Interesting & unusual RF/microwave boards

[D Straney]

Index:

• Raytheon mystery board (below)
• Rohde & Schwarz CMW 270 RF front-end module
• Remec Transmitter, ED-0421-0 15 Ghz
• Teradyne mystery board & modules opened
• Radar frequency multiplier
• Cubic Defense Systems mystery boards (large & small) & die shots from small board's hybrid module
• Mystery boards from L3 Communications
• Remec Receiver, 18 Ghz (& die shots from prev. transmitter)
• 2.2-2.3 Ghz Frequency Synthesizer made by L3/"Telemetry-East"

I normally get my "shopping therapy" impulses out by buying cheap interesting-looking electronics sold as scrap on eBay, and so besides the avionics in my previous posts, have also ended up with some pretty unusual boards with various bits of high-frequency circuitry.

Here's one to start it off that came through a slightly different route:
Raytheon mystery board
I got this from American Milspec with the intention of removing the RF modules to reuse in my own projects and maybe open one of the duplicates, but eventually realized I wasn't going to have the time to do that and was also moving soon, so sent it back.  Took a bunch of photos first though:


The connectors don't have any function labels, but the card itself is labeled with "1650062-103", which turns out to be a "Receiver-Transmitter Subassembly" (NSN 5895-01-234-0520) made by Raytheon, as part of the AN/URQ-33 radio set.  I couldn't find any references to the AN/URQ-33, but there was also a mention on parts-sourcing websites about the EPLRS, which turns out to be a military location-sharing system that happens digitally over UHF radio.

Let's look at the middle section, which is mostly symmetrical:

The center SMA connector feeds a power splitter, which on either side passes either through a mystery "KWE" module (haven't been able to find any info on these), or into the large "728000-112" modules (no info on these either).  This split signal likely(?) feeds the RF port of the mixers on either side, while the two LO signals come in separately through additional SMA connectors.  The IF signals then appear to go through the 728000-47 RF amp modules, and then the "Hughes"-marked bandpass IF filters @ 315 Mhz.  The 728000-48 limiting-amp modules probably follow these in the signal chain to process the filtered IF signals.

After this, things get much less clear.  The Merrimac modules on either side are listed as electronic phase shifters, but with a 1-6 Hz operating frequency; have to assume this is a typo.  The metal cans nearby are UA733 high-speed (140 Mhz) op-amps, which isn't fast enough to deal with the IF directly (assuming that the 315 Mhz BPFs are in fact the IF filters), but they would be fast enough for data encoded on a 315 Mhz IF.  There may be some kind of phase-decoding loop going on with each channel's combination of a phase shifter module, 2x UA733s and 2x SE521 high-speed comparators (<20 ns), to output a digital data stream.


So far, it looks like this middle section splits up an RF signal and feeds it to two separate receivers.  The LO signals are probably supplied by the identical side sections, which seem to be frequency synthesizers:

The large 728000-174 module is a VCO; no info on frequency range.  It seems to feed two separate outputs, one of which goes through the gold-plated bandpass filter (684-921 Mhz) mounted to the wall.

Programmable Frequency Divider
The column of ICs behind the VCO contains first an 11C90DMQB (equiv. SP8680) ECL frequency prescaler, then a mystery IC (JM38510/06006B), and finally 3x 4-bit ECL counters (10016DM).  This would be the programmable frequency divider portion of the frequency synthesizer; the ECL logic (rather than the "74LS" family used in the slower control sections) is needed to deal with the high 100s-of-Mhz LO frequencies.

There's two 8-bit shift registers (JM38510/00903B = 74LS164) near the rear D-sub connector, one of which has traces running to the ECL counters (through pairs of resistors, which do the TTL-to-ECL level translation; these ECL counters work off of negative voltages).  I think this is a serial control interface which uses 12 of the 16 shift register bits to drive the counter "preset" inputs and therefore set the output frequency, if the counters are chained in count-down mode and use the "underflow" signal both to preset the values again, and as the output of this programmable frequency divider.

Phase Comparator & Control Loop
Two more SE521 dual high-speed comparators sit near the D-sub connector; at least one or two of these must be used as the phase comparator, to produce an error signal between the frequency divider's output and a reference frequency coming in from outside (likely 10 Mhz on the D-sub connector).  The group of discrete components near the VCO is probably what generates the control voltage for the VCO and adjusts it based on the measured phase error.  The two JM38510/31401 one-shots (74LS123 equivalents) next to this section suggests to me that this is some kind of charge pump topology which increases or decreases the voltage on a "memory" capacitor in fixed steps, when an error signal is present.

The remaining 4 bits from the shift registers seem to feed the SNJ5445 BCD-to-decimal decoder, which has 10 different resistors connected to its output.  My best guess is that these resistors set the pulse time for the one-shots, and therefore let the control loop's slew rate of the frequency synthesizer be programmed (setting how fast an error signal can ramp up or down the VCO's control voltage).  The EPLRS Wikipedia page mentions that it uses frequency-hopping for security; because of this, fast frequency re-tuning and therefore fast LO re-programming is likely very important, and so adjusting the control loop gain on the frequency synthesizer would be important for minimizing the time that it takes to lock onto a new frequency.  The two separate receiver channels from the same RF input may even trade off back and forth between each other for fast frequency-hopping, to ensure no gaps in reception: one receiver could be listening while the other is already tuning to the next frequency.

Anyways, there's more photos in the full album, hope you enjoyed.

[Neomys Sapiens]

JM38510/06006B should be a OR/NOR GATE, DUAL 4-5 INPUT. At least that is what DLA says.

[D Straney]

Oh good to know, thanks.  If it's ECL too might be a way of getting a non-power-of-2 division factor from the prescaler chip, or just be for general control (clock enable, etc.) otherwise.

Next:
Rohde & Schwarz CMW 270 RF front-end module
This was from a piece of test equipment to test WiMax transceivers, a wireless standard that never fully caught on and isn't in widespread use anymore,  which is why no one else wanted this module.

It's got 3 type-N ports on the front which correspond directly to the 3 ports on the front panel of the instrument, heatsink fins on one side, and a control connector along with 4 coax connectors which must continue on to the other RF sections deeper inside the CMW 270.


Inside, you can see a control section with a small FPGA (Lattice ispMach) and some linear regulators (the Linear Technology DFNs in the 2nd photo), then a whole lot of RF switches.



This module makes very heavy use of the Peregrine PE42551, an SPDT switch spec'ed up to 6 Ghz and high-ish power for this type of thing (1W or so before it hits the 1 dB compression point).

There's no upconversion or downconversion happening in this module: its purpose seems to be routing signals and setting levels.  From looking at the front panel...

