Category Archives: Electronics

Electronics

Various

Measuring the Speed of Light: Fast light pulses, just about 10 m long

Leave a comment

Revisiting some circuits built 15(!) years back – feels more like 10 years back, high speed light pulse transmitters and receivers. These come in handy to measure the speed of light. Nowadays, I would rather send out some train of pulses with applied digital modulation, say, a pseudorandom sequence, and then calculate the time difference digitally, from the received signal. This would surely be be very much more sensitive, but also much less instructive than receiving light pulses discretely.

The first motivation years back was a request from a school to build several sets of light transmitters and receivers, so that students could measure the speed of light by determining the time of travel of light pules. It follows closely this wonderful article of Mr. Ehret, unfortunately, I only have it in German:

Bernhard EHRET: Messung der Lichtgeschwindigkeit mit Lichtimpulsen.
Journal: Praxis der Naturwissenschaften. Physik, Volume 41 (1992) 4, pages 17-35

I built several versions of the light transmitter, the first sending pulses of small power, at a current pulse amplitude that doesn’t hurt the LED for a long time. Good enough for some shorter distance measurements. As a reflector, a back-reflector used for light gates is recommended (cat’s eye reflector), approx. 30×30 cm will do fine.

The schematics all are similar: there is an input capacitor, a current limiting resistor to charge a pulse capacitor, and a transistor used as avalanche device to generate the pulse. The pulse duration is very much limited by the LED rather than the transistor.

Mr. EHRET recommended a BC546 resistor, and I have also tried this first. The collector-emitter breakdown of the BC546B which I used in the circuit was above 200 Volts, and with the energy stored at such voltages in a 4n7 capacitor (a FKP1 Wima pulse rated capacitor), the LEDs used were quickly destroyed.

These are the two test circuits, the received and the sender, similar to the suggested circuits from the journal article, but with voltage stabilization of the receiver (LM317T added to regulated the voltage to 18 V).

The receiver is a wide-band FET-input amplifier, the resistors at the source of the FET are quite important: together with the PNP transistor, these provide negative feedback to counteract the Miller capacitance effect of the FET. The amplifier circuit could be replaced by any modern fast transimpedance amplifier, but hard to beat the cost of a BF256 FET and a BF979 transistor. Both parts can be replaced by similar devices, eventually with some minor adjustment of the resistor values to adjust the currents appropriately.
The photodiode, rather than the BPW34 I used a fast Hamamatsu S5972, because I had several in stock from other projects. The S5972 operates at low voltages, wide spectral response also for visible light and has about 500 MHz bandwidth.

The transistor conductive breakdown will happen in less than 1 ns, but the pulse is about 43 ns long, because of the intrinsic properties of the LED, like, its inductance.

With the high breakdown voltage, it seems the light output is not much larger compared to the pulses generated by a 2N2369A at ~90 Volts, and there is considerably more ripple and noise at high breakdown voltages (white trance below is the BC546B circuit compared with the 2N2369A – yellow trace). Definitely, the classic 2N2369A (I have a stock of “JANTX” mil-spec tested devices and don’t recommend to use some copies or fake 2N2369 but rather metal TO-18 case old stock to be sure about the breakdown characteristics – modern versions and copies/fake 2N2369 may work well for common uses, but could have a completely different die inside, with random breakdown voltages).

The 2N2369A gives reliable breakdown performance, and the rate of the pulses can be controlled by the voltage applied to the circuit, adjust to about 20 kHz, which requires roughly 90 Volts.

This is the board, with the BC546B soldered, it is a rather modern version of that transistor. Maybe I should have tried some older stock BC546B, BC238B and so on, but whatever you use, the breakdown voltage should not exceed 150 Volts, otherwise, you are at risk of damaging the LED. With high efficiency LEDs and 90 Volts breakdown, there is no practically relevant aging of this circuit, but at 4n7 capacity and >100 Volts, some aging will show up eventually. The lifetime of the LED will likely also depend on the pulse current, but the test circuit was build with rather low inductance traces (short wires, barely 1 cm), so the currently will be close to the highest obtainable. The LED could be soldered in with shorter leads.

