James Wilson active probe

[ataradov]

I recently needed an active probe and I did not want to pay a ton. So, I looked around and found a project by James Wilson (https://jmw.name/projects/active-probe/) based on BUF802.

I wanted something that looks more like a probe, so I made my own version based on the same schematic. I removed the switching converter that generates the negative rail and just used a couple 9V batteries. This way I fully avoid any switching supplies in the system and gain a convenient way to provide the power.

I also made a 3D printable case for the probe and the battery holder.

I don't really want to publish it on GitHub, since it is not really my project, so I'm attaching it here in case anyone finds it useful.

I did not do thorough characterization, I don't have the equipment necessary for that and I just needed low capacitance probe, I don't really care about the absolute maximum bandwidth or extreme linearity.

If you are considering making one, one thing I would modify is add more vias under the pad of the BUF802. It gets quite toasty. I don't think the current design is unworkable, but more heat dissipation would not hurt.

[moffy]

Thanks for the project, I'll tuck it away if in the future I need an active probe, love the use of the lacing cord.

[ataradov]

love the use of the lacing cord.

It provides good strain relief and it is very easy to work with. Even a single unsecured knot holds very well, it just sticks to itself.

[twospoons]

Very nice! I would love to see some performance metrics.

[shabaz]

Nice. I still need to build your USB sniffer! That's been on my list for ages.

[ataradov]

Very nice! I would love to see some performance metrics.

I don't have a VNA and the best scope I have has 100 MHz bandwidth, so they sure would not come from me.

The original article has a lot of performance measurements and the high frequency front end is basically the same layout, so I can't see things change drastically.

[schmitt trigger]

Beautiful implementation!

A question, though: how did you solder the braid to the board without melting the inner polyethylene insulation?

[ataradov]

A question, though: how did you solder the braid to the board without melting the inner polyethylene insulation?

From a second attempt and the soldering is pretty shitty, as you can see. But it is good enough and the waxed thread does the rest of physical holding.

[bson]

Another suggestion: move the supply pins a little further back.  They're pretty close to the braid, while peeling the cable any less looks like it would only make soldering even more difficult (the braid being thick).  Maybe a small cutout to seat the cable in?

[ataradov]

The central pin that is closest to the braid is the ground, so it is the same as the braid. And I don't see a way to even intentionally connect +/- pins to the braid. You would need to try really hard.

[tszaboo]

(R1 should be larger, because 2.5V*2.5V / 50 Ohm = 125mW > 63mW)
Nice design, I like the cable attachment.
What was the consideration to not use a connector for RF?

[ataradov]

Yes, it is undersized for a full swing voltage. I doubt it will ever see that.in my case or in most practical cases in general.

The size mostly. I really wanted something that does not get in a way. Probing with rigid ground connection is already not fun. And I was willing to sacrifice potential high frequency performance.

Same for the spring loaded pogo pins. They are likely not optimal for frequencies over 250 MHz. But I pick convenience over absolute performance in this case.

The pogo pins could possibly be replaced with a fixed tip and a flexible ground leaf spring contact. It would probably be just as easy to use. But I don't have a good source for a spring contact like that.

[tszaboo]

UFL connectors, also called IPEX are super tiny. But for hand assembled boards, I guess it doesn't make a huge difference if you are hand assembling it anyway.
Or my current favorite, which is the board edge MCX connector. You can rotate it, since it's snap locked, though some RF expert will tell you that's a bad idea.

[ataradov]

The cable side of the U.FL connectors supports very thin cable, which is not going to be great for long runs, so it would have to be a short u.FL to SMA and then SMA to BNC for the rest of the run. There is no way to directly attach RG174 used here to u.FL connector.

This was for sure one-off project. It was done for one specific use and the next time I will need it is like going to be in months. If it was a product, I would either pick a more solderable cable or use a regular SMA connector.

[jbb]

Maybe someone could use a low temperature solder for the cable attachment? However, I don’t know how low temperature solders would hold up mechanically; they might be more prone to cracking.

[ataradov]

It soldered fine as is. There is probably a way to solder it better, or may be there is a more temperature tolerant cable. But for one-off, this is not bad.

People have been soldering cables to PCBs for a long time, I bet there is a technique that works better, but I have not tried to research it.

[electron_plumber]

What a coincidence, I just designed a very similar probe using the same chip!

