Finished: 16-05-2022

HP 8614A

0.8 - 2.4 GHz klystron oscillator

The HP 8614A was a technological marvel for its time. Introduced in the year 1962, this generator was based on a mechanically tunable klystron oscillator. The big selling point of this machine was that the power output was incredibly flat throughout the range of the instrument. Over its entire frequency range, the power level stayed within 0.5dB, eliminating readjustments of the power output when changing to a different frequency.

Other useful features of this generator include AM and FM modulation. Meaning that both the output can be pulsed with microsecond precision, but it can also be locked to a more accurate master oscillator. This made this generator a very versatile piece of kit for any RF laboratory.

First impressions

With a lot of vintage equipment, a sad but true reality is that it usually won't survive for longer than a couple decades without any intervention. Although the building quality of HP equipment is generally very high, not every component ages equally, and a significant malfunction will render it out of service. Once this device is turned on, the only use it currently has is being a very weak nightlight and a suboptimal muffled fan. There's no power output and the needle of the power indicator is about stationary. Let's have a look and see what's inside, shall we?

Troubleshooting

When troubleshooting, there are a couple tips that will make the likelyhood larger for a successful repair.

Before attempting to replace components, firstly:

Try to understand how the piece of equipment works. There's no use digging around in a piece of electronics if you don't know what it does, how it is operated, and what tricks were used to achieve its goals.
Equipped with this knowledge, now you can actually properly identify if the machine does or doesn't do what it is expected to do.
Then isolate all parts that work from the ones that don't by measuring. This prevents hastily made conclusions, resulting in replacing the wrong parts.
And finally, use common sense. Is the power chord plugged in? Is the power supply still working correctly? What is the condition of the electrolytic capacitors? To name a few examples.

To be frank, in my excitement I also struggle following up these tips myself...

A quick explanation about klystrons

As earlier stated, the signal generator is based on a klystron oscillator, a 'reflex klystron' to be more precise. Now what exactly is a reflex klystron?

At first glance, a klystron looks pretty confusing. Thankfully, it can reduced fairly easily to a lot of recognisable parts. Its two main components are the electron gun assembly, and the cavity.

Firstly, the electrons are boiled off a filament through thermionic emission with the supplyvoltage Vf. As the filament, or cathode, is brought to a really negative potential with Vc with respect to the anode, an accelerating field is created for the electron from the filament to the anode. Then the electrons speed through the cavity opening towards the repeller electrode. As this electrode is negatively biased using Vr with respect to the cathode, the electrons are repelled back towards the cavity.

The 'ALC'

One of the innovations in this generator was effectively using a klystron to generate the RF. The other innovation was the fact how linear the power output is of the generator over its frequency range. The mechanism that is used to achieve this is the 'Automatic Leveling Circuit' or 'ÁLC' for short. The block diagram can be seen below:

The working principle of the ALC is fairly straightforward. The signal from the klystron goes through both attenuators. Behind the ALC attenuator, there's a detector diode. The signal after this diode will be a negative voltage proportional to the power output of the signal after the attenuator. This signal travels to the meter amplifier, which bumps up the weak signal from the detector a bit, and drives the panel meter indicating the power the generator is outputting. This voltage is compared with a target voltage dialed in by the 'ALC CAL.' knob on the front panel. This error term is fed into an emitter follower and a set of amplifiers, changing the bias current in the attenuators. This feedback stabilises the output power after the ALC attenuator. Because the impedance of the diode is not the same for every frequency, the ALC loop is kept seperated from the RF output with a seperate pickup antenna in the cavity. This means that not only the RF output of the generator is kept stable throughout the frequency range of the generator, but also the impedance.

Diagnosis

As earlier stated, this generator does not do much besides functioning as an overengineered heater. The following behaviour could be observed: once the device is turned on, after about 10 seconds the power indicator shoots up to about -4 dBm. With the ALC CAL. knob, this can be turned up all the way up to 1 dBm. After another 20 seconds, however, the indicated power drops to 0.

According to the previous owner, the generator worked perfectly with an external detector connected to the output of the generator, inserting the signal back into the 'remote leveling input'. This would bypass the ALC attenuator and detector diode, meaning that the issue is probably located somewhere in that part of the circuitry. The generator is furthermore untouched and the electrolytic capacitors have never been changed.

Detector diode

Now that it's known that there's an issue somewhere surrounding the ALC attenuator and detector diode, it would be an easy first step to measure the detector diode whether it works properly. It is located inside the housing attached to the attenuator block in the centre of the generator:

After taking out the diode, it measured to be an open circuit. The failure makes sense. This is a point contact diode, meaning only a very thin and small whisker is attached to the semiconductor crystal. This might have either burnt up as an electrical failure, or it broke due to mechanical failure. Either way, this diode has to be replaced. Sadly, this is easier said than done... I wasn't able to find a partnumber in the manual, and I couldn't find anything with a similar package on the internet. I went to my go-to forum, 'het Nederlands Transistorforum', to ask whether people reconize the diode. Little people were able to help me and provide me with a proper answer... until one kind gentleman sent me a message telling me he had a NOS detector diode, the 1N21WE, laying around that he didn't need anymore! The diode was sent to me through postage, and the project could continue. Whilst the shape of the diode is completely different, the package is fairly standard and might it burn up in the future, getting a replacement shouldn't be too much of a hassle.

