HP8444A cavity oscillator

[Alan Bain]

I have an HP8444A tracking generator that is not working.

The fault is easily localised to the 1.55GHz cavity oscillator which was producing no power. A replacement RF device (original HP part) later, it is now happily oscillating but at 1.84GHz. Just in case it was my counter (5245L + 300-3GHz plugin) I was very kindly allowed to check this on an HP5342 (sadly I don't own one).

The thing is very simple, a length of coaxial transmission line shorted at far end and loaded to ground by a co-axial capacitance of around 4pF. Checking dimensions it looks correct to oscillate around 1.55GHz. Calculation details at end.

At this stage I don't know what else to try other than putting the cavity on a reflection bridge and checking an impedance sweep but my very ancient network analyser is also in need of repair!

But must admit I'm puzzled - has anyone more familiar with cavity oscillators than I got an idea?


The 8444A 1.55GHz oscillator is a coaxial cavity OD 38mm ID 12.7mm which makes Z0 around 65 ohms (and are close to optimum Q proportions). It's 12.7mm long with far end a short. Wavelength 1.55GHz is around 193mm. The far end is a short so Gamma_0 = -1 and the Smith chart equation is

Gamma = exp (-2 * j * beta * x) * Gamma_0

beta = 2 pi / lambda and x is distance from termination toward source (am I alone in having to derive this every time to work out the sign of x?)

So at 1.55GHz on 12.7mm of line at the open end Gamma = -0.68+0.73i
Gamma = (ZL-Z0)/(ZL+Z0)

So Z = 0.09+28i so admittance Y = 0.0001−0.0354i

So to add some parallel capacitance to cancel the imaginary part of admittance leaving a high impedance - ideal for resonating

Need 2*pi*f*c = 0.0354

So to resonate at 1.55GHz c= 4.6pF and I measure 4pF (using an elderly bridge only reading to multiples of 1pF).

[Jon_G4MDC]

I don't know much about these lan but that resonator seems very short to me.
You say the inner conductor dia is 12.7mm dia and that it's length is the same as the dia.

I have a similar oscillator with a phase lock circuit. Crystal is ~100MHz. It can be set to oscillate between 1400 and 1900MHz in 100MHz jumps as it locks to each multiple. The freq control is by threaded rod which winds out more resonator length.

The active part of the oscillator is loop coupled to the resonator.

It's a while but I think the resonator length went between about 25 and 35mm or so. I could find it and dismantle it to measure if it would help.

[G0HZU_JMR]

I don't know that much about cavity oscillators but would be interesting to see some pictures of the resonator structure and how it all connects together. I had a quick look in the service manual and it shows the schematic of the oscillator as below. A typical resonator might look something like the lumped equivalent below but I'm not sure if this applies in your case.

Which component was faulty/replaced?

The BJT (Q1) looks to be arranged in common base as a negative resistance generator and the varactor diode CR1 looks to be an old school leaded component. The lead length of this part might be fairly critical as I'd expect it to be operating past its series resonance. So it probably looks like a variable inductor when the bias voltage across it is changed. What you don't want is to end up with it being series resonant at 1550MHz as this will prevent oscillation in the correct place. So the lead length and the way it is soldered in place could be quite important.

[G0HZU_JMR]

I guess what I'm trying to say is that the designer of that circuit may well have chosen to solder that varactor diode in place with a certain amount of excess lead length in order to keep it away from self resonance. If this part was replaced it would have to be with a direct equivalent and it would have to be soldered in place with the same lead lengths as the original. The capacitance states 6.5pF at 4V on the schematic and it might be about 4pF at 12V so it would not be good if the series inductance was in the order of 2.6nH as this would cause a short at 1550MHz. I'm really just guessing but it would probably be better if the series inductance was several nH higher than this and this could be achieved with a few mm of excess lead length.

[Radio Wrangler]

C1 will be somewhat critical in that circuit as well as the length of track in the base connection. Along with the transistor's internal stray Cs, these things get gyrated to produce a negative resistance at the collector to make the tank oscillate. The -ve resistance has to be strong enough to take out the 51 Ohm collector feed resistor as well as the losses of the tank.

