EEVblog® Electronics Community Forum
Products => Test Equipment => Topic started by: JacquesBBB on December 01, 2015, 07:16:36 pm
-
This is a question following my earlier post
https://www.eevblog.com/forum/beginners/a-few-question-on-jw-pulse-generator-based-on-2n3904/ (https://www.eevblog.com/forum/beginners/a-few-question-on-jw-pulse-generator-based-on-2n3904/)
directly related, but I prefer to start a new thread uniquely (if possible ) devoted to the question of the
measure of the bandwith of the Rigol DS1054z with a Jim Williams pulse generator. The other post is devoted to the construction of pulse generators and I will continue to post there my various experiments on the matter.
I have read Tim answers in this first thread,
I have read
https://www.eevblog.com/forum/beginners/bandwidth-calculation-caveats!/msg807413/#msg807413 (https://www.eevblog.com/forum/beginners/bandwidth-calculation-caveats!/msg807413/#msg807413)
and I still do not understand.
The problem is that I find a rise time r_t of 1.2 ns on the Rigol DS1054z that corresponds
to a bandwidth BW = 0.35 / r_t = 292 Mhz , way off the advertised 100 Mhz.
Tim proposed that i this is due to sin(x)/x interpolation, or vector display.
I have done it again with dots display, and I have the same results.
Suppressing sin(x)/x interpolation is more difficult, as it is only possible (as far as I understand)
with 3 or 4 channels.
In this case, the sampling rate falls to 250 Ms/s. According to
http://m.eet.com/media/1140862/19209-263113.pdf (http://m.eet.com/media/1140862/19209-263113.pdf)
it is thus normal to have a smaller bandwidth
At this sampling rate, I found nearly the same curve, whether or not I use the
sin(x)/x interpolation.
The figure given by the Rigol measure is
4.05 ns without sin(x)/x - > 86 Mhz
(https://www.eevblog.com/forum/beginners/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=184968)
and
3.75 ns with sin(x)/x -> 93 Mhz
(https://www.eevblog.com/forum/beginners/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=184970)
but this is only due to the difference of maximum, not to a difference in the slope, and this
is minor compared to the huge difference at 1 Gs/s.
When I increase to 500Ms/S I get 2.8 ns - > 125 Mhz ( sin(x)/x on)
(https://www.eevblog.com/forum/beginners/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=184972)
and at 1Gs/s 1.4 ns -> 250 Mhz (sin(x)/x) on).
(https://www.eevblog.com/forum/beginners/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=184974)
Edit :
So my question remains : what is the real bandwidth of the Rigol DS1054z at full sample rate (1Gs/s) ?
How to relate the rise time to the band width ? All the notes I have seen are minor differences, of a few (max 20) %,
here we speak of a factor of 2-3.
-
Have you used a -20dB (or more) attenuator yet?
Tim
-
A liberated one is about 2.3 to 2.4ns.
I did a video on it here with a liberated DS1000 at 16:26.
http://youtu.be/mS3sCJd_GPk (http://youtu.be/mS3sCJd_GPk)
-
@Tim :
Yes, as you can see the amplitude of the pulse is now very much reduced to about 1.5 V.
I made an atenuator with two 68 R and 160R = 150R +10R.
according to http://chemandy.com/calculators/matching-pi-attenuator-calculator.htm (http://chemandy.com/calculators/matching-pi-attenuator-calculator.htm)
The atenuation should be 16.33 db with no reflection.
@Howardlong : I see you have 2.4ns on the Rigol DS1054z, at 1Gs/s . In the same setting, I have 1.4ns . You can see it in the picture above.
How do you explain that ? My pulse is certainly not as good as yours, but this should make my raising time even worse.
How can I get 1.4 ns ?
It would be interesting to have some other measures on the rise time of the Rigol DS1054z.
-
Get a leveled RF signal generator and increase frequency until amplitude drops to 70% of original amplitude. Then you know -3dB bandwidth. Simpler than using a pulse generator with unknown specs
-
Thanks for the tip - that works simple enough. On a MSO2072A I got about 130 Mhz at -3dB (based on 70% of the original amplitude measured at 70 MHz).
-
@Tim :
Yes, as you can see the amplitude of the pulse is now very much reduced to about 1.5 V.
I made an atenuator with two 68 R and 160R = 150R +10R.
according to http://chemandy.com/calculators/matching-pi-attenuator-calculator.htm (http://chemandy.com/calculators/matching-pi-attenuator-calculator.htm)
The atenuation should be 16.33 db with no reflection.