My amplifier design differs from Wilson's in a few ways.
* He has a voltage divider in the HF and LF path... mine has just 1 divider. I'm not really sure why you would separate it into two, as that ostensibly creates an error term.
* I appeared to use a larger trimpot -- I couldn't find the smaller ones anywhere. But my compensation is all done in the lower-frequency composite loop, where the parasitics are less important.
* I filled/capped my vias, and added 5 under the amplifier (compared to what appears to only be 1 on Wilson's probe) so hopefully my amp runs a little cooler
* I added a USB C connector (because I have a million of them) and a buck-boost so I can power off my oscilloscope's USB port, or optionally a 9V battery
* I went with angled pogo pins that achieve ~1.4mm pitch at the DUT -- I figured this would work well for gate-to-source pin pitch on SO8 FETs, and also numerous passives.
* I ordered two different pin lengths that I plan to experiment with, which is why there appear to be 3 pins in the CAD.
* I also added a through-hole for a little mini-grabber ground lead in case I want to use that instead of two pins

I _really_ wanted to embed the pogo pins inside a slot in the PCB... but JLCPCB couldn't manufacture a thin enough slot for me so I ended up adding pads to solder them, like you did. 

Mine should arrive from JLC in a few days! I have an SVA1032X and a Bode 100, so I should be able to characterize the response and will let you all know how it looks!

[jmw]

Oh hi!

* All my compensation is done in the lower-frequency composite loop, where the parasitics are less important. So, I'm hoping that allows me to get away with the two large'ish trim pots.

This piqued my interest - can you share what your input topology looks like? I've spent a lot of time tweaking the input network to squeeze the last bit of flatness out of it, and it's a tricky problem, especially in the crossover region.

I've gotten more comfortable with USB-C and might try a variant that uses that instead of a barrel jack, but it seems uncertain a PD trigger + buck-boost can net out cheaper than that (rather expensive but nice) charge pump from AD.

Edit: are those 0201 parts at the input?

[jbb]

Off topic, but I do wonder how feasible it would be to make a differential probe with two BUF802 and a high performance difference amplifier…

Edit: of course, I don’t actually need such a probe so this is idle curiosity.

[KE5FX]

Off topic, but I do wonder how feasible it would be to make a differential probe with two BUF802 and a high performance difference amplifier…

Edit: of course, I don’t actually need such a probe so this is idle curiosity.

LMH3401 and similar parts in the series are likely better for that, using external series resistors rather than more elaborate input buffers (see nctnico's probe for example).  CMRR gets harder to maintain as the frequency goes up.

[electron_plumber]

can you share what your input topology looks like? I've spent a lot of time tweaking the input network to squeeze the last bit of flatness out of it, and it's a tricky problem, especially in the crossover region.

Sure. My thinking was that potentiometers VR1 and VR2 allow me to compensate for manufacturing tolerances of the resistor networks and, indirectly, the HF caps C13 and C14. I was initially concerned about parasitics of the potentiometers. But by isolating VR1 from the HF path by R27, I think I've convinced myself this won't be an issue.... we'll find out soon.

it seems uncertain a PD trigger + buck-boost can net out cheaper than that (rather expensive but nice) charge pump from AD.

Getting power from a USB host simply requires the connector and two strapping resistors. The buck boost probably isn't necessary for the USB case. It only comes in handy if I decide to use the battery, which I have diode-ored with Vusb. To your point, I could probably optimize cost a bit more than I did. But I'll save the cost optimizations for work projects .

are those 0201 parts at the input?

Yes. I split up the input resistors in attempt to reduce capacitance and flux leakage effects. We'll see how well that works.

I just noticed your design is 2GHz. Nice! I aimed for slightly higher than 1GHz. I would need a nicer scope to make use of 2GHz probes .

[jmw]

Drop me a note when you get a chance to measure it, especially with the Bode 100 since it can see the crossover region really well.

As for why two voltage dividers in the design: a frequency-compensated voltage divider has a flat response and a single-pole input impedance. The parallel combination of multiple compensated dividers also has flat responses and a single-pole input impedance. So it's very easy to characterize and compensate. The single voltage divider, when you add the input capacitance of the OPA140, can't be compensated perfectly flat. But does it actually matter, especially with the added isolation of R14 and R23? Idk ... so I'm interested to see how it measures.