This is really the best choice, the 2n2369A!

The high voltage the drive the transmitter is conveniently generated by a YH11068A DC-DC converter, available from Aliexpress for just a few dollars. Even though this module has considerable noise, it doesn’t disturb the receiver, and because of the current limiting resistor in the pulse, has also no effect on the pulse performance.

It is such a nice experiment and easy to build, some small danger with the DC voltage, but most of that circuit is severely current limited. If you have a scope that is capable of ~50 MHz bandwidth, it is a nice experiment for nights outdoors, and even kids will be interested to set up the mirror and see how light travels at limited speed.

At the time, I did also experiment with laser diodes, red and orange diodes ranging up to 35 mW, but there is no particular advantage in using these. Focusing a laser beam over 10-20 meters of travel it also not an easy task. Better just to use a LED, and large lens of appropriate focal length (say, 200-250 mm).

In the archives I also found some of the earlier measurements (note the datecode: September 2011), using a 54720a scope with a 54721a plug-in (4 GSa/s with 1.1 GHz bandwidth). Some pulses are barely 25 ns wide when measures correctly, which is less than 10 meters of light. I.e., the light pulse is so short that it is just about 10 meters “long”. The light source was a HL6323MG laser diode, a 639nm, 30mW AlGaInP laser diode of Ushio corporation.

Various

Steep Edges: Revisiting some old tunnel diode circuits

Leave a comment

A phone call recently reminded me about some experiments a long time back, using some tunnel diodes to generate fast edges. Various circuits were built at the time, this is one of the final working versions. All soldered to just a piece of double-sided circuit board with two SMA connectors fitted.

The trigger signal, a reasonably fast square signal is provided through the BNC connector. The bias circuit is separated from the tunnel diode by a few nanoseconds of delay line, which helps to make the transition stable.

From these “old days” I still keep many tunnel diodes that were at that time ubiquitous from the left-overs of even older soviet times (made in the late 1970). There must have been several boxes of tunnel diodes that made it to the Western market. Maybe from some surplus military sale or give-away, because many of the diodes are “military grade”.

The diodes are for various currents, like, 5 mA and 10 mA, and there are both GaAs and Ge (germanium) types. For this circuit, I have the 1i305B mounted, a 10 mA fast-switching optimized Ge tunnel diode.

With a bias of about 8 volts, roughly 8 mA of current, I get the circuit to trigger nicely. At the BNC input, there is a protection circuit (a DC block and a 1N4148 diode to limit the amplitude). Tunnel diodes are susceptible to soldering heat, so be careful and use pliers to conduct away the heat from the terminals while soldering. I solder them “from the side of the board”, melting the solder on the board or connecting part, and then barely dipping in the pre-soldered terminal of the tunnel diode for merely a fraction of second, then cooling it with my finger. Slight burns may result but the diode will be safe.

The rise and fall times are nice, the triggering is stable and precise. I have sold all my super-fast oscilloscopes (used to own a HP 54750a 20 GHz scope and a even a 54720a 1 GHz real-time scope, but sold these several years ago after completion of certain projects that required these – I don’t want to start a museum of test equipment so I regularly sell equipment that is no longer needed, especially equipment that is expensive and prone to degradation).

For the tunnel diode, there is even a datasheet available, albeit in Russian.

Easy enough to read with Google Translate, but even without translation, the key properties are obvious: peak and valley current, current ratio, capacitance. Over the years I have used some of these tunnel diodes to repair old oscilloscopes and trigger circuits. Typically, to substitute diodes that are no longer available (certain types of General Electrics) or otherwise prohibitively expensive. So far, never any trouble finding a suitable Russian diode.

A quick schematic from the archives, sorry it is not written very clearly, but already more than 10 years old. Left SMA is the output – there is a 47 Ohms series resistor, and a small network at the outlet to suppress ringing. The delay line is just a piece of regular RG 174 (50 Ohm) cable.

