSA3021X TG -> Coupler IN
Coupler Coupled OUT -> SSA3021X SA IN
Coupler OUT -> BNC T on DS1054Z CH1 with 50 Ohm terminator on the free BNC T port

With this setup, the results look much more reasonable but still impressive: I find the 3dB point of the DS1054Z to be at 195MHz. By swapping the BNC end of my SMA -> BNC adapter cable between the BNC T at the o'scope input and the input of a second SA (Rigol DSA815TG), I verified the level was stable within +- 0.2dB. So this should actually be a pretty reliable test I guess. The only factor that could still lead to too high figures is the ancient BNC 50 Ohms terminator that I used on the BNC T (haven't got a BNC feed-through terminator, should put it on my shopping list...).

Okay, just checked the BNC terminator (with cable but less the scope input) with a return loss bridge to provide at least 25dB of reflection damping at 200MHz, so this should be good enough for this test.

Cheers,
Thomas

[nctnico]

At 100MHz you won't have reflections unless you use cables >50cm. The biggest contributor to the error is the impedance of the capacitor. With (for example) a 15pf capacitance in parallel the total impedance gets lower so at 100MHz a 15pf capacitor parallel with a 50Ohm terminator at the input results in an impedance of 45 Ohms. This is a 10% error at a mere 100MHz.

[tggzzz]

Can someone try to confirm this -- possibly cable reflections might have affected my measurements?? A similar test without TG level compensation and a 50 Ohm terminator at the "free end" of the BNC T already showed soemthing like 263MHz 3dB single channel bandwidth on my DS1054Z, so the result may also well be accurate.

Generator -> 2m of 52ohm cable -> T-piece to scope 1Mohm//20pF -> 2m of 52ohm cable -> 50ohm terminator. Scope: 10ns/div, Tek 485.

Note the "notches" due to the 20pF, and other reflections due to the 52ohm/50ohm mismatch.

[MrWolf]

Many thanks, TheoB! Good to see posts from someone who (a) knows what he is doing, and (b) gives a balanced view of the DS1054Z, putting things in perspective, mentioning strengths and limitations without much fanfare.

Slightly disagree, this thread is mostly about strengths. "Fanfare" usually starts when someone mentions limitations, some just wont shut up w/o good fight Has been going on since scope came out if you use search button a bit. So just for newcomers I'm gonna make short actually balanced summary here: This scope is indeed good if you have actual need the speed. That means sitting in 5...20ns timebases (or according zoom level). If your actual needs are located below these timebases, horizontal precision is heavily crippled compared to almost any other modern DSO. 

[MrWolf]

CH #	Sinc	Average	Risetime
1	On	Normal	1.5ns
2	On	Normal	2.11ns
2	On	Averaged	2.09ns
4	Off	Normal	4.61ns
4	On	Normal	2.83ns
4	Off	Averaged	4.53ns
4	On	Averaged	3.34ns

From this good table showing effects of sampling rate on bandwidth another thing becomes evident:
Z would be a real kicker if it supported ETS. Because with ETS it would have full analog bw on all channels on repetitive signals.   Did not the old models have it? DS1102* 25GSa/s, DS1052* 10GSa/s. Good stuff if you know how to use it 
Edit: Various scopes tested for risetime here (incl Rigol). Effective use of ETS vs RTS demonstrated (DSO7104B, 54831D):
https://youtu.be/mS3sCJd_GPk?t=15m36s

[rf-loop]

CH #	Sinc	Average	Risetime
1	On	Normal	1.5ns
2	On	Normal	2.11ns
2	On	Averaged	2.09ns
4	Off	Normal	4.61ns
4	On	Normal	2.83ns
4	Off	Averaged	4.53ns
4	On	Averaged	3.34ns

From this good table showing effects of sampling rate on bandwidth another thing becomes evident:
Z would be a real kicker if it supported ETS. Because with ETS it would have full analog bw on all channels on repetitive signals.   Did not the old models have it? DS1102* 25GSa/s, DS1052* 10GSa/s. Good stuff if you know how to use it 

Take true pure sine wave and measure true BW again. You are perhaps in the Fourier trap with edge what is undefined. Sure is that it is not pure step as in "school books". Also you need know that Z box Sin(x)/x function is real joke. (evidences published here in forum many times.
Edge risetime measurement can use for rough BW estimation but all calculus are based for step what follow fourier serie perfectly (and with pure gaussian BW shape). Now you measure so that you do not know nearly anything what is true signal shape exactly in BNC input connector. 
You really do not know how to "fool" scope so that it display faster risetime than true BW.  7 pints wink: Think harmonics what are effective in BNC input connector and think levels. Do you know if all what input see  goes like example: 1/3, 1/5, 1/7, 1/11  and so on...

Theory is nice. But, if practice differs from theory: Theory as it is used is wrong.

[TheoB]

From this good table showing effects of sampling rate on bandwidth another thing becomes evident:
Z would be a real kicker if it supported ETS. Because with ETS it would have full analog bw on all channels on repetitive signals.   Did not the old models have it? DS1102* 25GSa/s, DS1052* 10GSa/s. Good stuff if you know how to use it 

Yes ETS (Equivalent Time Sampling, I had to look this one up :-) ) is what my old Philips PM3320A used to have. Fine for repetitive signals. The scope had only 250Msa/s but a bandwidth of 200MHz. I must say, I'm more happy with my little Rigol. Much less noisy, smaller, real time sampling up to 1Gs, 24M deep memory, I love it  . Adding ETS would help in the case you need accurate timing with four channels. But that's only for risetimes/delays outside of the spec of the scope (7ns!!). So we cannot complain can we?