Turning a new detector case

With the new diode in my posession, some thorough eyeballing of the old casing and the new diode, and a fruitful afternoon of breaking my back behind a lathe in a friend's metalworking shop, this is the solution I came up with:

Firstly, a collar for the new diode was turned and glued to the exterior. This reliefs the strain from the contacts within the attenuator casing. The connector for the diode was replaced with a new one so it could properly be attached to the attenuator module. Now the new detector case could be turned. A couple compromises had to be made in the case's design because I only had access to a 4 jaw chuck simply because it was permanently stuck in the lathe. This resulted in the lack of the ability to grip a workpiece with a smaller diameter than 5 centimeters. This meant that every part had to be completely turned in one sitting. Besides that, there were no tools around to turn big threads on the lathe using the autofeed. This meant that the cap of the case had to be mounted with setscrews rather than one big thread.

A rookie mistake

WHAT?!? You want to tell me that an American piece of engineering uses... imperial threads?? Who on earth would have guessed such a ridiculous thing?? Well... I sure didn't, and this is a mistake that bit me back quite a bit... Not only did my feedthrough capacitor in the detector have imperial threads, it even seems to be the NORM!! Calm down... it's gonna be alright...

So after tapping M5 metric threads in the cap of my freshly turned detector case, a new feed-through capacitor had to be carefully fabricated. This was done by taking a METRIC brass screw, drilling a hole all the way through, and soldering in a 10 pF solderable feed-through capacitor. The lead had to be extended with an extra bit of copper, and the silverplated spring was moved from the imperial to the metric capacitor:

Further testing

After that fiasco was finally resolved, the detector could be put together, and the attenuator assembly was built back into the generator:

Looks pretty acceptable right? I thought so too. Though, it's not looks that make this generator function better, we are going for functionality! Sadly, it still did not work and the generator was still not outputting any signal. Let's look for further issues.

Replacing electrolytic capacitors

Like I mentioned earlier, I suffer from a severe case of excitement every now and then. This caused me to skip a lot of much simpler steps in the troubleshooting process. Although a big problem got solved here that had to be solved anyways, there were a couple more banal issues to be resolved. Whilst all power supply voltages measured somewhat within broad tolerances, there was one I missed that was off by a severe amount. The filaments of the klystron, together with a couple tubes in the scquare wave oscillator for internal AM modulation, are fed using DC rather than AC in the rest of the instrument. This voltage is created by taking the negative 320V supply and feeding that through a seperated secondary winding of the power transformer. This adds a 12.6V peak-to-peak AC voltage on top of that supply. This gets further rectified and is regulated using a zener voltage reference. Stupidly, rather than adding -6.1V to the original supply voltage, only about 1/3rd made it through with an obscenely large ripple. This is clearly the work of a faulty capacitor. After quickly and dirtily grafting in a new capacitor, the issue was resolved, meaning the next crucial step of this repair should be getting rid and replacing every electrolytic capacitor.

Whilst this might appear a bit unfortunate, old electrolytic capacitors don't have an eternal life expectancy. They only last a couple decades, and whilst there's a lot of discussion whether to replace or not to replace ones that work fine, I rather not risk any short-circuits. The extra effort pays off in the long run! Besides this, the filament in the klystron has been operating on a very low voltage, meaning it will still have quite some life left. Let's get rid of those crusty old buckets, and make some mounting plates for the modern replacements:

The mounting plates were fabricated from regular PCB material. The traces were painted on by hand with a matchstick and some lacquer. Then they were etched using sodium persulfate. The rivets of the old capacitors were drilled out of the chassis, and these new mounting plates were pop riveted back in.

With the big cans replaced and the majority of the work done, replacing the other capacitors is done in no-time. Whilst the original capacitors are axial, I still decided to replace them with radial ones. Although the footprint differs, the pros outweigh the cons. Axial capacitors might be better shielded when the negative terminal is connected to ground, the radial ones are simply much cheaper. They are so cheap, in fact, that for the same price of the cheapest axial capacitor I can get my hands on, I can get a radial one that has over 10 times the life expectancy from a better known brand (in this case, Europe Chemi-Con). It is very important to note, however, to get ones with long leads. Else it can be quite a chore having to extend them before putting them back into the PCB. Besides this, it never hurts to insulate the leads using some shrink tubing or the insulation of a stripped wire:

Quick calibration

Having replaced the capacitors, it is important to check all voltages and do needed readjustments. The correct voltages, tolerances, and measurement points are conveniently listed in the manual:

For the -350 volt supply, it was easier to simply measure the -670 volt line in the power supply. Let's establish the boundaries:

The -670 volt supply should be between -663 and -677 volts.
The -320 volt supply should be between -315 and -325 volts.
The -212 volt supply should be between -207 and -217 volts.
The +20 volt supply should be between +19.9 and +20.1 volts.
The filament supply should be between -6.05 and -6.25 volts.