There are also datasheets for a variety of other tunnel diodes available in the Manuals’ archive, contact me in case you can’t find them.

Various

Broncolor Primo: the photo flash is flashing again

Leave a comment

A very unusual and dangerous repair, a defective photo flash. This is a professional unit, for use in photo studios. Generally, these units work by charging a big capacitor bank to 300-400 VDC, and then discharging this energy in an instant through flash bulbs. This unit has 10 pcs. of 2450 µF capacitors, charged to 360 VDC. About 1600 Joule, which makes this unit very dangerous, I would strongly discourage anyone not familiar with power electronics to even open the case.

For its size, it is quite heavy unit, and has a 16 Amp fuse, it will recharge quickly, drawing substantial power for a short time.

Inside are 12 large capacitors, 2 for the charging voltage stabilization, 10 for the capacitor bank. After duly checking the caps and their safe discharge state, I tested them all, each individually. 3 were bad: 1 completely disconnected at the terminals, 2 with no capacity, worn out.

Flash capacitor need to withstand high discharge current, so we cannot just use any ordinary cap but need to source “flash capacitor” – found reasonably priced one from China, because there were other faults with this unit besides the capacitors, there was no reason to by expensive caps first, without knowing if this unit can be fixed at all. Unfortunately, there are no schematics available.

Studying the electronics, there is a primary thyristor (TXN1012) switch DC stage, a type of coarsely regulated power supply. This had a blown transistor in its control circuit. Failed by arcing. Fortunately, I was able to still read the color rings under a microscope, and replaced the part and an associated Zener diode. Also the thyristor and a MOSFET in the thyristor were replaced (the MOSFET tested good, but I didn’t want to take a risk).

The replacement caps have 2000 µF. 3x 2450=7350, 4x 2000=8000, so I decided to install 4 of the 2000 µF capacitors (2450 µF were not available easily). It results in about 5% higher energy in one bank, good enough.

After some hours of complicated failure search and repair, some very careful tests (checking if the caps load symmetrically, which they did), finally the green light of the “flash read” LED was lighting up.

The flash worked – but only for a short time, then: SMOKE from the flash box. Expecting the worst, opened it up right away. Surely, first removing the cables from the capacitor bank.

Inside of the flash box, 4 Xenon? flash bulbs, with spiral trigger electrodes. The high voltage trigger transformers are right inside the flash box. The smell is awkwardly familiar: an exploded Rifa safety capacitor. 0.1 µF, 1000 VDC.

Fortunately, it failed open, as it is supposed to, lots of bad smell but no damage or fire. There is one capacitor for each bulb, a total of 4 (2 sets).

I cleaned up the mess, ordered 4 original Rifa (now KEMET, but they still print “Rifa” on these), and soldered it all back together. The other 3 were electrically still good (tested for leak resistance several MOhm, and isolation test passed at 1000 VDC), but had many cracks so I replaced all capacitors.

Eventually, all is working again. There is a built-in studio light, quite fancy unit. Hope the repair will last for a while!

Various

ANENG MH15 Isolation Resistance Meter: a remarkable deal

Leave a comment

Typically, I don’t report about test equipment acquisition unless these are associated with repairs, but this time, I will make an exemption. I used to have an old isolation tester, but it has been playing up, and with the analog instrument, difficult to repair. So I checked about more recent instruments, mostly, to check electric installations and motors and similar, for isolation after repair. This requires a 1000 VDC test, because some isolation defects tend not to show up at low voltage.

After a quick search, I found an almost unbelievable offer, just about 25 EUR! Shipping included!

One week later, a badly packaged box arrived, in a plastic bag, but without major damage.

Inside, a handy soft case.

The instrument, the plastic is of good quality. The red plastic is a little hard so it will not absorb too much shock when you drop the device.

Immediately, I got some of the highest value resistors I have in stock, a Remix 1000 MOhm +-5 % resistor. This is good to 10 kV, so there is no problem or leakage when doing a 1000 V test (other resistors may be have voltage rating of 300 V or 500V).

