Amplifier Research 50A220 RF amp teardown

[D Straney]

This is a 50W RF amplifier, which claims to work from 10 kHz - 220 Mhz.  It died at work driving some badly mismatched loads, from what I understand, but from the lack of smoke, obvious burned bits inside, etc. it seems it went pretty quietly.


The entire thing is pretty much one giant heatsink, with an RF board on top, and the power supply underneath.  I didn't dig down to the level of the power supply, as that would've been some pretty involved disassembly on a piece of equipment I don't own, and the power supply is almost guaranteed to be boring - just a transformer, bridge rectifier, and some capacitors.  Here's the interesting part though, the RF board!


A closer look, you say?



Luckily, since it's just a 2-layer board with easily-visible through-hole components, it's not that hard to figure out how everything's connected (especially as it's all 80% ground plane, as I'd expect from anything reasonably high-frequency).  RF amps are not something I have a lot of experience with though, so this interpretation is going to be shaky in a lot of places, but here goes...

There seem to be a few linear regulators at the top-right edge (1) for biasing, and maybe for the actual DC collector/drain power, but that seems less likely.  There's also what looks like a thermal cutoff switch (2) to kill the power if that giant heatsink gets too hot.  For the RF parts, a schematic should make it easier:

(For what it's worth, I have no idea if the transistors are bipolars or FETs, but I've drawn them as FETs just because)
RF input enters through coax on right (3); goes through a DC-blocking cap then back-to-back diodes for limiting, and gets AC-coupled to the 1st stage (4), with an AR153 transistor.  1st stage has some bias fed from a trimpot, and some feedback through a (large brown) resistor and a messily-wound inductor to the gate.  The drain is loaded down with a variable cap, and fed with bias from a series large inductor (marked 1.4mH) and a small inductor (the toroid) in series.  The large inductor is probably for low frequencies (it's supposed to operate over a really wide frequency range...1.4mH = +j88 ohms @ 10kHz) and the small inductor (the toroid) is probably for high frequencies where the SRF of the large inductor makes it no longer inductive.  The cap at the input to the large inductor provides some local decoupling.

1st stage output is AC-coupled to some kind of matching network.  2nd stage (5) has a different transistor (AR156) but otherwise similar; has the same feedback structure, variable drain cap, and dual-inductor load.

Input of the 3rd stage doesn't have any elaborate matching network, just a variable cap on the gate.  The DC bias, feedback, and load inductor is the same as the previous two stages.  Transistor is now AR157.

After that, my guesses get far more vague as my lack of knowledge really shows itself here.  The dual transistors and their connections suggest a class-B 4th stage ( 8 ), although it could be 2 out-of-phase class-As to cancel symmetric distortion or something; there's a couple trimpots which seem to set the DC bias for each transistor independently.  The right-most toroid wrapped with coax (7) I think is a balun to take the up-until-now single-ended signal and produce a differential signal to drive both transistors' gates out-of-phase.  Each transistor has its own local negative feedback through a tan resistor and a DC-blocking cap.  After the 4th stage, the signal path becomes symmetric and looks differential for a bit (notice the top and bottom toroids).  The drains get loaded with an interesting combination of differential (black cores) and common-mode (toroid at bottom-left, (9)) inductive loads.

I have no idea what the dual set of cross-connected transformers (marked as "?" on schematic) terminated in a triple capacitor is doing.

Finally, the top-left pair of toroids (10) probably are another balun, taking the signal from differential back to single-ended this time.  It took me a bit to get the weird side-by-side winding arrangement there, but I think it's just a more vertically-efficient way to stack two cores on top of each other: this gets you twice the core area for the same magnetic path length, which gives more inductance while keeping the saturation current the same.  The returned-to-single-ended signal then goes through some more AC coupling and variable-cap loading before leaving at the top-left to go to the output connector.

It's interesting that the transistors have part numbers (ARxxx) which match the manufacturer's name...wonder if they're custom parts, or at least custom-branded?  Couldn't find anything about these parts online.  The other numbers (9321, 9238, 9339) look like date codes in a [2-digit year, week number in the year] format, and if this is true, would indicate that this amp is from the early 90's, which wouldn't surprise me given the construction style.

[coppercone2]

better suited in rf section, you will get more responses there

[D Straney]

Ok thanks, hadn't seen any teardowns in the RF section so thought maybe that was frowned upon.  Seems I can't delete my own topic to re-post (and wouldn't want to do two identical posts), any chance of a move to the RF section from the mods, please?

