R&S FSL Spectrum Analyzer teardown and confusion

[D Straney]

Ok, this one has been a long time coming (as you can see if you look at the dates on these photos).  A while back, I got the chance to explore a Rohde & Schwarz FSL spectrum analyzer (with an impressively wide range of 9 kHz - 3 Ghz) that wouldn't turn on.


Here's the top and bottom sides, with the cover removed:

The top side has the CPU board, plugged into the main board, with the power supply at the back end.


The bottom side normally has the RF module sitting up front (I had removed it here), and a large open space for a battery option at the back.  You can see that the RF module plugs into an SMA connector from one end (which goes straight to the front-panel input connector via some rigid coax) and a ribbon cable to the main board at the other end.



The CPU board is likely based on an Intel processor, definitely has an Intel chipset, and runs an embedded Windows XP (which honestly made the UI a pain to deal with) from a CompactFlash card.


The main board seems to have all the "in between" functions: interfaces to the front-panel display and buttons, power supplies, the 10 Mhz reference clock, and data acquisition connecting the RF module to the CPU.  Sadly I couldn't take off the CPU board to get a better look at the main board - the CPU-side connector was an unsoldered press-fit type, which when I started gently pulling on it gave signs that it would rather pull out of the CPU board than de-mate from the main board's connector.  Didn't want to turn what could be just a power supply issue into a much bigger/more expensive issue, so I stopped there.

Anyways, the thing we really wanted to see was the RF module!



Here's the circuitry in all its naked glory:



I haven't been able so far to figure out everything that's going on here, especially because I'm not an expert in microwave electronics.  In general, I'd expect a modern spectrum analyzer to have some input attenuators for range selection/protection, maybe some different selectable band filters, one or two IFs worth of frequency-synthesizer LOs / mixers / filters / amplifiers, then direct digitization of the last IF.

Well, we can see what looks like some selectable attenuators on the input here, with a relay and some Peregrine PE42551 RF switches:

Relay seems to select whether the input goes into the attenuator or the bypass path, and the ICs select whether the output is connected to the attenuator, the bypass path, or a mysterious path which goes to the other side of the board, through an IC I can't identify, and then into the Inner Mysteries of the inner layers (doesn't come out the top again):


There's also a nice distributed-element filter connected to an unpopulated connector, probably for one of the options we didn't buy like a tracking generator.

The IF (or something at least) definitely gets digitized at the end: there's an LTC2206-14 14-bit 80Msps ADC that looks like it's fed by a single-ended-to-differential converter in the SOT-23-5 package, with its outputs going through buffers to the ribbon cable back to the main board of the instrument.


What's in between the input attenuator and the ADC though?  That's the tricky part...

The LO (at least one of them?) is at least surprisingly easy to spot.  There's a VCO, based on (unsurprisingly) a single-transistor LC oscillator with a group of 8 varactors for tuning.  It's nice and easy to figure out the connections because they're all directly on the top layer:

There's also some power filtering and linear regulators to the right of the VCO (LT1964 low-noise LDOs, to be specific).

To the left of the VCO is the rest of the frequency synthesizer:

There's an HMC439 phase detector at the left, meant as part of a "build your own PLL" sort of affair.  It likely gets its input through the custom-looking "PD_DIV2" chip at the right, which is probably a programmable divider.  The HMC439 feeds a pair of tuning currents back to the VCO, and if you look back at the VCO shot, you can see a pair of signals coming up from vias together and going into the 5534 dual op-amp just above the VCO, which should turn them into a tuning voltage for the varactors.
A transformer just below the HMC439 seems to turn the divider's single-ended (I mean, uh, unbalanced) output into a differential (I mean, uh, balanced) input that it needs.

There's a bit of other weirdness just above the programmable divider that I couldn't figure out.  The VCO's transistor (and all the others marked "N9" seem to be the BFG540W.

After this though it all gets murky as far as I'm concerned:
There's one section with a bunch of tuning-fork-looking power dividers at the left, and 3 almost-identical sections which could be for different frequency ranges?


There's some filter blocks:


There's this weird shit over here, which seems to be transmitting something through a patch antenna (?) and picking it up with a slot antenna () on the other side?


