TDR sampler/pulse generator "pod" (TDR-G3)

A "pod" containing fast pulse generator and sampler. This "pod" should serve as the basis of TDR-G3. The other part will be the main board.

Public Chat
Similar projects worth following
This "pod" contains MAX3798 pulse generator (programmable laser driver promising 26 ps risetime), ADCMP582 used as sampler and SY89321 as LVPECL to CMOS converter. The board serves as both pulse generator and sampler. It can also be used as oscilloscope frontend, however it cannot be used fro triggering.
The project is inspired by work of loxodes, Ted Yapo and Darwin Sabanovic (although the last one never shared his project).

The "pod" contains MAX3798 for pulse generation. The datasheet promises 26 ps risetime into 50 Ohm termination. Its modulation current can be programmed in 512 steps thus allowing for changing the output power of the generator.

The generated pulses are fed into resistive network (special three terminal resistor for this purpose) and then into DUT. The mix of incident and reflected impulses are fed back through the resistive network into ADCMP582 which serves as sampler.

The PCB fits into SFP+ socket. By building a main PCB with 2 pulse generators with variable time offset and one DAC, this can be a complete TDR. By converting its output to frequency domain, it can serve as crude VNA able to measure S11 parameter When using two pods, it can serve as both TDR and TDT. Because both are complete transceivers, it is possible to use a device with 2 of these pods as full 2-port VNA, which can measure S11, S21 and also S12 and S22.

Gerber files. Please note that the stackup is Isola 7628 4-layer 0.2 mm stackup with 1 mm final thickness as used by JLCPCB.

Zip Archive - 313.34 kB - 03/08/2020 at 09:29




Adobe Portable Document Format - 84.91 kB - 03/08/2020 at 09:29


  • An assembly job and first mistake found

    MS-BOSS05/04/2020 at 18:21 0 comments


    the board finally arrived. So I assembled it (very funny when it comes to QFNs) and now... well, nothing. I have to design the motherboard which will utilise this.

    The first mistake I found while assembling it is that R110 shouldn't be there. It can be simply shorted together when hand-assembling. Somehow, I forgot there is internal 50Ohm termination inside the ADCMP582 since I have a design laying around which uses the ADCMP576 which has no fancy internal terminations and I got these two mixed up.

    There is one trick I use always when assembling anything with QFN or QFP packages. First I solder all the solder pads on BOTH the PCB and the chip, then I apply solder paste (manual), then solder it using heat gun. Then, the failure rate gets under maybe 5%, most of which are shorts, unsoldered pads are gone. And removing the shorts is easy with soldering station and a bit of flux. However, since I didn't make anything in months, I forgot to use this trick and I was fighting with 1 unsoldered pad for 20 minutes...

    Ready to solder once coffee is made.

    The problematic solder pad after being already solved (the leftmost one at the bottom row).

    Cleaning in IPA in ultrasound cleaner:

    Clean board with soldered SMA connector (Rosenberger 32K243-40ML5, a middle-class connector). There is one trick to solder the central pin (believe me, it is to be soldered, the high-class SMAs which can be easily reused have a springy pin, not a stiff pin) - first position and center properly the SMA on the feed line (the centering is quite important), then slightly tighten the screws, position again, tighten and check whether it is correctly placed. Then, using a solder-paste applicator with narrow tip, apply a blob of paste on the pin and the feed line, then using the tip, try to make sure the paste is applied to both parts. Then, solder using heat gun.

    That's all for today :)

  • First prototype was sent to the fab

    MS-BOSS03/08/2020 at 09:01 0 comments

    Hi folks,

    I thought that after building my first standalone TDR (now called TDR-G2), I should build another one which would fix its issues. The main issues were:

    • High noise floor
    • Limited bandwidth (about 2 or 2.5 GHz)
    • Limited risetime (around 270 ps)
    • Slow sampling
    • Very bad SMA connector and its footprint (almost completely correctable by SOL calibration)
    • Many other issues caused by focus on low price

    This time, the reflectometer will use ADCMP582 as one-bit sampler. Because its threshold will be controlled by a DAC, it will be possible to approximate the measured value. See Ted Yapo's oscilloscope (here on HaD) or loxodes's TDR (on GitHub) to see how well it works.

