Professional Scope Purchasing Criteria

[TunerSandwich]

Also LeCroys claims are "warranted"....which is a worst case scenario.  Obviously to characterize the accuracy a probe must have been used for the measurement.  So if an active probe (PA) wasn't used to make those claims..... then we can only get "better".

Those specs are pure scope specs and do not consider probes. These are "at port", i.e. a fully parametrized signal is laid to the input (usually via a calibrated direct RF connection, i.e. no probes).

If you connect a probe you have to take it's specs into account as well, i.e. if you connect a probe with (just to take some figure) say 2% of warranted accuracy then your warranted accuracy won't be better than that. If however your probe has a warranted accuracy of say 0.5% then the warranted accuracy for that setup will be the 1.5% of the scope.

With newer active probes (i.e. AP-033/034, HFP Series) which have their own internal calibration system (which calibrates the probe's amplifier section) the scope basically performs an (almost) end-to-end calibration when it does it ReCal. I'm not sure about your AP-020 probe, though, which is pretty old (it's from the old 9300 Series days, but the fact alone that a pretty modern LeCroy scope still supports their nearly 20 year old probe is just great!), as I don't know if it does have self-cal functionality.

Just wondering, for your scope screenshots, why do you use a camera and not just a screen dump? You can reconfigure the printer button in the upper right corner to create a screenshot instead and save it to whatever storage you want to. Quicker and better to read. And you can still use a printer i.e. in LabNotebook.

I assume the specs aren't including a specific retail probe.....but they include a measurement discrepancy either way.  Whatever probe/signal path....was used to measure the input amplifier has characteristics of its own.

From what I have read the ap020 does self calibrate, as well as auto compensate.  It does provide an end to end calibrate solution (have verified this by measuring).  I did notice some consistent noise from it's FET though (90-120 ish MHz).  Obviously the correct way to read all of this and calibrate the REF is with a differential measurement, into a a pre-amplifier....with some good isolation. 

My guess is that the input PA/A on the scope is characterized with a broadband signal (weighted noise @ known amplitudes/bins)?.  There is still going to be error in the measurement chain.  We don't have enough information to know for sure.  However the ap020 accuracy (stated) beats the scope input (as far as I can measure/decode/extrapolate)

P.S.  I could do screen-dumps if you guys prefer.  I mainly don't do them, because I have the camera out already to snap the test rig.....also they are extremely small (xVGA I think). 

[nightmechanic]

Thanks for the elaborate answer,

Using your DMM analogy:

Just like with a DMM....where the accuracy is stated as "full scale".  If the "scale" is 0-10 V the accuracy might be 0.001%.  If the scale is 0-100 the stated accuracy usually goes down an order of magnitude....for example, possibly from 0.001% to 0.01%.....

However let's say that the circuitry was in the DMM to have a "rescaling" from 90-100V.....then we could re-gain an accuracy of 0.001%, as the "scale" would be the same width....

the claims aren't for DC "offset".....as DC offset can be easily nullified etc......

Lets say I did just that (rescaling) using some ”rescaler” module that adds a DC offset to my input, then my overall measurement accuracy would be my, better, DMM accuracy + my rescaler accuracy which , as far as i understand is exactly what happens in a scope when the offset is not 0 - the overall ground referenced measurement accuracy is the sum of the gain accuracy (usually specified in % of full scale in use as you mentioned) + the DC offset accuracy (either the scope’s or the probe’s as in your case I think).

I do not see how the DC offset can be nullified (other than measuring an accurate enough reference at the DC offset setting you are using and subtracting the error).

For relative/dual measurements such as peak-peak etc. then, of course, the only thing that counts is the gain accuracy as you mention.

so, for example...if I AC couple the input and take a reading....on my specific scope/bench the sDEV in the measurement is the same as it would be when DC coupled....

I agree, however, (short term) sDEV will quantify the noise (source + scope). and I think that scope noise is (should be) only part of the accuracy specifications, other wise there wouldn’t be any margin for any drifts at all (time , temp etc.).
Long term sDEV may not be the same due to DC offset drifts.

Also LeCroys claims are "warranted"....which is a worst case scenario.  Obviously to characterize the accuracy a probe must have been used for the measurement.  So if an active probe (PA) wasn't used to make those claims..... then we can only get "better".

This is why manufacturer specs are generally only guidelines....what matters is what YOU can measure in YOUR lab environment.   

I completely agree, I think manufacturer specs are worst case when you are doing nothing wrong, when you are doing everything right you can get better results but you need to verify that somehow (as you have shown for your lab setup).

P.S.
the probe I am using (ap020) has a dynamic range of 5V.....so the maximum scale I can represent with it is 1V/div....8V total....however only 5V of that scale is usable, within the stated accuracy specifications.....

And the AP020 has a +/-20V offset capability, do you know if there is an accuracy spec for it somewhere? (as I wrote before, the only thing I could find is the 2% attenuation accuracy spec)

Thanks again,
Ran

[nightmechanic]

As you can see the accuracy of the scope drifts as the vertical scale becomes wider.....

That is what the term "full scale" means.....the accuracy has very little to do with total DC offset....

Your examples clearly show the meaning of ”Full scale” accuracy, however, I would imagine the DC offset error being the same for all the different scales (as it has the same value), so I do not understand how you can rule out any error related to DC offset?

The DMM's are NOT as accurate as the scope, despite their LONGER integration times.  Because they are reading DC, and not all of the signal energy is DC.  The AC component can't really be read by the DMM's...even in AC mode, because the max bandwidth of the AC range on the DM3068 is 200Hz.....and it's unknown on the Fluke 289, because they don't state it (and I can't be bothered to actually measure and extrapolate.....that's why i have a scope).

I am definitely missing something here, DMMs (in DCV mode) are intended to measure the DC component and only the DC component as accurately as possible, if your are looking for something else (ripple etc.) then it is  definitely the wrong tool, so why do you think the scope’s MEAN function gives you a more accurate result?