@Howardlong : I see you have 2.4ns on the Rigol DS1054z, at 1Gs/s . In the same setting, I have 1.4ns . You can see it in the picture above.
How do you explain that ? My pulse is certainly not as good as yours, but this should make my raising time even worse.
How can I get 1.4 ns ?
It would be interesting to have some other measures on the rise time of the Rigol DS1054z.
If there is ringing or overshoot on the signal (not visible due to other effects such as sample rate) that would produce such a scenario. Also I am not sure what construction techniques you have used. If you are simply using a few wire ended resistors dead bug style that might well explain any overshoot and ringing. We haven't sen the construction method for your JW pulse generator either. I am not saying they are wrong, but when you talked about stringing four resistors together for the attenuator I wasn't sure how it was done, and layout will be fairly crucial.
Does your JW pulser have a pulse stretcher on it such as a piece of coax? It looks like it does from the scope traces.
As an alternative to the attenuator, you could also try a single 51 ohm resistor at the scope end of a BNC cable. If you use right angle terminations, you can, if you're careful, place 0603 or 0805 resistors inside the RA plug.
FWIW I noticed the same effect you have some time ago on the Rigol and I never managed to explain it away fully. However I've learned a lot since then in the realm of high speed time domain measurements, so maybe now's the time to figure it out.
I am not sure why the Rigol doesn't allow you to switch off sin(x)/x, on 1 or 2 channels, it's annoying.
-
Get a leveled RF signal generator and increase frequency until amplitude drops to 70% of original amplitude. Then you know -3dB bandwidth. Simpler than using a pulse generator with unknown specs
This is "botter en touche" (kick into touch, evade the issue, dodge the issue, pass the buck (translation from http://www.wordreference.com/fren/botter%20en%20touche (http://www.wordreference.com/fren/botter%20en%20touche))).
The question is : what is the smallest rise time possible for the Rigol DS1054z ?
This is what I want to understand. The fact that the specs of the pulse generator are not precisely known should not be an issue. Whatever you do, as far as I understand, you will never be able to measure a rising time with smaller value than the rise time of the oscilloscope.
measured r_t > oscillo r_t with the approximate value measured r_t = sqrt( oscillo r_t^2 + pulse generator r_t^2)
-
@Howardlong :
The construction of my pulse generator is on
https://www.eevblog.com/forum/beginners/a-few-question-on-jw-pulse-generator-based-on-2n3904/ (https://www.eevblog.com/forum/beginners/a-few-question-on-jw-pulse-generator-based-on-2n3904/)
as said in the beginning of this thread.
Yes I put a coax, but could improve it by changing to more adapted impedance.
Yes I terminated with a T and 50R plug on the scope.
FWIW I noticed the same effect you have some time ago on the Rigol and I never managed to explain it away fully. However I've learned a lot since then in the realm of high speed time domain measurements, so maybe now's the time to figure it out.
I am pleased to see that someone else wants to understand.
I am not sure why the Rigol doesn't allow you to switch off sin(x)/x, on 1 or 2 channels, it's annoying.
I have verified that it is written so in the Rigol docs. (Edit) I agree that this is very bad. They should add the possibility to
remove this interpolation in a firmware upgrade.
In addition, I have a question :
When rise time is mentioned, the amplitude of the signal is not mentioned. How does the rise time depends on the amplitude of the step signal ?
-
I am pleased to see that someone else wants to understand.
Others do too :)
In addition, I have a question :
When rise time is mentioned, the amplitude of the signal is not mentioned. How does the rise time depends on the amplitude of the step signal ?
One simple point is that if any of the front-end components are driven out of their linear range, then all bets are off. It is possible that a narrow strong pulse could exceed such limits.
That's true of analogue scopes as well, and Jim Williams has written good articles on how this has to be understood, and which (old) scopes are less susceptible to it. I can't be bothered to search for a reference, but it does everybody good to real all of Jim Williams output :)
-
From
https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/msg800232/#msg800232 (https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/msg800232/#msg800232)
Probably variation took the bandwidth safety margin a bit higher than usual. Mine with the "hack" tests well over 200 MHz with the risetime method, very close to the theoretical upper limit of their frontend design.
(https://misc.c4757p.com/DS1Z_QuickPrint8.png)
I see that other have found a resul for rise time (1.5ns) that is very close to the 1.4ns I got with my pulse generator.
Does anybody else has some equivalent measure that he can post here ?