I'd be nitpicking if I'd comment negatively on these values. They are thoroughly within the specified calibration values! Now the voltages are properly dialled in, and the filament voltages in the square wave generator part are back to specs, let's do a brief check and calibration to see whether everything works as intended.

According to the manual the duty-cycle should be between 45-55%. This is what I managed to crank out:

These repelled electrons hit the wall, and get absorbed into the edges of the cavity. The centre of the cavity is folded inwards resulting in two parallel circular plates with a hole in the middle, acting as a capacitor (purple). Due to the skin effect caused by the high frequency signals, the inner surface of the lobes in the cavity effectively form an inductor (green). Therefore, a cavity can be simplified and approached as an LC tank!

Oscillations will rapidly accumulate inside the cavity, resulting in a rapidly fluctuating electric field parallel to the electron beam in the centre of the cavity. An accompanying magnetic field will also be generated around the centre in the plane of the cavity. The fluctuating electric field will either accelerate (red) or decelerate (blue) the electrons in the electron beam, bunching the electrons together in small groups. When the voltage of the repeller electrode is tuned correctly, the electrons will be sent back to hit the cavity wall at (3/4 + k)th (with k being 0,1,2...) the period of the desired frequency. This way the klystron can be operated in multiple repeller modes, increasing the frequency range!

As now the beam current is fluctuating exactly at the tuned frequency of the klystron, the oscillations are further kicked up like a swing and becomes self-sustaining. The fluctuating magnetic fields are eventually picked up by a pickup loop so the signal can be used outside of the cavity. By changing the physical size of the cavity, the frequency can be changed of the oscillations. For the 8614A, the inductance is changed by lengthening or shortening the cavity. The repeller voltage has to be changed accordingly, to maximize the output power. The amount of output power can be dialled in by pushing the pickup antenna further into the cavity.

PIN modulators

Okay, at this stage I can't really hide from this anymore. Let's address the elephant in the room. This generator is STILL not doing what it is supposed to do. The detector is replaced with a fresh new working diode, yet there is still no output whatsoever... There is an uncalibrated output present on the generator, being a third probe in the cavity that sends the signal out directly to be measured at the front. I got to borrow a crystek branded RF detector from my workplace, and after a quick and dirty test, attaching the output to my oscilloscope, the klystron appears to be life and kicking and well within tolerances! Phew, what would I have looked silly if I did all this work to find out the only irreplacable part in the whole generator would be broken...

I ended up hooking up the biasing input of the PIN modulator in the ALC to ground, and hooking up my oscilloscope to the output of the freshly made detector. This should theoretically result in atleast somewhat of an output voltage, but it could be quickly concluded... no power whatsoever. My one but biggest fear became reality... Replacing the detector diode already was a bit of a pain, but repairing the modulators is atleast an order of magnitude worse. Ah well, we came this far, it would be anticlimactic to conclude the repair attempt here and let the generator collect dust before it gets scrapped for its high aluminium content!

PIN modulators, how do they work?

We have to get to know the dragon before we can slay the dragon. First you take it out for tea, have a chat, laugh a bit, earn its trust, and then you strike to swipe some of its loot! Now let US grab a cup of tea and have a look at these pesky PIN modulators. If you'd be here, I could talk your ears off your head about PIN diodes, but I think it would be best to let HP's engineers explain this matter to you:

So to summarize really quickly, in this setup the PIN diodes are being used as voltage variable resistors. The higher the current through these diodes, the better they start to conduct, the less power moves through the transmission line. The high-pass filters on either end of the modulators are used to filter out any DC component introduced by the biasing of the diodes, the low-pass filter in the biasing line is used to keep RF out of the sensitive ALCircuitry.

Now, in this manual is stated that this is the only non-servicable part and when there are any issues with this module, it should be sent back to HP. And whilst Agilents support on older equipment has been pretty fantastic, even supplying these maintainance manuals through their database for free, it is to be expected that they won't be servicing this part anymore.

The reason why these modulators aren't home-servicable is the nature of the used PIN diodes. They were still in their early stages and the silicon is placed in the transmission line without any kind of protection whatsoever. There are some humidity absorbing beads present, but they are all deteriorated and amber coloured after 60 years of existing. The diodes probably broke down due to humidity, hence the modulator stopped working altogether. Let's be a bit mischievous and break into the modulators anyways, the diodes need to be replaced after all.

Onslaught

It feels a bit unholy knowingly doing unreturnable damage, but even if this doesn't work out, it will still educate. Firstly a bunch of screws had to be cranked open. These were all hex key screws... some imperial size... so I ended up filing down a metric hex key so it fit snugly into the screw. Wait... you thought the unholy part was ruining the PIN diodes? Nah, they were already ruined, converting a perfectly fine metric hex key into an imperial one is the unholy damage I was talking about! All the felonies I am committing to repair this machine...

But I digress... Let's have a look at the guts of the modulator:

Now it gets really interesting. Here we can see some very clever old-skool engineering going on. Let's have a little closer look at the diodes:

Now let's pay some attention to the distance in-between each and every diode. In the explanation is stated that the diodes should be 1/4th wavelength apart. Is that the case?