If you have a 2ns rise time you get 2 samples at 1Gsa/s. But since the scope does not trigger on the analog input signal, the two points are positioned somewhere randomly on the edge. If you capture, say 10 times, you have 20 points along the edge. The issue is now that the scope needs to decide from each set of two points found on the slope where the trigger moment is. With lower sample rate, the edge start to wobble (see earlier screenshots). There are often no samples on the edge anymore:
Vector display:

Dots display:

Still the scope triggers on the transition. This increases the jitter a lot.
With ETS a scope must trigger on the input signal itself and then take a sample somewhere on the edge with a known but random delay for each capture. More expensive scopes might trigger before the digitizer. I think the Rigol does everything in the digital domain.

[rf-loop]

More expensive scopes might trigger before the digitizer. I think the Rigol does everything in the digital domain.

Imho not today. Lot of scopes (today) do digital side trigger what have lot of advantages compared to analog side pathway old trigger system as example in Rigol DS1000E and many many similar.
(yes, analog sidepathway trigger system _Can do_  also very good but it is not easy and cheap (Example in some very advanced LeCroy models))

Read example Rohde&Scwarz RTO about this and about "oversampling" and fine interpolation for perfect horizontal positioning.

With well made full digital side trigger system trigger jitter is really small. Look example how Siglent works.
No one want old crap cheap scopes analog trigger system. These exist only in museum.

 

[MrWolf]

Adding ETS would help in the case you need accurate timing with four channels. But that's only for risetimes/delays outside of the spec of the scope (7ns!!). So we cannot complain can we?

Well DS1054Z is all about operating outside of spec And it's not so much about complaining, but supplying young players with best possible information about all the options and possibilites in the world of scopes so they could make informed choice.
ETS is indeed not very well known since it was mostly high-end DSO stuff. Now high-end slowly moves off the ETS, not because it's bad but because high-end has now good enough realtime sample rates. For example cannot find ETS in Tek MDO4024C (200MHz) spec, but it has 2.5GSa/s on all four. For 0.25GSa/s all four scope it would be most welcome. It is not expensive technology in itself, can be found on scopes cheap as ~100€.
So ETS is like warp drive for scope. You do not have much control but can go to places no amateur has gone before, just point in right direction before launch Another clever tool in toolbag so to speak...

[saturation]

This thread has gotten busy!

This data is very interesting, but is this right, memory depth is <= 30 pts?? can confound the readings markedly.  Any chance you can repeat it what you did on the link and keep memory depth consistent throughout at the highest the 1054z allows? 

https://www.eevblog.com/forum/testgear/rigol-ds1054z-bandwidth/msg1095425/#msg1095425

Enjoy.

[TheoB]

This data is very interesting, but is this right, memory depth is <= 30 pts?? can confound the readings markedly.  Any chance you can repeat it what you did on the link and keep memory depth consistent throughout at the highest the 1054z allows? 

Yes, that's just the time you see times the sample rate (5ns*12*250M=15 points from left to rigth). I indirectly choose the sample rate by enabling more channels. In vector mode it really looks ok, but that's the interpolation. Has nothing to do with samples measured :-). The two screenshots in my previous shows that more clearly.
The maximum at 5ns is 5 samples per division (1Gs/s) or 60 samples per trace. But averaging helps here.

[saturation]

Some general comments:

Tr  vs RF generator:  yes, as mentioned small Tr errors lead to large estimates in frequency response, so great care must be made in measurement.  But a fast edge is easier and cheaper to obtain today than a confirmed flat RF source. 

You can easily obtain a >= 1ns edge using any clock from fast digital logic to test systems with bandwidths to ~500 MHz. 

ideal vs real: the more instruments used deviate from ideal, the less the idealized calculations apply.

A pulse test assumes the system response is uniform throughout, for all timebases and amplification, thus an ideal DSO.

Since the Tr is assumed constant, it can easily be confirmed how far the DSO remains constant regardless of the timebases selected and amplification. 

If the rise time calculation varies by timebase, and/or vertical amplification it becomes difficult to generalize one setting for all the more the deviation occurs from the original data point, and it needs to be confirmed by an RF generator approach.

Or, you can apply a statistical approach to the response deviation and determine the degree of uncertainty in the measure you can tolerate.

[saturation]

Thanks Theo, I understand.  Do you know or have the published datasheet rise time of your EH 122 Pulse Generator?

This data is very interesting, but is this right, memory depth is <= 30 pts?? can confound the readings markedly.  Any chance you can repeat it what you did on the link and keep memory depth consistent throughout at the highest the 1054z allows? 

Yes, that's just the time you see times the sample rate (5ns*12*250M=15 points from left to rigth). I indirectly choose the sample rate by enabling more channels. In vector mode it really looks ok, but that's the interpolation. Has nothing to do with samples measured :-). The two screenshots in my previous shows that more clearly.
The maximum at 5ns is 5 samples per division (1Gs/s) or 60 samples per trace. But averaging helps here.