...there's two external RF channels per module, with the middle connector fixed as an output and the other two connectors switchable between outputs and inputs.  A big part of the circuitry here seems to be just switching between Tx and Rx functions for each of these connectors and routing the signals appropriately to those smaller dedicated Tx and Rx connectors for each channel, that feed the mixer & IF stages.

The other big function here is applying gain and attenuation as necessary.  Most of the RF switches involve selectively bypassing attenuators or amplifier stages.  For example:

There's also two sections, one for each RF channel, that look like the biggest power boost, with 2-stage amps right next to large contact areas between the board and the enclosure - I'd bet money that this is what the heatsinking from those fins on the module cover is mostly intended for.

The non-RF section in the middle (partly visible in the next photo) has what I'm guessing is biasing & control for these power amp sections.

I have to admit I'm confused by the circuitry right at the external connectors though:

There's a resistive pad directly on the connector, so it's burning power even when used for Tx (fair I guess, if you can make the thermals work, to improve the impedance match from both ends? add a tiny bit more ESD robustness?).  But there's what look like they'd be final output amp & pre-amp transistors for Tx and Rx respectively: the "O3" device is used in two places, and the "O5" device in the other 3.  I can't see any DC biasing on the inputs or outputs though, and you can see on the middle dedicated-Tx output how the "O3" device's pin connects directly to the type-N's pin, without any DC blocking in between.  The pinouts also wouldn't make any sense as transistors: look at the right-hand connector in that photo, and notice how the same pin (on the same "O5" device) goes to the connector for both the Tx and Rx paths.

Only possibility that makes sense to me is if these are low-capacitance anti-parallel diode pairs to ground, specifically made for RF signal protection.  I've never seen something like that in this package, but then again I'm definitely not a full-time microwave guy so it could be a thing?  I couldn't find any semiconductors listed in the normal places with "O5" or "O3" markings that had the correct package, so this is just my best guess.

Oh also, let's just take a second to admire nicely milled RF enclosures and all the little "tunnels" that connect the different sections:

[D Straney]

Remec Transmitter, ED-0421-0 15 Ghz
This is a cute little transmitter module that fits in the palm of your hand.


I'd say it looks like someone removed the connectors, but from lack of mounting points it looks like there were never connectors to begin with.  I'm guessing this could've been part of some sort of rooftop-microwave comms system.

Once we look inside, the layout is pretty clear and the ports are nicely labeled.

Here's a nicer shot, catching the light:


The actual signal path is fairly straightforward:

The LO and IF signals come in separately and get mixed together, pass through a serious bandpass filter to remove all the mixing products except the desired output frequency, and the result then gets 3 stages of amplification with progressively larger amplifiers, until we reach a waveguide transition so that the output can be fed to an antenna (likely some kind of horn shape).  Various DC bias paths can be seen too, feeding the different amplifier stages.

LO path
The LO signal gets special treatment, as it's particularly important to the whole system: a strong LO signal is needed to forward-bias the diodes correctly in a passive mixer and make it work correctly (notice how there's no DC power to the mixer die), and harmonics or other "dirtiness" on the LO signal will create phase noise and other undesirable things for a communications system.  Therefore, the LO gets, after a resistive pad presumably to improve impedance matching, a bandpass filter and its own amplifier to boost the signal right before entering the mixer.  Adding an extra amplification stage here also has the desirable result of keeping the LO signal clean for the whole system - because amplifiers work (mostly) only in one direction, this adds a lot of isolation between the incoming LO signal, which probably is fed to and used by multiple other boards, and the passive mixer whose own imperfections could otherwise inject some of the IF and upconverted RF signals, as well as harmonics of all these, back into the LO for the whole system.


Limiter
There's also an interesting section between the 1st and 2nd power amp stages, which I'm not sure about but have some guesses.  You can see in the labeled "Limiter?" section from above how the signal is split along two paths (with a standard Wilkinson divider), passes through two small mystery sections, and then is re-combined afterwards.

Inside the cavities are what look like most likely individual diodes.  A DC continuity check shows that the signal is continuous from one side of this section to the other (the diodes aren't in series with the signal), and a diode test mode on the multimeter shows a 0.6V junction from the signal path to ground, but not the other way around.  This all suggests that each cavity contains two diodes in parallel, with their anodes connected to the signal and their cathodes connected to ground.  The bond wires from the PCB trace probably jump down to the first diode's anode, connect across to the second diode's anode, and then back up to the PCB trace on the opposite side.

I'm guessing this would be to limit the power output of the whole device by limiting the signal level going into the final 2 power amp stages, however, I'm not sure why it doesn't use bidirectional clipping with anti-parallel diodes to ground.  I briefly considered the possibility that these could be varactors for frequency multiplication, but it seems very unlikely given that the geometry of the power splitter & combiner are the same (this type uses a quarter-wavelength section) and a filter would be needed afterwards to select the appropriate harmonic and avoid a "dirty" output full of the original signal and all other harmonics.

Splitting up the signal before clamping it with diode(s) serves two purposes, though:
1. The signal gets clipped at a higher power level, because clipping starts when 1/2 the power (1/4 the amplitude) becomes comparable with the diode's forward voltages, instead of when the full power & amplitude becomes comparable with the diode's forward voltages.  Selecting different diodes can also vary the clamping level (/max. allowed power) somewhat, but only over a limited range.
2. Placing identical elements at an odd multiple of a quarter-wavelength apart cancels their effects on the transmission line impedance: see https://www.microwaves101.com/encyclopedias/quarter-wave-tricks#pindiodes for an example.  The two cavities look horizontally separated by roughly 1/4 wavelength (judging by the geometry of the power splitters), and so this could be a way of removing the effects of the diode capacitances on the transmission line impedance.  I don't have the time or network-theory background to know for sure if this works through a power splitter, though, but just putting it out there as a possibility.

Output power sensing
After the final power amp stage, there's a directional coupler (the jagged section) which picks off some small fraction of the output signal to the antenna.  The port at the bottom left has a series diode to ground (if you look closely you can see the diode symbol in copper), which acts as a peak detector and rectifies the output, to turn the sensed output signal into a rough DC indication of RF power.  This section is stuffed full of wideband quarter-wavelength stubs (the paper-fan-looking shapes) which serve as RF shorts to ground, to keep as much as of the RF signal as possible out of the power-sense output so it can remain "clean" DC; there's also some more filtering with discrete resistors downstream.

The external system can then monitor either the output transmit power, or the power reflected back to the final stage amp (which could be used to shut down the transmitter to prevent damage if the reflected power was unusually large, for example).  I believe the way it's connected (with the white RF termination resistor to an RF ground at the top-left port) makes this a reflected power monitor rather than an output power monitor, but I'm not a microwave expert, so take that with a grain of salt.

