RACAL-DANA 9084 synthesized signal generator

[synx508]

As I was preparing to leave the DDRC rally I thought I'd take a look for some tantalum capacitors so I could repair a RACAL-DANA 9084 that I hadn't bought yet. It's always nice to combine two rally buys like that.
Having spent £17.50 on more capacitors than I really needed in not quite the value that I'd eventually need I bought the signal generator for the asking price of £25, knowing that it would be a nightmare or a wonderful character-building challenge.
The seller assured me that it was blowing a fuse in the power supply and that this "was good" because "it meant that the problem was in the power supply". This was a very poorly constructed theory but I didn't argue because I already knew that it would be full of tantalum capacitors in various states of distress. Then he dropped the bombshell that "it worked when I injected power into the motherboard directly". My investigations suggest that this was probably "worked" that was limited to some of the front panel LEDs lighting up and the LED display showing some numbers, but that's good enough for a lot of sellers.

Arriving home I decided to check what the 9084 really did, so I fetched a 2A fuse (okay, 1.6A since I have a big bag of those) and fired it up. The fuse blew immediately, so I fired it up without the motherboard attached and noted that +24V was missing. This turned out to be a shorted tantalum and a destroyed TO-3 shaped 7824 regulator. I replaced the tantalum with one from my rally selection and the 7824 got swapped for a TO-220 version.

I needed a manual, luckily there is one available for free* online.

* it's on a website that wanted me to join but I didn't want to, so I didn't and found a way around its annoying ads

The 9084 is internally divided into three groups of RF screened boxes that sit atop a motherboard, with a chassis at the rear holding power supply parts and the OCXO and a front panel with a sandwich of two front panel related PCBs and a display PCB, so I removed a lot of screws to get a better look at all of this.

I checked a few tantalum beads on the +15V rail and found a few that were now two leads and a pile of dispersed ashes and a few that were shorted. so I snipped the shorted ones and reattached the motherboard and decided to measure all the rails - something that was cut short by more smoke and a blown fuse. This time I'd seen that -15V was missing, so more tantalum capacitors were snipped and a few were replaced. Then the cycle repeated several times and after a while I had a display showing 0.0000000MHz and some front panel LEDs, some of which responded to switch presses in the correct way, some of which didn't. Also, effectively no RF output but the audio signal generator used for modulation was working and controllable which was a good sign. The OCXO was cold but producing approximately the right frequency, so that was nice.

I think most people would've given up at this point because it was clearly going to be a lot of work and I had underestimated the complexity of the instrument and some of the interesting ways in which it could fail, too.

Then the 24V rail failed again, my rally tantalum had let me down, but it turns out that I was an idiot using 25V parts on a 24V rail (like RACAL did), so that was easily fixed.

Dismantling the front panel more to investigate the not-working LEDs revealed that they had failed open - this is often a very bad sign - memories of the seller "injecting power" came back. I replaced these with what I had to hand, which included some clear lens green LEDs that are so bright you can read by them. This worked but was a stop-gap and I've ordered some proper diffuse green LEDs. With all the LEDs available it was clear that the logic for band selection was very broken - as all the band LEDs were lit. It couldn't be in every band at once and indeed it wasn't. The band switcher works by starting a clock when you press a band button, then activating each band's LED and switch line one after another until it finds the band that has been selected. It was not finding a band and it was starting on its own, so something was stuck electronically (I ruled out the switch early on). I decided to remove the front panel board and operate it from a bench supply where it behaved very well indeed. The fault disappeared when a scope probe was placed on the outputs of a faulty 4049CN chip, so when I tested it on the bench I couldn't see a problem!

This mysterious fault turned out to be a theme, in total three 4049CN chips from the same manufacturer all had one or more bad units in them. The funniest was the 4049 that is used to flash the display when one or more of the PLLs is out of lock. This is supposed to flash at a few Hz but I measured it, it was "flashing" at 4.3MHz. A 4049 from a different manufacturer was absolutely fine.

With the front panel rebuilt and its logic performing more as the designers had expected I was able to switch bands and enjoy none of them working.

The manual places the schematics in approximately the same order as the signal flow, which is nice, so I started at the beginning. The reference/source board, I changed a tantalum on this and it worked perfectly - giving me a 1MHz, 10MHz and 1kHz output, all derived from the OCXO. I extracted the 1kHz comb loop and fired it up with external signals and it worked perfectly. The same wasn't true of every other board!

Most boards needed tantalum capacitors but a few had spicier faults like the diode switched filter board which had one unit of a 4041AE (it's a buffer chip with complementary outputs) that wasn't producing any output. I replaced the bad unit with two units of a piggy-backed 4069UBE but have ordered a proper replacement 4041UBE which should arrive in a few days.

