GW Instek MDO-2000E

[MrW0lf]

Usual much loved 2000E looks...



...with unusual feature:



...and one trendy option (too much):



http://www.gwinstek.com/en-global/products/Oscilloscopes/Digital_Storage_Oscilloscopes/MDO-2000E_Series

+interesting read to FFT fanboys:
http://www.gwinstek.com/en-global/Download/DownloadFile/DownloadFile/download%23_%2301_Oscilloscope%23_%23MDO-2000E%23_%23Brochure%23_%23PDF_BH_MDO-2000ESeries_E.PDF

[nctnico]

Interesting! This way you can have a complete test solution from a single piece of equipment. IMHO it seems geared towards educational situations.

[MrW0lf]

Reading manual:

Range Start: -250MHz ~ 499.995MHz
Stop: 500Hz ~ 750MHz
Span: 1kHz ~ 500MHz

Span:RBW
5000:1...1000:1 auto or manual

Up to 4 trace types (Normal, Average, MinHold, MaxHold) at same time, source is any (one?) analog ch.
I was thinking for some time now that something like this can and should be done (FFT with more conventional SA style approach)

Edit: Something does not compute. Why max start 499.995MHz... Maybe they mixed up Stop and Span in manual.

[saturation]

GWI really leverages its Zynq development, adding "modules" to create forks in the DSO line.  Like nctnico, I do see this in particular focused on the EDU>> factory floor market, as with an AWG, DMM, DSO, PSU and SA rolled into one, really saves on school bench space but some of the specific instruments are not outstanding for the overall cost [ by list EUR1700 for the 4 ch 200MHz top line model], in particular:

The SA, spurious response is near that of the software FFT in the 1000B, even if overall its better than an FFT
I'd hate to damage an entire DSO using what is spec'd to be a $50 DMM, or lug the whole around for just a DMM
The PSU is like those in powered breadboards, and at most for TTL, somewhat limited for experimenting with robotics motors

[MrW0lf]

Spectrum analyzer demo on the MDO-2000E:

Not very technical... all shown could done with regular FFT also.

[SWR]

Interesting! This way you can have a complete test solution from a single piece of equipment. IMHO it seems geared towards educational situations.

I've just ordered two for our embedded SW engineers and an extra one for my home lab.
Desk space does get crowded with a couple of 27" monitors.

It should be quite nice for debugging HW abstraction layer and the HW it self.
The AWG is nice for input signal simulation and the HW guy can simulate CPU output signals before the SW is ready.
The only missing thing that would be nice to have is an electronic load to test the outputs.
But to be fair - 6 instruments in one is quite a bargain for €1410.

It's also a fairly cheap LabVIEW all-round I/O device for automated testscripts.

Best regards
Soren

PS: I haven't seen any comments on the built in self calibration using the CAL output on the back.
It would be a pretty good business case to include a bit of redundancy and good diagnostics to have a reliable CAL output.
I would gladly pay $100 extra for this if I'm saving $1000/year in external calibration and don't have to be without the instrument for several weeks.
Do any of you know about the reliability of this output?
It would be really nice to be able to do self calibration in 20 minutes any time you like check the accuracy (and skip the calibration bill & being without the instrument).

[rhb]

The problem with the self cal is the lack of traceability.  If you have enough scopes, buying a scope calibrator and sending it out for cal makes good sense.  With the Instek 2000E line, Leo's pulser and an FFT will get a lot of data.  The only shortfall is the higher ranges on the attenuator which are also the most problematic.

I've been meaning to see what the cal signal looks like.  That might make a good evening amusement.  I really don't feel like doing any heavy lifting and I'm back to "waiting on parts".   In this case I need to order some too. I'll see how long that takes.

[SWR]

Yes, traceability is key if you CAL output is not fail-safe.
The big question is: how much would it cost to disrupt the calibration industry?

You need to build in some redundancy and good diagnostic coverage on the CAL output, but it might not be terribly expensive to do that.
Adding $10 worth of HW would not be a problem because the function is worth much more.