Anyways, I can dig up the eBay info if anyone's interested in re-purposing this kind of thing, the seller had a bunch of similar microwave transceiver blocks and were open to offers so I got it for something like $20.

[D Straney]

On second thought, I think the "limiter" is actually a variable attenuator.  You can see a DC control feeding into this section just before the power splitter, from a little bit of circuitry and a port labeled "VVA" (maybe "voltage for variable attenuator").  This most likely is used to put a variable forward bias current into the diodes to ground, which creates an variable small-signal resistance to ground depending where the diode is in its I-V curve, and produces a variable attenuation of the RF signal.  The same scheme is used all the time in variable RF attenuators such as the Mini-Circuits one I took apart here: https://www.eevblog.com/forum/rf-microwave/a-look-inside-some-mini-circuits-modules/msg4779338/#msg4779338  This would explain why the diode connection to ground is only in one direction, and the variable impedance to ground would provide another reason to place the two diode pairs a quarter wavelength apart and maintain a 50 ohm system.

[D Straney]

Teradyne mystery board


Because of the amount of gold plating and metal/ceramic packages here, I assumed this came from some kind of aerospace or military application, but a closer look showed that it's from Teradyne, a manufacturer of automated test equipment:


At a slot in the side, you can see the PCB layer stackup, which has microwave substrate for the outer layers, and likely plain fiberglass for the inner layers - this saves some cost or other constraints compared to an all-microwave-substrate stackup, when the inner layers are only carrying low-frequency control signals:


Let's take a moment to admire the microwave section more closely:

There's also no shortage of relays:


Mapping out the connections shows that it's roughly structured as a downconverter with 3 selectable bands, followed by programmable IF gain (implemented as a chain of selectable attenuators), and an optional section that looks like some kind of phase modulation decoder.  You can tell it's a downconverters because the signal after the mixers is no longer routed as microstrip in the same way, and it starts being switched by normal relays instead of solid-state RF switch ICs, which means the IF frequency could be anywhere up to very roughly 100 Mhz or so.  Note that there's also two different selectable IF filters; seems to be a lot of flexibility built in, which makes sense for ATE.



A lot of it is straightforward even if complicated, but the maybe-phase-decoder part is what I'm not sure about.  I couldn't find any info on the ESCQ-2-10.5 part, but from resistance measurements between the pins the best I can guess is that it's a transformer, or maybe a lumped-element power splitter.  The same signal getting sent along two paths before getting multiplied in the mixer (U20) is interesting.  At steady-state, the result would be independent of input phase, so I can think of only two possibilities for what this is supposed to do:
1. The fixed many-stage LC filter on the top path acts as a delay, and so the output of the mixer functions as a "temporary phase change" signal, changing from its steady-state value whenever the new signal has propagated through the phase shifter but not through the filter/delay yet.
2. Depending on the shape of the LC filter's response, and how the phase shifter behaves over frequency, this could be some kind of FM demodulator?
Ideas?

Overall, given what Teradyne does I'm guessing this is used as part of a microwave receiver used to quantify the output signal of either a RF IC, or some kind of system like a radio, as semiconductors aren't the only application they make ATE for.

[D Straney]

Frequency multiplier for radar
Here's a module from the AN/APQ-120 fire-control radar, used in the F-4 fighter jet aircraft, from the late-60's:



Let's just take out the screws and pop the top & bottom covers, and...

Oh.  Well shit.

Luckily the foam was soft, so after a lot of digging with a screwdriver and knife like the world's worst archaeologist, I found an interesting-looking box with a lot of different compartments and carefully-routed cables.  The bottom side has most of the action...

...while the top side has mostly power distribution and accessible points for the (many!) trimmer capacitors.  Can only imagine how long it must've taken to calibrate one of these in production:


There's an input coax connector, and output coax connector, and a circular connector with DC power input.
The simplest and least interesting part is the power supply compartment, which has what's likely a voltage regulator (or 2?), fed from the circular connector.

A couple outputs from the power supply exit through RF feed-through filters in the bottom of that compartment:


It turns out there's 5 different sections to the frequency multiplier: let's walk through them one by one.


Stage 1: amplifier/harmonic generator
First, we have an amplifier fed directly from the input coax, visible at the top of the first photo below:


This is a single-transistor RF amplifier, most likely designed to saturate its output on the high and/or low swings and produce a harmonic-heavy output signal (normally the exact opposite of what you want with an amplifier) - this is where some of the frequency-multiplication comes from.  Not exactly sure what the two trim caps are for.  I thought about trying to trace out the exact circuit here, but it ended up being too frustrating between the compact double-sided layout and the remaining bits of foam.

Stage 2: filter
Next, the output of stage 1 runs to a filter section, which is supposed to pick off only the desired harmonic from the output of stage 1 to pass on further, and suppress all the other unwanted harmonics.  Unfortunately with the way it was stuck deep in the cavity here, I wasn't able to pull this one out non-destructively:


There's 6 filter stages, each with a transmission line to ground (which has its highest impedance, and therefore highest output, at the frequency where a the transmission line length is a quarter wavelength), and a trim cap to ground to adjust the effective transmission line length slightly by adding a variable capacitance in parallel.  Each filter stage is coupled to the next one by alternating film caps and inductors, which I'm guessing suppress some of the lower frequencies.  The small glass-encased device is just a 50-ohm resistor.

(The bottom is all ground plane, nothing exciting to see there:)


Stage 3: amplifier(s)
This is another dual(?) amplifier with two transistors, to boost the low-amplitude high-frequency harmonic that's been isolated from the first stage's output.  There's two transistors here, one with 4 pins:


Stage 4: amplifier(s)
In this stage, you can see the power levels increasing, as both transistors are stud-mount power devices.  When fully assembled, these are mounted to a metal frame with a solid thermal connection to the box for heatsinking.



Stage 5: fast diode & microwave filter
With stage 4 removed, you can see the full extent of stage 5 underneath:
(Before)

(After)


This is the most interesting part, and also the highest-frequency one, which consists of another harmonic generator followed by a filter:


The signal from stage 4 comes into some kind of matching network...

...which then descends through a transmission line made from a solid strip of metal...

...to a step-recovery diode (SRD) or similar type of microwave diode, which generates extremely fast rising or falling edges when switched on or off, creating many harmonics of the input signal:


There's an interesting little adjustable slider which fits over the transmission line that leads to the SRD, adding a capacitance to ground at a variable position.  The nut seems to have been soldered in place after calibration to keep it from moving:


Anyways, after the SRD, most of the length of the stage 5 module should consist of bandpass filters that, just like stage 2, pass only the desired harmonic from the SRD's harmonic-rich output and suppress all the unwanted harmonics.  I wasn't able to get the module option to look at these, but because the AN/APQ-120 is an X-band radar, these are probably filter cavities tuned for somewhere in the 8-12 Ghz range.

