Modeling parasitics in active probe input

[jmw]

After shelving it for a bit, I've resumed my active probe project using the BUF802 chip. I'm running into a lot of problems trying to tune the input to the BUF802 because of parasitic effects. I made a board that terminates the input and feeds it to the BUF802 with a /5 divider. The minimal stub trace after the termination and shared ground plane should be nearly an ideal tip and ground connection with super low inductance. I packed everything as close as possible with 0402 passives. There's a secondary input path for DC accuracy using an opamp that brings the overall input impedance to roughly 1 MΩ || 1 pF. Here's the overview schematic and testing results of the frequency response:




So what I've got so far is the transfer frequency response is the yellow trace. The green trace is moving \(R_D\) between the input termination and divider - even worse. The brown trace is with increasing the value of \(R_D\) in the original position. There's a resonance near 2.5GHz that won't go away even with huge \(R_D\) that tanks the bandwidth. The datasheet for the BUF802 has this to say about S21 peaking: "keeping the trace length minimum is of prime importance to ensure no peaking occurs in the S21 transfer function of the BUF802. The trace inductance can form a resonant circuit with the input capacitance of the BUF802, causing peaking in the S21 response. Add a small resistor in series with the DC blocking capacitor to dampen the LC resonance created by the trace inductance and the input capacitance of the BUF802" ... which I've done my best to follow. With 0402s, the total length from the termination resistor pad to the BUF802 input is about 7mm:



I'm a little stuck:
1) a lot of times I work backwards from lab measurements, try to add parasitics and see if I can get a simulation to look like the bench result. But after adding an L in series with \(R_D\), what else could be the source of the parasitic inductance that's causing the peaking in the frequency response? At 1-3 GHz, this part of the circuit seems electrically small enough not to require transmission line models & EM simulation, but is there something in between LT-SPICE and Sonnet (which I can't afford anyway) I can use to model the physical parts of this layout?
2) what's up with the oscillations (about every 90 MHz) in the frequency response? Is this a reflection (from what, though)?

[moffy]

You have what looks like  a reasonably flat response out to 1.75GHz which is pretty remarkable. The input is quite high impedance, the 1.6M/Trimmer cap/402k, if I was looking for parasitic effects that is where I would look especially the trimmer cap. You could try replacing it with a small, select on test, cap to see how the response varies, just a thought.

[PartialDischarge]

For HF probes (>500MHz or so) input impedances are kept low, usually 100kOhm or less to have fewer problems with the response.

[jonpaul]

For wideband use > 500 MHz...GHZ we use TEK P6136  Zo probes or just DIY at zero cost.

Can make 500 ohm 10x or 5000 Ohm 100x.

See Tek Circuit Concepts oscilloscope probe circuits

https://w140.com/tekwiki/images/6/62/062-1146-00.pdf

Jon

[Marco]

Shouldn't Rd be behind the blocking capacitor?

If you can accept a not precise factor gain, why not go pure capacitive divider for the BUF802 path and use resistive trim on the LF path to get that flat with the HF response. Minimum components in the HF path.

[David Hess]

The ground plane needs to be removed from the high impedance circuits.

[jmw]

Removing ground plane below the input is something I want to try next - this moderately increases inductance but sharply reduces capacitance on the pads and tracks, yes? I wish there was a way to simulate this easier and more mechanically. I just used a trace impedance calculator to manually add elements for every pad and trace and it started to mimic the tested result, but really litters up the schematic with a ton of trash. When I adjusted the numbers for no ground plane on layer 2, it's much flatter.

I tried a capacitive divider earlier and it didn't work well. The compensated resistive divider is easier to analyze and tame. I also tried bodging a fixed capacitor in place of the trimmer (C8) but the change was pretty minimal.

[Marco]

The compensated resistive divider is easier to analyze and tame.

That remains to be seen with some bits of spring steel above a separate ground plane with common mode interference, rather than a near ideal 50 Ohm impedance source.

[jmw]

Removed the ground plane below the input path ... definitely better results! There's a resonance at ~1.7 GHz which could be some residual stray capacitance that could be improved, but this is a good result after so many tries. Frequency response is nice and flat (the jump at 1 MHz is where I switched from oscilloscope to spectrum analyzer).

Next step: remove the termination and input coax coupling and try things in a more pencil-shaped board using pointy tips.

[jwet]

Forgive me if you've already got this but its not on the schematic.  The input to the BUF02 needs some kind of DC path somewhere.  The input Z is 50G, a high number but not infinity.  See Figure 6-25.  Your schematics don't show supply for BUF02 but input bias current depends on CM.  The test circuit shows a 10 meg R going to "in bias".  Without this R,  the input FET's will find their bias based on the leakage current differences between the two input diodes.  Your schematic don't show where these diodes clamps pins go either.  The test circuit shows them reverse biased to Vcc and Vee with 1k R's.  Maybe nothing.

[jmw]

Yeah, I left off some of the plumbing for clarity. They are present, mostly unmodified from the reference designs.

[MathWizard]

Just looking at how various probes are actually built and work. Makes me wonder how people did so much in the past, like working on TV's and all the oscillator and filter circuits in them.

I might try cutting up some TV coax and making a 1k probe, but I'd have to tape it onto the scope.

[T3sl4co1l]

Just looking at how various probes are actually built and work. Makes me wonder how people did so much in the past, like working on TV's and all the oscillator and filter circuits in them.

I might try cutting up some TV coax and making a 1k probe, but I'd have to tape it onto the scope.

Well, one thing is they didn't.   For example, a typical AC voltmeter was a low-capacitance (low microwave capable) thermionic diode installed in a probe body:
https://frank.pocnet.net/sheets/140/2/2-01C.pdf
AC is rectified into the probe's body capacitance (there's also a ground clip from probe body to chassis) and the DC reading goes to ones' traditional VTVM (so, 10MΩ input resistance or thereabouts).  You can't tell the frequency or waveform, but you know frequency from your hopefully calibrated signal generator, or from certain handy references (like a crystal controlled marker generator), and waveform is going to have a hard time being anything but sinusoidal on the usually narrowband RF and IF circuits you're aligning.

Radio receivers can be used as crude spectrum analyzers, if you don't mind that you're hand-plotting tuned frequency against say AGC voltage (which for a lot of radios, will be roughly logarithmic with input voltage).  Scopes were definitely out of, ahem, scope -- even if you could swing for the top of the line, Tektronix only had a full wideband scope of some 85 or 100MHz in 1960 or thereabouts (nearly all tube design).  Bandwidth rose quickly with the introduction of silicon transistors and ICs, and more esoteric methods where applicable (like direct deflection input, with MCP and storage in the tube, to view single-shot events in ~ns time scales).

Tim