[TheoB]

Thanks Theo, I understand.  Do you know or have the published datasheet rise time of your EH 122 Pulse Generator?

I just have the instrument lying around. No datasheet. Could also not find anything on the internet. It supposed to be able to generate pulses as narrow as 2ns. But I cannot measure that with the rigol scope to confirm. At work I have all the tools I could ever dream about, but at home it's just for hobby  . So I could measure it, but it's heavy and I have to carry it a long way then.

[David Hess]

Thanks Theo, I understand.  Do you know or have the published datasheet rise time of your EH 122 Pulse Generator?

Even more importantly, does it have the specifications for aberrations?

One reason I have a real sampling oscilloscope is for calibrating my fast transition reference level pulse generators.  Without this, a transient response and bandwidth measurement using a fast edge is of questionable accuracy.  These measurements of the DS1054Z input bandwidth are not consistent with earlier measurements using a leveled signal generator and probably reflect the poor quality of the signal source.

[TheoB]

Thanks Theo, I understand.  Do you know or have the published datasheet rise time of your EH 122 Pulse Generator?

Even more importantly, does it have the specifications for aberrations?

One reason I have a real sampling oscilloscope is for calibrating my fast transition reference level pulse generators.  Without this, a transient response and bandwidth measurement using a fast edge is of questionable accuracy.  These measurements of the DS1054Z input bandwidth are not consistent with earlier measurements using a leveled signal generator and probably reflect the poor quality of the signal source.

Agree. I clame no accuracy. With the limited resources I have I just share my observation. That's what this thread started about:

Anyone else noticed this difference in bandwidth as a function of vertical gain?

Most likely answer is that the bandwidth is limited as some amplifier is enabled for gain settings <= 200mV

[David Hess]

Anyone else noticed this difference in bandwidth as a function of vertical gain?

Most likely answer is that the bandwidth is limited as some amplifier is enabled for gain settings <= 200mV

I do not remember which one it is but there is a variable gain amplifier used to drive the digitizer similar to the National LMH6518 used in the DS2000A series.  Whichever chip Rigol used (LMH6514 or LMH6515?), I doubt that it would cause the bandwidth to change with vertical gain in such a low bandwidth design.

The single ended to differential conversion stage however has a pair of switched equalization networks and for some reason a tail current adjustment.  On one hand while that *could* cause changes in bandwidth, it seems unlikely.  On the other hand, it is not clear why the switched equalization networks exist or why the tail current is adjustable.  My guess is that the tail current is used to adjust the common mode voltage going into the variable gain amplifier and the equalization has something to do with the separate bandwidth limiter stage.

If the bandwidth really is changing at different volt/div sensitivities, I would consider it a flaw unless it is listed as part of the specifications and even then, it is a rather annoying behavior.  In some high bandwidth designs, there is just no good way around it but at 100 MHz, it should not be an issue.  I do not know if anybody has made reliable measurements of the DS1054Z bandwidth over a wide range of volts/div settings; while it is straightforward to do, it takes a leveled signal generator, reference level pulse generator, and almost always some DC to RF attenuators to do properly.

[nctnico]

If (some) gain is needed then it is hard to make a good (flat frequency curve, low distortion) amplifier even at 100MHz.

[rf-loop]

Also Rigol total junk Sin(x)/x can be one trap.

Just do not trust any thing what you see on Rigol display.

Of course BW shape can be also different with different V/div settings. I do not remember exact V/div step(s) where it change input circuit.

Even with high quality RF generator with perfect flatness it is not so simple to measure BW shape. This is because there is not available 50ohm impedance. There need be calibration grade precision splitter and calibrated level meter assembled in scope input. (or generator what have leveled output "head" what can directly connect to scope input.)
 And if there is 15pF (+ 3pF) input capacitance parallel with terminator. It is far away from 50 ohm impedance specially if use pulse edge what have say example < 1ns risetime. And more fun. There is not 50ohm resistance and 15pF capasitance. There is "magpie's nest" network what include all, series and parallel connected resistances, inductances and capasitances and then "miracles". Whole system  can interact how ever with unknown fast edge and how it is connected.  Do not try accurate things with "GHz"  RF if really do not have knowledge and enough experience with any things but near DC. (near DC is  all <100MHz) There can drop in trap more fast than you can blink your eye.

It is NOT 50ohm impedance if you use feed thru terminator ot using T and terminator.

Edge risetime based BW wondering is... may I say. Total waste of time (in this case and with these all user errors and unknowns) . All results are wrong and more miracles to peoples who do not have any real experience with RF things. More fast edge may lead more severe and well hidden errors, example due to ringing and due to interact with system weakness. In this case this Rigol some joke design. Note that there can be also interaction with this bullshit Sinc.

Here is one example what this "1000Z-box for nice images"  can include
 https://www.eevblog.com/forum/testgear/rigol-ds1074z-weird-signal-level-problem/msg563208/#msg563208

In theory and in practice. Say example if there is real 3 - 4ns risetime in some scope. How easy it is fool rising speed test result using adjustable rise speed edge and (not visible ringing - overshoot) if other person do not know what method is used he/she stay fully fooled.  It is very easy to get total junk result than near truth if do not know signal. As long as anybody do not know what is exactly in scope input connector these wondering with strange results and accurate numbers looks like just playing fun.