Ran

[TunerSandwich]

Thanks for the elaborate answer,

Lets say I did just that (rescaling) using some ”rescaler” module that adds a DC offset to my input, then my overall measurement accuracy would be my, better, DMM accuracy + my rescaler accuracy which , as far as i understand is exactly what happens in a scope when the offset is not 0 - the overall ground referenced measurement accuracy is the sum of the gain accuracy (usually specified in % of full scale in use as you mentioned) + the DC offset accuracy (either the scope’s or the probe’s as in your case I think).

I do not see how the DC offset can be nullified (other than measuring an accurate enough reference at the DC offset setting you are using and subtracting the error).

For relative/dual measurements such as peak-peak etc. then, of course, the only thing that counts is the gain accuracy as you mention.

I agree, however, (short term) sDEV will quantify the noise (source + scope). and I think that scope noise is (should be) only part of the accuracy specifications, other wise there wouldn’t be any margin for any drifts at all (time , temp etc.).
Long term sDEV may not be the same due to DC offset drifts.


I completely agree, I think manufacturer specs are worst case when you are doing nothing wrong, when you are doing everything right you can get better results but you need to verify that somehow (as you have shown for your lab setup).


And the AP020 has a +/-20V offset capability, do you know if there is an accuracy spec for it somewhere? (as I wrote before, the only thing I could find is the 2% attenuation accuracy spec)

Thanks again,
Ran

The input PA gains accuracy with increasing levels of attenuation.  I am not sure the amplifier class in the MXi, but almost every amplifier gains resolution (above noise floor) with less input drive (reduced THD). 

DC offset is a easy factor to measure, and then reject....through AC coupling....you could easily route an AC coupled line....inject a differential signal (buffered) etc etc etc....lots of methods for removing DC in a measurement...

HOWEVER, keep in mind that the specifications aren't for DC offset accuracy.  They are simply for "DC full scale".  We have to make some assumptions that the DC offset error is going to be a reasonable percentage of the full scale discrepancy. 

sDEV is the only useful measurement for near 100% DC vs near 100% AC coupled.  It's not a perfect benchmark, but it's a ballpark that lets us know the total DC error......it's a well educated guess.  If the sDEV was wildly different we could assume that there is an inaccuracy in DC component right?

The reason I recommended a traceable collection of references to the OP, was for this very reason.  Before we take ANY measurement, that is critical in characterizing a warranted product.....we should verify the signal path with a reference of known traceability.  Yup it's a pain in the ass.....but nowhere near as bad as the bad will caused by something avoidable. 

I have a spec sheet for the ap020 somewhere around here.  I do remember choosing them for these specific kind of measurements.  They have the bandwidth and accuracy needed, to match (better actually) the scope input.  Naturally they should, as they set-up proper tip to end unity gain staging.....(within a reasonable degree....I will try to find that spec sheet).

[TunerSandwich]

Your examples clearly show the meaning of ”Full scale” accuracy, however, I would imagine the DC offset error being the same for all the different scales (as it has the same value), so I do not understand how you can rule out any error related to DC offset?


I am definitely missing something here, DMMs (in DCV mode) are intended to measure the DC component and only the DC component as accurately as possible, if your are looking for something else (ripple etc.) then it is  definitely the wrong tool, so why do you think the scope’s MEAN function gives you a more accurate result?

The DC offset error is not stated by the manufacturer as a DMM would state.  It's not wholly relevant to the use of the scope.  The purpose of the scope is to resolve "noise" (always AC) above DC.  An easy way to tell the DC offset error, would be to feed absolutely pure DC into the scope....with no error from the source.....and a virtually infinite vertical amplifier.  Obviously pure DC wouldn't show up on the scope screen at all.  So if I ground the inputs, or terminate them into 50ohm, let's say, then I should see the closest possible marker of total DC offset error. 

I think the context here has been forgotten.  This all relates to measuring DC-DC switch-mode POL.  Also other reference signals.  The AC riding on the DC carrier is always relevant, as it's energy.  When you get into precision regulation, this becomes highly critical.  Especially inductance and AC coupling from the PCB.  If I put my DMM into AC+DC mode (common) it will miss all of the AC component, beyond it's sensitivity. 

Ripple is a misleading term.  It suggest that somehow the deviation in energy is decoupled from the DC component.  It never is.  The purpose of using a scope to characterize broadband energy is to measure the sDEV from the baseline DC component. 

here is a screengrab of 50 ohm terminate DC input.  You can get a very good idea of what the total inaccuracy of the front end is this way.  Obviously the 50 ohm terminator is subject to some inductance etc....but it's a hell of a lot more resolving of total energy than a DC coupled DMM. 

The MEAN is going to be as close to the DC component as we can get.....obviously removing ALL deviation from the front end is impossible, and there is going to be some DSP/processing error as well.....but you can see just how miniscule it is.  Since I can't fully characterize the 50 ohm terminator, and BNC....I can't get a mean of 0V.....which is theoretically impossible anyhow.  However look how much loser we get to the pure DC component, with narrower and narrower scales.  Do it trust the scope precision down to mV?  I sure do....especially when the 50 ohm terminated mean is uV

[nightmechanic]

Thanks again for taking the time to answer,

HOWEVER, keep in mind that the specifications aren't for DC offset accuracy.  They are simply for "DC full scale".  We have to make some assumptions that the DC offset error is going to be a reasonable percentage of the full scale discrepancy. 

There is a DC offset accuracy spec:
"Offset Accuracy: Fixed gain setting < 2 V/div: ±(1.5% of offset value + 0.5% of full scale value + 1 mV)”
(from the Xi users manual in the Lecroy website)

Ran

[TunerSandwich]

Thanks again for taking the time to answer,

HOWEVER, keep in mind that the specifications aren't for DC offset accuracy.  They are simply for "DC full scale".  We have to make some assumptions that the DC offset error is going to be a reasonable percentage of the full scale discrepancy. 

