EEVblog #199 – Tektronix MDO4000 Oscilloscope Teardown

Part 2 of the Tektronix MDO4000 Mixed Domain Oscilloscope review.
This time, a world exclusive tear down.
What magic lurks inside the die-cast shielded RF spectrum analyzer circuitry?

Be Sociable, Share!

    About EEVblog

    Check Also

    EEVblog #801 – How To Design A Digital Clock

    EEVblog #801 – How To Design A Digital Clock

    The next step for Ahmed Mohamed How to design your own digital clock from 4000 ...

    • Squonk


      Dave, you’re the Man!

    • jarrod

      That really is a thing of beauty. the rf section is briliant. I wonder if thats where EE is heading with transfer speeds between chips getting faster and faster.
      I reckon that beaded line is a 1/4 wave transformer or something.
      Good on techtronics for getting this out to you

      • jarrod

        And of course thanks for putting together such a great teardown :)

    • JRR

      That RF section truly is pornographic.

    • Fox

      Yeah, great teardown, and the RF section is a really awesome construction.
      In my opinon the beaded line is some sort of unbalanced to balanced converter to use the maximum performance of the ADCs, because as far as i know, today almost all ADCs have differential inputs.
      Great idea from Tektronix to put an spectrum analyser into a DSO, but obviously the scope needs some performance tweaks.

    • Ciccio

      Thanks for showing such a brilliant engineering work.
      The golden plated surface of the external layer of the RF PCB is full with maybe thousand round “holes” or vias or dimples or whatever they are. I cannot see clearly, but I’ve never seen something like.
      Anybody out there may explain me what they are and what is their purpose?

      • Ian Barnes

        They are ground plane stitching vias.

      • Zimbo

        To expand on what Ian said… those vias stitching the copper pour on the top layer to the ground layer make the the copper pour to appear electrically as essentially just one big solid chunk of metal. The idea is that you don’t want large (relative to the wavelengths involved) pieces of metal just “floating” (not tied to ground or some other low-impedance node), as they can act as capacitors to form resonances that “suck out” bits of the microstrip signals a bit, they can couple signals to some other microstrip that you don’t want them to, etc.

        That all being said, they probably did overkill it a bit here — at 6GHz, their wavelength would be in the ballpark of 25mm (in the board material), and hence to get a ground connection every 1/10th of a wavelength you’d stitch every 2.5mm. Additionally, you usually really only need to stitch around the peripherally of the copper pours as well, since that’s where the strongest current densities around. However, since vias are the next best thing to free, I expect it was easier to tell the PCB layout guy, “just throw in a ton of vias everywhere!” and call it good rather than bothering to calculate any of this. :-)

    • Seb

      Thanks Dave vor this brilliant teardown. I think the big micro with the metal housing is an Xilinx FPGA from the Virtex series or someting else. the metal shielding on the analog board is impressive. this is not a standard cheap metal can EMI shiel. it’s a havy duty custom made shieldung:)quiet nice.

      what i’m wondering iss the big fan inside the tek. did you find any dust filter oder something else at the housing ?
      when i take a look into a atx powersupply from an office pc oder a pc which is running in a lab (not such on, running in a high clean area), its heavy dusty. now take a look in to the future when the tek runs 2 or 3 years in a lab. i think theres the same problem with the dust like in the PSU. the component heat rise up and reduce the performance or kill some components by the thermal death.

      i hope tek will spend some time in development to release a passive cooled unit.

    • Patrick

      That is one sexy piece of electronics, how long dis it take for the teardown?

    • Frank

      Dave: The unknown mounting on the back of the machine seems to be a standard VESA mounting for screwing the thing to a monitor arm or a wall mounting. These VESA mountings are very common on TFT displays.

    • Bambur

      Dave, thanks! Would you post high-resolution photos of the boards please, if you have taken any of course? Thanks in advance!

      • EEVblog

        Yep, coming up on my Flikr account shortly.

    • Rob

      Thanks Dave, good one. Wonder how much carry over from MSO2000 series boards and firmware – they also have slightly odd user interface. Could explain the odd vertical soft key placement.

      Looks like more than a sprinkling of off-the-shelf minicircuits RF parts – mixer (SIM-762H+) and lots of LTCC filters (white ceramic 1206 sized things). Synthesized LO looks like standard Analog Devices part (ADF4350).

      Frank beat me to it to suggest backplate mounting was for VESA arm to get the thing off the bench.