With unknown pulse edge there in scope input ((it is not same as pulse output).  It is same as we measure BW shape using RF generator what flatness is highly unknown.

Unknown edge is roughly same as RF generator unknown sweep flatness in  this case. 
If you do not know generator levels with different frequencies  how you use it for define scope BW. 

[MrWolf]

While Dave dissects their bodies, I seem to have taste for their brain contents This puppy's brains are now properly torn apart and splattered across some graphs. Initially just wanted to check if there is risetime difference <=200mV/div & >=500mV/div. Kink there is in graph. But noticed that Sin(x)/x causes noticeable overshoot.

Teaser image from PDF:


Did risetime test across vertical setting ranges in 1GSa/s and 250MSa/s Sin(x)/x ON/OFF modes.
OFF mode looked a bit more logical than wavy CGI in ON mode.





So I did the second test using Testec TT-DE 112 950MHz demodulator probe as scalpel.

20, 40, 60, 80, 100, 120MHz sine.
350mVpp and 3500mVpp signal levels.
1GSa/s Sinxx=ON; 250MSa/s Sinxx=ON; 250MSa/s Sinxx=OFF

- Looked at gen output after test cable & 50ohm pass-thru using demod and Agilent U1272A. No level dropoff with rising freq, gen is good then.
- Did run first testset (1GSa/s) with demodulator and scope after pass-thru. Level dropoff detected with rising freq just like it should be over scope input.
- Compared if demodulator affected scope readings - near zero effect.
- Did run all testsets (3x2x6) w/o demodulator, recorded voltages and risetimes.

Adjusted demod amplitudes to start of each testset 20MHz reading in Excel.
Graphed demod voltages with scope voltages.

Teaser image from PDF:


Noticed:
- Sinxx has heavy effect on risetime only when looking at step response, not when looking at stable sine.
- 50mV/div and 500mV/div have so large relative voltage differences with Sinxx=ON that seem to belong to different scopes. 500mV/div overshoots big time when looking at normalized graphs.
- With Sinxx=OFF normalized response graphs are almost identical.

Conclusion:
Everything happening with Sinxx on just pure CGI. Probably Sinxx=OFF is not a filter, but (more closer to) true response that gets hack-o-boosted big time. How else to explain massive sine Vpp overshoot, especially in 250MSa/s case?

Think I had my way with this thing now. Maybe should cook it instead of christmas turkey 

[guenthert]

But noticed that Sin(x)/x causes noticeable overshoot.

Uhm, I guess that has been discussed to death elsewhere already  and isn't specific to the DS1054Z.  You might know that the signal is a square wave and hence consider the flatter graph more faithful, however given the limited sampling rate, the oscilloscope cannot know this.  For the available data and input bandwidth the sin(x)/x interpolation is the most faithful representation the oscilloscope (any oscilloscope) can give you.  Connecting the dots (samples) using straight lines implies higher frequency components which weren't (couldn't be) sampled or passed the input low pass filter.  Showing those is kind of lying.

I like the compromise the analogue discovery software (WaveForms 2015) chose: it offers Sin(x)/x ("smooth") graphs as option, but defaults to straight lines, however if the displayed graph contains few data points (much fewer than horizontal pixel?) than those are explicitly drawn as a hint to the observer that those lines in between are just made up.

[MrWolf]

Uhm, I guess that has been discussed to death elsewhere already  and isn't specific to the DS1054Z.

Gasoline engine is not specific to Ferrari either. Can be found in Robin Reliant. 

For the available data and input bandwidth 

According to available data (provided in the PDFs) Z box behaves in somewhat faithful way only in <=200mV/div vertical settings. This is confirmed by taking independent readings directly from scope input with live signal, using non-obtrusive probing. In >=500mV/div setting it artificially boosts the signal to look like more high bw scope. Once again programming, not hardware failure.

the sin(x)/x interpolation is the most faithful representation the oscilloscope (any oscilloscope) can give you

Most faithful representation is given by excessive sampling rate, either RTS or ETS. Once gain - real data.

[nctnico]

the sin(x)/x interpolation is the most faithful representation the oscilloscope (any oscilloscope) can give you

Most faithful representation is given by excessive sampling rate, either RTS or ETS. Once gain - real data.

Even better: understand the math behind sin(x)/x and you'd know/understand you don't get any extra information from an excessive sampling rate where fs/2 lies beyond the range of the anti-aliasing filter.

[MrWolf]

Even better: understand the math

Even better: understand the reality (has to do with experimental data etc  ):
ETS (raw data) superiority vs RTS+Sin(x)/x demonstrated in real word, on hightech hardware:
https://youtu.be/mS3sCJd_GPk?t=20m52s

[nctnico]

The only thing I hear in that video is could be... could be... could be... could be...  which means that he is making a nice video but has no idea what he is yabbering on about.

[MrWolf]

he is making a nice video but has no idea what he is yabbering on about.

Not everyone is Dave, he just shows it best he can. Anyway from your point of view Agilent did include ETS just because they are stupid and do not read EEVBlog.  Dunno, maybe, think gonna go make some cookies now 

[rf-loop]

the sin(x)/x interpolation is the most faithful representation the oscilloscope (any oscilloscope) can give you

Most faithful representation is given by excessive sampling rate, either RTS or ETS. Once gain - real data.