There is a DC offset accuracy spec:
"Offset Accuracy: Fixed gain setting < 2 V/div: ±(1.5% of offset value + 0.5% of full scale value + 1 mV)”
(from the Xi users manual in the Lecroy website)

Ran

It's incomplete.  We have to go by what we measure in the environment we are calibrating in.  Specs are always incomplete and shouldn't be be the benchmark for accuracy....if we can do better than them.  If we do worse than them, then we can turn to them as a benchmark for solving a potential problem.  I.E.  my scope is 'better" than "spec".  Awesome, carry on.  My scope is "worse" than spec....uh oh better investigate (usually something on the users end....and usually a blindingly stupid error by a human). 

So if we add up our measurements AND the device functions flawlessly over a reasonable test period.....then we can have 2 markers for successfully meeting or beating "spec".  I think a lot of people get buried in "specs" and ignore measurements (with traceability/repeatability) and averages. 

I am curious to see what LeCroy says it's "best" case scenario specs are.  If we had those, and we beat them, then maybe we could look further into the matter and say "something doesn't add up"

[TunerSandwich]

Here is another shot of the same measurement....BUT the pre-processing interpolation is set to 5 averages, sin x/x and leCroys own "ERES" filter.  Which is simply a noise filter, that interpolates an effective "dynamic range resolution" of + 3 bits.  I left the setup window open so you can see it's effect and configuration. 

In addition the input now has a 20MHz filter on it.  So the ERES + input filter = an effectively "steeper" roll off at the filter point. 

I also have a zoom window open, so you can actually see the waveform.  i wouldn't trust the accuracy of the zoom window that much.  It's obviously interpolated and not a representation of the true dynamic range of the scope.  The measurements are on the C3 waveform (not the zoom) so I do trust them, to a degree.  I don't believe for a second that the scope is uV accurate though.  I don't have a reference tight enough to make that determination.  If we wanted to verify that, the keithly source meter would have to come out and everything would have to go into an isolation enclosure....and the environment would need better control.  I also don't own a signal path I would trust from source meter out to scope input amplifier.

Obviously I draw a line at what is "overkill" for my needs.  However it's not overkill for something like brain research.  Especially neuron or micro tubual mapping/research.  Or highly advanced computational r&d.....and who knows, maybe someday we will have a need for that level of precision in DC systems.  As we near 99% efficiencies and try to best them, it becomes very relevant.

I digress.....here is the screen grab

P.S. one critical thing to remember in any measurement, is that there is ALWAYS an AC component in any signal.  Sometimes big, sometimes very little.  So is a DMM measuring in DCV truly measuring ONLY DC?  No.  If I have a DC reference that is MORE accurate (remember accuracy and resolution are two different things), and by more accurate I mean LESS total AC inductance than the known DMM inputs/leads....ok then I can also say that the DMM is "less accurate" than the REF.  Obviously this traceability then extends to a meter that is KNOWN to be more accurate than the REF (usually the meter the REF was calibrated on).  I know that the REF I used in this test is LESS accurate than BOTH the dm3068 AND the MXi, as 5.00000 on the 3068 = 5.003(ish) on the scope.  I did this on purpose (like I explained above).  Why?  Because it highlights a scenario where the scope was used to find a problem, in a DC component (dominant) that the DMM could not resolve.  I certainly hope this is as well established as I have considered it to be.  This is the protocol for standards and testing.  Well at least the way I have been practicing the trade for quite a while.  This method has worked well for me.  Let's not forget the original post here was asking for criteria, for choosing a "professional" scope...for the very specific kind of work I have demonstrated and applied methodology to.  Obviously everything here is specifically applied to that kind of work.  I have broken down nearly every stage he will go through in development, and why specific standards and tools are necessary.  I like to give thorough answers 

[nightmechanic]

(I hope the OP is still interested in whats going on.. )

So,
You have shown the noise level (sDEV) and the zero offset DC value, both look low and accurate respectively, however, looking at the scopes stated offset accuracy (as incomplete as it may be, thats all I have to go with with regards to theoretical/worst case accuracy limits):
"Offset Accuracy: Fixed gain setting < 2 V/div: ±(1.5% of offset value + 0.5% of full scale value + 1 mV)”

There is a (large) error component related to the offset value, so in your example it would be 1.5% of 5V ~75mV, obviously your scope seems much better than that due to the correlation with the DMMs, but if you want to validate how good it really is you need to do so at that operating point, which as you mention is a different and difficult measurement problem.

Due to the above, I am still having a hard time accepting that the scopes MEAN value is a more accurate ”DC only” value than a (accurate) DMM, but that is my problem 

I am not in anyway challenging the fact that large bandwidth is required for such analysis (I am far from being knowledgeable in this field, from my little experience, high frequency artifacts from DCDC converters are a problem and high bandwidth as you mention is required to understand and solve these issues), it is just the DC accuracy that I am struggling with. In the past we have done similar measurements using AC coupling (to minimize scope errors) to see all the non DC energy, and used an accurate DMM to determine the ”MEAN” value, may be a mistake on our part, although I am not yet convinced...but then again, my problem

Thanks,
Ran

[TunerSandwich]

(I hope the OP is still interested in whats going on.. )

So,
You have shown the noise level (sDEV) and the zero offset DC value, both look low and accurate respectively, however, looking at the scopes stated offset accuracy (as incomplete as it may be, thats all I have to go with with regards to theoretical/worst case accuracy limits):
"Offset Accuracy: Fixed gain setting < 2 V/div: ±(1.5% of offset value + 0.5% of full scale value + 1 mV)”

There is a (large) error component related to the offset value, so in your example it would be 1.5% of 5V ~75mV, obviously your scope seems much better than that due to the correlation with the DMMs, but if you want to validate how good it really is you need to do so at that operating point, which as you mention is a different and difficult measurement problem.