[coppercone2]

I find it interesting because I have a 100-500MHz 50W generator that looks similar but this one is shifted down in frequency.

I have yet to tear mine apart though. I will be looking at adding something like this at similar power levels to my amplifier rack one day. Need a better lab though, once you start getting into rack mount equipment it really needs a decent sized alone room.

[KJDS]

I've not looked closely, but will guess that the output stage matching is a guanella balun

[dmills]

Thats not really a matching network at the output of the first stage so much as a low frequency attenuator to keep the drive at low frequency under control.

Also, the output side of the final balun should have a cap or two in its ground leg to avoid a DC short.

Apart from the massive low frequency ferrites and series inductors to get down to 10kHz, it is a very conventional design.

[D Straney]

Had assumed it was a matching network because of the reactive components, but the positioning of the cap and inductor makes sense now - as frequency goes up the shunt impedance gets larger and the impedance bypassing the resistors gets smaller, tending towards a straight pass-through.  Will have to look into the details of a guanella balun as well.

With the part about the DC short, those fixed caps I drew in parallel with the output probably then are actually AC-coupling the coax shield to gnd at that point.

Appreciate the commentary all around! I'm trying to pick up some RF circuit basics (from a lower-frequency analog background) mostly by understanding the workings of the simple building blocks first, but doesn't hurt to see an actual full design all put together either.

One thing I was wondering about was having a directional coupler at the output or not; I feel like with the Ghz-range amps I've seen photos of, there's almost always a directional coupler at the output to dump reflected power into a big dummy load nearby.  Would an amp like this lack something similar because (in my very limited understanding) making a decent low-loss directional coupler that can work well down to 10 kHz (and over such a wide bandwidth) is really difficult and it's easier to just up the ratings on the output parts?

[dmills]

At Ghz frequencies a 'circulator' becomes practical to terminate the reflected power into a resistor, but as you say this is much less of a practical option at HF.

I would bet that AR have current limited the power supply (to cope with the low Z case) and just massively over sized things to deal with 100% reflected power and stay within the voltage limits of the devices, you can do that in a 50W design, it is rather less practical at a kW or so.

The cross connected coax windings are a classic wideband transformer technique, looks like 4:1 to me, search term 'transmission line transformers'. The trick here is to have sufficient magnetising inductance at 10kHz without using more then about 1/8th of a wavelength of cable at 220MHz.

House numbered parts are not a surprise, probably really old Phillips or Motorola stuff.

Regards, Dan.

[coppercone2]

I thought it goes like this (for coaxial things):

Directional coupler: used for directional sampling signal at low power (more so then a splitter/T). This has no mainline directivty I think, it is omnidirectional, the only directivity it has is in regards to its sampling port. Mainline being the signal that passes through the non sampling ports. A bootleg way to think about it might be a 'high power impedance'. If you terminate it, it won't really do much to the main signal. I think you are supposed to think of it as a transformer. You can't really trip the circuit break in a home distribution by shorting out a little transformer because of impedance.

circulator : used as a building block i.e. transmit receive antenna, uses magnetic or active circuit. This one delivers full power to the next port in a equivalent circuit minus small loss with high mainline directivty. So it effects the signal power, its the whole enchilada being coupled, not a 'high impedance' sample.

isolator: special circulator that has a load directly soldered into one of the ports acting as a full load. Often same model as a circulator but with a little block screwed into one of the ports rather then a coaxial connector. But also can be a closed model. Relies on circulator for its equivalent circuit.

So I think if you want to deal with reflected POWER for PROTECTION you need to use a isolator/terminated circulator. If you need to deal with reflected power for sampling purposes you can use a directional coupler but it wont offer protection. I think the terminology you used might be wrong. Keep in mind when you order them, you get a spec called isolation WITH a circulator but that does not make it an isolator. It's giving you the isolation of a circulator in dB but it does not have a load. When you buy on ebay look at pictures carefully. I got a bunch of isolators that I thought were circulators.

You can actually build a low power circulator using op-amps (there is a design article from 1992, I got the parts but I have yet to make the PCB). I don't really see why you can't make a power active circulator that works from low to high frequency, but it may be very difficult. The op-amps that work in those frequencies are pretty weak. You would need to buffer them (i think) some how with some kind of microwavish composite amplifier (not sure what the hell would happen practically)?