Finally, at the stage 5 output, you can see some kind of ceramic block, and the electrode which picks off the signal for the output coax.

The output signal then travels through a short right-angle bit of rigid coax, to the output connector on the module.

Hope this was interesting.  Let me know if you have any special insights.  Sorry for the unusually bad quality of photography here, with everything being under workbench lights and covered with bits of foam.

[eb4fbz]

A lot of it is straightforward even if complicated, but the maybe-phase-decoder part is what I'm not sure about.  I couldn't find any info on the ESCQ-2-10.5 part, but from resistance measurements between the pins the best I can guess is that it's a transformer, or maybe a lumped-element power splitter.  The same signal getting sent along two paths before getting multiplied in the mixer (U20) is interesting.

That’s a custom part, but I bet it is a 10.5MHz 90deg splitter. That would fit your theory about that section acting as a phase detector.

[D Straney]

Makes sense - then I'd imagine the variable phase shifter would be tuned to match the fixed lumped-element delay in the other path, so that the (averaged) output of the mixer would be zero in steady-state, then go positive or negative briefly when the phase changes (for as long as it takes to propagate through the delay line).

[D Straney]

Teradyne mystery board (continued)
I decided there were enough interesting-looking RF modules on this board that I was willing to open a couple to have a look inside.  For comparison, I picked both a generic amplifier (APS388) used in a few places, and the SML1 limiter, which is supposed to clamp the signal level to keep it below a certain threshold.



You can see that each package contains not just a single die, but an entire hybrid circuit with multiple bare dies on a ceramic substrate.

Let's take a closer look under the microscope... (shoutout to my local hackerspace, Artisan's Asylum, for letting me use their equipment on public nights)

SML1 limiter




According to the datasheet, this is supposed to limit a signal to between 0 and -20 dBm, depending on the supply voltage / control voltage.  The circuit is pretty simple, with 3 separate methods of limiting the signal's maximum peak-to-peak voltage.

• Anti-parallel diodes with a series resistor for "softer" clipping, directly on the input
• Anti-parallel diodes for "harder" clipping, directly on the output
• A diode bridge in the middle, AC-coupled with a capacitor on either side because of its DC bias

Because my impression is that the limiting threshold is supposed to be adjustable, I think the bridge in the middle does most of the work, with the anti-parallel diode pairs as backups.  The way the bridge works is similar to a diode bridge sampler.
When idle, the input and output of the diode bridge sit at Vcc/2 with all diodes conducting a quiescent current set by Vcc, R1, and R2 (which are equal).
When the input signal goes positive, D3 and D4 come into play.  D3 and D4 have roughly similar forward voltages, meaning that the output voltage rises along with the input voltage.  However, because of the polarity of D3, the input signal can't directly deliver positive current to the output: the positive current to the output can only come from R1.  R1 forms an AC voltage divider with the 50Ω output load (shown on the schematic), and therefore sets a maximum output signal level based on the ratio of resistances, and Vcc.  When the input voltage is too high, the R1 current (Vcc - Vin - D3's Vf)/R1 no longer has any current "left over" from supplying the output (via D4) to conduct through D3, so D3 turns off and there is no longer a path from the input to the output.
When the input signal goes negative, the same thing happens but with R2, D5, and D6.

Here's the diode bridge up-close:

...and the elaborate set of series and parallel sections that can be selectively bypassed with wirebonds to set the value of R3:

The output anti-parallel diodes are pretty straightforward:


APS388 amplifier



The star of the show here is Q1, which is the RF amplifying transistor.  (I have no idea whether it's bipolar or a FET)  The giant core, L2, is a 1:1 transformer on its collector/drain; I don't know why a transformer was used instead of a bias inductor, but I'm sure there's a reason, just haven't thought through it in detail as I haven't designed a class-A RF amp before.  The resistor running straight down the middle (R1) seems to provide negative feedback.

The bias circuit is worth looking into in more detail here: it seems to aim for a fixed collector/drain current on Q1.  Q2 and Q3 are wired as a differential pair which essentially compares the Vbe of Q2 and Q3.  Q2 is diode-connected, so that the voltage divider formed by R5 and R6 sets a specific voltage at the emitter of Q2.  R4 acts as a current sense for Q1's DC collector current.  When the collector current is too low, the voltage across R4 drops and because Q3's base is "fixed in place" by Q2, Q3's Vbe increases.  This drives a larger current through Q3, and increases the base bias to Q1, therefore increasing its collector current and providing stabilization.  The same works in reverse if Q1's collector current decreases below the setpoint.  C5 provides heavy bandwidth limiting to this Q1-collector-current-regulator circuit, ensuring that Q2 & Q3 will only work against DC changes in Q1's collector current, and leave the actual RF signal alone.

Another way to think about it is that Q3 is an emitter follower, with a fixed setpoint created by the R5 & R6 voltage divider plus Q2 diode-connected to compensate for Q3's Vbe.  Q3's emitter voltage across R4 sets a fixed supply current.  The excess supply current that Q3 has to conduct to keep its emitter voltage "in place" is redirected to the RF transistor's base to close the Q1-collector-current-regulation feedback loop.

[D Straney]

Cubic Defense Systems mystery board (large)
Here's one of a couple modules I got from here that were being sold for parts << $20.  "94987", which pops up everywhere, is the CAGE code for "Cubic Defense Systems", for what it's worth.


Because signal paths are nice and easy to follow on single-layer RF boards, we can see that a signal comes in at one end, passing through one of the bottom-side-metal-can amplifiers(?) and then a circulator...

...passes through a couple discrete-transistor amplification stages, with tombstone'd SMT power filter caps...

...and then goes to a directional coupler, where some of the signal gets picked off:


A second signal input goes through an amplifier of its own, with some big blobs of foam to hold inductors in place...

...and then goes to a coupler of its own:


By the way, those small metal boxes on each input & output are confirmed to be filters, with very "ship-in-a-bottle" construction.  This one looks like maybe a bandpass:


Both output signals sampled from those directional couplers then travel to a middle section, where they're mixed together in the "CF049" module, and then boosted by the 2nd bottom-side-metal-can amplifier before being sent out over yet another coax.  Opening it the "CF049" module got messy, so no good photos there, but it had 3 tiny magnetics and a tiny diode bridge inside: a double-balanced diode mixer.