With a careful design it should be possible to build a fail-safe CAL output with enough redundancy and diagnostics to detect when something is wrong with the CAL output signal. The circuit should avoid any common cause failures, be able to detect any possible failure and deny self calibration in case of a failure in the output or the diagnostics circuitry.

They might not have gone to this length with the GW Instek design, but I can't even imagine how much value there would be in disrupting the calibration industry.

[rhb]

It would not be hard to disrupt the calibration service industry.  Whether or not that is good is another question.  Watch the video of Dave's visit to the Keysight cal lab.  However, $10 of HW would allow extending the cal interval out to several years.

I've not investigated what a Fluke scope tester does, but I did know  the procedures for a Tek 465 and had acquired most of the kit for that.

Most of the traditional analog scope tests can be replaced by some precision input signals and DSP routines.  For example, while rise time is an important aspect of BW it does not quantify the flatness of the spectrum.  But with a signal such as Leo Bodnar's 40 pS pulser  the spectral flatness can be recovered to ADC precision via an FFT.   The traditional method was a signal generator and a selective level meter in addition to a Tek 106 pulse generator.

As an example, a simple RC circuit and some mathematics are sufficient to characterize and correct ADC errors. The cost would be the flash to store the code.  However, the non-recurring engineering cost is not trivial.  What I'd consider a proper state of the art job would run $20-30k.  That's based on  using  a sparse L1 pursuit solution. It would only take me a few days to prototype the code, but implementing a production version is a lot more time consuming.  You have to make sure you have found and tested *all* the edge cases.  It can take days to identify the edge cases and days more to figure out how to test them.  And there is no such thing as self documenting code.

I wrote a couple of 15,000 line libraries that had no bugs reported in almost 20 years of use before being retired because it was obsolete.  This was part of a 500,000 line port of old FORTRAN code from VMS to Unix.  The first year in service we had fewer than a dozen user submitted bug reports.  It asymptotically  approached zero because we had a large regression test suite which ran every time we built the system on multiple platforms.

The Instek uses the Zynq 7010 for the main functions.  There is a Spartan 6 which appears to be for handling the LCD and front panel chores.  I believe the Siglent also uses the Zynq.

I  managed to pick up a GDS-2072E for $244 in an Amazon NOS clearout.  I'm assembling a development environment consisting of a Zybo Z7-20 and a BeagleBoard X15.  As I have no FPGA experience I got a MicroZed specifically for the Avnet tutorials.  All of this is for the purpose of developing an open source firmware for Zynq based instruments.  That will take a few years, but I need a good project.  There is no work for people like me in the oil industry (too old and over qualified) .

As it happens I spent the last 3-4 years studying sparse L1 pursuits which is a more general description of compressive sensing.  There are no instruments on the market that offer that yet.  A prototype was done by Nicholas Tzou at Georgia Tech for his PhD which he finished in 2014.  There appear to be many advantages to applying sparse L1 pursuits in a DSO besides reducing the data volume.  It's interesting and I think that an  FPGA compressive sensing IP has good commercial potential.  People will tell you they love you for free, but if they give you money, you know it's true.

I've posted a bit about this, but all it's attracted is a bunch of derision from people who do not know and will not read the mathematics.  Aside from not knowing the work of Candes and Donoho at all, they don't actually know Fourier, Wiener and Shannon well.

[SWR]

It would not be hard to disrupt the calibration service industry.
...
Most of the traditional analog scope tests can be replaced by some precision input signals and DSP routines.  For example, while rise time is an important aspect of BW it does not quantify the flatness of the spectrum.  But with a signal such as Leo Bodnar's 40 pS pulser  the spectral flatness can be recovered to ADC precision via an FFT.

I think it would be a pretty good business case to do this as a scope manufacturer.