Even better: understand the math behind sin(x)/x and you'd know/understand you don't get any extra information from an excessive sampling rate where fs/2 lies beyond the range of the anti-aliasing filter.

Yes but if manufacturer do not follow any rules and name some kind of "smooth" function as Sinc  after then you can not trust any thing.
This joke violates rules. And more fun. it do not care real sampled data for Sinc draw. They want produce just "nice art image".

One limited wink what is going there is here:

https://www.eevblog.com/forum/testgear/rigol-ds1074z-weird-signal-level-problem/msg563208/#msg563208

After junk Sinc function first published by Rigol with DS1000E ancient model they have not repaired and learn anything.

1. Sinc (Sin(x)/x) function result need: Curve draw via real sampled data points. Rigol flush samples off out like if they are garbage.

If do small home work it can easy see that Rigol Sinc is joke what can result "what ever".
Example if look Siglent and and lot of others, even old Tektronix DSO (example 2440 or 2230 2430) draw Sinc tightly through the sample points (of course).
It can confirm very easy example stopping Siglent acquisition and then turn display mode dots or vectors and then Sinc on.  It can immediately see  that all draw go always via true sample points. 

Rigol do not. And even more bad, in many cases user can not even check because Sinc (better say "smooth" or what ever but not Sinc) can not switch on and off. Also it draw fake dots between sample dots. 

[David Hess]

No one want old crap cheap scopes analog trigger system. These exist only in museum.

Ya, who needs equivalent time sampling when they can have Gibb's phenomena distorting their transient response!  100 MHz DSOs with 2 GS/s ETS sample rates and 500 picoseconds of jitter were junk!  And 20 GS/s with 50 picoseconds is even worse!

As a practical manner, analog trigger jitter in a 100 MHz oscilloscope is easy to make insignificant.  Even in a 300 to 500 MHz oscilloscope, it should be insignificant.

When a digital trigger is used, then it appears to produce zero jitter on the triggered edge however this is misleading.  The sinc reconstruction error caused by aliasing on a fast edge (wobulation) produces its own jitter which gets added to the edge of the following cycle and the rest of the waveform.  To put it another way, aliasing makes the trigger point ambiguous.  You can see this on the old LeCroy marketing videos where even when looking at a single edge, only the trigger point itself is noise free which to me looks as weird as Gibb's phenomena.

If do small home work it can easy see that Rigol Sinc is joke what can result "what ever".
Example if look Siglent and and lot of others, even old Tektronix DSO (example 2440 or 2230) draw Sinc tightly through the sample points (of course).

The old Tektronix 2440 series is a great example of this because it can draw the sinc interpolated signal while also highlighting the real sample points and they always line up like they should.  The 2230 and 2232 however do not support sinc interpolation at all which makes sense given their relatively slow 8088 and 80C188 processors running at 20MHz/3=6.7MHz and 100MHz/7.5=13.3MHz respectively.  The 2440 uses an 8MHz 6809 (system), 4 MHz 6805 (controls), and separate custom waveform processor.

If you drive the 2440 with a fast edge and do not use equivalent time sampling, then the Gibb's phenomena also shows up when sinc interpolation is used which is to be expected given a rise time to support 300 MHz operation and only a 500 MS/s sample rate.

[rf-loop]

If do small home work it can easy see that Rigol Sinc is joke what can result "what ever".
Example if look Siglent and and lot of others, even old Tektronix DSO (example 2440 or 2230) draw Sinc tightly through the sample points (of course).

The old Tektronix 2440 series is a great example of this because it can draw the sinc interpolated signal while also highlighting the real sample points and they always line up like they should.  The 2230 and 2232 however do not support sinc interpolation at all which makes sense given their relatively slow 8088 and 80C188 processors running at 20MHz/3=6.7MHz and 100MHz/7.5=13.3MHz respectively.  The 2440 uses an 8MHz 6809 (system), 4 MHz 6805 (controls), and separate custom waveform processor.

If you drive the 2440 with a fast edge and do not use equivalent time sampling, then the Gibb's phenomena also shows up when sinc interpolation is used which is to be expected given a rise time to support 300 MHz operation and only a 500 MS/s sample rate.

Sorry my old memory. It was Tek 2430 what I had years ago as also 2440. 2430 have also Sinc as 2440.  I have never used 2230.


Generally:

With fast edges (or what ever waveform)  as example rectangle/square wave. Corners "wobbling" is just aliasing because some harmonics in signal goes over Nyquist wall. (or better say, too near it). 
Still today I can see peoples who admire and want tune up frequency response like more is better. Bullshit. Less is better. Scopes are for inspect unknown signals, not known sinewave inputs.

I have not seen any entry level or bit over this level new digital scope what have acceptable analog side filter system. Still they design front ends like in cheap analog scopes where slow decaying gaussian BW shape is fully ok.
It is not ok in digital scopes in real time "one shot" sampling mode. (ETS is different - but only for repetitive signals). It looks like digital peoples are designing oscilloscopes and no one really care about analog front end before ADC.    Same can see in some scopes what have 50ohm  input. Not even close real 50ohm impedance, exept with DC. Perhaps designer do not know what is difference between 50ohm resistance and 50 ohm impedance in case that we handle any other than DC -  ELF - VLF. 

