High Speed (500MHz+) Current Source, 10mA

[luky315]

I need a fast current source, but not with much current. 10mA would be great, but 2mA are also ok for most of the intended measurements ("build in test equipment" for sensor development). But the most important points are that it should have over 500MHz of bandwidth and a decent (<1%) accuracy. The "load" (or DUT) is ground-referenced and the maximum voltage is <3V, so this should be relatively unproblematic.
Which would be the best topology for this use case? is a (modified / improved / advanced) Howland current pump with fast amplifiers (OPA855 or similar) in principle able to work that fast? Or is another topology better?

[jbb]

What’s the sensor?

I’m hardly an expert, but the thought occurs that up around 500 MHz you’re getting into the realm of controlled impedance transmission lines.

Why not excite your sensor with a 50 Ohm RF source (and use 50 Ohm coaxial cable for connections)? I make out 50 mA into 50 Ohms to be +21dBm, which should be achievable with an RF power amp.

For benchtop testing, a Vector Network Analyser (VNA) might be a good tool for measuring the sensor behaviour. It combines an RF signal source with RF measurement equipment (magnitude and phase). Probably not practical to build one in though!

If necessary, you can add DC biasing for your sensor using a Bias Tee.

[luky315]

It's a nonlinear sensor with a resistive component andin series with a (lousy) voltage source. We are talking about something experimental and  unfortunately I can't disclose the details on the sensor yet.
A simple RF source is not really working, it has to be a "real" current source, naturally with some constraints regarding the maximum vltage on the output (here 3 V)

[Marco]

High side PNP current source with an inductor then a SiGe current source on the bottom?

[PCB.Wiz]

A simple RF source is not really working, it has to be a "real" current source, naturally with some constraints regarding the maximum vltage on the output (here 3 V)

You will never get a "real" current source as the parallel impedance of even 1pF is in the low hundreds of ohms at 500Mhz.
A higher voltage into a (good) resistance, may be as good as any multi-GHz transistor. 

[Whales]

It sounds like you want an inductor.

Set it up with an approximate current source (perhaps even a resistor, depending on your application) and it will do its best to keep the current constant regardless of external load and voltage changes.  Especially for low timespan (high frequency) signals. 

You will want to choose the design and materials of the inductor well for 500MHz+.  Some ferrite mixes will work better than others.  Also make sure the inductor has low parasitics and is shielded (to prevent emission & reception).

[ArdWar]

Maybe start with defining what do you actually mean with "500MHz of bandwidth"?
Do you actually need current source that can modulate the current at 500Mhz?
Or do you "just" need a current source that can supply into 500Mhz variable load with constant current?

[T3sl4co1l]

Sounds like an application sorely in need of a bridge method. Try redesigning it with that in mind, and see what happens.

Tim

[exmadscientist]

If you do end up having to go down this nightmare path, the book "Analogue IC Design: The Current-Mode Approach" (ed. Toumazou, Lidgey, Haigh) has a lot of interesting stuff. (There used to be a PDF floating around but I can't dredge it up now. Insert the usual comments about the good old days here.) It is a slog of a read and most of it is the usual academic... stuff... (I think this is one of those staple-together-chapters "books" academics do) but especially in chapter 4 (and kind of in 3) there are some ideas that are applicable. They're a bit far out, but you're going to need far out, because this is a hard design.

How much output impedance are you thinking will be needed? That is where the cascodes really start to come into play.

[SiliconWizard]

A simple RF source is not really working, it has to be a "real" current source, naturally with some constraints regarding the maximum vltage on the output (here 3 V)

You will never get a "real" current source as the parallel impedance of even 1pF is in the low hundreds of ohms at 500Mhz.
A higher voltage into a (good) resistance, may be as good as any multi-GHz transistor.

Yep. A current source at these frequencies is not going to be a practical solution due to even just parasitic capacitance (and inductance).
A (carefully designed) bridge is probably the way to go as T3sl4co1l suggested.
But we don't have details so we don't know why the OP thinks the sensor can only be excited via a current source.

[jbb]

You get a cascode and you get a cascode and you get a cascode. EVERYONE GETS A CASCODE!

:-)

[Marco]

The really high speed VCSEL/laser drivers seem designed as current sources, with an output impedance as high as possible rather than 50 Ohm matched.

That's what this most closely resembles, unfortunately most the available ICs are wirebond and/or digital.

[Marco]

You get a cascode and you get a cascode and you get a cascode. EVERYONE GETS A CASCODE!