At a first glace, very nice result. -2% within the tolerance of the part.

Checking the voltages, 1000 V is fairly accurate.

500 Volts, even better.

Some more deviation at 250 and 100 Volts.

The specifications are quite good. But eventually, the instrument is not used much to measure resistance, but to check for conduction when 1000 Volts are applied. Normally, it will either show >2000 MOhm, or a spark will fly and the high-voltage isolation test has failed.

For completeness, also checked a 200 Ohm, 0.1% resistor. 1.5% deviation is the specification of the MH15, and deviation found is 0.25%, very good.

All in all, a great instrument for the occasional user, and one more reason to not skip isolation resistance tests.

Various

L33 Thermal fuse: inner workings

Leave a comment

Recently, I had a project that required a reliable thermal fuse. There was little space to accommodate the classical axial versions, so I did some investigations and settled for the L33 type fuses. These come in various temperature versions, here, the 130°C limit, rated 2 Amps, 250 Volts.

The main components are the 2 wires, a plastic case, and some resin. Having never studied one open, I disassembled a few good ones.

Clearly, the resin is filled to certain level. At the top, there is a bridge between the wires, made by low-temperature melting alloy (having tin, bismuth, indium and related metals).

The alloy is quite substantial, likely to be able to handle 2 Amps of current.

Triggering it with heat gun, the alloy melts, and there is enough space in plastic case that it forms a drop, effectively interrupting the circuit.

While the devices studied showed very good consistency of construction, a little overfilling with resin may result in the thermal fuse not opening, I hope the manufacturer has this parameter under strict control. For the 10 pieces I have, the weights measured on a precision balance where quite uniform at least. But if you get such critical parts from some doubtful sources, better you do some tests first and be sure they are reliable.

Various

Blaupunkt OSTIA Home Radio: revamping an old beauty

2 Comments

As a family heritage, we ever had an old radio from my great-uncle (brother of my grandfather) named Modestus, and ever since I can remember is had been standing in the bedroom of my parents. Eventually, it was no longer used there and moved into my mechanical workshop where it still serves to play background music while I am operating machines.

It was build some time in the early 70s, and is based on germanium transistors – to be precise, 11 germanium transistors. The sound is not bad given the relatively simple circuit. However, in the last year it must have suffered some degradation of the FM tuner, because this tends to drift, and the reception is not clear always. Especially in winter, switching it on after a “cold” start, it needs some re-tuning after about 30 minutes of operation. A little inconvenient. Rather than spending a long time trying to fix an old FM tuner, I decided to take another approach – adding a new digital (PLL) tuner.

In my stock of old parts I had a no longer used PCI TV card, which incorporates a Philips FM1216MK tuner, a combination TV and FM tuner use a TSA5523 PLL, and can operate from a single +5 V supply (because of an internal DC-DC converter).

The card is a combined ISDN-TV-FM card. The tuner can be easily desoldered. Control is by i2c bus, two wire interface. Some libraries exist, but I didn’t use those. Rather straightforward to send the bytes needed to set the frequency and to do some more configuration needed. The tuner has a stereo decoder, but I operate in mono mode – there is only one speaker in the OSTIA radio.

A quick setup with a i2c LCD added for debugging. Using a Arduino Nano3 board clone with an ATMega168p microcontroller. But any microcontroller will do.

Now, integrating the new tuner to the OSTIA – my objective was to not destroy the old beauty, integrate minimally invasive. A first attempt to use the build-in transformer failed, because it could not provide the roughly 200 mA current needed for the Philips tuner.

To feed the audio signal, I cut a bridge on the board (which carries the FM audio from the old tuner), and injected the audio from the new tuner via a 100 nF foil capacitor.
For control of frequency, there is no an incremental encoder on the back of the radio (I rarely change the station if at all), and when you push on that encoder, the last frequency set is stored in EEPROM. The LCD has been disconnected, not needed during operation.

Finally, the OSTIA back at its accustomed place in the workshop. Reception is good and stable now, all frequency locked to a small quartz crystal.