Some scope may have 100MHz label and then real time samplerate 250MSa/s. But analog side may have example -3dB over 150MHz and perhaps -6dB over 200MHz and even 300 MHz still leaks to ADC inputs. This is wrong. And looks like no one care. Designers comment is only "so what"... and here in forum many peoples try tweak and hack for more analog BW. 

Best modification is not rise analog bandwidth but reject it and/or modify bandwidth shape more steep like "brickwall" some amount under Nyquist wall.   When samplerate drops, corner f need also drop.
Construction where every input have its own ADC is more easy. Samperate do not drop if more or less channels are in use.

Who is first company (in entry level and bit higher level scopes) who really understand how important is analog front end and its BW filter - and also do it. But, it need also explain to entry level noobs why this is important.
If scope have (well made) ETS mode for repetitive signals, in this case BW shape can be "wide open".
Some times it is fun how they add some digital filter systems to scopes but not any half word about much more important analog side filters. They can not be replaced with the filtered digitized signal.

With well designed analog front end and its BW shape and then perfect made Sin(x)/x interpolation  there do not exist any form of signal aliasing. As long as analog BW and its shape is ok for used samplerate and signal reconstruction using Sinc.

Who is first manufacturer who make well designed analog front end for real time sample mode oscilloscope. Of course other method is rise ADC sampling rate. 1GSa/s for 100MHz scope with slow decay gaussian BW shape is ok if anything over 300 - 400MHz  dot leak from input connector to ADC.   

[David Hess]

With fast edges (or what ever waveform)  as example rectangle/square wave. Corners "wobbling" is just aliasing because some harmonics in signal goes over Nyquist wall. (or better say, too near it).

With perfect sampling, the sinc filtering of a fast edge should only produce the Gibb's phenomena which is phase coherent with the input signal and looks like preshoot and overshoot.  I think the actual "wobulation" sometimes seen is caused by intermodulation distortion in the digitizer which results in additional frequency spurs which are also aliased.  In both cases, increasing the sample rate or bandwidth limiting the input signal helps.

I have not seen any entry level or bit over this level new digital scope what have acceptable analog side filter system. Still they design front ends like in cheap analog scopes where slow decaying gaussian BW shape is fully ok.

It is not ok in digital scopes in real time "one shot" sampling mode. (ETS is different - but only for repetitive signals). It looks like digital peoples are designing oscilloscopes and no one really care about analog front end before ADC.

Some of the very early Tektronix DSOs included four pole 24dB/octave Gaussian filters which should have helped with this but I do not know if their unfiltered input bandwidth also had a 4 pole Gaussian response.  On an analog oscilloscope it hardly matters except for noise measurements.

Same can see in some scopes what have 50ohm  input. Not even close real 50ohm impedance, exept with DC. Perhaps designer do not know what is difference between 50ohm resistance and 50 ohm impedance in case that we handle any other than DC -  ELF - VLF.

Most designs implement a switchable 50 ohm feedthrough termination in front of the high impedance buffer which adds significant input capacitance compromising the 50 ohm input impedance.  At 200 MHz and below this is not a problem but above that it can be; my 300 MHz 2440 handles it well enough.  Some oscilloscopes route the input signal around the high impedance buffer when in 50 ohm mode like the 350 MHz Tektronix 485 and dedicated high frequency oscilloscope usually lack a high input impedance buffer.

Best modification is not rise analog bandwidth but reject it and/or modify bandwidth shape more steep like "brickwall" some amount under Nyquist wall.   When samplerate drops, corner f need also drop.

...

With well designed analog front end and its BW shape and then perfect made Sin(x)/x interpolation  there do not exist any form of signal aliasing. As long as analog BW and its shape is ok for used samplerate and signal reconstruction using Sinc.

Except for raising the real time sample rate or using equivalent time sampling, there is no good solution for this in a time domain instrument.  If the filter shape does not have linear phase, then the filter itself will effectively create the same problem which rules out filters which have higher performance in the frequency domain.  The best solution I have seen is in some early Tektronix DSOs which implemented 4 pole Gaussian bandwidth filters.

[rf-loop]

With fast edges (or what ever waveform)  as example rectangle/square wave. Corners "wobbling" is just aliasing because some harmonics in signal goes over Nyquist wall. (or better say, too near it).

With perfect sampling, the sinc filtering of a fast edge should only produce the Gibb's phenomena which is phase coherent with the input signal and looks like preshoot and overshoot.  I think the actual "wobulation" sometimes seen is caused by intermodulation distortion in the digitizer which results in additional frequency spurs which are also aliased.  In both cases, increasing the sample rate or bandwidth limiting the input signal helps.

I have not seen any entry level or bit over this level new digital scope what have acceptable analog side filter system. Still they design front ends like in cheap analog scopes where slow decaying gaussian BW shape is fully ok.

It is not ok in digital scopes in real time "one shot" sampling mode. (ETS is different - but only for repetitive signals). It looks like digital peoples are designing oscilloscopes and no one really care about analog front end before ADC.

Some of the very early Tektronix DSOs included four pole 24dB/octave Gaussian filters which should have helped with this but I do not know if their unfiltered input bandwidth also had a 4 pole Gaussian response.  On an analog oscilloscope it hardly matters except for noise measurements.

Same can see in some scopes what have 50ohm  input. Not even close real 50ohm impedance, exept with DC. Perhaps designer do not know what is difference between 50ohm resistance and 50 ohm impedance in case that we handle any other than DC -  ELF - VLF.