:-)

So something like this?

[mawyatt]

A little trick we utilized a few decades ago with SiGe Bipolar Cascode Stages (IC designs) to extend the BW is placing a small inductor between the Collector of the bottom NPN to the Emitter of the Top NPN of a Cascode. Should work fine at lower frequencies with slower transistors as well!!

Analysis shows this as a "Pi" implementation of a transmission line lumped element equivalent, absorbing the capacitance seen at the bottom transistor Collector and Top transistor Emitter into the lumped element equivalent. This works really well, and gives significant BW extension without additional DC Power consumption

Recall we posted something on this awhile back but can't remember when/where!!

Edit: Found it!!

https://www.eevblog.com/forum/projects/some-free-bw-in-a-cascode-amp/msg3718969/#msg3718969

Best,

[Marco]

I don't think the base emitter capacitance or speed will be relevant, getting a discrete circuit stable at all will be the problem.

This is with BFP420 because MPLAB Mindi had that by default, NXP and Infineon have much faster Ft transistors now in slightly better packages.

[David Hess]

I need a fast current source, but not with much current. 10mA would be great, but 2mA are also ok for most of the intended measurements ("build in test equipment" for sensor development). But the most important points are that it should have over 500MHz of bandwidth and a decent (<1%) accuracy. The "load" (or DUT) is ground-referenced and the maximum voltage is <3V, so this should be relatively unproblematic.

Which would be the best topology for this use case? is a (modified / improved / advanced) Howland current pump with fast amplifiers (OPA855 or similar) in principle able to work that fast? Or is another topology better?

Forget using anything with multiple stages.  What is needed is a simple transistor current source, with a cascode transistor added to remove Miller feedback, increasing impedance at higher frequencies, and likely some inductance in series to raise the impedance at even higher frequencies.

For a grounded load and positive current, this would normally require PNP or p-channel transistors, however NPN or n-channel transistors can be used to sink current below ground with a resistor to the positive supply supplying the positive current.

A differential pair can be used to create a variable current source with high adjustment bandwidth.

[mawyatt]

A properly implemented Cascode provides excellent Output to Input Isolation, so stability should be relatively easy to maintain under varying loads. Simple in both Single Ended and Differential implementations, they are very popular for these very reasons.

Best,

[Marco]

BTW what does Tina-TI think a Howland will do for a fast opamp with all 100 Ohm resistors driving say a 1n4148? I just can't handle that UI and Microchip unfortunately does not make RF opamps to use in MPLAB Mindi.

[luky315]

The Wyatt current source is interesting, but the use of Aluminium Resistors for temperature compensation makes me a bit uneasy 😅

[mawyatt]

You can also use Copper, Gold, Silver, Zink

Best,

[Marco]

A properly implemented Cascode provides excellent Output to Input Isolation, so stability should be relatively easy to maintain under varying loads.

The non ideal high side current source causes issues. In the sim I added an inductor to steady it at high frequency, but in the sim it causes a load dependent dip near 100 MHz. You can just use say a 24V rail and a 1k resistor instead of a PNP current source with inductor, which will be much more well behaved, but your accuracy goes to shit.

It's a shame TPTB won't allow us to buy PNP SiGe transistors, like you could use in integrated circuits.

I'm still not sure the Howland can't work, at least they can use PNP.

PS. maybe a high side NPN current source isn't so bad either? The simulator likes it (curves for a load resistor of 10 and 100 Ohms, amplitude holds steady despite the significant voltage excursion across the load at 100 Ohm).

[mawyatt]

It's a shame TPTB won't allow us to buy PNP SiGe transistors, like you could use in integrated circuits.

We had no PNP SiGe devices available in the SiGe BiCMOS processes we utilized, only NPNs. Not sure of any IC process or discrete that is/has SiGe PNP.

Best,

[Marco]

TI says they use PNP SiGe. Opamp design gets really annoying and power inefficient without complementary transistors.

Regardless, with SiGe discretes a high side NPN fixed current source and a low side modulated NPN current sink with or without cascode seems a good enough option to try at least.

[T3sl4co1l]

Don't overlook the gyrator, effectively an emitter follower coupled to a series load resistor, thus acting as a current source at high frequencies; an active inductor.

Real layouts might cost as much stray capacitance as you're saving over a PNP, or peaking the capacitance and just dealing with it.  Or even, since the laser diode has such a low impedance, not giving a crap about the capacitance at all, incremental resistance and stray inductance dominates performance of the output node.

Tim