(Look at the right-hand edge)

There's two back-to-back power splitters/combiners in a configuration I don't understand at all (see top-left):


There's even some mystery Atmel parts, with custom-looking part numbers that don't turn up in any search results:


Anyways, if anyone feels like explaining what all the in-between pieces actually are, go crazy, here's the link to the full photo album of board shots:
https://flic.kr/s/aHsmLYteis
Just let me know what you find.

[Garan]

Those are great pictures. Here are some comments

According to the FSL datasheet, the FSL is a triple conversion analyzer. It upconverts 9kHz ~ 3 GHz up to 7158* MHz for the 1st IF, then down converts to 838 MHz and finally downconverts again to 48 MHz before it is sampled by the LTC2206.

I strong doubts on whether the first IF is really at 7158 MHz for a 3 GHz analyzer. The photo evidence just doesn't support it. Nevertheless, I'll start the analysis assuming 7158 MHz is correct.

With these frequencies, we expect to have the LO frequencies to be
LO1 = 7158 to 10158 MHz
LO2 = 6320
LO3 = 790 MHz, note that LO2 = 8 x LO3.

The LO (at least one of them?) is at least surprisingly easy to spot.  There's a VCO, based on (unsurprisingly) a single-transistor LC oscillator with a group of 8 varactors for tuning.  It's nice and easy to figure out the connections because they're all directly on the top layer:

This is likely the 790 MHz VCO for the third mixer.

After this though it all gets murky as far as I'm concerned:
There's one section with a bunch of tuning-fork-looking power dividers at the left, and 3 almost-identical sections which could be for different frequency ranges?

The 3 almost identical sections are the VCOs for the 1st LO. To keep the phase noise low, the required tuning range for first LO is broken up into 3 different spans so that each VCO doesn't have to cover a wide span. Notice the different line lengths used in each section. That is the resonator for the oscillator. Judging from the size of the resonators, the VCO is not operating at 7.1 ~ 10.1 GHz, but probably at half of the required LO1 frequency (3.579 to 5.079 GHz) and there's probably a frequency doubler somewhere.

But I can't find any evidence of a frequency doubler chip anywhere, so I don't think the LO1 and 1st IF is really at 7158 MHz. I think the 1st IF is at 3998 MHz and the LO1 VCOs span from 3.998 to 6.998 GHz.

There's this weird shit over here, which seems to be transmitting something through a patch antenna (?) and picking it up with a slot antenna () on the other side?


(Look at the right-hand edge)

There's two back-to-back power splitters/combiners in a configuration I don't understand at all (see top-left):

The weird shit is a microstrip to slotline transition for the first LO signal. Essentially it's a balun that feeds the differential LO signals into the mixer diodes on the other side.

The outputs of the 3 VCOs are combined along the left edge and the split into two paths, one to the HMC441 and the other path is for the tracking generator section.

[TheUnnamedNewbie]

I've had a first look at the PCB. It's hard, not having the board in front of me makes it a lot more complex. I'm also seeing a lot of 12-pin devices marked:

837
9901
81

Judging by the fact that they seem to have 3 RF input/outputs, I'm going to guess they are a mixer of some sort, though I can't find anything online in my quick search right now.

I am going to say that the first IF is actually at 7158 MHz. Keep in mind that while this might be a 3 GHz model, it also comes in 6 and 18 GHz versions.

The LO path (or what I think is the LO path) has 2 HMC441 amplifiers, which are 6.5 GHz to 13.5 GHz.

I might spend more time looking at it later, esp if some other people can chip in and might get me going again.

[D Straney]

Appreciate the answers.  Yeah I was wondering where that 3rd connection to the input switches goes, if it's a source for self-test/self-cal or something similar.

Luckily enough I actually still have the RF module (lab was closing down, management showed all signs of scrapping the whole instrument so figured I might as well keep the interesting part as a learning experience - told them to give me a call if they wanted it back for repairing the instrument or better "for parts" sale value or whatever, but never took me up on it).  So I'll go through and do some continuity checks on the different coax via arrangements to get a better idea of what's connected to what, and post annotated photos when done.

Couldn't find part numbers for a lot of the mystery QFNs, but yeah the "837" seems like either a mixer or switch (although I don't know why they'd bother with a different switch model, if the one used for the front-end can handle the full freq range well).  The "424" ones seem like attenuators maybe, from the connections (2 RF, plus a bunch of DC).