Most designs implement a switchable 50 ohm feedthrough termination in front of the high impedance buffer which adds significant input capacitance compromising the 50 ohm input impedance.  At 200 MHz and below this is not a problem but above that it can be; my 300 MHz 2440 handles it well enough.  Some oscilloscopes route the input signal around the high impedance buffer when in 50 ohm mode like the 350 MHz Tektronix 485 and dedicated high frequency oscilloscope usually lack a high input impedance buffer.

Best modification is not rise analog bandwidth but reject it and/or modify bandwidth shape more steep like "brickwall" some amount under Nyquist wall.   When samplerate drops, corner f need also drop.

...

With well designed analog front end and its BW shape and then perfect made Sin(x)/x interpolation  there do not exist any form of signal aliasing. As long as analog BW and its shape is ok for used samplerate and signal reconstruction using Sinc.

Except for raising the real time sample rate or using equivalent time sampling, there is no good solution for this in a time domain instrument.  If the filter shape does not have linear phase, then the filter itself will effectively create the same problem which rules out filters which have higher performance in the frequency domain.  The best solution I have seen is in some early Tektronix DSOs which implemented 4 pole Gaussian bandwidth filters.

Yes, all this is, in practice, quite complex and need do some compromise.

Of course there is many solutions and all these
Rising real time samplerate related to BW.  (in low gfrequency scopes today quite easy)
ETS is still useful in some cases but only if quality is good
Better filters in analog front end.  Too often designers forget this in low end scopes!
Doing as perfect Sinc as possible.  Even most of low end scopes they try - but as seen, some not even try.

Here is examples and not for argumenting if Siglent is good or bad. Only just for example for some random readers who have total lack of knowledge about these things. (Examples made using Siglent SDS1102X+ but this is not important here)

First image:
Some rough image about BW shape.  (but not for evaluate this naamed scope real BW or BW shape, only for example for this discuss)
Note that this scope is named as 100MHz.
Image is made using slow sweep from around 1MHz to 360MHz linearly and sweep time 6s. One horizontal div 30MHz
There is vertical cursors what show around -3dB level related to low freq start.
Then there is marker A. It is around 100MHz position. (model name freq)
Then there is marker B. It is around 250MHz. If this scope is used both channels on, samplerate is 500MSa/s what means that Nyquist wall is 250MHz.
We can see that even with this frequency leevel have not dropped even 6dB. All what ADC get over this frequency when 2 channels is in use produce aliasing. And this need understand that also example square wave harmonics go very easy over this if there is enough fast risetime.

Sine wave aliasing is what it is  all know it.

Here is example about other form of aliasing. Corners wobbling.

Second  image:
1GSa/s and not visible aliasing. Also it can see when look rise time measured statistics (look sdev)

Third image:
sample rate is lowered to 500MSa/s. (two channels in use, CH2 trace is top of image)
Now we can see high amount of aliasing.  Also risetime statistics show it.
But quess what, if we take now couple of one shots we can find also waveforms what show "better risetime" . With 500MSa/s there can find single shots what have 1.9ns risetime and with 1GSa/s fastest risetime is 2.1ns. Fun, is it.  If I do some "trics" using example transfer line ringing I can get really weird results.  Here in these examples signal is known. Source is health Tektronix 284 pulse generator what produce around 70ps risetime without here meaningful overshoots and it have also quite flat top enough time.  But, as can see in first image, analog front end filter is not ok for 500MSa/s.  If analog front end have more steep filter so that 100MHz is not attenuated and -3dB is example 150MHz and then 200MHz is highly attenuated - corners wobbling (risetime variations) is away.
 
After then think Rigol 250MSa/s... and its wide open analog front end... and still some peoples want it even more wide BW.
When you see something you need next think is it DUT or scope what do it. Experienced people may easy regognize things quite easy (but in some cases it also need extra thinking and time more or less)

Of course analog scopes are naaturally totally free from these aliasing things what ever channels are in use and what ever horizontal speed. (this is why I still use analog scope in some repair/service situations wwith really unknown signals. I need make money, not spend my time for fun)




This image is NOT VALID for evaluate this used oscilloscope real BW or its shape. This is because sweep generator level accuracy (flatness) is not checked with sufficient reliability. But this image can use for this subject in this message.


Fast edge from TEK 284. Scope 1GSa/s and Sin(x)/x ON
No visible aliasing.



Fast edge from TEK 284. Scope 500MSa/s and Sin(x)/x ON
Strong visisible aliasing.  Because there is frequency components over Nyquist limit what are not enough reduced before ADC
(look BW shape image.)

[David Hess]

I know of at least one person who put a motorized control on their SG503 or SG504 leveled generator so they could do accurate oscilloscope and probe sweeps.

I wish someone with a Tektronix 284 pulse generator or equivalent like a PG506 which is acceptable up to 100 MHz would run these tests on a DS1054Z before and after the bandwidth hack.  The best test I have seen shows a problem.

[nctnico]

IMHO there is not much use in trying to use an oscilloscope beyonds it's specifications especially a lower end one without dedicated 50 Ohm inputs. For starters there is the capacitance of the typical 1:10 probes which is a huge load for frequency components in the 100MHz region. And even if you use external 50 Ohm terminators the input capacitance of the scope will screw things up. At the end of the day you are just stacking errors on top of errors when trying to use an instrument beyond it's limits.