The distance between each diode assembly is about 28 millimeters. If that's 1/4 wavelength, then the entire wavelenght should be 28 * 4 = 122 millimeters, or 0.122 meters. To calculate the frequency, knowing the wavelength, we can use the formula: f = c/λ. We know that the wavelength, λ, is 0.122 meters. c is the constant of the speed of light, which is 299792458 m/s. This gives us a frequency of 2.46 GHz! This is at the highest end of the range of which the generator can put out.

Now let's zoom into the filters at either end:

These assemblies aren't just diodes, they're a whole collection of different components integrated in one tube! At the top, there is a tiny blue chip with a golden dot in the middle. This is actually the diode itself, bare, and unprotected from the elements. The tube itself acts as one of the plates of a capacitor. It is directly connected to the anode of the diode. The cylinder is coated with a non-conductive layer. In the cylinder there's a small resistor with one end sticking out. Once this cylinder is inserted in one of the holes of the assembly, the resistor touches a little springy plateau. This ensures that when the transmission line is mounted on top of the diodes, the golden electrode mounted on the chip gets pushed actively against the transmission line and makes proper contact. The entire tube the cylinder slides into is actually completely grounded, automatically making up the second plate of the capacitor, and connecting the resistor to ground. Very clever!

The filters at the input and output are highpass filters. We know that the higher the frequency, the lower the impedance of a capacitor and the higher the impedance of an inductor. This means that more RF signal will be diverted through the output in this circuit, whilst less will be moved towards ground. The opposite is true for the DC biasing signal. The traces in the picture that would be on the top PCB in the assembly are turned transparent so either side can be seen simultaneously. C2 is constructed by a piece of dielectric insulating film between the transmission line on the top PCB and the conductor on the bottom PCB. The signal therefore travels from the top PCB to the bottom PCB. On the bottom PCB, theres a thin copper strip that goes to the groundplane. This is the inductor. Finally, C1 is actually constructed from two capacitors in series. The traces are insulated by the same piece of dielectric film. The signal first moves back to the conductor on the top PCB and almost immediately goes back to the bottom PCB, where the signal is then taken out of the module by the soldered lug of the connector.

Finally, lets look closely to the low-pass filter:

This part is a little more obscure. I marked the most obvious part in the schematic, being the inductor. They employed a really funky trick here. To make proper contact with the transmission line, a spring is pressed against the conductor directly under the transmission line. Now, we all know a spring is a wound piece of wire, so it can also double as an inductor! But that's not the clever part, it's the fact they also thought of dropping in a chunk of ferrite to ramp up the inducatnce of the spring that boggled my mind. I'm fairly sure the engineer who came up with this must have mentioned this life-hack to every collegue throughout his entire career. And rightfully so! As for the rest of the filter, I think the pad that makes contact with the other end of the spring probably forms a capacitor with the case, but from there I couldn't really see anything interesting anymore without being destructive. The final capacitor in this filter is probably between the conductor of the wire coming out and the case, using the insulation as dielectric, but that is just all speculation.

PIN modulator repair

So now we know how the modulator functions, an attempt can be made to fix this part as well. We know the diodes have to be replaced somehow. Let's not be bothered by thinking up a way to mount the diodes to the cylinders, and first be concerned with what specs have to be dealt with. In the explanation of the modulator, the writer mentions that the resistance of the PIN diodes is 5000 ohms naturally. Once the forward bias is 1/2 mA, the diode should be 30 ohms. That's about it. I think it would already be an achievement to find a single diode that comes close to this single spec, so I am not going to be picky about all the other figures you can mention surrounding diodes. After a search that felt like eternity, I found a suitable diode that was in stock at a supplier in my area: the BAR64. These work up to 6GHz, so they are well within specs regarding that, but most importantly, the forward resistance of the diode is about 30 ohms at a forward bias of 1/2mA. I decided on getting the BAR64-02V specifically as it comes in the SC79 package. The smallest package I could get a hold of. An added big benefit of replacing all of the diodes is that the modulator will now be moisture resistant!

New resistors

To document the resistor and capacitance values of the cylinder assemblies (in the case I ever want to build my own attenuator), I noticed that the resistors actually drifted pretty severely out of tolerance. This is actually an excuse to replace them, as because they are so tiny, I lost one, so I might as well just replace them all. The original use of shortly clipped through-hole resistors I think is a little bit dodgy. In this case I don't think it is an unnecessary luxury to use a modern substitute. Therefore I decided to go with a set of military-grade thin-film MELF resistors. Let's first look at what values are needed, and what size the resistors should be:

It can be seen that the engineers decided on three different values for resistors. 2 x 4k7, 2 x 3k3 and 3 x 680R. They all have a silver band for tolerance, meaning that the values should be within 10% of the given markings. Here it is also more clearly visible that the resistor are clipped off through hole assembly carbon resistors. Rather than using carbon resistors, I am going with MELF resistors instead. These resistors are actually very similar to regular through hole metal-film resisors. They also consist of a resistive material vapourised onto a ceramic cylinder. And these also have end caps crimped onto either end to make proper contact with the resistive layer. Though, the big difference is, rather than attaching leads onto these endcaps, the metal is actually kept exposed, resulting into a cylindrical SMD resistor:

Now this is a large improvement, as the point of contact with the resistor's conductors are significantly bigger. This means it is mechanically less stressful to the resistor. Knowing the measurements of the original resistors, being 1.6 mm by 5.5mm, Vishay's MMA 0204 series resistors were most easily accessible to me (don't worry, you'll read later mostly physical dimensions are the biggest constraint, almost everything will do). Whilst this is all good and dandy, there are also some cons using these. I could only get a hold of the standard series MELF resistors. These are laser trimmed into a spiral shape. This means that these resistors also act somewhat as a small inductor, and whilst at low frequencies this does not really cause any issues, at RF it can become problematic. Whilst Vishay does produce a series of MELF resistors dedicated to RF, these are incredibly hard to find and almost impossible to obtain for an individual like me. If this does not work, the plan B would be to use an SMD chip resistor instead. It is easier to find these in different trimming patterns because they are significantly more widely used than MELF resistors. Vishay's R&D department published a useful paper regarding the differences in frequency response between different trimming patterns. A quick glance shows the following details:

In the top graph it can be seen that the bigger the physical size of the resistor and the higher the resistance, the worse the performance at higher frequencies. Whilst no explanation is given why the performance reduces in the paper, my best guess is that in both cases, the larger the resistance/physical resistor size, the longer the path is from the one terminal to the other, resulting in a larger inductance. As for the way the resistors are trimmed, the most beneficial way is to use Edge Sense trimming. This is for the same reasons as resistor size/value. Rather than trimming notches in the film, lengthening the distance between both terminals, the film gets shaved away from the sides, keeping the path as short as possible. This results in the lowest inductance.

These MELFs are laser trimmed with a spiral shaped groove. This is the worst possible trimming regarding the relatively high inductance. The origial carbon resistors used in the original design outshine the metal film resistors in this aspect, because the construction of a carbon resistor is simply a straight rod of carbon with legs pressed into either end:

For now, let's just try out how the MELFs perform. If it doesn't work (which is highly unlikely as we will discover in the next section), we now know the critera how to find a substitute that will work better!

Capacitor measurements

Like mentioned earlier, knowing the capacitance of the cylinders is not critical to this repair. It is just interesting to know how they relate to the resistor values, and they might come in handy if I ever decide to build my own voltage variable attenuator in the future. They will also provide a better understanding regarding the protection resistors in series with the PIN diodes. Whilst these cylinders are marked with differetly coloured dots, it is not a conventional method of colour coding. This means that the capacitance of these cylinders have to be measured somehow. As I didn't have a functioning LCR meter at hand, I decided to use a signal generator (HP 4204A), an oscilloscope (HP 54501A) and a resistor (3k6) to determine the capacitance. The process was really easy. Using the resistor and the unkown capacitance, a first-order low-pass filter could be constructed. A 1 MHz signal was fed into one end of the resistor. The other end was hooked up to the oscilloscope, and the amplitude was dialed in to be exactly 2 volts:

To measure the capacitance, a shim was 'carefully' constructed by clipping off the end of a tiewrap. This replaces the resistor in the cylinder, and prevents the cylinder to be grounded by the resistor. Then the case of the attenuator assembly was connected to ground, and the lead of the resistor that is connected to the oscilloscope, was pressed against the top electrode of the cylinder. The cylinder has to be fully pushed into the attenuator assembly. The resistor's value was chosen in such way, that the amplitude of the signal sagged by a measureable amount when the top of the cylinder was being touched.

For every colour code, the amplitude measured at the output of the filter varied a little bit. This could be noted down in a table. To find the correct capacitances corresponding to these voltages, the same process can be done, but for capacitors with known values. Using a set of high accuracy capacitors as reference, these were the following values found using this method:

I understand my handwriting is poor. Let me clear it up with properly typed out text. From left to right the capacitances are as follows:

68pF - 47pF - 39pF - 56pF - 39pF - 47pF - 68pF

And the corresponding resistances:

4k7 - 3k3 - 680R - 680R - 680R - 3k3 - 4k7

Now these values are known, we can get a better understanding how the 'protection' part works stated in the description in the manual. Let's take the centre frequency the generator can provide, 1.6GHz. Then we randomly pick one of the branches with the values 47pF for the capacitor, and 3k3 for the resistor. The impedance of the capacitor can be calculated using XC = 1/(ω C) = 1/(2 π f C). we know that f =1.6GHz, and C = 47pF. This means that the impedance of the capacitor is about 2 ohms! This is significantly smaller than the resistor parallel to the capacitor, which is very interesting. This means that for low frequencies, the impedance of the RC-pair is very high (XRC = R ). For the RF produced by the generator, the impedace of the RC-pair is very low (XRC = XC ≈ 0 ). This will protect the diodes for excessive DC current, whilst for the RF component of the input signal, the working of the diodes is almost completely unaffected.