Overall, the block diagram looks like this:

The signal flow vaguely suggests that one of the signal paths is a transmitted RF signal, and the other is an LO signal, and the "monitor" circuitry in the middle is downconverting the RF-Tx signal to an IF to make sure that the output amplitudes of both RF & IF are in a normal range etc. etc.
I'm not sure why the top signal path has the most-heavily-heatsinked amplifier (the metal-can one) coming first in the signal chain, and followed by a circulator.  It might be performing some non-linear function, like a frequency multiplier?

Let's look inside both those metal cans to see what's happening there, because I couldn't find anything useful with the part numbers, and also because finding miniature accidental-hybrid-artworks inside mystery boxes is fun.  As a quick side note, you can see some creative power routing on the bottom of the heatsink here, with grooves milled into it where discrete wires are run for distributing the supply voltage.


Here's the metal can that boosts the mixed output:



It's a pretty standard-looking RF amplifier in most ways, a common-emitter (Q1) with a large spiral coil that forms a tapped auto-transformer for stepping down the voltage from the transistor's collector.  Q3 and associated components form a biasing circuit that attempts to regulate the DC collector current of Q1: R5 works as a current sense resistor, and so when Q1's collector current is too low, the voltage on Q3's emitter increases.  This increases Q3's Vbe (because its base voltage is fixed by the R6/R7 voltage divider) and increases Q3's collector current, which increases the base(/gate?) bias on Q1.

The strange part though is Q2.  I have no idea what this is doing here.  It has a larger die than Q1, and the patterning I can see on the surface makes it look more like a high-frequency transistor, rather than a very "normal" BJT like Q3.  The connections don't make sense, unless it breaks the standard pinout of "bottom of die = collector/drain".  If it really does have the base(/gate?) connection where I showed it, then it seems like it would operate off of a kind of "current sense" for Q1.  Any insights here would be appreciated.

Let's take a look at these nice spiral inductors, and resistive traces in the biasing circuitry:



Now, here's the metal can the precedes the circulator:



The biasing is much simpler here because it appears to be a common-base amplifier, with Q1's DC setpoint determined only by the R2/R3 voltage divider, and R1 (ignoring Q1's Vbe, that is).  The DC-blocking capacitors here are little vertical single-layer flat plates, rather than the big SMT multi-layer ceramic caps in the previous metal can, which suggests that the operating frequencies here are higher.  The step-down auto-transformer/tapped-inductor L1 here makes sense, as a common-base amplifier has a very low input impedance equal to the incremental resistance of Q1's Vbe: so here you get the best performance by stepping down the 50Ω(?) input to match the much lower transistor impedance.

So, I'm still not sure why the whole board's signal chain is laid out the way it is.  This device isn't a mixer or anything else obviously non-linear, and it seems to be used as an amplifier rather than a frequency multiplier: even if Q1 was driven heavily to the point where it saturated each cycle and produced a lot of harmonics, you'd expect to see a lot of filtering following it, to select only the desired harmonic and suppress all the others.  I have to admit that I'm also not sure what the benefits are of a common-base vs. common-emitter amplifier are in an RF context where impedances are fairly constant; the tradeoffs are much more obvious at low frequencies, but I'm not a full-time RF guy.  Might have something to do with noise figure, or specific types of transistors available?

There's nothing else particularly noteworthy in this metal can, except that the inductors are rectangular(!!!) instead of circular:

Fuck, I love microscopes.
You can also see some trimming (or parameter selection?) of one of the resistors, with wirebonds selectively bypassing sections of it to tune the resistance.

Anyways, hope you enjoyed the close-ups.

[D Straney]

Cubic Defense Systems mystery board (small)
This is the second of the Cubic Defense mystery RF boards:

You can see there's two identical channels of.....something.

There's a few pieces of semi-rigid coax running over this board, suspended like an elevated railway on some metal posts.  Two of them actually terminate on this board, but the other(s) just seem to be passing through.



There was actually a pair of boards: one marked with "Rev. L", and the other with "Rev. N".  They look just about identical, until you cut open the hybrid modules:

Rev. L (now with EvilMonkeyz) has a more complicated circuit inside the hybrid module:

...while Rev. N has been simplified somewhat:


Next microscope-trip to the local hackerspace, I'm hoping to get a look at the dies inside my Rev. N hybrid and see if I can get a part number from each one.  Meanwhile, though, here's the circuitry on the rest of the board:

It looks like an RF signal comes into each hybrid, when then produces a low-frequency differential output of some kind.  The obvious function would be a power sensor / level detector, but the hybrid internals look a lot more complicated than I'd expect for a simple function like that.
One clue also comes from the lack of local DC feedback on the LM118 amplifier: this means that either...
(1) it's being used as a comparator to produce a digital output: not unheard-of, I guess, but doesn't make sense that it has local AC negative feedback, or
(2) each channel here is being used as part of a larger control loop; for example, automatic gain control (AGC) for the level of the RF signal coming in.

Additionally, there's a couple common connections between the two hybrids, and it looks like they may be supplied only with a negative voltage.  I'll have to see what ID'ing the hybrids' internal dies turns up.

[D Straney]

Mystery boards from L3 Communications
Kenton (Evilmonkeyz) and I swapped some duplicate scrap a while back, and so I ended up with these mystery modules.  Based on the construction style and part numbering, they look like they were originally part of some kind of aerospace or military radio system made by L3 Communications (CAGE code 06401, which shows up everywhere on these).




The many wires and coax cables branching off each of these boards, along with their irregular shapes and small size, suggests that each one lived in its own shielded compartment separate from the others.  Would've loved to see what the whole system looked like intact before it got taken apart and the pieces ended up on eBay.

Amp and filter board
Let's get the simplest one out of the way first.



This looks like just a couple amplifiers (one of which is missing), followed by a filter module.  The filter helpfully lists its frequency (12 Ghz) and manufacturer (12855 = Smiths Interconnect).  The "wj" on the amplifier module suggests it's made by Watkins-Johnson, a big radio manufacturer:


There's also a directional coupler on the output for sampling the output signal and bringing it off-board somewhere.  The very non-coaxial-looking wire connection exiting at the top-right (very much not appropriate for a 12 Ghz signal) suggests that the 2-terminal gold device there is some kind of power sensor - this could be just as simple as a fast diode and a capacitor.  It's also interesting to see the "pass-through" ceramic block in the lower-left, where maybe an attenuator would've been placed in a different design variant.


