Anyone make PCBs with Othermill?

If 10pf resonates with the probe tip at 350MHz, the inductance is 20.67nH.

Fr = 1 / (2 * pi * sqrt (10e-12 * 20.67e-9)) = 350,065,968.699 Hz

You are not going to get that with a straight piece of wire 7mm long.

Maybe you are using a ground lead that is 10" long?

I'd be very surprised if that circuit went much above 100 MHz.

You can buld narrowband amplifers by hand for frequencies above 10 GHz. But you are going to have a hard time hand-wiring a broadband amplifier for frequencies much above 100 MHz and expect it to be flat.

Buy two Minicircuits 8 GHz ERA-1 for $1.37 (Qty 30) and stick them in series on a good ground plane.

Put a 450 ohm microwave resistor on the end of a 50 ohm coax and connect to the first ERA. Verify it is terminated in 50 ohms. Use capacitive coupling between stages, LF cutoff frequency of your choice. Trim the gain at the outout to 0 dB and you are done.

DC coupling is more difficult.

Why not use an opamp? ADA4817 maybe.

Bootstrapping is hard to do at gigahertz frequencies, because the strays bite you. There isn't really such a thing as "high impedance" above a few hundred megahertz, except in very special cases. Your average chip resistor has a parasitic capacitance of 0.08 pF, give or take, so a 1k resistor sitting on its own has a corner frequency of about 2 GHz.

The ATF38143's input capacitance is below 1 pF, so with a nanohenry of parasitic inductance from the leads and (very small) probe pins, that comes in around 7 GHz, with a Q of about 1 at 50 ohms.

It's meant to work with DC-coupled, 50-ohm scopes such as the TDS 694C. You can roast the input resistors if you put too much voltage on them, so for a general-use probe, an attenuator makes sense. (Scopes are pretty noisy in any case.)

Cheers

Phil Hobbs

Is that for the high frequency side?

Only CAD $5.79 at Newark. I didn't realize they were so inexpensive.

But I'm seriously thinking of a split channel. Some low frequency op amp for the DC, and perhaps a nice Hittite (ADI) to get to 30 GHz or so.

Please be assured I pay careful attention whenever you and Phil talk about splitting the channel into high and low frequency and the difficulties in recombining them. An old HP journal article also talks about this technique. I think the rolloff was at around 1 MHz, but I forgot the instrument it was used in.

I already looked at the noise figure for the ERA-1+: 5 dB. That's not bad for a cheap 8 GHz version to get started. Maybe it's time to stick it in LTspice and see how it would work.

Overload recovery is going to be interesting.

The bootstrap is not necessarily to get a high impedance from the point of the circuit. It will be in a 50 ohm environment anyway. The problem is the current through the ground wire. This creates a voltage that lifts the probe off ground and puts stray currents on the outside of the shield.

With the 34143, I don't anticipate having any components between the gate and the test point. The source is right there, so the connection to the guard is very short. The probe diameter is as small as possible so it can get into tight spaces. This means the actual probe tip can be very short and still hit the connection when it is next to a thick ic or some large component such as a heat sink.

I assume the 34143 is about the same. 7 GHz is all I'm trying to reach with the first version. I looked at Avago's site to see if I could find if they sell bare dice but had no luck.

Aside from microwave radar and avionics transponders, I don't see much need for high voltage on circuits running at GHz frequencies. Mostly low level oscillators and some fast logic.

That's the noise I am talking about when the signal is attenuated. That's why the 34143 is so attractive.

I was also hoping to use the probe with my HP 8566 and also start work on a microwave VNA. It would be nice to be able to look at a filter and see if it was really doing what it should.

The fast Hittites are 50 ohm distributed amps and insane power hogs, and a real nuisance to bias and use. Probably not what you'd want in a scope probe.

Splitting shouldn't be a big deal. A scope probe isn't expected to have 0.1% step response flatness. Somebody, LeCroy I think, makes a scope with multiple front-end mixers, multiple IF amps and ADCs, and a recombiner.

The ERAs make good pulse amps. They tend to not be very close to 50 ohm input impedance, usually less.

Good luck getting a Spice model! MiniCircuits occasionally changes the recipe (or buys them from someone else) now and then anyhow, and everything changes.

Why not use a passive (resistor) probe and put an amp downstream in a box if your scope needs it? A little sigal averaging can get the s/n ratio back.

The Caddock resistors, at 450 or 950 ohms, make good clean passive probes.

We sometimes build passive pickoffs and SMB connectors into our PC boards to let us probe things that would be difficult otherwise.

Only the resistor goes in the probe. The amplifier is in a separate box.

Maybe a pain, but it's the only way I know to get to 30 GHz or beyond.

Yes, a nightmare to align I'll bet.

That's exactly what I am talking about. But the probe is also for other things like a spectrum analyzer, VNA, and anything that needs wide bandwidth, low noise and flat response.

I worry about the length becoming part of a wavelength at the high frequency end. I don't think the response would be very flat.

What about a microwave trimmed surface mount? Fig. 6 for a MCT 0603 shows a 220 ohm is flat to 20 GHz in

Yes, that is the obvious solution for a released pcb. But when you are in development you need more flexibility. Or even if you get a wierd fault on a bad board that is not covered by your standard pickoffs.

Good to 6 or 8 GHz.

