Amplifying, filtering, general op amp tasks.. at higher frequencies (10s of MHz)

[Infraviolet]

I'm interested in trying some optical comms and sensing, I've made circuits for these purposes before at <10KHz and at 100KHz, but I'd like to try going to "very"* high frequencies in my current design. 10s of MHz.

I'm a while off starting much on this project, but would like advice on how one amplifies detected signals when working at this sort of frequency. Op amps I've typically used have GBWP of 10MHz or 1MHz, and those with high enough GBWP to do significant amplification at 10s of MHz look to be high priced (>£10 for single op amp chips) highly specialised chips with not many options any of which may go out of stock at a moments notice.

For this sort of stuff are those higher performance op amps often used, or is it better to go back to transistor fundamentals for this sort of thing? Given their prices I'd be very surprised if many manufacturers of any mass market device that needed to amplify fast signals would want a thing like tha on the BOM, so are non op amp techniques more common at this sort of frequency range? 10s of MHz is, for small sized circuits, not as high as typical radio frequencies, but do RF design considerations start applying? Does one borrow filter designs from radio receiver tunng circuits?

Thanks


*Very high compared to circuits I've worked with before, I know that there are lots of argued definitions of what high freq, very high, ultra high... mean

[Kyle_from_somewhere]

Once you're up to Mhz you've been in radio land for a long, long time. You might be thinking of FM 87-108 as Radio, but you're already in shortwave.

There are some opamps that can do at least 5Mhz, but that's something I discovered by accident playing with a prebuilt cmoy type headphone amp. Like you I eventually decided that it's not worth the money it costs as a hobbyist to hunt down RF opamps. You run into a lot of datasheets for stuff that just isn't for sale anywhere.

Transistors can do a lot, cheaply, but I've not done much with them.

[T3sl4co1l]

A few basics to get out of the way:

DC is where you can ignore all the pesky AC effects in a circuit.  That is, the reactances are large/small enough to ignore, given some threshold for ignorance (voltage error from expected response, stability, etc.).

Parasitic AC effects depend on length.  Wire length, loop size, component body length, everything.

You can make a circuit up to some 100s kHz, low MHz, just fine on solderless breadboard -- with more difficulty if it's switching (lower impedances, faster edges), or needs precision or sensitivity (you're probably not going to pull off a fully featured AM radio IF strip).

Along the way (between "DC" DC, and this sort of range), you'll probably need to reduce wire lengths (go from fly jumpers to short pieces of solid/tinned wire), keep current paths close together, etc.  This can be challenging to do on breadboard, but is feasible for a lot of circuits.

Want to raise frequency?  Shrink dimensions.  There's very little you can do on the breadboard in the sub-100MHz range -- a few things, but mostly very simple; even digital logic (say 74HC or ECL) tends to oscillate due to delays and spongy supply impedances.

So improve those impedances.  Use wider ground conductors.  Ground plane.  Build over copper clad, either cutting out pads with an annular cutting tool, or scoring with a knife or scraping with a graver.  Or cut up chits of PCB and solder them down for elevated islands.

This extends to low GHz with some effort.  You can always add shielding where needed, just solder down more copper clad -- though the circuit may not be so easy to update/maintain with shields in place... (You can make enclosures, joints, slots, screw mounts, etc. just as with any mechanical construction, given some effort; play around and see what works for you.)

Want to raise frequency?  Shrink dimensions.  You won't be able to handle much at this point, making deadbug construction a challenge; fortunately, PCBs with tight tolerances are cheap and plentiful, and automated/turn-key assembly isn't that bad either (albeit still not something a hobbyist might spring for, but a no-brainer for professional protos).  No problem using 0402s, or even less; some problem if you need to swap them out, maybe.   Plenty of.. well, you won't really be doing much direct RF work these days, probably just mounting a chip and testing antenna structures, unless you're building a module as such -- another ease-of-use optimization available these days.

And then PCBs run out in the... pfft, something like 100GHz, where structures are just too small to build, and it's very hard to avoid using optical structures instead -- that, or everything is monolithic (directly on chip, where small feature sizes are easy).