PLL board



The main feature of this board is the Qualcomm Q3036M, which is a mil-spec version of the Q3036 PLL controller.  I couldn't find a datasheet for this chip directly, but there's one for the slightly upgraded version, the Q3236, which mentions being backwards-compatible with the Q3036: https://www.qsl.net/n9zia/omnitracs/q3236.pdf

This contains everything for a frequency synthesizer except the VCO (for flexibility), and the actual analog portion of the control loop, between the frequency/phase-detector output and the VCO input.  This second part seems to be provided by the Philips 5534A op-amp next to it.

You can see a bunch of slightly-unusual passives:


...and a couple metal cans on the bottom:

Wasn't able to find data on either of these, but my guess is that one of them is a stable oscillator that provides the reference for the frequency synthesizer.  I may saw them open soon to have a look inside.
(Edit: the smaller can is an LM126 dual-tracking voltage regulator, probably used to generate bipolar analog supplies for the PLL circuitry.  The larger one looks like the reference oscillator - I sawed off the lid but was immediately greeted by a foam filling...no easy decap here)

VCO board #1


The board is extremely simple, with just a closed VCO module, and a pass-through trace for something unrelated.  The module here is listed as a "SAW Oscillator" with an 862.15-862.85 Mhz tuning range at this semi-sketchy parts sourcing website.


Let's open it up and have a look inside:




You can see the rough arrangement from the schematic: there's an amplifier with a frequency-dependent feedback loop, forming the oscillator.  The oscillator's output gets split off and also goes through a separate amplifier to buffer the output signal, followed by a bandpass filter to clean up any harmonics.

The main source of phase shift (and therefore what mostly determines the oscillator frequency) in the feedback path is the SAW filter, in the gold can.  SAW stands for "Surface Acoustic Wave" (SAW); it's essentially a mechanical delay line but which can propagate even multi-Ghz signals along the surface of piezoelectric material, transmitting at one end and receiving at the other.  The delay is set by the distance between the transmitter & receiver, and the propagation velocity, so it's known for being stable with temperature, and if I remember correctly, being more resistant to vibration and shock than a quartz crystal oscillator.  I'm not an expert on these, though, and they're an entire rabbit hole of physics on their own.

The same structure with two coupled inductors pops up a lot in this circuit: I haven't gone deep enough into the RF world that I recognize all the common circuits off the top of my head, but it sure looks like a lumped-element transmission line at a specific frequency.  A couple articles (1 and 2) explicitly call out this circuit as a lumped-element version of a coupled-line coupler, which makes sense from looking at it.  This is used as a power splitter on the output of the oscillator, to send half the power back through the feedback network, and half to the output-buffer amplifier.  However, I don't understand the use of this coupler in its two occurrences in the feedback network.  My best guess is that it's doing something like generating 180°-separated signals to feed to the tuning varactors, and then recombining them, to cancel out any asymmetries generated by the non-linearity of the varactors - haven't worked through the math though.

In the output filter, the tiny sections of wire across the capacitors to ground are used as inductors: this makes sense when you look at the values.  For a parallel LC circuit to resonate at 862 Mhz (and therefore allow the fundamental output frequency to pass, while shorting others to ground), 15 pF → 2.3 nH.  (It's convenient that these RF caps have the values written on them)  2.3 nH is a very plausible value for that short loop of wire.

I'm not sure what type of transistors are used here, as strangely there's no DC biasing on their inputs.  They might have some kind of output-to-input biasing resistance added internally, or they could be something like JFETs.

VCO board #2


This is similar to the other board, except the VCO module's output gets fed to a Merrimac FDF-4A-750 frequency doubler, amplified by an (upside-down) transistor, and then sent off-board.


Oops, looks like the VCO module here took a dent:


This VCO module is an HO1313 model made by RF Monolithics Inc., and the patent number referenced on the case (US4760352) refers to an oscillator tuned by, again, a SAW filter.  The HO1300 and HO1301 are 700 Mhz and 750 Mhz SAW oscillators in the same series, which look similar even down to the pinout.  The difference, besides presumably the frequency, is the "hi-rel" packaging here which might explain the lack of available datasheets if they gave it a custom part number for the special application.  I'm going to assume that like the other VCO, though, this oscillator works somewhere in the high-100s-of-Mhz / low-Ghz range.

Let's open up this one too and compare the internals:




This one is much simpler than the other, based around just a single amplifier: the µPC1677.  If you scroll to the 2nd page of that datasheet though, you can see it's not just a single transistor, but a whole 8-transistor amplifier circuit, 4 of which are diode-connected likely for DC biasing.

This VCO circuit is different in that it lacks a separate output buffer circuit, and it also uses a lot of winding PCB traces instead of inductors: I haven't measured lengths and substrate thickness, but this suggests it may be working up in the low-Ghz range as opposed to the other VCO.  It also has a resistive power splitter to divide the oscillator output between feedback network and output pin, which is much lossier (dissipates half the output power) but I'm guessing the lossiness also might make it less susceptible to "load pull" on the oscillator output, making the output buffer amp unnecessary? (I haven't worked through the math on that either though)

The varactor-tuned portion of the feedback network is interesting, with 6 identical back-to-back varactors fed the same bias voltage: this amplifies the capacitance variation, and therefore the tuning range, of a single one of these varactors by 6x.  The back-to-back configuration also I think should help cancel out some of the asymmetric non-linearities (= even harmonics) created by the RF signal moving each varactor back and forth slightly along its tuning curve.  The SAW filter here also gets its own supply voltage, which suggests that there's either some kind of electrostatic thing in there that needs a DC bias, or that it has its own internal amplifier (either to drive the transmitter, or the amplify the signal from the receiver).

Anyways, hope you enjoyed - let me know of any mistakes I made interpreting the VCO circuitry.

[D Straney]

Cubic Defense Systems mystery board (small) - continued
(Continued from here)
As part of the same local-hackerspace visits where I got some rough die shots from the aircraft voice-warning-generator's hybrid modules, I managed to get some nice views of the dies inside this mystery RF hybrid module from before:

It looks like there's a lot of wirebonded resistors(?) to ground, scattered around on various signals.




Unfortunately this didn't help much: I have no idea what these dies do, as there's no obvious large-scale structures or markings that correspond to searchable part numbers.  But they sure are pretty!

[D Straney]

Remec Transmitter, ED-0421-0 15 Ghz
(Continued from here)

I got the chance to put this module under a microscope recently, and so just wanted to show what some of the bare dies look like up-close.  The most interesting are the 2 cascaded amplifiers that boost the signal just before the transmitting antenna:


2nd-to-last amplifier:

Signal flow is from left to right here, with DC power coming in from above and below.  You can see two transistors on this die, arranged as two cascaded stages, getting larger from left to right (the rectangles in the silicon itself, each with a barbell shape on the metal layer).  There's a small spiral inductor in series with the input, at the left edge, which is probably for impedance-matching to the 1st transistor's gate/base; all the other spiral inductors seem to be "RF chokes" for DC biasing & power though.  There's many small gaps in the DC connections, which I think are most likely silicon resistors that you just can't see here because of the suboptimal lighting.