Certainly this radio now has no longer just 11 transistors – maybe 500 transistors now!

Various

NE555 Watchdog Timer: the ESP32 needs some oversight

6 Comments

The readily available ESP32-DevKitC boards have served me well in many application, but there are some issues with one of the circuits that is up all throughout the year in my house to record moisture and temperature levels. Occasionally, like, every few month, this ESP32 gets stuck, so the web server running on that ESP32 is not responding anymore, and the logging of the data will stop (red marked portions in the plot).

The root cause of that relates to the current pulses drawn by the WLAN circuits of the ESP32, and despite connecting a good USB power supply, proper cables, and capacitors, it seems that there are occasional issues that I haven’t been able to solve be capacitors, better power supply, or software restart-features. I added the later, but the ESP32 freezes to a level that any software reset triggered by the code won’t work. Shortly disabling the power converter on the ESP32-DevKitC (of the on-board 3.3 V regulator – its EN/enable pin is pulled high by a resistor) will restore the function and get the circuit started again.

As I am not always around watching this circuit, I added a good old trusted NE555 timer, which will send a reset pulse (by pulling the EN signal low through a Schottky diode), and the capacitor is shorted by the small MOSFET, as long as the ESP32 is sending a pulse (this is send every 10 seconds approximately, for a few milliseconds) — if the ESP32 gets stuck, there won’t be any pulses, and the NE555 will then reboot the ESP32 every other minute by cycling the power to the ESP32.

Micro-Tel 1295 Precision Attenuation Measurement Receiver (2nd unit), Microwave

Another Micro-Tel 1295 Precision Attenuation Measurement Receiver: irresistible green

3 Comments

I am trying hard to resist the temptation of buying more test equipment, but the Micro-tel special green color has a hypnotic effect on me, and combined with the right price, I could not resist to buy one more Micro-tel 1295 receiver. These are very capable 0.01~40 GHz fundamental-mixing receiver (fundamental mixer until 18 GHz, above that, harmonic mixer), with very large range, like, 120 dB, and 0.001 dB attenuation resolution. Ideally suited to calibrate attenuators or to check antennas, etc.

The unit – offered as non-working – arrived very well packed. Unfortunately, many people send sensitive equipment in some thin cardboard boxes. This particular equipment cost close to 85 kEUR in 1989, plus mixers. Also, it has long been under export control from the US, because of its unique range and accuracy.

Bubble wrap, other fibre wrap inside.

Finally all in foil.

The defect, it doesn’t show any reading on the display, and both the HI and LO leds are on, which is abnormal. The 1295 has a 12 dB range bolometer detector, any signal below 0.5 dB or above 12.5 dB will light up the LO or HI lamp, and you would need to select another 10 dB step of the IF attenuator (a high precision 30 MHz attenuator), or let the automatic attenuation selector do the job.

There are many boards inside, but all nicely numbered and with instructions in the manual.

The HI and LO level detection is done on the A3B2 assy.

According the the schematic, U2, a MC1458 generic dual opamp is switching the LEDs and providing signals to drive the automatic attenuation selector.

A quick check revealed that U2 is defective, so I replaced it quickly, and this already solved the issue and brought back the display.

Another trouble related to unstable lock of the 2.3 GHZ auxiliary LO that is used for the 0.01-2 GHz range (which uses a two-stage down mixing).

Fortunately, I had a spare 2.3 GHz from my parts unit (which I bought years ago – a partial unit – while I was living in the US). That part was missing one of its covers, and had also some issues earlier, but I had fixed it a while back just for curiosity. Now I can fix the unstable 2.3 GHz removed from the unit during next winter. It has a 2.3 GHz VCO, a 100 MHz local oscillator and a PLL inside.

After calibrating all the oscillator frequencies, which went without trouble, I noticed that the top 120 dB attenuator was 0.04 dB off, well, not a big deviation, but I would rather have the unit working perfectly. So I removed the attenuator for further study.

It is build with really high quality relais, more than USD 50 (each!!), and some precision resistors.

Nothing could be found wrong with the unit by visual inspection.

Also I used the VNA to check the attenuator, and all seems well working.

All the 3 segments, virtually equal at 10 dB each.

Finally, I put everything back together, a little clueless, but, now, for some reason, all is working and stable. Maybe it was some lose connector, or other strange effect that is now gone. All attenuators calibrated perfectly, using by HP 3335A level generator (which has a top-accuracy attenuator).

Finally the 1295, working just perfectly fine. Maybe better than ever before.

Interestingly, as with all of these Micro-tel devices, the side and top/bottom panels were painted with various kinds of special military paint – some with a rubberized paint that will dissolve into some gluey substance over time, some with a type of “abrasive” paint, other already re-painted in forest green.

The paint has very large and hard grit, almost like sandpaper. But I will leave it untouched, it seems the original looks for this serial number range (the 1295 seems to have been in production for 10+ years).

Now, a little gallery of all my Micro-tel 1295 receivers: the first two, part of my frequency-locked attenuator calibrator (can measure reflection and transmission at the same time).

One as part of an E-band (60-90 GHz down-converter).

Any now, already two spare units in perfect calibration.

Still, in the basement, a box of spares… likely I won’t run short of receivers soon. Maybe even buy another one should it come around.

Microwave

HP 11517A aka 08747-60022 Harmonic Mixer: a little study of a very intriguing device

Leave a comment

As part of a HP R8747A 26.5-40 GHz reflection/transmission test unit (for the 8411A network analyzer; 6300 USD in 1973 — about 40 kEUR today), I got two HP 08747-60022 harmonic mixers, one didn’t seem to work right, the diode has just 0.2 V voltage drop. These were fairly fragile devices, only designed for 1 mW of power, and very static sensitive, point contact devices.
In addition to the regular 11517A, the 08747-60022 has a bias connection (needs about 1.5 V DC bias, center positive).

The main unit can work from 12 to about 40 GHz, with a set of adaptor waveguides.

The unit can be taken apart, all precision machined.

The diode is pressed in, on some holder (haven’t tried to remove it from the case).

There are several other precision parts, a coaxial resistor, held on to the diode with a spring.

There is also a spacer, with a very flat capacitor, a DC block. The spacer is modified to connect the center conductor to a surface at the perimeter (used for DC bias).

Further up, there is a low-pass, machined from a single piece and gold plated.

The N-type connector, stainless, is screwed on.

The DC bias uses a small 1.5 kOhms resistor, and a custom connector, so that the resistor is pushed onto the spacers’s connection.

Here, a quick schematic. Seems a lot of engineering went into this device…

Finally, we need a microscope to study further here, the diode (the square), about 0.25 mm side length. It is isolated from the case by an air gap.

The diode is made by a contact junction, a small tungsten(?) whisker.

Probably, this whisker needed some adjustment during manufacturing. I tried to adjust it a bit, but this didn’t change the diode characteristics of my broken unit, unfortunately (anyway, these devices are more for study than for use; currently using only cartridge-based diode mixers).

Also, looking into the waveguide of the assembled mixer, with good long-range optics, I could get a shot of the actual point contact in action. Very interesting historic technology.

Millimeter Wave

Fixing a really-high-frequency phase locked oscillator: a noisy opamp and a little bit of sticky tape

4 Comments

Recently I got hold of a really marvelous bit of kit, an assembly of millimeter wave phase locked oscillator, all the way from 60 to 89 GHz. It is already 35 years old, but all based on solid state Gunn/Impatt oscillators, so there is little aging. For each frequency (there are 5 discrete frequencies), there is a full assembly of reference upconverter (100 MHz reference to some intermediate frequency in the 6-12 GHz range), a voltage controlled fundamental oscillator (e.g., bias-tuned Gunn diode), the necessary isolators and splitters and occasional attenuator, along with a harmonic mixer. The IF is 100 MHz or 200 MHz, depending on the unit.
Only one of the oscillators has been more than 20 kEUR in 1988 currency, and there are five such assemblies on this plate, along with control equipment and power supply…

Only one of the sources is playing up, not achieving phase lock, it is the 79 GHz oscillator. Notably, some of the small screws are missing and the lid of the oscillator cavity has been removed, which points to prior repair attempts. The Gunn has a protection network, a 2.2 µF capacitor in parallel with a ~10 Ohms, ~470n series network — this is to protect the Gunn diode from the inductance of the bias current supply (when the diode snaps-off according to its characteristics, it will induce a very fast current spike that could increase the voltage at the diode above its damage threshold).

