DC coupled 2.7 GHz Active Probe Project - Now Available!

[tszaboo]

I've seen that probe from Andrew Zonenberg before. He has some really nice other projects also. I think the main issue with that probe is the tip capacitance is actually quite high for a resistive probe, and the bandwidth stops at 2GHz. Not sure why it was necessary to use those Vishay FC0402 resistors. Under 2GHz I don't think a normal 0402 resistor would have been an issue, and they are miles cheaper.

You made me do this. 
But yeah, you piqued my interest, I designed this small board to be able to measure passives with the VNA. I was always disappointed by the inductors in RF matching, and wanted a way to measure them, this gave me the extra push to make this.
My issue now, how do I calibrate the VNA to 50 Ohm on a resistor, and then measure the DUT which is a resistor.

[joeqsmith]

Why have components mounted on the thru, open and short?  For the load, why wouldn't you use a standard?  Are you concerned the length of the load will made a big difference?

[tszaboo]

Why have components mounted on the thru, open and short?  For the load, why wouldn't you use a standard?  Are you concerned the length of the load will made a big difference?

I don't intend to mount those components, but I placed the ~3mm long transmission line and the component footprint anyway.
I guess I could use an FC0402 50Ohm as the load. Normally the length shouldn't make a difference, but from what I've seen, the closer you can make the calibration look like the real world condition, the better.
Plus, I only have a female cal kit for for SMA (at work).
But you know what, it's an interesting question as well. I'll calibrate it without the components on it and mount some 0402s and see if it makes a difference.

[nctnico]

Why have components mounted on the thru, open and short?  For the load, why wouldn't you use a standard?  Are you concerned the length of the load will made a big difference?

I don't intend to mount those components, but I placed the ~3mm long transmission line and the component footprint anyway.
I guess I could use an FC0402 50Ohm as the load. Normally the length shouldn't make a difference, but from what I've seen, the closer you can make the calibration look like the real world condition, the better.
Plus, I only have a female cal kit for for SMA (at work).
But you know what, it's an interesting question as well. I'll calibrate it without the components on it and mount some 0402s and see if it makes a difference.

With an OSL you will be calibrating at the connector. So whatever is different (like the trace not having a perfect 50 Ohm impedance and FR4 loss factor) for the real measurement will show up. Then again, I'm missing the DUT position for a reflection based measurement so how useful are the OSL parts on your board anyway?

But it does make sense to calibrate the board from an OSL calibration in order to determine whether the error is as expected c.q. up to what frequency your board is useable. But the through measurement will be the most useful anyway because the error is low over a much wider frequency range.

[joeqsmith]

Board would be taken out and reference plane would be at the end of the transmission line.   I assumed they were making multiple boards and populated the test parts on the second one.   

[JohnG]

FC resistors may be overkill, but they are nice and accurate. If you get them, you should look at the datasheet and consider the difference between the "wrap-around" termination and the "flip-chip" termination, shown in the figure labeled "Internal Impedance." (Good god, I hate this graphic designer trend of eliminating section and figure numbers. I curse them...). It shows a few % increase in impedance at 2 GHz, and a lot more at 6 GHz, for the "wrap-around" versus the "flip-chip".

https://www.vishay.com/docs/60162/fchpseries.pdf

If you use a more cost-effective thin film resistor, you can get the same benefit by mounting the resistor upside-down, i.e. with the resistive element adjacent to the PCB. It can save a few hundred pH. If you are measuring your boards and feel so inclined, you can measure this difference pretty easily at a few GHz.

Hope you find this useful.

John

[tszaboo]

Why have components mounted on the thru, open and short?  For the load, why wouldn't you use a standard?  Are you concerned the length of the load will made a big difference?

I don't intend to mount those components, but I placed the ~3mm long transmission line and the component footprint anyway.
I guess I could use an FC0402 50Ohm as the load. Normally the length shouldn't make a difference, but from what I've seen, the closer you can make the calibration look like the real world condition, the better.
Plus, I only have a female cal kit for for SMA (at work).
But you know what, it's an interesting question as well. I'll calibrate it without the components on it and mount some 0402s and see if it makes a difference.

With an OSL you will be calibrating at the connector. So whatever is different (like the trace not having a perfect 50 Ohm impedance and FR4 loss factor) for the real measurement will show up. Then again, I'm missing the DUT position for a reflection based measurement so how useful are the OSL parts on your board anyway?

But it does make sense to calibrate the board from an OSL calibration in order to determine whether the error is as expected c.q. up to what frequency your board is useable. But the through measurement will be the most useful anyway because the error is low over a much wider frequency range.

I'm planning OSLT, the measurement is not a reflection based it's through.
https://coppermountaintech.com/measurement-of-electronic-component-impedance-using-a-vector-network-analyzer/

According to this article, measuring 100 Ohm and above, I should use the through method, and around 50 Ohm, I can use the reflection
based on a S11 measurement.

Board would be taken out and reference plane would be at the end of the transmission line.   I assumed they were making multiple boards and populated the test parts on the second one.   

Exactly.
I'm going to open another thread for this, so we don't pollute lasmux's excellent design with off topic stuff.
https://www.eevblog.com/forum/rf-microwave/measuring-passive-parts-with-vna/new/#new

[RFKev]

S11 & S21 for the blade probe using the PNA.  My homemade PCB waveguides are going to be useless.    Starts out at 500 ohms as and drops off rather quickly reaching 100 ohms at 6 GHz.   Mag isn't very stable  (when holding all the parts in my hands, moving them around...).  Then there is de-embedding the interconnect.   Consider it all just a gross measurement to give us some idea how it behaves.

What was the measurement setup and METAS VNA Data Explorer configuration to get the probe impedance with frequency?

[joeqsmith]

S11 & S21 for the blade probe using the PNA.  My homemade PCB waveguides are going to be useless.    Starts out at 500 ohms as and drops off rather quickly reaching 100 ohms at 6 GHz.   Mag isn't very stable  (when holding all the parts in my hands, moving them around...).  Then there is de-embedding the interconnect.   Consider it all just a gross measurement to give us some idea how it behaves.

What was the measurement setup and METAS VNA Data Explorer configuration to get the probe impedance with frequency?

I showed the screen shot of the METAS setup and used S11 to measure the impedance.  Guessing that is not what you are asking.  Maybe consider posting a more detailed question. 

[RFKev]

S11 & S21 for the blade probe using the PNA.  My homemade PCB waveguides are going to be useless.    Starts out at 500 ohms as and drops off rather quickly reaching 100 ohms at 6 GHz.   Mag isn't very stable  (when holding all the parts in my hands, moving them around...).  Then there is de-embedding the interconnect.   Consider it all just a gross measurement to give us some idea how it behaves.

What was the measurement setup and METAS VNA Data Explorer configuration to get the probe impedance with frequency?

I showed the screen shot of the METAS setup and used S11 to measure the impedance.  Guessing that is not what you are asking.  Maybe consider posting a more detailed question.

I can see the settings along the top of the METAS screenshot. Was there anything configured in the "Setup" dropdown?  My main question is how was the probe interfacing with the VNA.  Did you touch the probe on the port 1 connector, or probe a transmission line?  If you probed a line connected to port 1, was it terminated with a 50ohm load?

[joeqsmith]

S11 & S21 for the blade probe using the PNA.  My homemade PCB waveguides are going to be useless.    Starts out at 500 ohms as and drops off rather quickly reaching 100 ohms at 6 GHz.   Mag isn't very stable  (when holding all the parts in my hands, moving them around...).  Then there is de-embedding the interconnect.   Consider it all just a gross measurement to give us some idea how it behaves.

What was the measurement setup and METAS VNA Data Explorer configuration to get the probe impedance with frequency?

I showed the screen shot of the METAS setup and used S11 to measure the impedance.  Guessing that is not what you are asking.  Maybe consider posting a more detailed question.

I can see the settings along the top of the METAS screenshot. Was there anything configured in the "Setup" dropdown?  My main question is how was the probe interfacing with the VNA.  Did you touch the probe on the port 1 connector, or probe a transmission line?  If you probed a line connected to port 1, was it terminated with a 50ohm load?

No, just S11. Everything is defaults except the top.   My comment about de-embedding the interconnect was the main point.  Ideally, I would have a custom jig for this.   You can't... well, er... you could probe an RF connector directly but nah....   I have a connector that matches with my home made standards and touched off on it.  This gets up roughly the correct reference plane.  Still, I expect some error.  There is nothing attached to the test connector besides the probe (which is what I am wanting to measure).   Think of it this way.  If the test connector were terminated to 50, then I  place the probe in parallel with that, we are not going to measure 500 ohms but rather 500 in parallel with 50, product over sum....