Due to the above, I am still having a hard time accepting that the scopes MEAN value is a more accurate ”DC only” value than a (accurate) DMM, but that is my problem 

I am not in anyway challenging the fact that large bandwidth is required for such analysis (I am far from being knowledgeable in this field, from my little experience, high frequency artifacts from DCDC converters are a problem and high bandwidth as you mention is required to understand and solve these issues), it is just the DC accuracy that I am struggling with. In the past we have done similar measurements using AC coupling (to minimize scope errors) to see all the non DC energy, and used an accurate DMM to determine the ”MEAN” value, may be a mistake on our part, although I am not yet convinced...but then again, my problem

Thanks,
Ran

I still have to trust what my bench results say...as they are a reference.  So let's say I was out by X, and my device under test exhibited acceptable sDEV from "spec".  Then I have to accept, that if those results beat a prediction....they offer greater accuracy than the prediction/simulation.  I would trade accuracy for resolution any day, as long as the resolution keeps my device at optimal operating specifications.  I would trust those results, far above what any manufacturer datasheets tell me.  Unless there was some completely wild discrepancy...and if that happened, the chances of a large cross section of production devices working well is seemingly low (yield).

Keep in mind, that the DMM may represent DC accuracy, with a steeper filter slope/isolation from AC....BUT the AC is, in reality, there.  If the AC is low enough in frequency for the DMM to capture it, and add it to the DC (many DMM have AC+DC measurement for this reason), then you more fully characterize the total power quality of the device under test.  That is a quantitative rule in formulating "power quality".  It's always the sum of the total components at the point of load. 

Also, a DMM is a mean value across the integration time, minus the total counts equation.  Obviously the higher the total polling rate, the more accuracy you will have.  Obviously the higher the total integration time is, you will lose accuracy, but GAIN resolution.  This is why I like to have 2 DMM, that can both do trend/min/max/avg readings.  I can set each at an offset integration time (one always < other) and try to see a higher degree of accuracy, by margining BOTH results (over time). 

Here is another screengrab....with the same interpolation pre-processing on the signal.  HOWEVER, this time I calibrate the scope against the reference.....so now the scope IS the reference.  Do you see my point now about calibrating at least ONE device in the t&m chain to an accurate reference?  Remember my REF is the LEAST accurate device in this signal chain.  That is often the case with a consumer good that is being tested.  So let's make a scenario here...

I calibrate ALL devices in the test system against a known stable reference.  I then measure the total offset between the entire system....and then use that sDEV as a re-calibration point for the measurement output.  It's a vastly more accurate way of obtaining a larger cross section of results, that meet "spec".  The other way will have a larger statistical deviation from one point of known reference.  This way i maintain traceability across each piece in the test system + each device tested, with a realistic sDEV across the total "yield".  Obvious failures will be more obvious.  Devices that exceed "spec" can be pulled and analyzed...to gain insight on WHY they are "better" and that can possibly lead to implementing those discoveries....to further yield a higher standard.  Traceability is king, above any manufacturers spec book. 

NOW, if the min max sDEV was supplied for each unique piece of test gear...and all items could be verified on site (of the test) then I would be MUCH more inclined to believe those statements OVER my measurements.  Especially if the test gear used to verify the system was provable as "more traceable" than the system.  This is why a maintained set of reference standards are important.  At the end of the day, you HAVE to have your own internal references, as daily or weekly calibrations against external standards can kill you in downtime and buy in (if one set of gear is always out for cal, you need two sets....).

The most accurate reading available...is not the DMMs or the scopes (notice the plurality there).  The most accurate reading is the mean of the sDEV average of everything in the test chain....minus/plus the results from the hardware under test. 


I hope that lends an answer as to why I would say I "trust" the mean result on my scope, more than the mean result on my DMM.  The scope is verified as being far more accurate than the datasheet says.  The DMM is, however not classified in the same way.  I am not sure my point about accuracy is being understood.....or it's quite possible that I have poorly articulated it.  The accuracy of my scope is higher, in relation to total power quality, than the DCV reading on my dm3068.  The specs for the dm3068 are even more "incomplete" than on the MXi.  Is the DC accuracy "better" than the dm3068?  That is not provable without some more investigation into the dm3068.  However the ref being used says that if I calibrate against my scope reading, I have more real world representation of the total power quality at the point of load.

I also hope OP is still around, as these tests all mirror the exact scenario his device will run into.....but I also think some other people might gain some insight from this discussion.  It's quite an examination about core fundamentals of oscilloscope usefulness.  I think i have also shown some tests, that others can do to see just where their t&m sDEV curve stands.  I view that as quite important and a good investment of time/effort. 

BTW the measurement discrepancy between the scope and the dmm is 0.009276V (DC-5ishMHz.  Oddly enough that is virtually identical to the discrepancy between the sDEV of the Fluke 289 and the sDEV of the dm3068.  Too close to be coincidence.  I think the dm3068 needs to be checked against a standard.  I need to send out my 5v lab standard to be checked.  If the 3068 has drifted that much since cal something isn't right?  I did recently move it and change the environmental control average.   

[nightmechanic]

I am missing something - is this the same MAX6350 reference that you have shown before or a different reference? if it is the same did you move it to match the scope/why doesn’t it match the dm3068 anymore?
why do you trust your reference more than your DMM?
In any case, as you mentioned, it does seem that the discrepancy between your dm3068 and your reference is too high - one of them is out as far as I understand (the dm3068 error at 5V should be +/-300uV or less if I am reading the specs correctly).

I completely agree with the traceability statements:
As long as you have a traceable stable (to me it means low noise, low drift, negligible HF contents) reference hopefully with the same voltage as your DUT (e.g. your lab 5V ref), you can figure out your equipments’ offsets (and noise levels) and they will be all traceable to that reference.

And you have shown (traced) that your scope is more accurate than your DMM at 5V - great.