So, given that perspective -- what you're really after is reducing coupling between traces (what ground plane does), so you can extend the approximation that currents stay in their own wires -- and reducing size, so the self-inductance (or capacitance) of those wires is less as well.  And when you can't even do that, use transmission line techniques: terminated lines look like a fixed resistance, independent of length.  Each step gains you a good, Idunno decade or so, of frequency range -- so with these methods in mind, you should be able to do 10s of MHz just fine.

Tim

[Kleinstein]

At 10 MHz it is not just the question of using OP-amps versus discrete transitors or other special RF amplifiers. The other point is than from abou 1 MHz on inductors and small tranformers become viable circuit elements. This is especially true for narrow filter with a high Q.

[nfmax]

Because of limited GBW, you will generally need to use more (OPAMP) stages, each with lower gain, than you are used to. Also, PSRR  can become dire at these frequencies, so plan to use lots of supply filtering, including series inductive elements such as ferrite beads, and parallel elements comprising both large and small capacitors.  You should feed the power in at the ‘output’ end, so stages that work with the smallest signals have the most filtering.

Another recommendation for using LC filters in preference to active designs: far easier at these frequencies!

[T3sl4co1l]

Because op-amps start to get a bit annoying, especially if you need a lot of gain (like you would for a radio), going to more basic designs is common -- hence transistors or MMICs versus op-amps.  MMICs being little more than transistors with internal biasing networks (and matching networks, for the higher frequencies) anyway.

The tradeoff is, for op-amps, you get very accurate outputs, and the high amount of negative feedback affords low distortion.  With the lower stage gain of typical RF amp designs, you get less accuracy (more manufacturing variance, less flat bandwidth) and more distortion (only local NFB is possible).  RF amps generally might not mind distortion, either, as harmonics can be filtered off entirely (and IMD, tolerated), so keep that in mind if you're looking through reference designs.

Overall, working in the SW band isn't too bad; signals stay in wires okay without too much precaution, you have free choice of op-amps (they're a bit pricier, or limited in supply voltage, but still plenty available) or discrete amps (granted, more and more RF BJTs are going obsolete -- PNPs almost entirely gone; MOSFETs are mostly medium to high power LDMOS, and high spec PHEMTs etc. are so screaming fast you might not even want to touch them (fT > 30GHz!)).  RF parts are preferable of course, but you still get plenty of gain from even just your ordinary like 2N3904 (~300MHz fT) that you're really only worried if you need to make a complete IF strip (say 80dB gain, so you need all of, like, three of them; such a drudge, eh? ).  And, yes indeed, inductors have reasonable Qs, even in quite small parts -- consider purchasing a kit of say 1206 chip inductors (or THT if you're that way), even air core are reasonable.

Also, decomp op-amps and current feedback op-amps are available.  Decomp means it's not stable down to unity gain or whatever, you can't [ab]use it as a voltage follower -- but it also means you get 10x or whatever more GBW for the same supply current and technology node.  Big win when all you want is a bit of gain!  And CFB, they don't roll off as dominant-pole compensated (voltage) amps do, the flatness just kinda stinks at high gain (because, well, what do you expect, right?).  CFB are a bit different to use (the feedback current matters, instead of voltage; and there is some loading at the input) but similarly useful here.

So for stuff like driving LEDs and laser diodes, optical modulators, etc., you can do all the usual current-source tricks, and if nothing else you can put the impedance at the end of a cable (with suitable matching network) and drive it from any old dumb amplifier. And probably gain/flatness isn't a big deal, considering the optical path will have many more dB of error in it.

Tim

[Infraviolet]

Thank you all, that's been quite a bit and some good links for me to be reading over.

evb149, bandwidth vs centre frequency and moving signals down the frequency spectrum, any tips where I can learn more about options here?

T3sl4col1, decompensated op amps, any tips for part numbers, preferably RRIO.

[David Hess]

Modern operational amplifiers are available with sufficient gain-bandwidth product to do what you want, in both voltage and current feedback forms, as well as a few operational transconductance amplifiers and difference amplifiers.  All of the same low frequency circuits work but of course layout and AC parasitics become more important.  Surface mount parts considerably ease layout consideration since DIP parts start have problems at 100 MHz and higher.

In the past, fast operational amplifiers were called "video operational amplifiers" since that was their major application, but they existed in some form even in the 70s if you knew where to look.