The output loop and the 1MHz loop both needed minor inductor tweaks, presumably because of component age drift. The 9-10MHz oscillator wouldn't start because its supply rail was being halved by a resistive tantalum capacitor. And here's where I rant. RACAL-DANA chose to design this thing such that each of the many, many loops sends "out of lock" using individual open collector outputs then connected them all in parallel. This means the control logic only gets to know that something is out of lock - not what is out of lock. It's a demented solution in a synth with so many loops. I added out of lock LEDs to each loop, they're only visible with the screened boxes open, but they're really helpful.

There's no computer in it, which means that all the complex states are stored in PROMs. Many, many PROMs. There isn't even a clever FSM, all the counters and the possible machine states are mapped from a combination of the selected frequency, modulation type and range to a series of parallel BCD and not BCD lines. This is fabulous from a noise point of view but there are loads of interconnecting wires and the remote uses a 50-way connector.

So at this point I have the loops locked and with the correct filter being selected but the ALC isn't working, output is very low (too low for a counter!) and distorted. The fault is, of course, the BeO loaded RF amplifier. Luckily the BeO part can be removed entirely and as long as it's not run for too long it can be tested outside the generator with only a +15V and -15V supply. So I did that and found many bad tantalum capacitors but everything else seemed okay.

Before bed I thought I'd try injecting small ALC currents in to see if I could change the gain. I could turn it down but not up, seems it was turned up to full all the time - suggesting the amplifier itself wasn't providing enough gain. This corresponds with the power vernier having no effect, it's a DC control loop and it was already doing all it could.

In the morning I plugged the amplifier into my "out of the box" bench setup but omitted to correct the polarity that I'd been using the previous night to adjust the ALC. The red was in the black and the black was in the red and I had not noticed, though I was immediately curious about the current draw and smoke. Oops. Damage was limited to all the capacitors that I'd just changed and the TL082 op amps that I hadn't tested yet anyway. Somehow all the 2N5160 and 2N5109s were fine, even the BFR91 that drives all that was working. When I changed all the charcoal parts it worked perfectly with about 25dB gain and >20dBm output before hitting gain compression. I'd had a lucky escape that had also resulted in the board working better than it did before I set it on fire. Nice.

This morning I've reassembled the generator and done some performance tests. It's really clean and quiet, all the ranges below 60-104 use dividers to some extent so phase noise is impressively low (I can only measure this as residual FM at the moment, but it's very noticeably better than my Marconi 2019A**). AM noise is also reasonable, with the tiniest smidgen of mains hum that isn't enough to get the 8901B to show anything more than 0.01% AM depth.

I should add that everything about working on this generator feels like a test, it's all so annoying that it must be a deliberate joke. Funny stuff like the LEDs on the front panel "fit them by pushing them into their holes, reassembling the entire front panel, then soldering the leads?" YES. REALLY. (I didn't, I did it by eye because I had to do about 15 of them and it's not really possible to fit them dry, they just fall out, you'd have to tack solder them then resolder later but it's also impossible to move them around and adjust them so perhaps they used a jig at the factory?!). Flying leads on every board down into a screening can that you can't reach into, even with long nose pliers holding the plugs, meaning it's almost impossible to connect or disconnect any of the plugs. No extraction levers on the individual cards, you have to use a screwdriver or similar hooked holes in the PCBs that are only nearly in the right place to lever them safely. Every part of it made me want to scream into the void but I kind of enjoyed the challenge. I have heard from someone who used to work on these that many were beyond economic repair because technicians beat them with hammers out of frustration.

I will update in a few days when I rebuild the front panel with non-dazzle LEDs.

Here's the filter

[tggzzz]

Heroic indeed!

That logic implementation is interestingly old-skool! I wonder if they really did it to keep the noise down? Conceivable, but a bit perverse.

[Zenith]

Definitely character building, although I believe you could have talked the seller down to £10, as I doubt he would have wanted to take it home. Then you could have built your character for £15 less, which is also character building in itself.

Per ardua ad astra!

[synx508]

Heroic indeed!

That logic implementation is interestingly old-skool! I wonder if they really did it to keep the noise down? Conceivable, but a bit perverse.