Once you get up to these sorts of frequencies oscillators with a loop circuit start getting into difficulties and negative resistance types tend to take over. I've had enough fun in recent years trying to keep a loop type with a SAW delay line manufacturable at 1090MHz.

David

[G0HZU_JMR]

Yes, if Q1 was replaced it would have to be fitted exactly the same way as the original.

It would ideally need to be made by the same manufacturer as the original as well. Looking at the service manual this part looks to be in a leaded metal can with 4 legs. This is a bit of a surprise. I was expecting to see a pill type package like a BFR91.

The problem here will be that there will be a delicate balance between squeezing out adequate negative resistance from the BJT and avoiding resonances within the BJT. Adding base leg inductance will boost the negative resistance at the collector and adding shunt capacitance at the emitter will also boost negative resistance at the collector. There is a 0.5pF shunt cap at the base already.

Looking at the s-parameters of a few fast BJTS I'd expect the collector of a fast BJT to look like about -40R in series with about 0.8pF up at UHF in this circuit. Adding 0.5pF at the emitter will boost this to maybe -60R but the capacitance will creep up. Adding inductance in the base will boost the negative resistance even more. However, this increases the collector capacitance quite steeply and brings in the risk of package resonances that are normally just out of band up at maybe 2GHz or so.

Everything has to be fitted exactly as per the original or there is the risk any added inductance in the connections could pull down the frequency of these resonances such that they compete with the main resonator. Even an extra 1-2nH in the base leg can be significant at these frequencies.

[G0HZU_JMR]

I suppose there may also be a risk of squegging or outright oscillation due to the transmission line properties of the bias choke L1. This appears to be an airwound coil stretched out like a long spring. If the BJT can produce enough negative resistance to overcome the 51R in series with L1 then the circuit could prefer to become a transmission line resonator using L1 instead of the desired 1550MHz resonator.

If it does squegg here it could be that the frequency counter picks out the strongest 'mix' from the squegging. I simulated the circuit and this problem becomes possible if the electrical length of L1 is a half wavelength at the unwanted oscillation frequency. The proximity of L1 to ground will also affect its transmission line properties so this should be kept the same as original.

It might be worth trying to add a lossy ferrite bead on the wire connection after L1 if this can actually happen. It would have to be a ferrite bead that can damp effectively up at UHF. The picture below shows where the ferrite bead could be tried.

[Radio Wrangler]

To look at the cavity oscillator and to see what's really happening you need a decent spectrum analyser, probably covering to at least 6GHz.

This cavity oscillator is a derivative of the one doing duty as LO2 in the spectrum analyser. Hunt up the HPJ article for the 1250MHz spectru analyser plug-in and you may find some exlanations of it in there.

David

[Alan Bain]

It seems that the circuit topology was changed by HP. The varactor was moved from the position on your diagram (collector of Q1) to a separate magnetic coupling to the cavity.

Here are a few pictures showing the cavity and the circuit. It was indeed Q1 that I replaced and I did try and match the lead lengths to the part being replaced (I used one with the identical HP part number). But your thought on base lead length is interesting as it looks a bit long.

One other interesting observation, the varactor control voltage is in the correct range but it only shifts the frequency by about 0.4MHz rather than the 4MHz mentioned in the manual.

I don't have spectrum analysis beyond 1.8GHz but I was able to show it to an HP8566B which showed only a single output frequency (with reasonable wide phase noise bands) around 1.84GHz (which to me was consistent with this not being the cavity resonance).

[Radio Wrangler]

Originally Posted by G0HZU_JMR

Looking at the service manual this part looks to be in a leaded metal can with 4 legs. This is a bit of a surprise. I was expecting to see a pill type package like a BFR91.