[vk6zgo]

A "trap for young players " is trying to use standard Op Amp circuits as Tuned Radio Frequency amplifiers, usually as an attempt at a radio receiver.
Op Amps just love to oscillate, so placing an LC tuned circuit at the input & another at the output, with a nominally negative feedback loop between them is just asking for trouble.
If the two LC networks are not tuned exactly the same, you can get phase inversion which turns your NFB into PFB & your amplifier into an oscillator.

This was how "Tuned Plate, Tuned Grid" oscillators worked back in the day using tubes.
In that case, the Miller capacitance again,caused nominally negative feedback, but tune the input & output LC networks "just so", & you have quite a good oscillator.

[David Hess]

A "trap for young players " is trying to use standard Op Amp circuits as Tuned Radio Frequency amplifiers, usually as an attempt at a radio receiver.
Op Amps just love to oscillate, so placing an LC tuned circuit at the input & another at the output, with a nominally negative feedback loop between them is just asking for trouble.

Linear Technology published a couple examples of tuned IF circuits in their application notes.

[Marco]

Just use LTC6560 if you need a fast TIA, messing around with GHz opamps ain't worth the headache.

[vk6zgo]

A "trap for young players " is trying to use standard Op Amp circuits as Tuned Radio Frequency amplifiers, usually as an attempt at a radio receiver.
Op Amps just love to oscillate, so placing an LC tuned circuit at the input & another at the output, with a nominally negative feedback loop between them is just asking for trouble.

Linear Technology published a couple examples of tuned IF circuits in their application notes.

Hell, yeah, it can be done, but it is a trap for a few people who have ended up looking for help on the beginners tab.

[David Hess]

I am trying to think of why an operational amplifier of any kind would be used at IF and RF frequencies.  They are useful in baseband applications, but even there simpler transistor designs perform better in many cases; oscilloscope vertical amplifier chains did not go through operational amplifiers for good reasons. (1)  One disadvantage that an operational amplifier will always have is higher noise because the differential input stage increases noise by 3dB over a singled ended input.

How does the distortion, IP1, and IP3 performance of a current feedback or fast 2-stage voltage feedback operational amplifier compare to to a shunt or series feedback transistor amplifier?  MMICs use internal shunt feedback.  I do not have enough experience with current feedback and 2-stage voltage feedback operational amplifiers to have an appreciation of their performance in this area.

There is an active circulator/isolator design (2) which relies on positive and negative feedback and uses operational amplifiers to give good performance from DC to 100s of MHz, but I wonder how practical a transistor implementation would be.

(1) Where linearity matters, like in the horizontal signal path, oscilloscope amplifier stages use shunt feedback; vertical amplifier stages almost always use series feedback.  There are some exceptions though; HP sometimes used shunt feedback in the vertical amplifier chain, but I have no idea why or how it affected the performance.

(2) I have been thinking of using this to design a network analyzer which operates from DC to 10s of MHz.  I saw it decades ago when it was first published, before the internet.

[Infraviolet]

"I am trying to think of why an operational amplifier of any kind would be used at IF and RF frequencies"
Very novice thought here, might be quote wrong, that's why I've been asking about this, but if it is a WEAK signal you are trying to find, filter out from other signals of other frequencies and then send on for further processing, an op amp is what is usually recommend at lower frequencies. And you can do loads of different things with op amps, huge variety of diferent circuits each with a specific action which can then be put in to chains to perform actions on a waveform as it passes from one to the next. I know transistors can amplify, but usually things made of discrete transistors seem more buggy than integrated circuits, more sensitive to manufacturing variability if the parts?

[David Hess]

Very novice thought here, might be quote wrong, that's why I've been asking about this, but if it is a WEAK signal you are trying to find, filter out from other signals of other frequencies and then send on for further processing, an op amp is what is usually recommend at lower frequencies.

Operational amplifiers are not preferred for weak signals because input noise of a differential input is inherently at least 3dB greater.

[Kleinstein]

At higher frequencies as operational amplifier is no longer ideal (input capacitance, non zero output impedance). So the simple design rules as for a low frequency signal may no longer be valid.

The noise is another point - this is not only the OP-amp itself, but also the resistors used in op-amp circuits.