You don't need to find all possible errors on an instrument - just the ones that might go unnoticed by an operator like increasing voltage and frequency tolerances.
An LCD or a button that doesn't work would be noticed by the operator anyway.

You do need a precision input signal with redundancy and diagnostics, so that you can detect if it is no longer as accurate as required.
The test needs to uncover the same errors as the standard calibration procedure - no more and no less.

I would certainly fork out some cash for a function like that or choose an instrument over one that didn't have it.

[rhb]

It would not be hard to disrupt the calibration service industry.
...
Most of the traditional analog scope tests can be replaced by some precision input signals and DSP routines.  For example, while rise time is an important aspect of BW it does not quantify the flatness of the spectrum.  But with a signal such as Leo Bodnar's 40 pS pulser  the spectral flatness can be recovered to ADC precision via an FFT.

I think it would be a pretty good business case to do this as a scope manufacturer.

They are doing  it.  The self calibration and signal path corrections are examples.  They may be doing a lot more of it in manufacture than is widely known.

[nctnico]

Yes, traceability is key if you CAL output is not fail-safe.
The big question is: how much would it cost to disrupt the calibration industry?

A lot. The CAL output is nothing more than a 'known' signal an oscilloscope can use to do a quick verification of ADCs and input attenuators. It is a very nice to have and for some high-end oscilloscopes it can mean the difference between needing a lot of equipment (Tektronix TDS500/600/700 series for example) or just a BNC cable (Agilent Infiniium 54000 series for example) to adjust it.

Nevertheless a self calibration/adjustment isn't traceable to any standard. So even with a CAL output you'd need to verify the oscilloscope using known-good equipment which has a traceable calibration if you want to absolutely sure the oscilloscope meets it's specifications.

[SWR]

Nevertheless a self calibration/adjustment isn't traceable to any standard. So even with a CAL output you'd need to verify the oscilloscope using known-good equipment which has a traceable calibration if you want to absolutely sure the oscilloscope meets it's specifications.

Traceability to a known standard is only needed if it is possible for the reference signal to become inaccurate without being able to detect that it has become inaccurate.

If it was possible to design an accurate reference signal and have fail-safe detection of any deviation outside the specified tolerance, then you wouldn't need any traceability to a known reference as long as the signal is within specification. You would need to avoid common cause errors in the redundant diagnostics circuits.

The disruption is the removal of the need for traceability to a known reference.

[rhb]

You can't remove the need for traceacability. Calibration is the process of verifying the system against the international standards.  There are different classes of calibration, but what a calibration lab offers is instruments which they in turn send to the national labs for calibration.

If the instrument performance can vary over time, so can a built in calibration reference.  The real key to lowering calibration costs is a low cost calibrator that can be checked on a regular basis at low cost.  An ASCII file of measurements of one  of Leo Bodnar's 40 pS pulsers measured with a calibrated DSO combined with some DSP work will handle a very large chunk of the DSO calibration tasks.  There is no reason that having such a device measured on a calibrated instrument should be expensive.    Nothing is being adjusted.  All that is required is to connect the device to a 10 GHz or faster scope,  run a program on a PC and store the measurements in flash on the device.

Designing such a device is not trivial, but it's not that hard either.

[SWR]

You can't remove the need for traceacability.
...
If the instrument performance can vary over time, so can a built in calibration reference.

The big question is: can you build a reference that cannot vary undetected?

If you can build a circuit that can detect any deviation outside a defined tolerance range, you can use it for calibration if it is within range, and display a message that it has to be shipped for external calibration if it is outside the defined tolerance range.

I agree that a very well defined square wave can be used to calibrate the input channel of a scope over a wide bandwidth range. The key issue is to make a circuit that can detect a variation in this reference signal with a very high propability. The propability of failure detection must be higher than the propability of an instrument going out of spec between calibrations.

You would need redundant channels with separate power supplies to remove the possibility for common cause failures, but if this could be constructed to be cheaper than an external calibration you would have a good business case. Not having to be without the instrument and the ability to do a calibration any time you're suspicious of a failure would be an added bonus. 