At a guess, it may be a packaged version of the HP21 chip bipolar device. It got fitted into several different packages, some custom for specific products

David

[Alan Bain]

The only data I could find on Q1 was (source unfortunately not recorded by me, probably something found online) and not including any S-params

HP1854-0292

FT min 1.6GHz @ IC 10mA
HFE min 25 max 150 @ IC 2mA
VCE min 15V
PD max 200mw
IC max 25mA
C CB 1,5pF
NF max 3.5dB

Some S-parameters would have been nicer, especially

[loulou31]

Hi,

the source was, with a proposed equivalent that is available on Ebay :

https://www.electronicspoint.com/for...-repair.41055/

Jean-Louis

[G0HZU_JMR]

The nearest BJT I have here to that is the BFS17 in SOT-23. I have the manufacturer's s-parameter data for the BFS17 in common base although this model is really old (1990) and it is measured at 5Vce 15mA Ic.

If I put this model into Genesys the BFS17 can easily generate negative resistance to beyond 2GHz in common base. Of some concern is that it can generate about negative 75 ohms even at 1800MHz. This is greater than the resistance of the 'stopper' resistor inline with the spring coil L1.

If I put together a crude model of the oscillator using a lumped resonator I get the result below where the cavity is competing with the spring inductor. There is resonance and negative resistance at 1800MHz with the BFS17.

Note that I wound a copy of the spring inductor based on your image and took a 1 port model of it using a VNA from LF through 3GHz. So this should be a very good model.

For the simulation plot below I've measured at the cold end of the spring inductor and this shows that if the spring inductor is grounded at the far end (which it will be by the feedthrough cap) then the oscillator circuit can go unstable at 1800MHz as there is resonance and a net negative resistance at 1800MHz.

Having seen your images I wonder if you could try adding a 22R (or try up to 100R) series resistor inline with the existing 51R stopper resistor as in the image below?

[G0HZU_JMR]

See below for the simple model of the complete oscillator. It assumes a lumped equivalent for the cavity and uses my 1 port model of the spring coil L1. I should really have taken a full two port model of it but I think the 1 port model will be OK for an initial attempt at modelling this issue. This really is a 'page 1 of the logbook' model of the cavity oscillator so only really suitable for initial analysis.

L1 causes a problem at 1.8GHz because it can act as a transmission line and it is about half a wavelength long at about 1.9GHz according to the VNA. The BFS17 is close to self resonance at the collector at 1.8GHz due to the 1pF cap at the emitter and about 2nH in the base leg. This means it just needs something close to a short circuit at the collector at 1.8GHz to complete the oscillator circuit at 1.8GHz and L1 provides this. The 51R stopper resistor doesn't have enough resistance to offset the -75R generated by the BFS17 so the circuit can oscillate at 1800MHz instead of the intended 1550MHz. of course all this relies on the accuracy of the model and if I try a BFR91 in place of the BFS17 then the problem goes away. I think this is because the BFR91 is further away from self resonance at 1800MHz.

[G0HZU_JMR]

Having just spent some time playing with a real (Infineon) BFS17 on a VNA I don't think the BFS17 is a good choice here. It can also generate lots of negative resistance down at around 300-400MHz and this means it could also oscillate down here as well. The inductance of L1 will series resonate with the capacitance of the BFS17 at around 400MHz and it looks like the BFS17 can generate more negative resistance than the 51R resistance of the stopper resistor.

So L1 has at least two modes that can potentially interfere with the circuit operation with a BFS17. I think this circuit absolutely needs the correct transistor and it needs to be fitted correctly and probably with minimal inductance in the base connection.

It might be possible for a small ferrite bead at the far end of L1 to fix things but there doesn't seem to be much room for it. It could also cause microphony unless it was glued somehow to stop it rattling. To stop the BFS17 from misbehaving down at 400MHz the VNA suggested that a 120R stopper resistor would be needed in place of the existing stopper resistor. This might be too much for the correct DC biasing so the ferrite bead might be the only option for a BFS17. Therefore, the BFS17 probably isn't an equivalent that you can use to obtain an equivalent circuit model for the original BJT.

I suppose you could try to minimise the inductance of the base connection as much as possible to see if that helps with the replacement transistor you currently have. If that doesn't help then maybe experiment with some excess base leg inductance as this might shift the negative resistance down in frequency below 1.8GHz and this might stop the instability. But this is desperate stuff really...