    The pulse generator used is MAX3798 which can generate edges with 26ps risetime. Its modulation current can be set in 512 steps using internal DAC, thus allowing for setting the output RF power (if you have ever been measuring nonlinear devices with VNA, then you know how useful it can be). If the range proves to be too narrow, PE4306 or similar attenuator may be added later. This driver should be able to operate into 50 Ohm network without any funny impedance matching.

    The used SMA connector is Rosenberger SMA, which is reasonably expensive nad should be more than good enough for this application.

    The coplanar waveguides are not ideal, mostly due to the fact that there is one coplanar going from the connector, then transitioning into coplanar going into the splitter, then into the two coplanars going into the generator and sampler... It is impossible to reduce reflections to zero (also because I do not have nor can use any EM solver). However, the reflections should be small enough to be suppressable by SOL(T) calibration, which was already proven in the TDR-G2 where a cheap chinese SMA was used along with very bad footprint.

    The board was sent to JLCPCB for manufacturing. The stackup is shown on image below. It is the Isola 7628 4-layer stackup as listed on JLCPCB website. Manufacturing cost for 10 pcs of 1 mm 4-layer impedance driven 7628 stackup board was $14 not including shipping fee. The board finish is HASL, because I do not prefer ENIG (see my microwave tips to see why it isn't good for microwave devices). And it makes the boards 3 times cheaper. The shipping fee was $7 for China post.

    The board should fit into SFP/SFP+ connector. Its dimensions are 35x37.5 mm, thus not fitting into standard cage. Sorry for that, but it was not my intention to fit int othe cage. Mounting holes are for M3 screws.

    Sidenote: You may be interested in seeing this application note.

    ISOLA 7628 0.2 mm stackup coplanar (Er = 4.3):
    W = 0.315 mm
    G = 0.14 mm


    W = 0.365 mm
    G = 0.25 mm

    ISOLA 2313 0.1 mm stackup coplanar (Er = 3.64):
    W = 0.21 mm
    G = 0.162 mm

View all 2 project logs

Enjoy this project?



Armin wrote 10/06/2021 at 09:44 point

I currently work on creating a simple IR-UWB radar and this project requires the generation of gaussian pulses around 130 ps. The MAX3798 looks like a cool alternative to the classical SR diode pulse generators or logic glitchers. Can you say, which rise times and amplitudes you achieved with your design?

  Are you sure? yes | no

Dan Maloney wrote 03/09/2020 at 17:11 point

I'm interested in how the PCB finish affects performance, and why ENIG isn't appropriate for microwave. Hope you can join the Hack Chat this Wednesday at noon Pacific where we'll discuss PCB finishes.

  Are you sure? yes | no

MS-BOSS wrote 03/10/2020 at 12:25 point

I already published a few bits of information on that topic there:

I will try to join the chat. if possible. I am currently resolving some issues regarding the COVID and our makerspace...

  Are you sure? yes | no

James Wilson wrote 03/09/2020 at 03:22 point

ADCMP582 inputs have internal 50R terminations to the +/-IN_T pins, does this mean R110 is unneeded?

  Are you sure? yes | no

MS-BOSS wrote 03/09/2020 at 10:04 point

The LVPECL termination requires R110. If it were used with some CML gate as driver, it could be unnecessary (but not checked that the 582 accepts CML).

  Are you sure? yes | no

MS-BOSS wrote 06/09/2020 at 18:48 point

Hi, I'm sorry for my first answer - the R110 is indeed unneeded and is a mistake which was resolved during assembly. I was working on a design incorporating the similar ADCMP567 which doesn't have internal termination.

  Are you sure? yes | no

MS-BOSS wrote 03/08/2020 at 09:35 point

Hi all,
I would really appreciate if you wrote here any mistakes you spot on my PCB, because I am not an expert at microwave/high-speed design.

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates