1GHz active probe project feedback

[lasmux]

Hey,

I've built a single-ended active oscilloscope probe to analyse one of my other projects (an active quenched single-photon APD which will get its own post at some point), and it's turned out pretty well. I'm considering setting up shop and flogging it on Ebay. It would be really cool to get some feedback and comments! Hopefully I'm not totally barking up the wrong tree.

Specification
Bandwidth (-3dB): DC - 1.15GHz
Probe capacitance: <1pF
Input impedance: 1MOhm
Output impedance: 50Ohm
Attenuation: 20:1 into 50Ohm, 10:1 into 1MOhm
Rise time: 500ps measured on my 500MHz oscilloscope. It's probably faster than this.
Measurable voltage range: +/-15V
Absolute maximum input voltage: +/- 50V



(simplified schematic for now)
This device is powered with a 9V battery, and have also a low voltage indicator LED. The probes are swappable gold plated pins, so you can have either sprung-loaded pins, rigid pins, flexible wires, or even a soldered connection to the DUT.


(data points >100kHz measured on a nanoVNA, low frequency data points added manually by checking amplitude response on an oscilloscope)

100MHz square wave.

(captured on my Agilent 54111D 500MHz oscilloscope, square wave generated by nanoVNA)
Here the white trace is the signal directly measured by the oscilloscope, and the red trace is the probe measuring the same signal

I'm going to look into a nice enclosure next. I'm working to keep the cost down of the unit, but it'll probably end up costing around £140 as I don't think I'll get much volume of sales which just makes everything more expensive. And I have to calibrate every device myself. So the probe is quite expensive, but still vastly cheaper than any other active probes out there. And actually much cheaper than any GHz passive probes also.

Some specific questions:
Would it be preferable to sacrifice a bit of bandwidth to reduce the peaking at 800MHz? It's only 1dB but maybe that's too much? Although it could be fakery caused by the nanoVNA too!!
Would my test results on my nanoVNA be accurate enough for someone to buy this? The device is calibrated but it's obviously not a professional piece of equipment.

Thanks

[macaba]

Active probes are an interest of mine, on my todo list to try, so I wanted to congratulate you on a project well done. 

I follow azonenberg's work in this area, he has posted his adventures on trying to get a well behaved response, so it might be worth taking a look at his twitter/mastodon account for ideas.

[jwet]

I agree with the last poster- nice work.  Not to detract and partially out of jealousy- I offer my 2 cents.

I worked on something similar with an OPA656, basically a slower cousin of the 858.  I was never able to get very far past 150 Mhz, which was actually fine for my needs but I expected to get 500 Mhz.   I don't know how you arrived at your input impedance numbers- this is easy in simulation but much harder in the real world.  Something to try is to look at S11 looking into the input with your VNA.  At 1 Ghz, 1 pf looks like 160 ohms but its very difficult to get anything like this in the real world.  In my design, the input got squirelly on the smith chart over a couple hundred Mhz- things like loops and kinks indicating resonances.  Hi Z RF probes like the HP85024 got fractions of a pF input C which is kind of what you need- this generally takes hybrid type design on exotic substrates which is why they cost.   What you're competing with don't forget is just a 950 ohm resistor on the end of a coax- it has 20 db of attentuation too but probably is a higher impedanace overall in the end.  See what your probe S11 looks like.  Hopefully I'm wrong.  The S21 plot you show looks excellent!

[lasmux]

Active probes are an interest of mine, on my todo list to try, so I wanted to congratulate you on a project well done. 

I follow azonenberg's work in this area, he has posted his adventures on trying to get a well behaved response, so it might be worth taking a look at his twitter/mastodon account for ideas.

Thanks for the encouragement! I'll check out Azonenberg, from a cursery google, he looks like he knows his stuff!

I worked on something similar with an OPA656, basically a slower cousin of the 858.  I was never able to get very far past 150 Mhz, which was actually fine for my needs but I expected to get 500 Mhz.   I don't know how you arrived at your input impedance numbers- this is easy in simulation but much harder in the real world.  Something to try is to look at S11 looking into the input with your VNA.  At 1 Ghz, 1 pf looks like 160 ohms but its very difficult to get anything like this in the real world.

Ah, yes I looked at the OPA656 a while back. A very nice amp.

I agree with the last poster- nice work.

Thanks 
I just ran a couple of S11 measurements and I get around 110 Ohms at 1GHz, so not quite the 160, but not super far off. I'll see if I can improve that with some tweaks. I noticed that the N2795A active probe (1pF also) is also around 150 Ohms at 1GHz according to its datasheet, so I think that's a good target to shoot for.


And the smith chart



To be honest I don't have a lot of experience with interpreting smith charts, maybe should do some more reading... I think I see a 1.8pF load at 1GHz (it's around 1pF at 500MHz), but I don't quite understand the implications of the shape.

[Mechatrommer]

i'm also working myself on such a thing, just at very slow pace, many other projects to take care of... its a pity OPA858 only available from digikey, the shipping cost is suck blood.

[TimFox]

Nice project!
How much difference do you see in frequency response between 50 ohm and 1 megohm termination?

[jwet]

Regarding your S11 plots, its pretty well behaved.  You have a bit of squirellyness on your smith chart as I did.  To interpret the smith chart, the center is 50 ohms real.  Items below that equator are -jx, capacative and those above are +jx, inductive.

The numbers on the lower left kind of tell the story- your input looks like 72 ohms with a 1.8 pF cap, not that much different from 50 ohms really.

If you measure an HP85024 on VNA for S11- (I may do it as I haven't in a while), you'll get something like 50K real R and .3 pF (j1k ohm), it would be way over on the right rim.  The DC impedance may or may not matter to you- it can screw up biasing.

Try the old lab favorite of a 950 ohm resistor tacked on the end of a coax with the other end gong to your terminated scope.  You'll be surpised.

[lasmux]

Nice project!
How much difference do you see in frequency response between 50 ohm and 1 megohm termination?

At lower speeds there's no difference, at higher speeds you start to get a lot of distortion from reflections in the coax cable.

The numbers on the lower left kind of tell the story- your input looks like 72 ohms with a 1.8 pF cap, not that much different from 50 ohms really.

If you measure an HP85024 on VNA for S11- (I may do it as I haven't in a while), you'll get something like 50K real R and .3 pF (j1k ohm), it would be way over on the right rim.

Thanks for the insight in the smith chart.
Yeah, I think that probe is capacitively coupled though, and 3GHz bandwidth, so it's not pulling punches at high speeds! I'll have a fiddle with the circuit and see if I can improve the input impedance.

[Mechatrommer]

reflection can occur as low as 30MHz on normal length coax cable used as probe cable... so inevitably it must be terminated to get such a sane GHz BW claim... hence halved of ±1.5V output swing ie ±0.75V right into the scope. 1/10X means ±7.5V measurement range... ymmv.

[mr ed]

Scope probes use resistive wire to help knock down reflections. Try butchering an old probe instead of using a normal coax.

[TimFox]

reflection can occur as low as 30MHz on normal length coax cable used as probe cable... so inevitably it must be terminated to get such a sane GHz BW claim... hence halved of ±1.5V output swing ie ±0.75V right into the scope. 1/10X means ±7.5V measurement range... ymmv.

Done carefully, sometimes "series termination", with a matching resistance at the source and a high impedance at the destination, suffices for signals on coax.
In that case, the transmission is x1 instead of x1/2.

[lasmux]

I've had a bit of a play with the circuit, and got the impedance to around 150 Ohm at 1GHz, but the bandwidth of the probe dropped to around 750MHz. So I guess I was a bit overoptimistic with the input impedance being 1pF originally. With the changes it would be a 750MHz probe, 1pF input impedance I believe, which I'm still pretty happty about tbh.

I think I've just spent so long thinking about the amplifier response that I kinda neglected really optimising the input characteristics. I'm considering another version of the board but using 0402 passives on the input network, hopefully that might improve things a bit. Fortunately small PCBs are cheap 


Although the smith chart isn't so pretty now...

[jwet]

That does look better.  I would try to shoot for a real R > 1k, and the C as low as possible.  The DC stuff will keep you from affecting DC bias when probing.

Remember what you're trying to do, make something that can as unobtrusively as possible look at an operating circuit.  Hanging a 150 load on something is pretty invasive.

Using 0402's and a very tight layout might help.

I took my circuit a step further into the sort of the insanity zone before I declared it good enough for my temporary need.   I fully "bootstrapped" so it would float up and down with the signal and "eliminate loading". I ran the amp from two current source, one sourcing on top and one sinking on the bottom.  There was a zener across Vcc/Gnd to maintain the total.  This gave me a pseudo floating supply that was stiff across the part but Hi-Z to the rails.  Your 9v battery could be pressed to do the same- maybe better.  I then took the output and fed it back to a bypassed mid rail point divider across the zener.  This kept the input of the amp tracking with the input of the circuit.  Since there is no voltage delta on the front end, the impedance is "infinite".  This worked somewhat  and got the capacitance down lower but I kind of ran out of gas as my need was just for something like 100 MHz.  At these frequencies I had to spin a board for each interation and at the time, this wasn't cheap like today.  I'd have to resurrect the exact circuit from old notebooks- I might have some old Spice runs too.  I worked at Maxim at the time and a co-worker was a famous old engineer name Bob Underwood.  He developed the very fast open loop buffers that National made in the 70's- LH0002, etc. and had been at Maxim about 20 years doing similar but ASIC like HS things- brilliant guy.  I used to bounce ideas like this off him and he came up with the bootstrap idea.  It definitely worked but had some squirellyness somewhere- long ago...

After dinking around with this for a few months, I bought a probe on Ebay- it even worked (somehwat unusual for these type of delicate gadgets).

Have fun and keep posting.

[Mechatrommer]

reflection can occur as low as 30MHz on normal length coax cable used as probe cable... so inevitably it must be terminated to get such a sane GHz BW claim... hence halved of ±1.5V output swing ie ±0.75V right into the scope. 1/10X means ±7.5V measurement range... ymmv.

Done carefully, sometimes "series termination", with a matching resistance at the source and a high impedance at the destination, suffices for signals on coax.
In that case, the transmission is x1 instead of x1/2.

can you give example? so i can punch in the numbers in simulation.

[TimFox]

One example, from old-fashioned NIM electronics, is a current switch (e.g., NPN differential pair from negative supply) feeding two BNC connectors in parallel.
It was common practice to terminate one of the BNCs with a 50 ohm termination plug, and send 50 ohm coax from the other BNC to the load.
If the load was "bridging", say > 5000 ohms, it could be considered high impedance.
With no transmission line, a current pulse of 20 mA would make a pulse at the termination of 1 V.
Launched into an ideal 50 ohm line (source impedance = 50 ohms), the initial voltage across the source end of the line would be cut by 1/2 to 0.5 V;  that pulse would travel down the line (for simplicity, assume the line is much longer than the pulse width) to the far end.
At the load end, the termination R >> Z0 would make a positive reflection, increasing the pulse height at the load to the original 1 V.
(If the load were also 50 ohms, then the voltage would be 0.5 V.)
The positive reflection at the load would then travel back to the source, where (with ideal terminations) it would be absorbed in the source resistance.
With a mismatch at the source, a smaller pulse would reflect back to the load and give an undesired reflected pulse delayed from the desired pulse.

[Mechatrommer]

well that sounds "niche"..

[TimFox]

It's just a practical example.
I apologize for using an example from decades ago.
Engineers use this type of termination all the time to send digital, sometimes analog, signals down transmission lines with less voltage loss.
A switched current into a 50 ohm load is a practical way to get close to 50 ohms output impedance, as used in my -hp- 8013B pulse generator.

If you want a simple example for Spice, just use an ideal voltage source (sine, pulse, whatever) with a 50 ohm resistor in series to a transmission line with different termination resistances.
If you make the electrical length of the line much longer than the pulse width, it's easier to see what is happening.

In Spice simulations, you can play with different mismatches (50 ohms at source, infinity at load) to see how much effect they have.

[Mechatrommer]

In Spice simulations, you can play with different mismatches (50 ohms at source, infinity at load) to see how much effect they have.

did it with some 1.25meter length mini coax that i measured with a bit of resistance and capacitance and tried to simulate... 1kOhm termination clearly not good past 10MHz... you probably right, but i believe not in a straight forward way, ymmv...

[TimFox]

It may be easier to understand the process if you use narrow pulses instead of a frequency sweep.

[lasmux]

Right, I think I've gotten all the improvements I can get out of this board layout, whilst keeping the input resistance at 1MOhm. The bandwidth is at 850MHz, probe capacitance is down to just under 1pF (impedance at 1GHz is 167Ohm). The reduced bandwidth has actually cleaned up the frequency response quite nicely, a lot less peaking. The oscilloscope trace looks a bit nicer too as the lower probe bandwidth means the very fast signal edges don't cause as much ringing. I'm going to do another run of the board anyway as I'm going to clean up a couple of bits that need fixing for when I put it on ebay, so hopefully I can get the bandwidth back up to 1GHz, but either way, I'm pretty happy with where it is. Once it's been on ebay a for a bit, I'll post the full schematic, the various spice simulations, and go through the amplifier design choices.

[lasmux]

And we're back to 1GHz bandwidth, with about 1.05pF input capacitance (150Ohm load at 1GHz).

[riscy00]

I doing the active probe thing to do, can you share your schematic? thanks

[jonpaul]

At 1 GHz much simpler to make passive Zo probe at zero cost...

50 Ohm coax><BNC cable

Place 450 or 4950 Ohm 1/4 or 1/8 W avial R in series, probe with end of R.

Use on 50 Ohm scope.

10X or 100X cant blow up, BW can be 2..4 GHz.

j