What kind of monster would a 220MHz 50W active circulator be? 3 of those amplifiers?

Also, a materials related question. Does circulator size scale proportionally with waveguide size? I.e. if you make a ferite circulator with WR430, will the dimensions of the magnetic bits scale to a ciruclator made with WR2300? (7GHz vs 200MHz)

And I am pretty sure the circulator is a waveguide element, so you are limited by the waveguide propgation characteristics for a particular circulator, you would need to use filters/diplexers to make a wider band circulator so signals are separated into different circulators to get a wider band coverage.?

[D Straney]

Thanks, you're completely right, I was using those terms all wrong - a circulator is what I meant.

That op-amp circulator circuit was really interesting, understanding its operation was a bit of a mind-bending experience.  Looked up "lumped element circulators" but it looks like those are all tuned to a specific frequency band with a resonant L and C, the "widest-band" ones seem to just have multiple L and C pairs (kind of like what you said with the transmission-line ones splitting into multiple tuned bands).

[ocw]

I recently picked something similar--a 20 watt Ailtech/Cutler-Hammer/Eaton 1020 amplifier.  Its expanded rating is for 500 kHz to 225 MHz operation.  Attached are a couple of pictures of it.  The shown 13.5 watts output was produced by -2 dBm input.  I'll wait to open up the RF PA until it needs some repairs.

[KJDS]

Here's a teardown I created a long time ago of an AR amplifier

[D Straney]

Interesting to see both of those.

I'm impressed with the size of the power supplies in the AR amp.  Unsurprisingly I guess, the first 3 stages also look identical to mine down to the individual components (may be different values or transistors, but the same arrangement) and each of the paralleled output stages looks identical to my 4th stage.  I'm curious how the power combining works on the output with all the transformers and 50-ohm resistors.  I guess if the transformers have a high-enough impedance then the outputs could be just stacked in series?  Whether a high-enough impedance for that is realistic, I have no idea.

[T3sl4co1l]

Oh ha, just ran across this for unrelated reasons.

FYI, the schematic looks correct, with a couple of mis-seen capacitors, and I guess whatever the bias network(s) may be doing.  Not a bad job!

Between stages 1 and 2: seems to be a Zobel network sort of thing, looks like reducing gain at LF.  Amps usually have less gain at HF, in a slowly varying manner, so this helps flatten overall frequency response.

All stages have generous shunt feedback (or neutralization, a bit of both), which should give fairly low gain, but there's a lot of stages to make up for that.  And overall gain is what, 40 or 60dB or something I'm guessing?  So they're going to need a bit, while keeping the bandwidth up.

First error at output stage 3, shows a shunt cap; I think the coax shield is bypassed to GND there, and signal coupled to core.  The coax is at stage 4 input bias, so this makes sense.  And then, the two power amps can be biased independently, or maybe there's some resistance between them (56R?) so they vary together a bit, but can still be set over enough range to deal with offset between them.  It'll be class AB something or other; how much A, depends.  Probably not too shallow (class B ish) to avoid distortion, and, if it's on a huge metal plate heatsink kinda thing, it could even be fairly aggressive (up to maybe 50%, i.e. similar Pdiss ~ 2 x Pout; those packages are good for what, 100W or so dissipation..?).

And what are the caps, guessing they're pretty massive, a couple uF ceramic (dipped chip on leads) kinda thing?  They use them quite generously, and it seems they're not so bulky that HF performance is impacted.  Which seems reasonable enough, 220MHz is still a pretty long wave compared to what's in here.

So the 4th stage output coupling, the "?" network.  The input to it is cross connected coax, so they act in parallel.  If the coax is Zo = 50R, then that's 25R between output terminals (C-C or D-D whether BJT or FET), or 12.5R per transistor.  Which is easily 25W each out of say 25V supply, so I'm guessing supply is in the 28-35V range, something like that; and that'd be very typical for an RF amp like this.

Also the supply bias for that stage, I think the big inductor is differential mode, though it is hard to see from these angles.  And that makes sense, no need for CM filtering to a push-pull amp.