With this knowledge, the conclusion can also be pulled that the usage of the MELF resistors will actually not impair the functionality of the attenuator. The resistors only have to protect the diodes for DC current, and at such low frequencies, the influence of the inductance of the resistor is negligible! For high frequencies, the impedance of the inductor increases. This also doesn't really matter, because the very low impedance of the series capacitor overrules the rising impedance of the resistor's internal inductance!

The herculean attenuator repair effort

Once the diodes and MELF resistors were finally in my possession, a plan could be formulated to repair the attenuator. There will be another major benefit to doing this. The attenuator will now be moisture-proof! Because the silicon in the diodes won't be exposed to the elements anymore, and the porous carbon resistors are gone, the moisture eaters can be permanently removed without worrying about further damage to the device.

Let's first find a way how to replace the diodes. After a lot of thinking and testing, I found out that there is a lot of extra slack in the springs that push the cylinders against the transmission line. This means that extending the top of the cylinder with a small diode shouldn't have a drastic effect on the working of the attenuator. The following method was used:

First the little plateau was cut and filed off to give the new diode some extra space. Then with a good amout of flux, a droplet of soldering tin was added. With a pair of tweezers, the diode was pressed gently against the droplet, and the droplet was reheated. The cylinder was then dipped in isopropanol to give it a good clean. To fixate the diode, a collar of two-component epoxy glue was applied with a toothpick. Sadly, the heat of the soldering iron deteriorated the dielectric paint layer on the cylinders. To fixate this, a thin mist of varnish was put on whilst protecting the top connection with a piece of aluminium shielding tape. This finicky process had to be repeated a total of 14 times until every diode was replaced.

Now the diodes were in place, the MELF resistors could be prepared. As they happened to be 2 millimeters too short, a little extra copper was added to one side, and a bit of shrink tubing was used to increase the diameter of the MELF. This ensured the resistors sat stably in the cylinders:

After a bit of trial and error, the method I found that worked best to extend the length of the MELF was by filing the end of a 1.6 mm thick copper wire flat. Then I applied a thin layer of tin on the face of the wire, took the resistor with a pair of tweezers, and heated up the side of the wire. The resistor could then be pressed into the puddle of liquid tin and the soldering iron could be taken off. Then, using a small file, the sides of the copper wire were cleaned off, cut, and the face was filed smooth. Finally a small piece of shrink tubing could be put on.

Finally the modulator was put back together with the fresh new diodes and resistors!

Further testing

The modulator module was put back into the signal generator. After connecting all wires back up again, and extensive testing, the gutwrenching conclusion could be pulled that the signal generator was still not working properly. Nothing changed regarding the automatic leveling, but there were some slight improvements. The calibrated signal output did show a perfectly modulated signal when the "SQ WAVE" button was depressed. This atleast shows that the repair did not make things worse! Yet, there still was no output coming from the detector.

At this point I started wondering whether the pickup probes in the cavity were still working properly. I desoldered the biasing wire from the calibrated output attenuator, and clipped it to ground. This means that almost the full output power from the pickup probe should be sent to the front panel. There was a solid signal coming from the calibrated output probe like expected. When hooking up the signal from the ALC pickup probe to the input of the calibrated output attenuator, a strong signal was detected by the external RF detector. This concluded that the ALC pickup probe was still working properly.

With a freshly repaired modulator that is proven to work properly, and a strong signal coming from the pickup probe, attention was again put onto the newly made RF detector. The whole assembly was taken off again, and a female SMA connector was crudely attached to the output of the ALC modulator. The borrowed RF detector was then screwed onto the connector, and the bias wire was pulled to ground...

... now there's finally signal. This means that the new RF detector simply did not work at all!! In hindsight, I think this is mainly due to the tight space between the connector and the housing. These parts are so close together, it forms somewhat of a capacitor. As we have seen earlier under the head capacitor measurements, it doesn't take a very high capacitance at these frequencies to be very influential in the circuit (due to the low impedance). With a very low resistance to ground, it probably resulted in a very severe power loss.

Calling this unfortunate, I personally think is a severe understatement. But let's try something else! Instead of making the detector myself, the crystek one has done an amazing job so far. How about we find a way to mount it to the attenuator instead?

Detector replacement

At this point, you might be thinking... Is this generator ever going to work or what?? I admit, it has been taking a while, but let's not give up just yet! Whilst at this point, buying a scrap generator for parts might be the best solution monetarily speaking, one-to-one replacing parts isn't gonna teach us anything. Besides, I like myself a challange anyways!

To continue with this repair, an adapter plate was made from a piece of thin copper sheet. As the sheet was way too thin, using the back-end of a screwdriver and the hole of a spool of soldering tin, a bit of extra thickness was created. This was done by laying the plate on top of the spool, and rubbing the handle of the screwdriver firmly in the centre of the plate. This plate was then cut in a circle, and all the necessary holes were made. An SMA connector was soldered direclty onto the extending connector of the modulator, the copper plate was attached to the housing, and the connector was soldered stuck onto the plate.