Last amplifier:

This one goes full-on alien runes with the patterning.  Signal flow is from top to bottom.  The signal is split at the top edge, with a strange interdigitated coupler that I don't understand, goes through two identical parallel paths (one on the left, one on the right), and is then recombined at the bottom edge with another interdigitated coupler.
Each parallel path goes through 3 successively larger amplifier stages; at each stage the signal is split again for twice the number of transistors as the previous stage, so that by the end there are a total of 8 total transistors having their outputs combined.
You can see some very tiny spiral inductors biasing the gates for the final (bottom) transistors, but for the most part the RF blocking in the DC connections here seems to be done with thin meandering traces (for higher inductance) or presumably-quarter-wavelength lines followed by large pads on the metal layer (which are probably acting as capacitors).
I have no idea what the clothespin-looking structures do, that show up all over this die - the ones that look like an "Ω" with pigtails.  Anyways, I am very much not an expert on MMIC design, so definitely let me know what I'm missing!

Remec Receiver, 18 Ghz
Here's another telecom receiver to match the transmitter.  The frequency isn't exactly the same, but like the transmitter it has the same cute palm-size packaging and a similar construction style.


Like the transmitter, it's also very simple to follow the connections:


The general idea is that the RF signal from the antenna enters a first-stage amplifier (the LNA) right away, and then goes to a mixer, where it's downconverted to some lower IF.  The LO path has a couple filters and an amplifier to clean up the LO signal (and prevent its harmonics from creating too many stray frequencies in the IF output), and to isolate it from the LO input, which is probably shared between multiple modules.  The IF signal, which starts out differential from the mixer, is converted to single-ended by a balun and then goes through a couple amplifier stages of its own - you can tell how much lower the IF is in frequency, because it uses discrete inductors and amplifiers in standard plastic packages with pins, rather than distributed elements on the PCB and bare silicon dies with wirebonds.


One part I'm not sure about is the reason for the alternate IF traces that lead to the IF balun: it almost looks like different traces could be wirebonded in to change the delay of each one separately, but I'm not sure why that would be necessary - maybe to tune out any differential mismatch in the balun and/or mixer?  But the most-direct traces between mixer and IF balun are solid, so this isn't something that could be selected during production by selective wirebonding, unless the traces were actually cut first.

Except for this, there's nothing unusual here to figure out, unlike the diode-based attenuator on the transmitter module: it's all even more straightforward here.  So after a closer look at the RF, IF, & LO paths...



...let's take a look at the two dies that deal with the RF signal:


LNA:

Signal flow is from left to right.  The low-level signal from the antenna enters at the left side, and is split into two halves by another one of those weird interdigitated couplers (are these an array of parallel-edge coupled lines?), amplified by two symmetrical sections (one top, one bottom), and then recombined at the end.  Each of the two identical horizontal sections consists of two cascaded amplifiers, with spiral inductors connecting in between for DC biasing.  Here, you can see the dark lines in the silicon better which are likely resistances in the DC biasing paths.
Each transistor seems to consist of 8 different sections; I'm not sure if this is just larger junction/channel area for lower noise, or if each "transistor" is actually multiple separate transistors.  The 2 parallel amplifier sections probably helps keep the internally-generated noise of this amplifier low: because the internal noise generated in each transistor is uncorrelated with noise generated in another copy of that transistor, a common technique at lower frequencies is to put multiple amplifiers in parallel from the same input, and then add the outputs together.  The noise reduction scales with the square root of the number of parallel amplifiers, so 4 parallel amplifiers = 2x noise reduction, for example.
Again, this is not my field though and I'm not sure how transistor noise scales with different parameters, so I'm not sure what the tradeoffs would be of paralleling a bunch of small adjacent transistors vs. one larger transistor, or if there's even a difference.  Let me know.

Mixer:

This is a Hittite Microwave (now Analog Devices) part from 2006, and there's a lot more going on here which is much less clear to me than the amplifiers.  RF enters at the bottom, LO enters at the top, and IF exits from the right side.  There's no DC connections here so it's all-passive, and the RF & LO connections are single-ended while the IF connection is differential (but it may also support a differential RF connection, based on the bond pads?).
Here's some only-vaguely-informed guesses at how this works: I think the groups of 4x devices each are probably diode bridges(?), which suggests this may be a balanced mixer of some kind - I can't fully follow the connections here.  It looks like each of the spiral inductor pairs is at least reasonably-well-coupled to each other, and so these are acting as coupled-inductor baluns.  The RF signal goes to two separate halves (one left, one right) which seem to operate out-of-phase from each other.  Each half splits the signal into a differential signal with a balun, and drives its 4x-diode bridge with the differential signal.  The LO signal also is split differentially by the balun at the top.  The output from each diode bridge is then recombined by a balun, which is added to the one half of the LO signal, and the two out-of-phase halves of the chip generate the two IF signals.  Again, these diode bridges may be double-balanced mixers, but I'm not sure as I haven't managed to figure out the polarities of the diodes and trace the connections yet: I guess it's possible there's also multiple diodes in parallel for lower noise(?) or something similar.

[D Straney]

On closer inspection of the mixer die, I realized that each of the spirals is actually two inter-wound traces, like the planar equivalent of a bifilar winding - so it doesn't rely on the loose coupling of two side-by-side inductors, but instead gets very tight (and probably nicely broadband, with each winding segment acting as a coupled transmission line) coupling.

I think the mixer circuit is something like this, but there's parts I'm very unsure about and are probably wrong.
For example, R2a and R2b clearly connect almost-directly across the outer edges of the L2 (LO-side) windings; this winding doesn't seem to be center-tapped so if R2a/b are direct connections then they would basically be shorting the LO signals to ground: but they look very very short esp. compared to the other signal "bridges" within the silicon here (where two connections need to cross), and they don't look like the clearly-resistive sections elsewhere.  Then again, I'm also not sure that the "center-to-center" bridges in L2a/b don't connect to the metal-layer center-tap...but if the midpoint of both sides of the effective transformer were connected together that wouldn't make much electrical sense either.

The circuit doesn't quite make sense to me: I can see how the LO polarity selects which pair of diodes conduct in each individual diode ring, but in the schematic as I drew it, I don't see where the imbalance in series diode current, in each conducting pair, would come from that would produce a non-zero voltage on either IF pin.



Edit: corrected reversed resistor labeling in schematic

[szoftveres]

weird interdigitated couplers (are these an array of parallel-edge coupled lines?)