[nctnico]

You can't remove the need for traceacability.
...
If the instrument performance can vary over time, so can a built in calibration reference.

The big question is: can you build a reference that cannot vary undetected?

NO! Not with 100% certaintity.

[rhb]

You can't remove the need for traceacability.
...
If the instrument performance can vary over time, so can a built in calibration reference.

The big question is: can you build a reference that cannot vary undetected?

NO! Not with 100% certaintity.

Well, actually you can.  But you have to calibrate it at regular intervals to detect the deviation.  This is what the national and other primary standard grade labs do.  They constantly compare their standards against each other and other laboratories.  Comparisons to other labs are typically done with transfer standards.

Read the metrology thread.  There's a set of transfer standards that circulates among amateurs.

The existing built in functions can correct a lot of things, but they can't correct errors in the builtin references.

There's an old saying, "Go to sea with one watch or three."

[nctnico]

I think SWR is asking whether it can be done without comparing (calibration) to known-good references and then the answer is still no.

[rhb]

Which is what I said.  If you don't check it the variation can't be detected, but that's not a solution.

[SWR]

NO! Not with 100% certaintity.

I agree - not with 100% certainty, but then again calibration doesn't give you 100% certainty.
Your scope could go out of calibration on the way back from the calibration lab.
Although the propability that this is going to happen is quite small it is not zero.

I don't think you can ever achieve 100% certainty, but if you can achieve an acceptable high level of certainty on par with what periodic calibrations can deliver, then it should be an equally good solution - just at a much lower cost because you remove the manual work of doing a calibration and all the overhead of running a business that perform the calibration.

[SWR]

The existing built in functions can correct a lot of things, but they can't correct errors in the builtin references.

They might not need to correct it (fail operational) - it might be enough to detect it (fail safe).
All you need is a message that your instrument is no longer within the specified tolerances, and the message has to have a high level of reliability - not 100% but pretty close to that.

There's an old saying, "Go to sea with one watch or three."

Thats a fitting example using redundancy as a mitigating factor - diagnostics is another useful tool. If the three watches were of different construction (reducing common cause failure) that would further reduce the propability of all three failing the same way at the same time.

[SWR]

Most of the traditional analog scope tests can be replaced by some precision input signals and DSP routines.  For example, while rise time is an important aspect of BW it does not quantify the flatness of the spectrum.  But with a signal such as Leo Bodnar's 40 pS pulser  the spectral flatness can be recovered to ADC precision via an FFT.

Thank you for the reference - that's a really cool device and I've just placed an order for one.

As far as I could read in another thread it's got quite accurate TCXO timing, but there's no mentioning of the amplitude accuracy, so I assume that it doesn't use a precision voltage reference for the amplitude. That would only be for the gain calibration anyway, so it's perfectly fine for relative amplitude measurements over a wide bandwidth. It should be perfect for a 200MHz scope and I'm looking forward to playing with this.

[rhb]

You may not like your scope's step response after you get it.  I was nervous about accidental damage, so I covered mine with a big piece of heatshrink.  I cut X's for the connectors, slipped it on and shrank it.

One of the goals of my FOSS Zynq DSO project is to implement filters in the FPGA that correct errors in the step response.  The step responses of my MSO-2204EA and GDS-2072E are awful.  Interestingly, the 100 MHz BW response of the MSO is almost identical to the step response of the GDS.  I need to plot them together for a closer comparison.  When I do that I'll calculate the amplitude spectra.

I cut down a heatsink to fit my MicroZed and glued it on yesterday, so I'm about to start work on Vivado and Verilog.  The heatsink took the slow boat from China but finally arrived.

[SWR]

That's some interesting projects you're working on.

I think I'll just 3D-print a small housing that fits the unit.
Thank you for the tip.

[rhb]

Much nicer, but  you've got to have access to one.  Please consider posting the files to Leo's pulser thread.  I think others would be interested.