Upon close inspection, the 2.2 µF capacitor had been tampered with, bad soldering, only one end connected, and in reverse polarity (it is a high-reliability tantalum cap).

When checking the down-coverted output, there is no stability at all, it has a very strange FM modulation. What could be the root cause?

Checked a few items:
(1) Upconverted reference, 11.3 GHz (7th harmonic will be 79.1 GHz): it is a very clean and stable signal, well locked without any visible noise.
(2) Checked all the cables and connectors by pushing gently, no sign of any trouble, all unchanged.
(3) The supply voltages are all practically noise free.
(4) Also fixed the 2.2 µF capacitor at the gunn diode cavity. No particular effect.

(5) Disconnected all the phase lock, bias driver, and driving the oscillator from an external supply.

Finally, with just an ordinary power supply connected and the voltage ramping up (don’t ramp it up too slowly or let the oscillator sit in an unstable region or at the Gunn peak current for any length of time!), there is a stable (surely, non phase-locked) output, but none of the strange modulation. So it seems, the oscillator is good. But wait, with some wobbling and touching on the part, it is shorting out the supply. Hmm. Time to go a little deeper, and I decided to remove the biasing rod.

Just for explanation, the biasing rod is isolated from the cavity (the metal block holding the diode, with the waveguide cavity in the middle), and conducts the current to the diode, which is at the tip of a metal screw holder.

Under the microscope, the very thin isolating tape (looks like some PET/Mylar transformer tape) is quite damaged and some metal of the bias rod exposed. Also the spring holding down the rod on the diode (which is fixed by a nylon screw) can contact the case easily. So all was newly isolated, and the screw and spring positioned carefully. Surely with all the soldering at the capacitor, things may have shifted a little. At least now no isolation problems any more. Sometimes when switching the oscillators on the assembly, the 79 GHz signal is not coming up, the Gunn drawing much less current. But it can be fixed by just power cycling the assembly, so it seems to be some rise time issue of the power supply.

This is the driver circuit, it has a hybrid high-speed Kennedy electronics 722 amplifier, but the low frequency path is a common NE5532 low noise opamp.

After checking around the low frequency path, the opamp seems to introduce some noise, note that for this level of noise at GHz levels, a few millivolts are more than enough to cause a lot of disturbance.
Fortunately, the opamp was socketed, so I replaced it with a OPA2277, which has about the same noise compared to the (low-noise) NE5532, and much lower offset voltage drift, and 40 dB better CMRR, excellent low frequency characteristics, and lower gain at >1 MHz.

Now, with stable current drive, we can measure the output power (downconverted by an harmonic mixer driven by a 11.45 GHz source, 7th harmonic: 80.15 GHz — shown is the minus mixing product, i.e., 79.0 GHz correspond to 80.15-79=1.15 GHz).

You can see clearly, the oscillator has good power from 79.1~79.25 GHz, but falling down just around 79.0 GHz. Note that at such high frequency, every mW counts…

After some thinking, I decided to try to lock at 79.2 GHz, and as it turns out, it is working fine at that frequency. With the 11.3×6=79.1 GHz, and 100 MHz IF, it works out, and the PLL doesn’t seem to be affected, regardless if you use the minus or plus side IF.

The IF signal can be probed conveniently at a test port, it is pretty clean.

Also the down-converted signals look good (this is 10 MHz span at 79.2 GHz center, corresponding to 80.15-79.2=0.95 GHz downconverted).

Also the close-in side bands are good and a very clean signal.