    • huh

      As beautiful as unreachable, though as a humble hobbyist I wouldn’t have any use for 99% of its features.
      Anyway, that RF section surely deserves a Playboy cover, or better a close up high res photo in order to turn it into a giant poster:)

    • Carlos

      The board is full of HMC311 amplifiers, MA/Com SPDT RF switches, LTCC low and bandpass filters.

      There are a couple of switched attenuators, the first one (20dB) is switched in or out with the Panasonic canned relay.

      Alternatively to the RF input there is a switchable broad band noise source for system calibration.

      The transformer near U2113 HMC311 switchable amplifier is not a transformer but a wideband choke inductor, because that amplifier should work from 50khz up to 6GHz, so a simple inductor can’t be used for biasing.

      After this amplifier, the signal path is divided into 3: 50khz to 3.75GHz (lower path), 2.75 to 4.5GHz (upper filter) and 3.5 to 6GHz (medium filter) depending on the selected center frequency and span.

      Lower band (50khz to 3.75GHz) can be digitised directly, but the higher bands have to be downconvertered. There is a ADF4350 local oscillator with two LTCC band filter banks and amplifiers and the SIM-762H+ mixer. The bottom left corner switch selects the mixer IF output or direct low band path.

      At the upper left corner the signal is amplified (2x HMC311) and broaband power is sensed with a 70dB logarithmic detector for ADC power trigger acquisition.

      That beaded coax line looks like the input balun for the 2x HMC480 balanced amplifier.
      This amplifier is the 10GSPS ADC driver, hidden somewhere.
      It should be near, because a 5GHz clock signal is generated on the HMC429 4.45-5GHz VCO and amplified with another HMC311.

      Where is the 10GSPS ADC is a mistery, but the spectrum analyzer itself finishes here.

      I got some of the info (exact frequency bands and ADC speed) from the Fundamentals of the MDO4000 Series application note. Other data comes from my RF knowledge so it could be wrong.


      • EEVblog

        Awesome explanation, thanks! Can anyone add to it?

        • Jean

          If I have time, I’ll try to assemble the photos and put Carlo’s explanation, plus mine (but he already did a good job) on them.

      • Gerald

        Can someone make that explanation into a slide show, pretty please?

      • Pantelis


        Any insight as to why they would divide the channel into 3 seperate paths as opposed to undersampling the signal and using a tuneable bandpass filter to examine the part of the spectrum they may be interested in? As someone who has often thought about making a spectrum analyzer, the cost of precisely tuned oscillators and noisy mixers has always been a turn off.

        The board looks awesome though. Someone(s) must have spent a long time looking at HFSS.

        Also, for the cast aluminum plate, my local sales rep for Rogers Corporation has shown me demo circuit with stuff like that. Pretty cool stuff.


        • Carlos


          You have answered yourself. High Q, tunable band pass filters (for example YIG tuned filters) are heavy and expensive. Giga samples ADCs and very high speed logic/fpgas are nowadays readily avaliable at very cost effective prices, plus >3GHz spectrum bandwidth simultaneous capture is awesome. Compare this with 36MHz maximum BW in real time mode of the top end RSA3408A spectrum analyzer!

          This equipment, as an oscilloscope itself, already has very high speed ADCs and logic, so it’s not so rare that Tektronix decided to do that this way. Of course there are some tradeoffs, mostly dynamic range and mixing spurs, but i’m sure Tektronix did it well enough to tell it’s a real spectrum analyzer and not a garbage generator.

          Sorry for my english and regards from Spain!

        • Zimbo

          Undersampling still requires that you have analog bandwidth sufficient to get the signal cleanly to the sample & hold circuit in the ADC. I.e., at least 6GHz without any significant amplitude droop. It’s almost certain their ADCs don’t have that much analog bandwidth (if they did, it’d probably be a 2GHz scope rather than a 1GHz scope :-) ), so mixing down is was likely easier than finding a higher analog bandwidth ADC. (And depending on exactly how the RF Power trigger circuitry was done, probably a fancier power detector as well.)

          Note that the fastest real-time oscilloscopes in the world — those made by LeCroy, with 45GHz frequency response –, use this same trick: You just can’t buy an ADC with 45GHz of usable analog bandwidth today… but you can buy one with ~25GHz of analog bandwidth, so those fancy LeCroy scopes split the input signal up into a (very roughly — there are margins that need to be added here) DC-22.5GHz path and a 22.5-45GHz path and then mix the 22.5-45GHz path back down to DC-22.5GHz prior to digitizing.

          The math to actually reconstruct the original waveform in this scenario is somewhat non-trivial once you consider all the path non-idealities involved — this is one of the reason you’ll find, e.g., 1-2GHz CPUs in high-end digital scopes.