-
From
https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/msg800232/#msg800232 (https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/msg800232/#msg800232)
Probably variation took the bandwidth safety margin a bit higher than usual. Mine with the "hack" tests well over 200 MHz with the risetime method, very close to the theoretical upper limit of their frontend design.
(https://misc.c4757p.com/DS1Z_QuickPrint8.png)
I see that other have found a resul for rise time (1.5ns) that is very close to the 1.4ns I got with my pulse generator.
Does anybody else has some equivalent measure that he can post here ?
There is a thread that talks about Jim Williams pulse generators and related info here:
https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/615/ (https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/615/)
You can find rise time measurements done with various scopes in the thread.
fwiw, I just measured rise time from a pulse gen on a MSO2072A and got about 3.85ns. The Rigol spec (which is just based on a calculation) says that the 70 MHz scope should have a rise time of 5 ns and a 200 MHz version should have a rise time of 1.8ns. Rigol scopes tend to outperform their rated bandwidth spec so it's believable that a 200 MHz version tests in the vicinity of 1.5ns.
Be careful of going too far down this road - you might get enticed into achieving picosecond rise times - which could lead to a new scope :). It starts with just wanting to see sub-nanosecond rise times (999 ps or faster - just so you can see ps instead of ns); then you want less than 500 ps, then less than 100 ps and so it goes. Rise time measurements can be a gateway drug and the next thing you know you are on your way to having 20 oscilloscopes. :)
But this web site can not only create addictions it can also help cure or avoid them. You can checkout the posts from Jay_Diddy_B, Howardlong and others and live vicariously by observing very fast rise times.
On the other hand if you want to measure really short duration stuff you could wind up with one of these:
-
In addition, I have a question :
When rise time is mentioned, the amplitude of the signal is not mentioned. How does the rise time depends on the amplitude of the step signal ?
Hi Jacques,
The best way to understand this is first consider a rotating vector i.e., a sinewave, then ask yourself where do the highest slew rates(gradients) occur?. Naturally, this happens at the zero crossings, correct?The next question is, how to increase the gradient at the zero crossing? It turns out there are two ways of doing it, first is by increasing the frequency that one is obvious, then there is amplitude, i.e., if you increase a sinewaves amplitude the zero crossings(gradient(dV/dt) increase as well, which is not so obvious.
So, we have i=Isin(wt), we want to know what the gradient is at the zero crossings:
di/dt=I.w.cos(w.t), cos(0)=1
so di/dt=I.w.....where w=2.pi.f
Since a cpu performance throughput is frequency only, it makes sense to characterize the silicon die to 1.1V instead of 5V and then we can run our pc's at 1.8GHz!
Any joy?
Cheers
Commie
-
There is a thread that talks about Jim Williams pulse generators and related info here:
https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/615/ (https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/615/)
You can find rise time measurements done with various scopes in the thread.
I have read this thread, but did not find the information I wanted in it.
fwiw, I just measured rise time from a pulse gen on a MSO2072A and got about 3.85ns. The Rigol spec (which is just based on a calculation) says that the 70 MHz scope should have a rise time of 5 ns and a 200 MHz version should have a rise time of 1.8ns. Rigol scopes tend to outperform their rated bandwidth spec so it's believable that a 200 MHz version tests in the vicinity of 1.5ns.
OK, but my scope is supposed to be 100 Mhz, not 200 Mhz.
On the other hand if you want to measure really short duration stuff you could wind up with one of these:
If I find a good deal on these on ebay, why not ?
-
There is a thread that talks about Jim Williams pulse generators and related info here:
https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/615/ (https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/615/)
You can find rise time measurements done with various scopes in the thread.
...
On the other hand if you want to measure really short duration stuff you could wind up with one of these:
If I find a good deal on these on ebay, why not ?
Make sure the quoted price includes shipping (and preferably installation) costs.
-
I am pleased to see that someone else wants to understand.
Others do too :)
In addition, I have a question :
When rise time is mentioned, the amplitude of the signal is not mentioned. How does the rise time depends on the amplitude of the step signal ?
One simple point is that if any of the front-end components are driven out of their linear range, then all bets are off. It is possible that a narrow strong pulse could exceed such limits.
That's true of analogue scopes as well, and Jim Williams has written good articles on how this has to be understood, and which (old) scopes are less susceptible to it. I can't be bothered to search for a reference, but it does everybody good to real all of Jim Williams output :)
Yes,if you clip the top of the signal,the rise time is decreased,but,of course,the amplitude is decreased,too.
This is a legitimate way of decreasing the rise time of a signal,& is used,both to regenerate a signal with degraded risetime,& to produce a Quasi-square wave out of a sine wave.(in the latter case,both peaks are clipped).