I hope that lends an answer as to why I would say I "trust" the mean result on my scope, more than the mean result on my DMM.  The scope is verified as being far more accurate than the datasheet says.  The DMM is, however not classified in the same way.  I am not sure my point about accuracy is being understood.....or it's quite possible that I have poorly articulated it.  The accuracy of my scope is higher, in relation to total power quality, than the DCV reading on my dm3068.  The specs for the dm3068 are even more "incomplete" than on the MXi.  Is the DC accuracy "better" than the dm3068?  That is not provable without some more investigation into the dm3068.  However the ref being used says that if I calibrate against my scope reading, I have more real world representation of the total power quality at the point of load.

Regarding ”total power quality”, as you mentioned, it should be the DC value + AC rms value(s), the MEAN function however, tries to mimic the DMM operation in a way by mathematically trying to extract the DC only component and the quality (sDEV) of doing so depends on the number of samples used, noise levels (and as such the used bandwidth) potentially with an ”accurate” scope (with zero gain and offset errors) and lots of memory you could do better than the DMM, but it will just filter the DC component better and reject more AC power, without enough samples you get into the same issues you mentioned for the DMM, its a "DC + some of the AC” measurement, is this what you are looking for?

So now I have a(nother) question, when you send your lab reference out for calibration, that (as far as I know) is done using a (very expensive, traceable, lots of digits  ) DMM - an averaging, limited bandwidth DMM, do you consider that as a calibration reference for your purposes? do you assume that the AC contents of the reference is low enough to be negligible?

Thanks for you patience,
Ran

[TunerSandwich]

I am missing something - is this the same MAX6350 reference that you have shown before or a different reference? if it is the same did you move it to match the scope/why doesn’t it match the dm3068 anymore?
why do you trust your reference more than your DMM?
In any case, as you mentioned, it does seem that the discrepancy between your dm3068 and your reference is too high - one of them is out as far as I understand (the dm3068 error at 5V should be +/-300uV or less if I am reading the specs correctly).

I completely agree with the traceability statements:
As long as you have a traceable stable (to me it means low noise, low drift, negligible HF contents) reference hopefully with the same voltage as your DUT (e.g. your lab 5V ref), you can figure out your equipments’ offsets (and noise levels) and they will be all traceable to that reference.

And you have shown (traced) that your scope is more accurate than your DMM at 5V - great.

Regarding ”total power quality”, as you mentioned, it should be the DC value + AC rms value(s), the MEAN function however, tries to mimic the DMM operation in a way by mathematically trying to extract the DC only component and the quality (sDEV) of doing so depends on the number of samples used, noise levels (and as such the used bandwidth) potentially with an ”accurate” scope (with zero gain and offset errors) and lots of memory you could do better than the DMM, but it will just filter the DC component better and reject more AC power, without enough samples you get into the same issues you mentioned for the DMM, its a "DC + some of the AC” measurement, is this what you are looking for?

So now I have a(nother) question, when you send your lab reference out for calibration, that (as far as I know) is done using a (very expensive, traceable, lots of digits  ) DMM - an averaging, limited bandwidth DMM, do you consider that as a calibration reference for your purposes? do you assume that the AC contents of the reference is low enough to be negligible?

Thanks for you patience,
Ran

I think you missed the paragraph where I described re-calibrating the reference to the scope reading.  Again done to show the total discrepancy between the two devices.....and back up the prior measurements.

What i was saying is that my dm3068 is reading a tad low compared to the fluke 289.  Assuming we can trust the 289.  i re-calibrate against it, and the offset between the 2 (3068 vs 289) is consistent high and low. 

What I have shown is the total sDEV across the entire signal chain.  Also that the stability of the vref is very good....as the sDEV avearge has very little drift. 

MEAN on the scope is not an attempt to null AC.  It's the mean above and below the total measured DC offset.  The pk-pk is shown as higher frequency, but lower energy.....the applied filter removed most of the AC component. 

Power quality is a measurement window....in which all of the metrics are time correlated.  You will never deduce true power quality with DMM + scope, unless the DMM is somehow sharing the same sample clock.

The PMA software in the MXi does exactly this.  It also uses a current probe to calculate watts RMS + power spectrum (10MHz) + voltage waveform.  The current ripple is part of the equation as well.  Memory depth is there, samplerate is there.....far above and beyond the DMM. 

I assume the returned lab reference is traceable to a larger degree than my equipment....with an accuracy AND resolution BEYOND my needs.  I can't assume it is free of noise, so that must be verified.  When the ref returns, it will be checked against all available gear in the test system, and a new sDEV will be noted.  I will also check is against other known references.  For my purposes, it's a prerequisite to usable measurement data. 

If you go back through all the pictures, and make note of the readings, you will see that the total DC inaccuracy, across all tests (scope) is less than 3mV against the 3086 (which appears to be reading just a tad bit low, in relation to other ref/289). 

So if lecroy worst case scenario is being calculated (by you) as 75 ish mV DC offset + full scale error.....then Lecroys BEST case must be below 3mV total that i am measuring.  If it's that far off between best and worst....then it's hard to trust that the math is being done correctly to calculate the true measurement error stated from the datasheet (I do believe my math is correct, because it's almost exactly what we are actually seeing).  My scope is from 2009....so it's aged well (too well) if your math is correct, if my math and measurement is correct it's right around where it should be.  Have a look back at the total inaccuracy I predicted, even before posting all the screenshots and engineering the new test.  I don't think it's a coincidence that they are within 6uV. 

Let's assume everything I am saying is totally wrong.  Then we are still beating Lecroy datasheet spec by many orders of magnitude.  Not matter how we slice it, we are seeing under 3mV total discrepancy across 3 very precise devices, all checked against a highly stable vREF.....DESPITE the vREF resonant tank trap I built in (to illustrate those effects).  Also keep in mind that 3mV is a WORST case scenario against ALL of the tests.  Some show better (smaller scale dynamic range).

P.S.

Keep in mind that I didn't JUST show traceability against 5v.  I showed it against 20mV, all the way out to 20V, and a bunch in between.  I picked 5V to focus on, because it is the dynamic range of the probe....so in case any massive transients showed up, we could have factored them in as "usable".  Thankfully we didn't see any 5v swings from nominal offset, so our data is all good.