Maybe use a regular surfmount resistor or two in the probe and equalize out any parasitics in the amp.

How do you know? What do you have to measure it?

I'd go for a Vishay that is specifically designed and trimmed for this application. Then measure it to make sure.

Length problem again. Equalization would be tough to work with at high frequency. You have a hard enough time trying to keep things flat.

HP N2803A, 6:1, +/- 0.8V peak. Page 7 of

A Tek 11801 TDR/Sampling scope, around 30 ps net pulser+scope rise time. And Agilent specs the probe for 6 GHz.

If you push it into a scope and export the waveform to a PC, you can do the eq in software. It's the "deconvolution problem." That's a separate discussion.

Buy a sports car instead.

Sounds kind of a loose way to define the frequency response. First, you need a VNA to show the frequency response to look for dips in amplitude or phase bumps caused by missmatch. Next you need the attenuation vs frequency to verify the attenuator is flat. This is the same measurement as Fig. 6 in the Vishay document

Finally, you need a proper TDR to define the risetime and check for aberrations.

I don't see the value in paying for an undocumented component that is not intended for this application. The Vishay has around three times the bandwidth and is documented and intended for this work. If you run into problems, you can call them for support. If you called Caddock they would not have the faintest clue what you were talking about.

Then do it your way.

No problem. That's what I was planning on doing.

Then I will know what I have to work with, instead of guessing.

In a linear system, the step response includes all the information there is. No guessing required.

Your method is the shoddiest I have ever witnessed. You are guessing when you compare the response to another probe that someone said has such and such a bandwidth. You did not verify the claim, and you do not have the equipment to do so. You are guessing all the way.

The step response may have a great deal of information. But that assumes the system is linear, the stimulus is perfect and the response is captured perfectly.

None are true in your setup. You cannot extract the bandwidth, group delay, reflections, phase and amplitude ripple, frequency rolloff, attenuation, or a dozen other things you need to know if you are to use the probe in any serious work. You do not have the capability to do so.

Your claim is highly misleading. You are using a theoretical argument that has no relation to what you are doing. That is highly disingenuous and has no place in professional organization.

Instead of acknowledging there is a better source for a microwave resistor for a resistive probe, you will ignore it and continue pushing the Caddock as your solution. Just because that's the way you are.

I don't care. I am beyond wasting my time. Those who need to know anything about what you say can easily find the solutions elsewhere.

And I highly recommend they do so.

Yo, dude, dial it back a bit. You may be an RF god yourself, or at least so mebody who knows what one looks like, but most of the FET probes in the uni verse are used with scopes.

A scope lives and dies by its step response, so there's nothing shoddy what soever abot centering on that. In fact, a good step response with low edge artifacts and near-zero tilt is very hard to get--it requires flatness and phase linearity over multiple decades of frequency.

Saying that a perfectly factual statement is "disingenuous" and not "profes sional" and trying to smear a fine outfit that you obviously know little ab out--well, that suggests that you'd lost your temper, at very least, or wer e maybe drunk. So do dial it back.

Cheers

Phil Hobbs

I assume you have spent your entire life in an especially severe monastery, that uses no Microsoft products.

You are guessing when

The step response of a probe isn't a guess. TDR can also be used to characterize a probe's input impedance. My personal limitation is that my TDR gear has roughly 28 ps net risetime, so my bandwidth limit is around 12 GHz.

What's the amplitude of the test signal from your VNA? How will you use the VNA to characterize probe linearity?

The derivative of the step response is the impulse response. That can be FFTd to complex frequency response. Any reflections will be immediately visible in a step response. How does your VNA characterize reflections and locations? The usual way will be to do reverse FFT things to derive the step and impulse responses, which a time-domain measurement already has.

If I have a fast system with a reflective abberation, I can slide my finger or a pencil point along the signal path and zero in on the connector or via or whatever is reflecting, in real time. That's really cool.

A good probe has a clean gaussian step response and no nasty ringing or reflections. Bandwidth is then about 0.35/Tr. Tek published a book on probe design, part of their Circuit Concept Series. Do you have that one?

What claim? Misleading how?

You are using a theoretical argument

I've done GHz probing up to kilovolts, with Caddocks and surface-mount parts. We build passive and active probes into PCBs to snoop fast stuff like differential PciExpress. The stuff that we do is real, not theoretical, and people buy it. We work mostly in the time domain, which is just as valid as working in frequency domain, and better in some respects. Time and frequency domains are just an FFT apart.

Why are you getting so personal and insulting, when we have an interesting topic that we could discuss? I never insulted you, I shared some stuff that we do. If you disgree with what I've said, say why, technically.

The Caddock is a cool way to quickly make a 6 GHz probe, in minutes, that tolerates kilovolts, for a few dollars (or less, because they are good about samples.) Sure, if you want to lay out a board and use surface mount resistors to go faster, do that.

Above about 10 GHz, the skin loss in the cabling will get serious, and will need to be equalized out in hardware or software. A probe with hardline cabling can be awkward; trust me on that one.

The Tek book is a good start there... lots or radical equalization. But I'd go software these days.

Let us know how your 30 GHz probe works out.

The acid test, and a main application, is eye diagrams from differential signals. Some of the diff probes are designed to be soldered into the circuit to be snooped. Good replacement market.