The signal is clipped then amplified in a stage with better frequency response than that of the previous circuitry.
-
In fact, Tim (T3sl4co1l) was right.
It looks like the sin(x)/x, combined with some overshoot on the pulse generation is artificially increasing the slope of the step function on the Rigol DS1054z.
I have largely improved my pulse generator, closely following Jim William AN94
http://cds.linear.com/docs/en/application-note/an94f.pdf (http://cds.linear.com/docs/en/application-note/an94f.pdf)
(https://www.eevblog.com/forum/testgear/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=185748)
I have now a much cleaner pulse. and find a time rise of about 1.9 ns. This is still much better than normal specification for a 100Mhz scope, but quite in par with other measures mentioned above.
The measure is made with an home made 20Db attenuator and a 50R termination.
(https://www.eevblog.com/forum/testgear/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=185746)
-
I see that other have found a resul for rise time (1.5ns) that is very close to the 1.4ns I got with my pulse generator.
Does anybody else has some equivalent measure that he can post here ?
Be careful of going too far down this road - you might get enticed into achieving picosecond rise times - which could lead to a new scope :). It starts with just wanting to see sub-nanosecond rise times (999 ps or faster - just so you can see ps instead of ns);
But it's soooo close already. Only 0.4ns needed.... :popcorn:
-
Pulse Gen ...
https://www.eevblog.com/forum/buysellwanted/pulse-generator-kits-or-ready-to-use/
(https://www.eevblog.com/forum/buysellwanted/pulse-generator-kits-or-ready-to-use/)
-
Thanks, but I am not interested in buying a pulse generator in a black box.
Where are your shematics, pcb picture, gerber files (if any ) ?.
I am interested in building one with elementary means, and to share experience with others that would like to build one.
This is why I describe here https://www.eevblog.com/forum/beginners/a-few-question-on-jw-pulse-generator-based-on-2n3904/ (https://www.eevblog.com/forum/beginners/a-few-question-on-jw-pulse-generator-based-on-2n3904/)
the various problems I encountered in building this simple, but delicate device.
In addition, your pulse generator does not provide a plateau, which is, according to JW, not the best way to evaluate the performances of a scope.
-
In addition, your pulse generator does not provide a plateau, which is, according to JW not the best way to evaluate the performances of a scope.
According to basic mathematics it doesn't work at all. I really wish people would stop pushing these useless impulse devices as a way to evaluate oscilloscope behavior. The results are garbage, typically making the oscilloscope look significantly better than it actually is. Without a square rising edge, and a long enough flat top to allow the oscilloscope system to fully settle before the falling edge, the results are completely meaningless.
But no matter how many times people howl about this on here, other people continue to just ignore them. :palm:
To scatter a bit of signal with the noise again:
159mHz cutoff filter with a full rise, 0.35/rt = 160mHz: full_rise.png (https://misc.c4757p.com/full_rise.png)
159mHz cutoff filter with a truncated rise, 0.35/rt = 250mHz: truncated_rise.png (https://misc.c4757p.com/truncated_rise.png)
-
In addition, your pulse generator does not provide a plateau, which is, according to JW not the best way to evaluate the performances of a scope.
According to basic mathematics it doesn't work at all. I really wish people would stop pushing these useless impulse devices as a way to evaluate oscilloscope behavior. The results are garbage, typically making the oscilloscope look significantly better than it actually is. Without a square rising edge, and a long enough flat top to allow the oscilloscope system to fully settle before the falling edge, the results are completely meaningless.
But no matter how many times people howl about this on here, other people continue to just ignore them. :palm:
THIS
It boggles my mind that people still use these impulse generators to "measure" rise time. The only valid input for measuring rise time is a step, and it needs to be a well-behaved step too, flat-topped and without overshoot. Overshoot adds high frequency content that would artificially contribute to a faster rise time, because mathematically it is like an implulse added to a step. If you want to build a step-type pulse generator to measure the rise time of a 100 MHz scope, you will need a known-good 1 GHz scope to calibrate the step generator before you can use that to characterize the 100 MHz scope. There is no way around that; you need a fast, clean step.
These (im-)pulsers are a fun project to build and play with especially when a person has no other signal source on the bench which can push a scope to its limits. But understand, you can't measure rise time with them.
-
Overshoot adds high frequency content that would artificially contribute to a faster rise time, because mathematically it is like an implulse added to a step.