[electron_plumber]

As for why two voltage dividers in the design: a frequency-compensated voltage divider has a flat response and a single-pole input impedance. The parallel combination of multiple compensated dividers also has flat responses and a single-pole input impedance. So it's very easy to characterize and compensate. The single voltage divider, when you add the input capacitance of the OPA140, can't be compensated perfectly flat. But does it actually matter, especially with the added isolation of R14 and R23? Idk ... so I'm interested to see how it measures.

When you adjust the potentiometer in your feedback path, you're only adjusting your low-frequency gain?

I'd be happy to compare notes once I have data.

I was just reading through your blog and found the June 2024 update interesting. You mention that splitting the resistors up to reduce parasitics might improve your response. I came to the same conclusion during the simulation phase (I had just finished a wideband transimpedance amplifier design, where feedback parasitics are a common pain point, so my brain was "primed" to think this way). Did you get around to respinning the board with the large resistors split into smaller series elements?

A brief aside on resistor parasitics:

• There's a widely circulated Vishay paper on thin film resistor parasitics, but IIRC, Vishay made an attempt to eliminate the effects of stackup and ground plane on the measurement and focus solely on the component (which makes sense, given that is what they're selling).
• Horrowitz/Hill casually mention resistor parasitics in their chapter on transimpedance amplifiers, but don't provide much data.
• I tried modelling in hyperlynx, but it seems to treat ground planes as solid, regardless of whether there is a cutout. So, that was no help.
• I looked into sonnet software thinking perhaps I'd try a full EM model, and secured a trial license, but haven't gotten started.

I ultimately decided it might be easier to just measure. So, in addition to the probe design, I quickly fashioned the small test board below for both 0402's and 0201's. I didn't spend a lot of time planning the measurement, so perhaps my tools won't be sufficient, but I am hoping to use the "2x through" de-embedding procedure described here to tease out the package parasitics. (The pads are for SMA end-launch connectors).

[jmw]

When you adjust the potentiometer in your feedback path, you're only adjusting your low-frequency gain?

Yes, the HF path has fixed compensation: "In the probe board, using two capacitors in series allows for easier tuning of the output level. I found a combination of 1.1 pF and 1.3 pF delivered almost exactly −20 dB at 100 MHz." I tuned the HF path as close to −20 dB as I could, and use the pot to match the LF level for flatness.

I was just reading through your blog and found the June 2024 update interesting. You mention that splitting the resistors up to reduce parasitics might improve your response. I came to the same conclusion during the simulation phase (I had just finished a wideband transimpedance amplifier design, where feedback parasitics are a common pain point, so my brain was "primed" to think this way). Did you get around to respinning the board with the large resistors split into smaller series elements?

I tried splitting them up and also using 0201s between the 0402 pads with basically no effect. One thing that was effective is reducing the input bias resistor. 10 M is huge and will be disrupted by even miniscule amounts of parasitic intrinsic capacitance. 3300 pF + 1 M has the same pole frequency as 330 pF + 10 M. I chatted with the designer of the Thunderscope that uses the BUF802 in the analog front end, and he is keeping all the resistors below 1 M for this reason. For the resistor(s) spanning the compensation capacitor (the 1.6 M in my design, and R12, R13, R14 and R23 in your schematic), the parasitic capacitance actually becomes part of your compensation circuit, so you may be able to use slightly reduced capacitance (in my case 1.1 p + 1.3 p instead of 1.2 + 1.2 p).

Unfortunately, parasitic extraction seems to be a weak suit in all the EM software I have access to (openEMS and Matlab RF PCB toolkit), so measurement is probably the way to go. Happy to compare notes with you after you've got your boards back.

[electron_plumber]

Probes have arrived. Unfortunately, the VNA I ordered from Siglent was dead on arrival -- so I'll have to wait on bandwidth measurements.

I checked cross-over on the bode 100, though. Attached is about as flat as I can manage with my double-potentiometer tuning method. Not as nice as I'd hope, but not too bad either. Hopefully I can flatten that out a bit with some tinkering. Input capacitance is reading just under 2pF at 50MHz, which doesn't seem bad but is a little higher than simulation predicts.

Since ataradov mentioned thermals, I also attached footprint and case temperature while sitting idle for a few minutes. Might get warmer during use but seems reasonable to me.