Please save me the headache. I know, the soldering job looks absolutely horrendous. Even with a big fat soldering iron, the fact the copper sheet was directly attached to a massive aluminium heatsink made soldering near impossible. I'm not proud of this, but life can't only consists of peaks. Its the valleys that reminds you how wonderful the peaks are... right? Thankfully, it is atleast very solid and the bodge will be put away behind a cover.

With this done, the modulator module was mounted back into the generator for the millionth time. Turning the generator on and attaching the detector now provided a similar signal as before with significantly less noise. We're finally on the right track. Now there's one problem that shouldn't be overlooked: the output of the detector is positive, but we need it to be negative! This is inconventient, but let's fix this using some modern tech.

The analog inverter

The goal is easy, and the execution is not hard either. The output signal of the detector has to be inverted. This can be easily achieved using operational amplifiers. Any general purpose opamp will do, but the type I have a lot in stock of is the NE5532. This is a small 8 pin IC with two seperate low-noise opamps. This chip happens to be particularly popular in audio related devices, but that doesn't rule out its overall usefulness in other different types of circuitry. Because I have acess to two opamps, I decided on using the first one as a buffer and the second one as an inverting amplifier. Before showing the circuitry, however, a small elaboration on both opamp configurations would be helpful.

The first question we should ask is: what is an operational amplifier?

With its history in analog computers, this component has found its place in a LOT of different analog electronics for its versitility. It can be easily used for summing, subtracting, multiplying, dividing, differentiating, integrating, converting currents to voltages and back, and tonnes of other interesting things. The part itself has the following "transferfunction": V0 = g (V+ - V-). Where g is the open loop gain of the amplifying stage. This gain is generally very high (generally from something like 10000 to a million).

And whilst one can go very into the depth of the actual inner workings why an opamp does what an opamp does, for this particular post I dont think doing that would be interesting or necessary. In this case the following rudimentary and incomplete rule will suffice: 'an operational amplifier will try everything it can do to use its output to make the difference between both inputs 0'. This, ofcourse, can only be achieved when the output is in some way, shape or form connected to one of its inputs. This is called a feedback.

Now let's have a look to the most simple configuration of such a feedback: the buffer:

Using the earlier stated rule, analyzing the behaviour of the buffer is very easy. When a voltage of 5V is applied to the positive input of the operational amplifier, the voltage on the negative input has to be made 5V as well in order to make the difference 0V. As the output of the opamp is in direct contact with the negative input, the opamp's goal can be simply achieved by setting its output to 5V as well. This means that voltage Vin = Vout.

It is important to note that this conficuration does not work when the positive and negative inputs of the opamp are swapped. This is because the opamp's output will be positive when the error between both inputs is positive and vice versa. This means that in the case where both inputs would be swapped, rather than stabilizing, the output will end up drifiting to either the min or max voltage the opamp can supply (the power supply rails) because the opamp starts seeking for a stable outcome into the wrong direction. For instance, if 3V is put on the positive input, and 1V is initially on the negative, the error is +2V. The opamp will amplify this error, making the output more positive. In that case the voltage on the negative input rises, which makes the error decrease slowly until it stabilizes at 0.

Secondly, we can now take a look to the inverting amplifier:

In this case, two resistors are introduced into the circuit, and the positive input is hooked up to ground. This means that the goal of this opamp is to try to get 0V on the negative input using its output. In this circuit, a resistive divider is hidden. When R1 and R2 are the same, and voltage is applied to Vin , to ensure the negative input is 0V, Vo has to become -Vin! This means the circuit is able to invert the signal is put in. The gain can be set by changing the ratio of both resistors. This is according to formula g = - (R2 / R1).

Equipped with this knowledge, it can be very easily seen it does not take much to design a circuit that can invert the signal from the new(er) detector. The following circuit was used:

To briefly explain, the output signal of the detector goes into the first opamp. This opamp is used as a buffer, and mirrors the voltage put out by the detector without pulling a significant amount of current (which could interfere with the working of the detector). The second opamp is an inverting amplifier. The interesting part here is that a potentiometer is put into the feedback. No two resistors are manufactured the same, hence the potentiometer is used to correct the gain of the circuit, making sure it is exactly -1. The two capacitors are used for decoupling and have to be as close to the IC as possible. A balanced power supply of -5 and +5 volts is used to power the circuit. Knowing what to build now, the PCB could be etched, drilled and populated:

For ease of connecting and disconnecting, two SMA connectors were soldered as in and output to the circuitboard. Three small loops were made from resistor wire to be able to conveniently attach the wires of the power supply.