It's the only reason that seems realistic, according to a few replies I got on Mastodon or Bluesky this was built for EMC performance above everything else so it makes sense. The only clocks inside the enclosure that run continuously are synchronous with the signal. The others are, I think, asynchronous but only appear fleetingly during keypresses and the slow astable that flashes the display when its out of lock or has an invalid selection (like showing the AM depth on the meter when AM isn't selected).
The custom ICs are interesting, too, the front panel with the LEDs and switches has two that look as if they came from Ferranti. Presumably ULAs or something similar. These do the needlessly complicated keyboard scan and setting latching stuff but they're not the machine's main logic or the store that holds the current frequency. That is handed off to the PROMs and some TTL and another RACAL custom IC in a white ceramic and gold package.
The logic holds two states, the local state and the remote state, there's an optional IEE488 adapter and an optional external remote/store, that actually looks rather neat - there's one on ebay at the moment. It gives me the impression that it was designed for a specific role.
Another fun thing about the 9084 is that it has ECL "fast" dividers, TTL "slow" dividers and CMOS logic, as well as an incomprehensibly large number of loops. There are loads and loads of 4v7 Zeners acting as level shifters for the logic that is sitting on the -5.2V rail (no, it's not just ECL down there, they've managed to use all three types there too)

Here's just one of the loops, which itself is more than one loop and a bit confusing.

racal9084-transfer-loop.png

[tggzzz]

Heroic indeed!

That logic implementation is interestingly old-skool! I wonder if they really did it to keep the noise down? Conceivable, but a bit perverse.

It's the only reason that seems realistic, according to a few replies I got on Mastodon or Bluesky this was built for EMC performance above everything else so it makes sense. The only clocks inside the enclosure that run continuously are synchronous with the signal. The others are, I think, asynchronous but only appear fleetingly during keypresses and the slow astable that flashes the display when its out of lock or has an invalid selection (like showing the AM depth on the meter when AM isn't selected).
The custom ICs are interesting, too, the front panel with the LEDs and switches has two that look as if they came from Ferranti. Presumably ULAs or something similar. These do the needlessly complicated keyboard scan and setting latching stuff but they're not the machine's main logic or the store that holds the current frequency.

Presumably the key scanning is done with logic that has very slow transition times. As far as I can tell, the ULAs were effectively TTL devices, some designed to be low power and slow, some higher power and source/sinking 40mA.

Are the LEDs multiplexed? If so, a 18181818 display would have some interesting current steps!

That is handed off to the PROMs and some TTL and another RACAL custom IC in a white ceramic and gold package.
The logic holds two states, the local state and the remote state, there's an optional IEE488 adapter and an optional external remote/store, that actually looks rather neat - there's one on ebay at the moment. It gives me the impression that it was designed for a specific role.
Another fun thing about the 9084 is that it has ECL "fast" dividers, TTL "slow" dividers and CMOS logic, as well as an incomprehensibly large number of loops. There are loads and loads of 4v7 Zeners acting as level shifters for the logic that is sitting on the -5.2V rail (no, it's not just ECL down there, they've managed to use all three types there too)

The various ECL/CML families can be surprisingly low noise, since the current changes are less than in TTL and CMOS.

I'd expect several logic families. I'd also expect different logic levels, especially if JFETs are used as analogue switches. (Yesterday I learned that you can still get new JFETS, and more, from https://www.linearsystems.com/jfetswitches )

Some of my bench DVMs have TTL sitting at -19V; can't remember whether that is "Vcc" or "Gnd"". Either way, just clipping a scope probe on in the traditional way causes... disappointment.

[tggzzz]

Here's the filter

I'll bet that's fun to align, with lots of
39: adjust Ш to get the best ф
40: adjust Ѭ to get the best Ѿ
41: adjust Ґ to get the best Ꙭ
42: measure, and if not good enough, repeat steps 39-41.

[synx508]

The filter alignment is quite simple on paper, the HF end uses two variable inductors and the LF uses variable capacitors. There are only two things to tweak for each band and the suggested method is to connect a spectrum analyser with a tracking generator and then for each band tune for minimum output at a corrner frequency and a null at a notch frequency. It does warn of interaction, though, but this seems to be just between the two controls and not adjacent bands. Less simple in the practical sense because you can't tune it "live" as the adjustments are inaccessible when it's in its screened box and if removed from the box with an extender card it'd detune, making it pointless to remove it.

None of the digit displays are muxed by the way, but they're HP BCD displays so they use fewer lines than a 7-segment would. The LEDs for the buttons are semi-muxed, but one LED is lit at once so it doesn't need to scan, the clock just stops once the Ferranti chip has found the right line and the side effect is that the LED remains lit.

I feel I should mention that, as with other RACAL equipment that I've worked on, the RACAL soldering approach is "long bent-over leads with distorted ends that don't fit back through the very tight PCB holes so it's a good idea to snip the ends off or risk pulling the terrible through hole metalwork out".