[David Hess]

IMHO there is not much use in trying to use an oscilloscope beyonds it's specifications especially a lower end one without dedicated 50 Ohm inputs. For starters there is the capacitance of the typical 1:10 probes which is a huge load for frequency components in the 100MHz region. And even if you use external 50 Ohm terminators the input capacitance of the scope will screw things up. At the end of the day you are just stacking errors on top of errors when trying to use an instrument beyond it's limits.

It is not quite as bad as that.  I think a larger problem is verifying performance without having a reference level pulse generator and leveled RF source.  And if you build, buy, or repair either, how do you calibrate them?

Probes are specified with a terminated 50 ohm source so they see a 25 ohm source impedance.  This is unrealistic of course for most applications but the 15 picofarad shunt capacitance of a x10 probe isn't too bad at 100 MHz if the source impedance is low.  It is even acceptable at 200 MHz.  A ground lead if used is at least as big a problem.

As the source impedance or frequency rises, other probe types need to be used like low-z passive probes or active probes.  Sometimes I just design in an emitter or source follower or transconductance output to drive a probe or 50 ohm transmission line to the test instrument. 

If a feedthrough attenuator is used or internal 50 ohm termination, then an oscilloscope's high impedance buffer sees 25 ohms.  I get practically identical results between using an external feedthrough termination, internal termination, and x10 probe attached to a termination (1) at 100 MHz and 200 MHz bandwidths.  Of course I should get identical results because of how probes and oscilloscope inputs are calibrated.  They are suppose to generate identical results with 50 ohm terminated sources providing a 25 ohm source impedance.

(1) They make or made special coaxial probe feedthrough terminations which place the 50 ohm termination as close to the probe tip as possible.  These are naturally expensive or unobtainable (2) but up to 200 MHz, a BNC to probe tip adapter connected to a BNC feedthrough termination works well enough.  A better home made solution would involve a probe tip to printed circuit board adapter and a coplanar feedthrough termination.

(2) Find them in the aisle next to the female BNC connectors which have a built in disk resistor feedthrough termination.

[tggzzz]

IMHO there is not much use in trying to use an oscilloscope beyonds it's specifications especially a lower end one without dedicated 50 Ohm inputs. For starters there is the capacitance of the typical 1:10 probes which is a huge load for frequency components in the 100MHz region. And even if you use external 50 Ohm terminators the input capacitance of the scope will screw things up. At the end of the day you are just stacking errors on top of errors when trying to use an instrument beyond it's limits.

I wonder how anybody ever built and verified the performance of the worlds highest performance scope/SA/etc

[nctnico]

IMHO there is not much use in trying to use an oscilloscope beyonds it's specifications especially a lower end one without dedicated 50 Ohm inputs. For starters there is the capacitance of the typical 1:10 probes which is a huge load for frequency components in the 100MHz region. And even if you use external 50 Ohm terminators the input capacitance of the scope will screw things up. At the end of the day you are just stacking errors on top of errors when trying to use an instrument beyond it's limits.

I wonder how anybody ever built and verified the performance of the worlds highest performance scope/SA/etc

Apply math and physics like they use the 'Josephson effect' to produce a voltage reference.

[David Hess]

I wonder how anybody ever built and verified the performance of the worlds highest performance scope/SA/etc

Apply math and physics like they use the 'Josephson effect' to produce a voltage reference.

It was more like applied engineering because the math and physics only go so far.

At least for time domain instruments like oscilloscopes and reference level pulse generators, they used *real* sampling oscilloscopes.  (1)  A sampling oscilloscope can also be used to calibrate a leveled sine wave oscillator and may be the best (least expensive) option for a hobbyist but typically a thermal RMS based meter would be used for this.

The original reference level pulse generator from NBS (now NIST), which shares a very similar design with the fast rise/fall outputs of the PG506 and earlier reference level pulse generators from Tektronix, still relied on a sampling oscilloscope (1) to verify its performance (instead of the reverse) and its design was already an order of magnitude too slow for the highest performance instruments of that time.

I think I read that at some point Tektronix sent a 70ps Type 284 tunnel diode pulse generator to NBS/NIST so they could certify it and it could become a gold standard at Tektronix but I have no idea how NBS/NIST went about doing that.  Then this gold Type 284 revealed that the original S-2 sampling heads were actually 75ps instead of 50ps.  I think the problem was that the big GR-874 RF connectors were too large to support TEM-mode propagation above 4 GHz.

These days NIST uses electro-optical sampling and an online search will reveal all kinds of information on this subject.

(1) Check out the calibration instructions for a Tektronix 067-0587-02 which is used to calibrate the Tektronix 7104 1 GHz real time oscilloscope.  They include using a 4 GHz sampling oscilloscope to calibrate the reference flat pulse generator output.  The 1 GHz 7A29 vertical amplifier for the 7104 also requires a 4 GHz sampling oscilloscope for calibration.

(2) I can tell from some of the photographs in the documentation that one of the sampling oscilloscopes NBS used was a 7000 mainframe with probably a 4 GHz S-2 sampling head.  Some of the other photographs show an older sampling oscilloscope made by HP and it looks like they had a different third sampling oscilloscope as well.