I'm especially fascinated by that mixer structure, hadn't heard of slotline before.  I'm guessing that the RF mixer input is to the top-left of the 1st mixer diodes through the series cap, and the output is to the top-right.  Stubs at the top-left maybe are some combo of blocking the LO from going backwards into the previous stage (AC short with quarter-wave stub) [edit: on second thought, scratch that, would have to be really wideband stubs for such a wide LO range] and matching the mixer input impedance seen at the RF input (diode series R / capacitance + source impedance of the slotline bit) to 50 ohms?  Could try to work out something more solid if I could tell what the actual wavelengths were but not sure about the dielectric constant, as from looking at Rogers selection tables mu-r can vary from 2-10...will have to measure the input microstrip which should be 50 ohms, and try to back out mu-r from there.  The bits just to the right of the mixer diodes look like an IF bandpass filter I guess, where those lines shorted at one end match a multiple of a quarter-wavelength for the IF frequency?

[TheUnnamedNewbie]

You can actually estimate the dielectric constant by looking at a few structures we can estimate the frequency of.

If it is possible, could you take some more pictures of the VCO section and so on, maybe with a ruler in frame to have a reference of size? My guess would be that the dielectric constant of this substrate is gonna be in the range of 3-4, as this is where a lot of your mid-range priced substrates are at (Megtron, I-Tera, RO4003, RO4340, etc). So I would start from there. We know the ball-park frequency bands of the IF, so that could be a good place to look for quarter-wavelength matching structures to figure out permittivity, and from there figure out the bands for some of the other filters.

[D Straney]

Yeah definitely, will get those VCO shots with a ruler.

Meanwhile, here's probably the easiest way to follow the connections: as a GIMP image.  Took new pics of the top and bottom, trying to get the field of view as "square" as possible, then did a flip and some scaling.  It's not perfect as there's a little skew even with the top-left/bottom-right corners lined up nicely, but it works pretty well.  Connections I found with a continuity check are on a separate layer so they're constantly overlaying everything.  Just zoom in and pan around the image, and turn the visibility of the "top side" layer on and off to flip between top and bottom.
https://www.dropbox.com/s/3dtifizy25kcuju/rf%20tracing.xcf?dl=0

Edit: Ok, here's some more photos:

[D Straney]

So, haven't spent a ton of time looking at the newer pics yet, but some things I noticed:
Test input: that add'l possibly-for-test-or-calibration input to the RF switches at the front-end goes to a long trace on the reverse side which goes to....an unpopulated IC.  Maybe there is a test source that's needed to actively correct flatness at the higher frequency ranges, doesn't look like it would be linked to one of the hardware options; TV trigger would pull from the IF I'm guessing.

Mixer distributed-element stuff: with a dielectric constant of 3-4, the 7.2 Ghz IF should have a quarter-wavelength of 5-6 mm, which nicely matches the length of the shorted-at-one-end stubs to the left of the mixer feed in the ruler pic above (which backs up the "IF bandpass filter on output" theory) and the open stub to the right (which would make this a notch filter keeping the IF out of the RF input).

Doubling mixer: here's a slightly-out-there theory as well.  I haven't looked at the VCOs in detail yet but Garan doubted the 7.2 Ghz IF due to the resonant lengths in the VCOs, suspecting that they were half the required frequency and there was a frequency doubler somewhere.  Anyways, the mixer looks at first like a single-balanced switching type, with the LO essentially gating the RF signal:


However, what if instead of the dual mixer diode having a series arrangement, the two diodes were in a common-cathode (or common-anode) arrangement instead?  In that case, it should function electrically like a "non-switching" single-diode mixer (with the LO shifting the small-signal attenuation throughout its cycle to create the multiplication-like action...only "pure" multiplication with an x^2 device instead of e^x I guess)...

...except that the LO frequency would get doubled at the same time?  Of course with the un-balanced mixer you lose the LO cancellation in the output that a single-balanced mixer supposedly gives you (not that I doubt it, just haven't proved it to myself yet) but maybe that's a perfectly fine tradeoff with where unwanted stuff lines up in the triple-conversion arch.  Or maybe this arrangement would actually remove the LO from the output - again, haven't fully thought it through.

I don't know how plausible this is though.