[schratterulrich]

Your probe project is very impressive. It certainly took a lot of effort to achieve this high bandwidth.
I came across this post because I probably took a similar approach to building an active probe.
My intention was to get a probe with very high input impedance and low capacitance to test sensitive nets like oscillator signals and to get an high impedance amplifier for 50 Ohm instruments.
I chose the ADA4817 operational amplifier and got the following characteristics for the active probe. (With direct SMA connection)
500 MHz bandwith
660 ps Rise Time
1,3 pF input capacitance
95 MOhm Input impedance

[lasmux]

Your probe project is very impressive. It certainly took a lot of effort to achieve this high bandwidth.
I came across this post because I probably took a similar approach to building an active probe.
My intention was to get a probe with very high input impedance and low capacitance to test sensitive nets like oscillator signals and to get an high impedance amplifier for 50 Ohm instruments.
I chose the ADA4817 operational amplifier and got the following characteristics for the active probe. (With direct SMA connection)
500 MHz bandwith
660 ps Rise Time
1,3 pF input capacitance
95 MOhm Input impedance

Hey, cool probe there Nice linear response out to 300MHz or so!

Thanks for the feedback. It definitely did take a lot of effort, and even more learning. I am currently waiting for the 9th revision PCB to arrive (!!!). Fortunately active probe PCBs are small and relatively cheap.

[hpw]

Fortunately active probe PCBs are small and relatively cheap.

And please do not forget any 2GHz differential mode(s) 

[LooseJunkHater]

What is the estimated price for the probe?

[lasmux]

What is the estimated price for the probe?

It'll be around £160/$200, excl shipping. I'm thinking of just sticking with a 2GHz version, and not bothering with an artifically worse/cheaper 1GHz model. I'm building a (hopefully) final iteration at the moment which will likely improve most of the specifications, then go into production.

[LooseJunkHater]

What is the estimated price for the probe?

It'll be around £160/$200, excl shipping. I'm thinking of just sticking with a 2GHz version, and not bothering with an artifically worse/cheaper 1GHz model. I'm building a (hopefully) final iteration at the moment which will likely improve most of the specifications, then go into production.

I'm not personally interested in the single-ended probe, but I'd absolutely love if you came out with a variable attenuation dual-ended probe. $200USD is a pretty great price! I'd imagine it would be difficult however for it to reach 1GHz. I wonder if you'd be able to make minor changes to your circuit to allow for it.

[lasmux]

After a bit of radio silence, I have a large update with big improvements to the probe:

I have worked through another two iterations of the probe as there were various specifications that I thought could be improved upon. We're at V10 now.

I also was in contact with Picotech, who generously offered to lend me a fast oscilloscope (Picoscope 9404-16, 16GHz sampling scope) for proper time-domain measurements of the probe. This has been a huge pleasure to use, and I am very grateful to them for lending me such a high specification oscilloscope. I unfortunately have to ship the scope back this week. So yes, big thanks to them! Buy their oscilloscopes!

I now no longer use the optional resistive ground lead to reduce the DUT loading, instead the signal pin is now always resistive. As such there's no longer a trade-off between probe bandwidth and reduced DUT loading. You get both high bandwidth and reduced loading

So the main new specifications are:
Bandwidth: 2.7GHz
Rise time: 130ps
DUT loading: 120Ohms worst case (at 1.6GHz).
Tip capacitance: 0.7pF
Return loss on a 50 ohm transmission line -15.4db max
Noise: 350uV RMS output, or 7mV RMS referred to input (20x probe)

Now for the performance plots.


Showing -3db at just over 2.8GHz. Linearity is still very good.


Tip loading is now minimum 120Ohms, and stable out to 4GHz at 200 ohms. My VNA is unstable beyond this.


Step response test from signal generator with 350ps rise time. Probe response is very close to the signal generator, and probe loading has not modified the signal at all. Note red and green lines are almost perfectly overlapping. Note that red and green traces are saved waveforms so they can be overlaid with the probe signal. The probe has a 5.4ns latency.


Rise time measurement using fast edge (70ps) from fast signal generator. This edge has a bit of ringing, so I'm uncertain how much of the probe ringing is because of the source signal ringing vs ringing because the source rise time is so much faster than the probe rise time. Probe rise time around 130ps which corresponds to a 2.7GHz bandwidth. Note probe has only a minor loading on this very fast signal. The rise time is slowed by around 6ps and there is some reduction to the peaking.


Nice flat-topped low-frequency square waves (1khz).


Test setup for oscilloscope measurements. Signal generator routed into oscilloscope (channel 4) via intermediate coplanar waveguide board. Active probe measures signal on coplanar waveguide and is read by oscilloscope channel 2.

I'm pretty sure I've said this before, but hopefully for real this time. I am looking at starting production of these. If anyone wants one, send me a PM to get yourself to the front of the queue
The cost will be between £180 and £200, depending on how much of my time I end up spending on each probe. Hopefully more on the lower end of that. The probe tips are hand-made, and there may need to be some tuning on each probe if the frequency response isn't as good as it should be when I test them.