Those are Lange couplers (essentially 90deg hybrids), and the arrangement is called "balanced amplifier".

https://www.microwaves101.com/encyclopedias/lange-couplers
https://www.microwaves101.com/encyclopedias/balanced-amplifiers

[D Straney]

Telemetry Frequency Synthesizer
Here's an interesting frequency synthesizer module I picked up for cheap: it's made by L3 Communications/Telemetry East (whoever that is). Between the construction and the manufacturer I suspect it's for some kind of aerospace application, but don't know for sure.



According to the label, output range is 2.2-2.3 Ghz in 0.5 Mhz steps (200 steps total).  The specific frequency is selected by 8 DIP switches under a cover on the side:



Above the output connector, there's also an optional spot for an analog modulation input, which isn't used on this particular unit.  The larger circular connector carries power and control signals.


There's an AD590 2-terminal temperature sensor stuck to the side, maybe to monitor case temperature during qualification testing:

These are pretty classy sensors, a step up in price and complexity from a simple thermistor or thermocouple.
Edit: just realized the logo on the label here looks like Sandia National Labs)

Internals
Of course the goal is to open it up though. After removing a lot of screws and breaking the epoxy used to seal the case against the outside environment, I took off the top and bottom lids. You can see that it's a very nicely made two-sided assembly, made with multiple compartments from a single block of metal.  Looks like someone sneezed all over the inside, though, with the generous application of glue for vibration resistance.



This is the purpose of each section:

The control section and VCO form the PLL and frequency divider that make up the core of the actual frequency synthesizer functionality. The output of the oscillator goes to the opposite side via a short piece of coax, and passes through two stages of power amplifiers, before exiting through a circulator to isolate any impedance mismatches from the amplifier or load.

Control section/PLL
The big gold-capped chip had no recognizable part number, so I popped the top to have a look at the die:



It's a Motorola MC145151 frequency synthesizer block. This contains the programmable frequency divider, and phase comparator for the PLL. A set of wires from the DIP switches run directly to its pins, so these probably set the frequency divider parameters directly. The 2nd output from the VCO goes through a short coax to this mystery SOIC-8 IC, which is probably a prescaler that divides the output frequency down to a range that the MC145151 can work with:


There's some op-amps and analog stuff in the left half of the control board, which probably make up the feedback section of the PLL, turning phase comparator output pulses into an analog control voltage to drive the VCO's varactor control input. Gain and phase shifts would also be applied here as necessary to ensure the PLL's control loop is stable.

If we lift up a glued-down trimpot and capacitor over here, we can see another ceramic and gold IC underneath. My best guess is that this contains the oscillator supplying the reference frequency to the PLL:


VCO

This is a varactor-tuned common-emitter oscillator, followed by 2 stages of amplification.  The oscillator output is picked off from the emitter.


I'm not too familiar with microwave oscillator topologies, but the oscillator's frequency-selectiveness seems to be created by a resonator on its collector: a transmission line followed by the varactor capacitance.  Because the collector transmission line is < 1/4 λ, you can look at this very roughly as a series LC circuit, or more accurately as a transmission line stub whose effective length is changed by adding lumped capacitance from the varactor.
The feedback portion of the oscillator probably happens through the transistor's collector-to-base capacitance.  This means the transistor's base needs to not have a direct RF short to ground, and it looks like the components on the base have a role in this, allowing the DC bias through while presenting a high(-ish) impedance at the base, to avoid squashing the Ccb-coupled feedback signal.  The base transmission line is < 1/4 λ, and so being an RF short at the left end (due to the 27 pF cap) its impedance looks inductive @ 2.2 Ghz from the right end.  Together with the 2x 3.9 pF capacitors this forms roughly a parallel-LC resonator to create a high(er) RF impedance, looking out of the base terminal.

A small amount of the oscillator's output signal is picked off from the emitter through the 2 pF cap, put through a pi-shaped resistive attenuator (probably to isolate the oscillator even more from amplifier stage 1), and then put through two amplifier stages.  Neither of these amplifier stages are notable, except for the curved transmission line on the 2nd collector: otherwise they're both classic Circuits 101 common-emitter amplifiers, with AC-coupled inputs/outputs, and emitter-resistor degeneration for DC bias stability, which is bypassed with caps for max. AC gain.

Most of the signal from the 2nd amplifier is sent to the other side of the module to the 1st RF power amplifier, while a small amount is picked off through the 0.1 pF cap to the PLL's prescaler.  0.1 pF sounds tiny, but it's 723Ω @ 2.2 Ghz, so together with the 51Ω resistor this forms a roughly 15:1 attenuator.

(The transmission line lengths here are not precise: I estimated them by measuring a distributed-element-only collector-DC-bias trace on the 2nd power amplifier and assuming that was 1/4 λ and that the ceramic PCB material was the same)

Power amplifiers 1 & 2


Nothing unusual here. The first power transistor takes its gate (base?) bias from a resistor divider at the edge that possibly includes a thermistor. The second power transistor has a DC short to ground on its input, so is likely either an MMIC that handles its own biasing, or a class-C amplifier for efficiency. I'd expect to see much more filtering in the class-C case to attenuate harmonics, but maybe the mixer/modulator/etc that uses the synthesizer's output appreciates a square-ish LO signal, and has its own filtering before the antenna.

The power amps both get their DC power from the...
Power supply section


This has an LT1084 adjustable 3A regulator mounted to the wall for heatsinking, plus 3 separate metal-can transistors. It seems like this is used only for the RF sections, from what I can tell (and the digital/control circuitry gets a 5V power supply from outside) but I'm not sure - it's all packed too tightly, with too much glue and overlapping wires & components, for me to trace all the connections in a non-destructive way.  Between all the floating through-hole components, reworks, and wiring, each of these modules must've required a LOT of hands-on assembly time.

Output section
Finally, after exiting the power amplifiers, the RF output goes through a circulator with a dummy load, and some input & reflected power measurement, before leaving the module.

The power measurement is done, not with printed microstrip couplers, but with these janky-ass loops of wire glued to the RF traces. These loops of wire lead to little ceramic daughterboards that contain peak detectors to produce a DC voltage (very roughly!) corresponding to RF power - the output voltages go straight to the external circular connector, for monitoring by some other piece of equipment. The coupling factor between the RF traces and loops of wire is going to be unknown due to the, umm, highly variable geometry, so this is probably intended just as a "RF present" vs. "RF not present" indication, definitely not any kind of precision power measurement.

Hope this was interesting. I may try and re-use the RF section as a power-VCO for microwave projects of my own, if I can figure out the DC biasing.