[rx8pilot]

I am still here, soaking up a treasure trove of information. We have buried in software deadlines this past week.

The discussion is covering so many details I will probably still be referring to a year from now - while continuing to paint a picture of what to consider when deciding on a scope. After I get it, I have tons of learning experiments based on what I learned in this discussion alone.

Typing on my tablet is too slow.... To be continued....

[nightmechanic]

I think you missed the paragraph where I described re-calibrating the reference to the scope reading.  Again done to show the total discrepancy between the two devices.....and back up the prior measurements.

I didn’t miss it, just mis-read it

MEAN on the scope is not an attempt to null AC.  It's the mean above and below the total measured DC offset.  The pk-pk is shown as higher frequency, but lower energy.....the applied filter removed most of the AC component. 

MEAN on the scope is the sum of the sample values divided by the number of samples: it is the average value of the samples, i.e. the DC component in the measurement window, whether you add a DC offset to that does not matter, assuming you were measuring some 5V ideal reference, and you set your DC offset at 5V,  it would be your average DC offset + gain errors, i.e. the DC component of your DC errors (within the measurement window of course).

sDEV is the RMS level of the signal ”around” the above MEAN (if MEAN is the DC of the measured signal, then sDEV is the RMS AC voltage of the measured signal in the measurement window), this includes the pk-pk energy which in some cases is intermittent / higher frequency and in others (e.g. pure sine-wave ripple) describes all the energy ”collected” by sDEV.


There is no argument (only envy ) that your lab equipment correlation is amazing and the scope is much much better than the specifications in the manual.

(I still believe my math is correct, but it does not really matter, as you said, as you keep saying, what matters most is traceability, although it would be interesting to hear Lecroys view on this )

Thanks,
Ran

[TunerSandwich]

I didn’t miss it, just mis-read it

MEAN on the scope is the sum of the sample values divided by the number of samples: it is the average value of the samples, i.e. the DC component in the measurement window, whether you add a DC offset to that does not matter, assuming you were measuring some 5V ideal reference, and you set your DC offset at 5V,  it would be your average DC offset + gain errors, i.e. the DC component of your DC errors (within the measurement window of course).

sDEV is the RMS level of the signal ”around” the above MEAN (if MEAN is the DC of the measured signal, then sDEV is the RMS AC voltage of the measured signal in the measurement window), this includes the pk-pk energy which in some cases is intermittent / higher frequency and in others (e.g. pure sine-wave ripple) describes all the energy ”collected” by sDEV.


There is no argument (only envy ) that your lab equipment correlation is amazing and the scope is much much better than the specifications in the manual.

(I still believe my math is correct, but it does not really matter, as you said, as you keep saying, what matters most is traceability, although it would be interesting to hear Lecroys view on this )

Thanks,
Ran

Negative, mean is a calculated window above and below.  I showed the "helper" markers in a few screenshots.  I will grab another screen later focusing more on what the mean reading is.  It's going to become more complicated, as sample rate (time) is changed.  So obviously we both know that the mean is an average, and gains/loses resolution with sample rate.....BUT it isn't a boxcar average.  It is not throwing away events, in an attempt to smooth....and gain resolution or sample more quickly and gain accuracy.  Hence it isn't like a DMM....integration time correlates to horizontal scale (sort of)....but the total counts (quantization rate) is not the same.  If the DMM was capturing the same data, and doing a realtime average....then we could say they are similar. 

i am not sure if we have a wording discrepancy here or the lecroys use of the term is misleading. 

I certainly would like to see a lecroy BEST case scenario for the WR64MXi.  Come to think of it I would like to see those metrics in EVERY manufacturers specs

[nightmechanic]

Thanks again,

First of all a correction:
when quoting before the DC offset accuracy for the scope (1.5%x DC offset value etc...) and the 75mV figure, I missed the fact that it is not the scope's DC offset mechanism that is used (as far as I understand it is not possible to reach such a high offset at 20mV/div), it is the AP020 (probe) that is internally generating the offset.
So, the relevant spec is the offset accuracy of the probe, which I could not find online. (although I assume it is not zero - getting to that later)

Regarding the MEAN function, from the scope's manual:



It is not a boxcar average in the moving average sense but it is a single value (not a window) "simple" average of the time domain samples.

Getting back to accuracy,
Lets try another (hopefully viable) theory:
Lets take the dm3068 as the "reference" and from this assumption try to figure out (curve fit) the scope's 20mv/Div DC gain error and probe's offset error with the measurements provided in teh history of this thread:

DM3068        scope MEAN       offset (probe)
26.3319mV    27.88mV             25.00mV
2.99658         3.00037              2.992
5.0000           5.00475              5.0
5.00049         5.0049                5.004
20.00656       20.02997            20.0

The (Excel) result is that the scope's error (vs the DM3068) is about: 0.11% x DC_offset +40uV, the largest error of this estimation vs the collected data (occurs at the 26mV data point) is 1.5mV.

Just showing "my math" perspective, and again, no doubt that the scope + AP020 is a very accurate setup.

Ran

[TunerSandwich]

Thanks again,

First of all a correction:
when quoting before the DC offset accuracy for the scope (1.5%x DC offset value etc...) and the 75mV figure, I missed the fact that it is not the scope's DC offset mechanism that is used (as far as I understand it is not possible to reach such a high offset at 20mV/div), it is the AP020 (probe) that is internally generating the offset.
So, the relevant spec is the offset accuracy of the probe, which I could not find online. (although I assume it is not zero - getting to that later)

It is not a boxcar average in the moving average sense but it is a single value (not a window) "simple" average of the time domain samples.