The cross-connected coaxes are then wired in series, so that the white coax would be 100 ohms (if they are 50).  The shields just tie together (by coupling caps; actually, there should be no DC voltage between these so they could maybe be shorted, but I guess it doesn't hurt this way, and it's only three more caps).  Note that the CM impedance at this node is poorly defined, because it's the midpoint between two RFCs in series: the pink coaxes from amp to here, and the white coax from here to GND (output).  (Probably the orange cores are lossy enough that this ends up well enough damped not to matter; high-Q resonances in locations like this can lead to DM-CM coupling and thus affect frequency response, so this would be helpful if true.)  But the white coax seems to go straight to the output, so I wonder if it might be specialty 25 ohm coax, the pink stuff.  In that case, supply voltage may be lower, as the transistors would see 6.25 ohms load each.

Actually... I think the output must be class A.  Common mode supply impedance is very high, so as the supply current varies with load, it'll just flop around and fart all over your signal.  So probably these are running near limits, 100W or more quiescent, and you just eat all that tasty, tasty class A power dissipation in exchange for the low distortion.

As for what the orange cores are, that's a good question; they're quite square, and I've never seen a ferrite in that shade of orange.  They may well be stripwound steel.  Micrometals makes (or has made) cores in that shade before, I have a few.  And that would go with the probably laminated-steel E-core chokes from the input stages.  They have to be extra large, because not many turns can be afforded on them (coax length limited to fractional wavelengths, hopefully?).

And if the coaxes aren't different impedances, then the maximum length of the pink windings I think has to be much less than 1/4 wave, maybe 1/8 wave, at Fmax.  So, maybe all of 10cm.  And there's 7 turns on there, and I'm pretty sure they're getting more than 1.4cm length per turn, at least that's what it looks like.  So the coaxes must be different types.

There is one final out, which is the trim cap at the output; maybe that allows compensating for the mismatch of coax impedances or something.  But that seems kind of far fetched and probably it's just to tweak response a bit, along with all the other caps in the chain -- which will mainly be to do with the upper corner.  (Which, of the trim caps used, the first one (1st stage output) is maybe the most interesting: it's on an intentionally longer trace length, and the feedback resistor is connected there.  Probably this provides some peaking effect.)

And the output, the white coax coming off the pair of orange cores, it's not a cap network, but the shield and core are at VCC; that'll be a cap from shield to GND, and core to output (assuming this top-left fly lead goes to output, or maybe some more compensation networks, power meter or something).

----

Not mine, but I have pics and manual floating around, may be of interest:
https://www.seventransistorlabs.com/Images/AR60LA/
just give the index a browse.  Highlights:
https://www.seventransistorlabs.com/Images/AR60LA/AR60LA17.jpg Front panel
https://www.seventransistorlabs.com/Images/AR60LA/AR60LA15.jpg Rear open
https://www.seventransistorlabs.com/Images/AR60LA/AR60LA40.jpg Plate line detail, top
https://www.seventransistorlabs.com/Images/AR60LA/AR60LA60.jpg Plate line detail, side
https://www.seventransistorlabs.com/Images/AR60LA/AR60LA70.jpg Grid line (bottom)
https://www.seventransistorlabs.com/Images/AR60LA/D1000669_RF_board.jpg Schematic, finals

It's a distributed amplifier, using planar tetrodes (vacuum state!), single ended class A.  The grids are chained together as a lumped-equivalent transmission line, and likewise the plates; this allows the capacitances to cancel out (in the same way that the capacitance from one segment to the next of a transmission line is balanced by the inductance between them, and thus a wave propagates one way or the other along the structure at some velocity, and with some impedance).  Bandwidth goes up arithmetically with number of stages, limited by circuit losses (grid drive attenuates as it goes along, as the grid itself has some AC impedance: literally, work is being done upon the electron beam, as it's being modulated by the varying grid voltage!).  Tubes happen to be fairly good at this, so the 8-stage design is pretty effective.  MOSFETs aren't bad at this (though common power transistor types aren't good enough to go beyond some 10s MHz), while BJTs are pretty poor.  The highest bandwidth designs that exist today, use distributed MOS amps on InP, GaN or such, monolithic designs with bandwidth into the low 100s of GHz.

Oh sorry, those are air-cooled radial-beam types (see also 4CX250, etc.).  High performance, but not planar. https://frank.pocnet.net/sheets/079/8/8121.pdf
Not sure what voltage they were running at, the manual scans aren't complete it seems...
Planar types can go much higher, see https://frank.pocnet.net/sheets/140/2/2C39WA.pdf for example.

Tim