[Zenith]

I feel I should mention that, as with other RACAL equipment that I've worked on, the RACAL soldering approach is "long bent-over leads with distorted ends that don't fit back through the very tight PCB holes so it's a good idea to snip the ends off or risk pulling the terrible through hole metalwork out".

Racal boards are horrible in that way. The components can be packed together so it's hard access the leads to snip them off. I've sometimes crushed the components, although there may not be room to do that. As you say, the leads are a very tight fit in the holes and they bent the ends over - to make life more interesting.

[tggzzz]

It does warn of interaction, though, but this seems to be just between the two controls and not adjacent bands. Less simple in the practical sense because you can't tune it "live" as the adjustments are inaccessible when it's in its screened box and if removed from the box with an extender card it'd detune, making it pointless to remove it.

That is perversely awkward!

While we have grown used to such perversity with software, I'm surprised to find it in hardware. Maybe they had test jigs that enabled it to be fiddled with inside a fake case sufficient only to mimic any such detuning.

[synx508]

That is perversely awkward!

Oh yes, but not if you treat the whole instrument as a warped test. Want to remove the front panel eh? That'll be remove rear chassis, the top and bottom panels and slide the side trim back to reveal the front panel screws, then remove the attenuator knob and the RF output socket. Oh, you'll also need to remove the screws holding the mains power switch cable clips and remove the 50-way remote socket from the rear panel and pass it through its own cutout (it barely fits, it felt like a puzzle). Then disconnect from the motherboard several RF connectors for stuff like the output level vernier and the two external modulation inputs (did I mention that you can do AM and FM at the same time like a perfectly normal person?). Not so fast, now you need to remove the two ribbon cable IDC connectors from their sockets, thankfully these have the proper eject levers, but no keying (though it's hard to get them wrong). Now the front panel can be eased out from the two sides. The two front panel PCBs are stacked up on spacer pillars and the lower one with the keyboard has extra screws to impart that high build quality feel to the push-buttons. It almost did this, except all the extra rigidity screws were made of a very soft metal (some sort of terrible steel, I suppose) that had made sweet chemical bonds with the aluminium front panel - the heads turned, the shafts stayed put, still at least I didn't have to drill them out. Now I have several headless screws and screwless heads.

[Zenith]

It does warn of interaction, though, but this seems to be just between the two controls and not adjacent bands. Less simple in the practical sense because you can't tune it "live" as the adjustments are inaccessible when it's in its screened box and if removed from the box with an extender card it'd detune, making it pointless to remove it.

That is perversely awkward!

While we have grown used to such perversity with software, I'm surprised to find it in hardware. Maybe they had test jigs that enabled it to be fiddled with inside a fake case sufficient only to mimic any such detuning.

The only thing I've come across like that was an early model Hameg HM-1005 100MHz oscilloscope. Some of the boards were slightly different to the ones shown in the available service manual, and the shields over the attenuator and pre-amp had no access holes. They must have been set up using a jig in the factory, and it was assumed they'd be good ever after. That scope probably cost around £750 in the late 80s. The later ones were different and had access holes as normal. The Racal 9084 was definitely aimed at the professional market and I'd guess cost a lot more.

Professional users would very likely send equipment like that off for calibration regularly. Maybe it was a would be clever wheeze. An attempt to corner customers to send it to Racal for calibration, using their special fake case set up jig.

[synx508]

This morning I measured the OCXO's current draw while watching the frequency on the counter, my theory that it was cold was wrong. It's actually really well insulated and even after the specified warm-up time the outside of the oscillator remains cold. It runs from 5V and the cold current is 450mA, dropping to 160mA when it has warmed up.
I had tuned it roughly to the right place last week and after 15 minutes or so it had returned to the same spot and was holding that frequency without wobbles.

It is working fine. This is so unexpected.

I am now waiting for some sensible LEDs and a 4041UBE to arrive so I can unbodge the front panel LEDs and the filter logic buffer. If that goes according to plan I may take a look at the remaining issue with the "x2 mode" (turns off AM capability and uses the headroom in the amplifier to double the CW/FM power, actually a rubbish feature that I find a bit annoying but I should repair it) and maybe try to make the switch action a bit more pleasant.

[synx508]

Mouser delivered my parts (well FedEx, really), so the proper diffuse green LEDs are in, switches freed up a bit (though unfortunately at a cost of a bit of plastic snapping off and one of the switches is now drooping slightly but it seems to be cosmetic), brand news 4041UBE fitted and bonus surprise self-repair of the x2 power feature. It can produce 19.5dBm output in FM/CW mode, which feels like an almost dangerous amount of power.