Getting back to accuracy,
Lets try another (hopefully viable) theory:
Lets take the dm3068 as the "reference" and from this assumption try to figure out (curve fit) the scope's 20mv/Div DC gain error and probe's offset error with the measurements provided in teh history of this thread:

DM3068        scope MEAN       offset (probe)
26.3319mV    27.88mV             25.00mV
2.99658         3.00037              2.992
5.0000           5.00475              5.0
5.00049         5.0049                5.004
20.00656       20.02997            20.0

The (Excel) result is that the scope's error (vs the DM3068) is about: 0.11% x DC_offset +40uV, the largest error of this estimation vs the collected data (occurs at the 26mV data point) is 1.5mV.

Just showing "my math" perspective, and again, no doubt that the scope + AP020 is a very accurate setup.

Ran

Hi Ran,

Sorry for the late reply (under the weather today...literally).

I think now we are getting somewhere, with regard to the usable accuracy of our data. 

There is one critical missing factor in here.  I always take this into account, but obviously the Rigol specs don't (fully).  Let's talk about the injected discrepancy of the probes leading to the Rigol.  I take most of these readings with the input set to >10Gohm.  However, let's characterize the probe/leads from the dm3068.  I will work on that in a bit here, as well as show the way in which the MEAN is updated. 

The MEAN is a calculated vertical window....and can be a truncated, or non-truncated sampling across the horizontal timebase.  The best way to do this is enter a narrow enough timebase, in which we can see individual sample points (quantization rate) and then switch to "dots" and remove the vectors applied.  The next part of the equation would be to set the integration time of the DMM and scope time base to be the same (which I did for a majority of the tests).  The fastest possible integration time (not that simple, but let's call it that) on the dm3068 is 100uS.  So lets re-visit that.  I showed that earlier on, but let's do it with less noisy signal.

I am going to change over from the vREF to the Rigol signal generator.  It's DC offset error is stated I believe (can't be bothered to look it up), but regardless of what the claim is, we will clearly see it across all of our "calibrated/traced" devices.....and we can extrapolate the sDEV.  Again we are focusing on sDEV and not "hardline" stated specs.  I am still a believer in generating our own criteria, and not wholly ignoring the datasheets....but at the same time not taking them as "gospel" (only cross section studies of one potential use case scenario).

Also remember that some of the scope data, is best case/worst case.  Since the parameters and signal chain have changed, throughout the represented data, we need to focus in on a more standard methodology....that doesn't change.  I think we are getting way far away from OP's initial post though.  Everything I have shown up to this point was done to illustrate OP's potential use case scenarios.  I hardly think DC from a less than stellar sig/gen fits in to his scenario.....but maybe it does.  After all this is critera for a verification methodology...which should be done, regardless of the types of systems someone works on (unless those are non critical and the user couldn't be bothered with accuracy).  I do recognize that some folks just want to make LEDs blink at unpredicted rates and lumens, and there is nothing wrong with that.

I don't reply to a lot of threads here, but this on had really sparked my interest, because OP used the term "professional scope purchasing criteria".  I was drawn in, because I recently went through the same situation.  So my apologies to the OP if this next series of posts/tests seems off topic.

[TunerSandwich]

New test system:

Source is the Rigol DS2102A-S.  Signal is DC with 2V offset...HighZ impedance.

Signal path is Rigol BNC out, via 50ohm coax -->FFM BNC t-fitting --> Pomona BNC to banana (never bothered to measure it's impact, but it's certainly less than the probes leading to the vREF, in the last tests).

Scope probe is the same AP020, but with a BNC coupler into the T-fitting. 

I played with some 0 DCV offset signals, and will show those.....just keep in mind that at those lower levels the impact of the test system will be greater than the discrepancy in both the dm3068 and WR64MXi.  I have some fancy impedance matching cables I build, and fittings/adapters/terminators, that can help solve those problems.....but let's focus on the 1 VDC before we go there (keeps it less convoluted for now.....but the other test IS important...and we will get to it). 

The integration time on the DM3068 is set to 2ms (which loses resolution, but gains accuracy).

The MXi timescale is set to 200uS/div.  So 10 div = +/- 1mS from center = total 2mS window. 

I played with triggering the DM3068 from the scope, and the other way around.  We can get into that a bit later....obviously the most accurate way to gain a true sample to sample offset, would be through a proper bi-directional triggering sDEV.  The DM3068 can output vcomp pulses, which are triggers on the DMM integration (NOT sample) interval.  So we can essentially trigger bi-directionally, to gain insight on the trigger/acquisition delay.....but that is getting pretty complicated. 

here are screen grabs.....also screen grabs showing the "mean" measurement parameter options.  I will explain them further if needed. 

The second to the last picture shows the impact of the DMM BNC to Banana adapter, terminate into 50OHM. 

Also the last picture is showing the sDEV and statistical analysis (impact) of the DMM probes used in the previous tests, on the vREF.  Remember they were the quick clips connected to the ballast/load and was the basis of the vSense loop.

The total counts in the statistical study, on the DM3068 are rather low....I only ran the test for a few seconds.  So the average is going to be somewhat useless....BUT the SDEV should be good enough.  For example I left the terminated BNC/banana adapter on the DMM and left the statistical study running....it's up to around 100k counts now and the AVG = 365 ish nV, and is climbing....but the SDEV is right around 2.28uV, and dropping.  This is pretty common as the probes polarize, or the resistor in the 50ohm terminator drifts (i should really replace those with a 2 watt low tempco resistor). 

[TunerSandwich]

...and here is the result of a second test (same signal and signal path), with the MXi locked to the dm3068 vcomp/trigger out....I will include a waveform capture of the vcomp pulse, as it was used into channel 2 of the scope....and then channel 2 was hidden on the main test view, but the waveform was being used as channel 3s trigger.  So essentially vcomp out to c2 in...trigger on c2...so that c3 measurement is time correlated to the dm3068 capture (input). 

I am also including a chart of the results, so you can see the deviations from baseline....the clear winner here is the mean of the mean (MXi).  Remember that in 2 samplings the MXi can develop a mean of mean.  So we further increase resolution AND accuracy.  It beats the DC accuracy of the DM 3068.....