          • Pantelis

            Hey Zimbo,

            Took me a minute to check out the comments again. I understand what you’re saying with the ADC sample bandwidth and in now way would I expect any ADC with that much bandwidth. An idea I’ve had (for one day) would have a bandpass filter (preferably adjustable) that would be close to the bandwidth of the ADC which could be tuned to examine different ‘chunks’ of the spectrum. The undersampling would allow those chunks to be folded back into the sampleable realm of the ADC.

            Again…just an idea I have one day when I get sick of my job and decide to do electronics on my own. :)

        • Zimbo

          One other thing… I’d question whether or not there really was that much HFSS work involved: The custom shield divides everything into nice, tight cavities, and they’re using lots of ground stitching, took care to mitre the microstrip corners, etc. — At 6GHz, those techniques are often “good enough” without bothering to resort to HFSS. This is particularly the case when, these days, pretty much all the 1GHz+ scopes (from Agilent, Tek, LeCroy, etc.) use some form of digital signal correction to account for various small “bobbles” or frequency loss: If you have, e.g., some 1dB depth 10MHz wide resonance in your frequency response at 3.18GHz, it’s not worth the time and effort to spin a board and try to fix it directly; it makes far more sense to just calibrate it out instead.

          I *do* fully expect they used a 2.5D simulator to take a look at the layout’s behavior, though (something like Axiem or Sonnet or Momentum) — they’re rather faster and cheaper than HFSS (which is a 3D simulator); that’s often worthwhile even at 1GHz.

    • Carlos

      Forgot to mention that this was hard porn. Please not read if you are not +18!

    • Bill Clay

      Dave, is this a demo unit from Tek that you have to return after reviewing? This is regarding your last comment at the end of the vblog.

      Also, I noticed the traces in the RF section had all the corners cut off. What is the reasoning for this?

      • Carlos

        Bill, they are called mitered or chamfered bends. 90deg corner bends are totally forbidden on microstrip lines because they result in reflections and impedance distortion due to excesive shunt capacitance introduced at the corner.

      • EEVblog

        Yes, it has to go back.

      • Worf

        @Bill: If you watched Oscilloscope: The Movie (i.e., the review), at the very end Tek was wary of sending Dave the scope out of fear that he’d break it (because he always take it apart). You know, before Dave drop-tests it. (Spoiler. Go watch the movie for the results).

        That’s where the comment comes from.

    • Greg Uzelac

      Wow, I am in love all over again. That RF section was stunning! The close-ups were awesome. Thank you. Now, how do I tell my wife that I have seen other parts that are mind blowing?

    • chuck


      Kinda off topic, but you’ve mentioned that youtube is the where your videos will appear first (the blog’s rss feed is updated later)

      But what about itunes? How is that updated in relation to youtube/eevblog schedule?

      • EEVblog

        iTunes is updated several hours after the blog, I’m not sure how often it pings.
        The iTunes podcast is part of my posting to the blog, so it updated sometime after that.
        It’s last, sorry!

    • peet
      • Carlos

        You are right peet. It’s a diversity switch built with 2 PIN diodes in a SOT-23 package.

    • troubleshooter

      – I counted over a dozen RF switching parts in one of the signal paths.

      – There’s a VCO next-door to the signal input path.

      – The biasing and decoupling on two of the MMICs is reversed in layout.

      – and there’s a design by committee panic going on trying to identify the spurs going into the +17dBm mixer.

      …I’m not sure about this one

    • Pingback: MDO4000 RF “Reverse Engineering” « Pick and Place()

    • Jean


      I just made a post with a diagram of the RF section, using the info gathered from Dave’s pictures, the comments and my small RF knowledge:

      Don’t hesitate to add corrections or more info in the comments, so I can update the schematics.

      Greetings from France!

    • All American Pressure Canners

      An oscilloscope is an electronic instrument that is chosen to represent the voltage of an digital machine. It signifies an individual or additional electric potential differences in a apparent two-dimensional graph, with the horizontal axis representing time and the vertical axis showing voltage. It is utilised to diagnose the operating problem of any electric devices.

    • tmbinc

      I’m trying to solve the “ADC mystery”. Dave, which model did you tear down exactly? Those with 2.5GS/s or 5GS/s on the analog channels?

    • tinhead

      on the bottom side of the RF PCB, left top corner there is ADCMP562, you can clearly see traces going from the comparator to the PCB region where the mysterious ADC shold be located. So i assume TEK is using this comparator as “sampling head”, which is more than good enough for a spectrum analyzer.