[tggzzz]

Looks good Kudos to you.

Kudos also to PicoTech. Their attitude is like Bill and Dave, and that was instrumental (ho ho) in building HP into the great company it was.

If I was younger, it sounds like I would want to work for PicoTech

[lasmux]

Looks good Kudos to you.

Kudos also to PicoTech. Their attitude is like Bill and Dave, and that was instrumental (ho ho) in building HP into the great company it was.

If I was younger, it sounds like I would want to work for PicoTech

Yes exactly. They do seem like an excellent company in all respects! Really cool vibes from them.
I do also have a picoscope 2206B (50MHz bandwidth) which I use for all my embedded and lower speed development. It and its software are seriously good.

[2N3055]

I agree with Tggzzz, nice work on the probe...

I also have great respect for Picotech...
And 3 Picoscopes (actually 4, with retired one I don't use anymore)

[lasmux]

The noise calculation was a bit tricky as the scope noise was much worse than the probe noise level. For subtracting noise, I found the following equation :
𝑃𝑟𝑜𝑏𝑒_𝑅𝑀𝑆= √(𝑆𝑐𝑜𝑝𝑒𝐴𝑛𝑑𝑃𝑟𝑜𝑏𝑒_𝑅𝑀𝑆² − 𝑆𝑐𝑜𝑝𝑒_𝑅𝑀𝑆² )
So I'd measure the AC RMS noise of the scope without the probe plugged in, and then measure the AC RMS noise with the probe plugged in and tips grounded, and then perform the above calculation.

The noise level on the oscilloscope with nothing plugged in was 1.986mV RMS, with the probe plugged in and the tips grounded, this rose to 2.0258mV RMS, meaning the probe had a noise output of about 0.35mV RMS.

I think I had the scope set up right... I used the oscilloscope channel with the lowest baseline noise, which was channel 4. The scope was configured to operate with random equivalent time sampling, free-running mode, 5us per division, and 250000 points - for an effective sample rate of 5GS/s. When I selected shorter or longer time bases, it didn't make much difference to the final calculated noise level.

Probe noise without probe attached:


Probe noise with probe attached:

[jmw]

Nice work!

I've been slowly working on a probe based on the BUF802, with some similar specs: 10x attenuation, input impedance of 1 MΩ || 1 pF, Zmin ≅ 100 Ω. I've been simulating and testing one piece at at a time, and put together the first complete prototype this weekend.




It's hitting about 2.3 GHz, and there's no big resonant peak in the frequency response, so I'm probably even leaving some bandwidth on the table. I used a test jig PCB that fits against the front of the probe, but had to "cheat" a little by making another ground connection to lower the ground lead inductance. When held by hand as pictured, the bandwidth is closer to 1.6 GHz.

I'm trying to think of practical ways of keeping the right ergonomics and pushing inductance lower. Unlike the input capacitance that is more or less fixed by the probe design, ground inductance is mostly determined by the user's probing setup. A handheld probe needs to have a pointy shape to get into tight spaces, so there's always going to be a little bit of a loop. Most commercial probes have similar input impedance, so they must have some way to get super low inductance to hit their bandwidth number. And even Keysight mentioned their probe bandwidth numbers are measured using jigs that are not representative of most real world use. 

I suppose more attenuation makes it easier to get lower input capacitance; is this the reason you went with 26 dB attenuation?

[lasmux]

Hey, nice probe! Nice flat response

I've found that dealing with stray inductances and capacitances is quite difficult, and quite design/layout specific. Some designs were more sensitive to it than others. It's madeningly easy to set up some LC resonance somewhere which causes a big dip or peak in the response at some frequency. Maybe you have something similar at 1.6GHz and 1.9GHz? Those dips are just single data points however so it could just be noise?
Dealing with ground lead inductance is difficult as a lot of things you can try either make the probe more difficult to use, or damage the response linearity/bandwidth. I stuck with using a spring pin probe for the ground lead because it makes probing real circuits much much easier if one of the connections is sprung, even though this ground lead has quite a bit of inductance. You don't have to get the probe angle perfect to get a good connection.


I chose a 20x attenuation for a number of reasons. As you say, it makes it easier to reduce the input capacitance, also, the slew rate on the op amp I'm using is a bit slower so by increasing the attenuation it increases the bandwidth at which the probe becomes slew-rate limited for very large AC-amplitude input signals.

[joeqsmith]

Any progress with the latest revision?