Balanced power supply

To power this inverter circuit, a balanced power supply is needed that spits out -5 and +5 volts. There are a million ways to go about this, but I decided to go with what is probably the most popular design for a regulated power supply. I based the design around the 78XX and 79XX regulators. These devices are incredibly simple, you put a voltage in that is higher than what you need, and a regulated voltage comes out the other side of which the rest is dissipated by the regulator as heat. The following circuit was made:

Firstly, on the primary side, 230V AC is connected from an outlet. The power goes through a fuse for safety. A transformer with two isolated windings steps the voltage down to 6V AC. They are connected similarly as if a 12V AC winding were used with a centre tap. The output of this transformer is then rectified using a full bridge rectifier. The rectified power is smoothened out with a large capacitor before it enters the regulators. After the regulator, a resistor is used to pull a small current to help the device regulate, and two capacitors are then connected to filter out any remaining noise. Now a PCB could be made:

For the people wondering how I made these circuitboards in particular, I hand-draw the traces. I spray a puddle of spray paint into a small container, and I sharpen a dowel in a pencil sharpner to make a point. This I soak for a while into the paint and carefully draw on the traces. For a one-off circuitboard, I found this to be very effective, quick, and efficient. It might not look as pretty, but the process has never failed me in the past. I also tend to fill in all empty space with a ground-plane. The reason for this is to preserve my etching solution. The less that has to be etched away, the more PCB's I can make! After the etching, the paint can be easily taken off with steel wool.

I am aware the shape of the circuitboard is not particularly pleasing. It was terribly time consuming to find a spot where I could properly add a PCB into the generator. I wanted to optimize the surface area so I didn't have to pack all components awfully close together, but that resulted in a very wacky shaped PCB in a very odd location. The power supply was mounted behind the left handle of the signal generator:

Whilst being a very tight squeeze, it did work out in the end. One thing I am a bit displeased with is the fact that the PCB touches one of the feedthrough capacitors for the klystron assembly. For the rest, there were fairly little complications. The 230V was taken from the main power supply transformer of the signal generator. The wires were attached to pins 1 and 4 of the transformer specifically. This does mean the option to switch the system to 110 volts will not work anymore. Though, I wasn't planning going on holidays with my signal generator anyways... it waaaay too heavy!

Yet another round of testing

Alright, another round of soul-crushing suspense. Is this repair gonna work? Honestly, my hopes weren't amazingly high. Let's crudely connect everything we built up into the generator, dial in the potentiometer of the inverter, and turn it on...

no smoke yet... tubes are warming up and glowing...

And what now?? The power output needle sways fiercely to the right?!?! Turning the ALC knob moves the needle around steadily?? You know what THAT means? THE GENERATOR IS ACTUALLY WORKING PROPERLY. Or better detailed, the ALC is able to do its job again!

Seeing this really took a lot of weight off my shoulders. All the time invested into this generator did not go up in smoke (both figuratively and literally). The next step now is to tidy everything up.

The last couple steps...

Now the parts arrived, the inverter PCB can be finally permanently built into the generator. The powersupply leads had to be extended a bit, and some general tidying up wouldn't hurt. The cable between the output of the detector and the input of the inverter was made of an old semi-rigid RG-402 cable. This cable definitely didn't have to be this fancy, but I snagged a really great deal of about 4 euros per meter of the stuff, so I might as well use it just for kicks.

The signal generator was closed again, and reassembled completely. After turning it on... everything still worked! Using a second detector on the calibrator output also showed fairly linear operation over the entire spectrum (regarding amplitude). This means that the rebuilt modulator is working fine together with the ALC. The square wave modulation also did its job, and the physical attenuator seemed to work perfectly. About any system I could calibrate with my current collection of labequipment is now perfectly in spec.

Finishing touches

The generator is physically in great condition, especially for the age. There's a small chip in the paint here and there, but the face of the generator is scratchless. A quick wipe-down with isopropanol quickly got rid of any dirt, grime or markings. One thing that drew my eye instantly, however, is the yellow plastic of the indicator bezels. They really jump out being a completely different tint as the lacquer of the face. To fix this, I used a method I saw being used a lot in vintage computer restoration called 'retrobrite'. Whilst being a commercial product, it is also fairly easy to do the same with stuff you can get from the local pharmacy. I used a solution of about 2-3% hydrogen peroxide, put the bezels with the liquid in a container and put them in the bright unobstructed sun to shower the bezels with UV. After two days, the bezels looked like they bleached a bit, let's have a look at how they turned out:

The difference might not be very obvious at first, but in real life it is incredibly noticable! I don't know how it would have originally looked like, but I hope and think I came quite close and will definitely use this process again in further projects.

Another small issue I noticed, after attaching new legs and a support bar to put the generator at a readable angle, is that the needle of the panel meter has a little bit of friction around the 0 dBm mark. It is still perfectly useable when the device is put flat on the table however. Besides, I am a bit too burnt out on the project to take everything apart again for a drop of oil. I will address this at a later time when I open it up in the future for the calibration of the mechanical dials (once I have a serious powermeter and frequency counter). As for now, I can live it.

Final words

This was a long lasting project, but a very enjoyable and rewarding experience. It is the first piece of RF gear I introduced to my lab, and I am very excited to dig down and do some fun projects with this generator! I learned a lot about the intrinsic working of various RF parts such as the detectors and voltage controlled attenuators. It is also fun to actually own a working reflex klystron, and to read up about a lot of theory written by the talented engineers from the pioneering days of this technology.

Now owning this piece of kit, the follow-up project will be building my own RF amplifier using another ancient technology that's still used to this very day: the Travelling Wave Tube (TWT). The tube is currently already shipped and on its way all the way from America!

When you spot any mistakes, things I missed or general improvements, please send me a message and I'll try to correct them asap!