Obviously this is using 50ohm coax, so the bandwidth is limited by the cable...








VCOMP TRIGGER SIGNAL FROM 3068






iF WE WANT TO GET REALLY PRECISE, WE CAN FACTOR IN THE MEASURED ERROR FROM THE BNC TO BANANA ADAPTER ON THE 3068 (i DOCCUMENTED THAT IN THE POST ABOVE THIS ONE)

[TunerSandwich]

Here is a test of the same signal etc....BUT with the absolute tightest lock possible to the VM comp out of the dm3068.  After I characterized the signal....I used it into the ext trigger input of the MXi.  It does achieve tighter "lock", than the ch. input...

Both devices were put on "hold/stop" simultaneously....in which case the dm3068 holds the capture, but keeps the VM comp trig out running....this assures that the MXi is still getting the trigger signal....just in case of a split second delay in the trigger line. 

It's not sample accurate triggering/timebase.....which is why I was saying earlier that using a separate DMM and Scope to calculate power quality is not productive.  It lacks the lock of a sample accurate timebase....

[David Hess]

......
Failing that you can use 2 normal probes and engage the DSO maths. 

Won't CMRR be vastly worse with that method?

The DC CMRR can be adjusted with the variable controls for the vertical inputs assuming your DSO has them.  DSOs which implement the variable gain function after the digitizer (most or all DSOs now?) will perform much worse than an analog oscilloscope or old DSO because of quantization noise when making differential measurements.

The AC CMRR can be tweaked with the probe compensation adjustments.

The greater problem is the common mode input range which will severely limit sensitivity.  Also some DSOs are atrociously slow when doing simple math like invert and add.

[David Hess]

Well, OK... but what you'd see on a 300 MHz scope screen looks nothing like the actual shape of signals that exist on a bus capable of 800 MHz. The probe alone will wreck the shape of the signal beyond all recognition, and the low pass filtering effect of the scope ensures you don't even see the wreckage properly.

A standard probe is the wrong tool for these kind of jobs even if they are rated for 300MHz they simply cannot be used at that frequency in any real circuit. An active or low-Z passive divider probe is the only choice and as a rule of thumb you'll need such a probe for measuring signals over 100MHz.

When using passive probes, as a sanity check I sometimes attach the probe and then attach a second identical probe to see how the signal changes so I can evaluate the loading effect of the first probe.  A suitable capacitor to simulate the tip capacitance can work for this as well.

Occasionally when I know it will be a problem, I just design in a little buffer to drive a probe or coaxial line.

[David Hess]

The part that I could not deal with is not having buttons and knobs to manipulate the interface. Mousing around and clicking up/down arrows to change timebase or v/div is a show stopper. For professional audio and video, many companies have control surfaces that provide a more intuitive interface for software based recording and editing. The surfaces have LEDs, displays, knobs, buttons, etc that mimic how it was done in the old days. You can map these functions to the keyboard which is helpful, but still too cumbersome.

I hate GUI implementation which simulate physical interfaces when it detracts from usability which so far has been always.  Skeuomorphic interfaces were a blight when the Greeks and Romans did it (triglyphs) and they are a blight now.

There might be a good oscilloscope graphical user interface but I have yet to find it.  To add to the pain, the ones I have evaluated (at least the Tektronix ones) were full of bugs and the keyboard and mouse take up more workbench space.

Does anything like that exist for USB scopes? Although they are not perfect, they will likely be contenders in the near future and the human interface seems to be the biggest limitations for me.

I have not seen anything like this.  When I do napkin DSO and other instrument designs, I sometimes end up with a separate optional "control head" which controls the base unit though the high speed interface which would otherwise go to the PC and may or may not include a display.

[TunerSandwich]

DSOs which implement the variable gain function after the digitizer (most or all DSOs now?) will perform much worse than an analog oscilloscope or old DSO because of quantization noise when making differential measurements.

That depends on how the dithering (if any) is applied.  Quantization noise can be shifted to orders of magnitude above the measurement range, with proper dither.  Sony did this very effectively with DSD.., as the THD can be virtually eliminated from the "usable bandwidth". 

The holy grail would be to implement an A/D and pre-filter into the probe body and send direct 1 bit digital (at ludicrous sample rate) to the scope input.....a steady state would allow for almost complete reduction of the noise in the step response.....and the signal would be virtually analogous to the analog input.....

I do recognize that is wholly impractical though, but a man can dream 

P.S. food for thought....it is also plausible that the noise of an analog front end would be greater or equal to a digital signal, depending on the age and spec of the input amplifier.....that is a very interesting potential scenario (I am thinking a test is in order).  I should build some filters and test the 2225 against the WRMXi.  Not that the 2225 is some fantastically low noise scope, but it's all I have...

here are some comparisons.....the 2225 is limited to 5MHz (x10 is pulled), the DS2102A is limited to 100MHz, and the WRMXi is limited to 600MHz....as you can see the rigol is "beating" the "noise" on the 2225.....but it's hard to say if that is actually true, as I am not sure how much of that noise is being eliminated with post A/D filtering/dither.....Rigol seems to be playing around with some trickery in DSP, as I highly doubt the quality/noise of the actual pre-A/D amplifier is better than the LeCroy.  However I do wonder if the noise on the 2225 is because of it's inputs being "out of spec.....that is something I will have to investigate...

FWIW...IF we can trust the represented waveforms on the display (measurements), I have to say Rigol has done a fantastic job in eliminating noise from the trace....as it's 20 times the bandwidth of the Tek 2225 (as it's setup).....and same from LeCroy as that is 6 times the bandwidth of the Rigol......obviously keeping in mind, that the LeCroy vertical limitation is 2mV/div, and the other two are set to 500uV/div

[egonotto]

Hello,

the Rigol and many other DSO does not show the noise  at output from ADC if the input is set to ground. They shows only a useless straight line.

Set the input do DC and short the input, then you see the noise.

Regards
egonotto