Along these lines, here's a third simulation. Same filter, but a step with overshoot. The overshoot "just happens" ;) to be calibrated such that it's not even visible at the output of the filter, and yet...
0.35 / rt = 642 mHz
Holy shit, my filter is fast!! :scared:
overshoot.png (https://misc.c4757p.com/overshoot.png)
-
My Pulse Gen based Jim W. Pulse Gen (very simple)
-
Yeah, we know. ???
-
Pulse RT 255ps
-
Did you read one bit of what I wrote?
-
The pulse should look like this one if you want to get correct rise time.
-
These (im-)pulsers are a fun project to build and play with especially when a person has no other signal source on the bench which can push a scope to its limits. But understand, you can't measure rise time with them.
Even if you could, what would it tell you about the 'scope that sine wave attenuation doesn't?
-
Well, it would give you an inflated number for the size of your EE-peen, no?
JW pulse gen = EE viagra?
-
Here is what I got using a Tek TDS460A normally 400Mhz, but limited to 100Ms/s.
It is not a very responsive machine, but its the highest BW I have.
I used it to calibrate the pot to attenuate the overshoot, following JW AN94.
I have no other way in my lab to generate signal of frequency higher than 20Mhz.
(https://www.eevblog.com/forum/testgear/pulse-generator-rise-time-and-rigol-ds1054z-bandwidth/?action=dlattach;attach=185984)
-
And like I posted earlier,
https://www.eevblog.com/forum/beginners/bandwidth-calculation-caveats (https://www.eevblog.com/forum/beginners/bandwidth-calculation-caveats)!/msg807413/#msg807413
And what does it matter?
Clearly, only signals with comparable waveforms can be compared with simple rules. (Duh?)
The circuit in my simulation has largely the same risetime, but reduced ringing/overshoot, if I tweak the frequency response to be softer: closer to a Bessel response, which will be more typical of oscilloscopes (e.g., Tektronix's famous vacuum tube distributed amplifiers from the early 60s). However, I'm more interested in the frequency response, so I want those extra dB of flatness out to the cutoff frequency, which introduces ringing without improving the step response much.
(What I find impressive, really, is those Tektronix amplifiers must've been good for 500MHz of gain-bandwidth! In tubes, no less! The reason they topped out around 100MHz, at the system level, is a combination of factors: intentionally softened frequency response (for clean step response), required cascading of stages (two 100MHz LPFs cascaded do not have 100MHz BW, but more like 70MHz -- and they needed 3-4 stages in the oscilloscope to get enough gain!), and needing gain at all, GBW being what it is.)
Tim
-
From https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/msg775096/#msg775096, (https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/msg775096/#msg775096,) I think this picture paints a thousand words.
Hi,
Like Tim said, I think the pulse from your pulse generator is too short. Here is a simulation of a filter being tested with a short pulse and a long pulse:
(https://www.eevblog.com/forum/blog/eevblog-306-jim-williams-pulse-generator/?action=dlattach;attach=175681;image)
You would think the short pulse indicates higher BW, but the filters are identical.
Try adding about 18 inches of 50 Ohm coax in parallel with the capacitor in your pulse generator. This will length the pulse to about 4ns.
Regards,
Jay_Diddy_B
-
I have seen this recent post
Below You will find step response of my hacked Rigol DS1054Z (100MHz). This test was performed using Tektronix 284 70ps pulse generator to check if there is any ringing.
Pictures for time base settings: 5ns/div and 50ns/div.
Results: quite good.
Edit: Especially for Fungus I'll read the rise time measurement shown on the upper picture: 3,2ns.
(https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/?action=dlattach;attach=185571)
The observed rising time is here 3.2 ns. Much larger than the one I have of 1.9 ns.
The used pulse generator seems to be of high quality, and it would be nice to have other examples to understand if there is some variability on the DS1054z scopes.
-
Pulse RT 255ps
Sigh.
|O
-
These (im-)pulsers are a fun project to build and play with especially when a person has no other signal source on the bench which can push a scope to its limits. But understand, you can't measure rise time with them.
Even if you could, what would it tell you about the 'scope that sine wave attenuation doesn't?
Rise time for one thing.
Seriously though each is a valid test in different ways. One difference is that a levelled sine generator that goes to GHz is a very complicated and expensive piece of gear. A pulser is significantly simpler.
-
Even if you could, what would it tell you about the 'scope that sine wave attenuation doesn't?
each is a valid test in different ways. One difference is that a levelled sine generator that goes to GHz is a very complicated and expensive piece of gear. A pulser is significantly simpler.
But the results are difficult to measure/compare.