[Alan Bain]

I think there may have been some more redesign by HP. I went and looked at that resistor in the collector to add some extra and found when they moved the varactor they seem to have changed the collector resistance to 562 ohms. A scan from the manual section attached. This is not the first time HP manual changes have caused me much perplexity.

I did find a suggestion a philips A2149 was an equivalent (https://www.parttarget.com/5961-00-4...4-92AC8E39F4CC) but no more RF data

[G0HZU_JMR]

That's interesting, thanks. 562 ohms in series with L1 fixes the stability issues for for me on the simulator so I'm now a bit confused why it is still oscillating up at 1.8GHz. The original design with 51R in series with L1 was really twitchy so I'm not surprised it has been changed to a larger value of resistance. I'm also not surprised to see the varactor diode get relocated.

On the simulator, the change in the emitter capacitance from 0.5pF to 1pF was quite significant in terms of how it affected the impedance at the collector at 1.8GHz.

Does the cavity screw adjuster make any difference to the oscillation frequency? If something was wrong with the varactor diode it might be able to bridge/change the cavity resonance but this would have been happening before the BJT failed.

I've got a fair bit of experience using common base BJTs in an RF oscillator but normally this is with a printed resonator or a skinny coaxial cable type resonator up in the GHz+ region. I really don't have experience of a cavity oscillator like this. I'd be tempted to tweak the screw to see if the oscillation frequency changes. If it does change then maybe something has happened to the resonator components. I doubt the dielectric constant of the capacitor material has aged/changed. Have you seen it working correctly before it failed?

[Alan Bain]

Following another lead mentioned above (thanks David) I looked at the LO2 in the 8555 (I don't have one of these, the 8444A had a long life with later SAs) and I see a similar cavity but with a pair of common base transistors in the oscillator and the usual "not field repairable" note. It's sad HP did this so often with cavities (think the 8640 sig gen or the plugins for the 5245) and these are often the parts which need attention (it's doubly frustrating when there is also a long detailed description of how some very vanilla linear psu works).

It is worth noting I have a device marked as HP1854-0292 obtained new old stock from a reputable source. But the oscillator came to me non working and there are signs someone had screwed the tuning slug right in (so it marked the cavity base) which would be consistent with trying to get the frequency down.

[G0HZU_JMR]

To give a bit of confidence in the modelling I built up the BFS17 circuit in a similar dead bug style but without the cavity. I left the stopper resistor at 51R and the simulation plot below predicts it could go unstable at around 400MHz and/or 1800MHz. This is because the plot shows a net negative resistance and resonance at both of these frequencies. When I built it, the circuit was unstable at both frequencies at the same time but I could make it swap between them by changing the collector voltage.

See the brief youtube video below.

https://www.youtube.com/watch?v=CCZ4qxYnHRs


This shows it oscillating at either 400MHz or 1800MHz or both as I vary the collector voltage back and forth. So the s-parameter data is fairly good I think. This probably demonstrates why HP increased the stopper resistor to 562R as this kills both these modes.

However, playing some more with Genesys the simulation does also show one other way it can go unstable at 1800MHz (even with the 562R resistor in series with L1) and this is if there is a hidden but direct series acceptor circuit to ground at 1800MHz somewhere at the common node at the PCB/screw connection. This would be the same as providing a short to ground through this common node at 1800MHz. I'm not sure what could create this but it would only require a few pF and a few nH in series to achieve this at 1800MHz. It's highly unlikely but there could be the equivalent of this hidden somewhere in that node as it pokes a bolt/screw through a hole via a grommet.

The second simulation plot below removes L1 and the 51R stopper resistor completely and it shows that there is a resonance within the BFS17 at 1800MHz caused by the base inductance and the emitter capacitance and the internal capacitances within the BFS17. The plot shows that there just needs to be an RF short connection to ground at the collector (at 1800MHz) to complete the requirement for oscillation at 1800MHz. It might not be the same with the correct HP transistor but I suspect it may be able to do something similar.

[G0HZU_JMR]

If the black grommet in the access hole can act as a capacitor dielectric to ground then the hex screw at the access hole can act as part of a variable capacitor and series inductance to ground. Can the hex screw be screwed out a bit? Does the oscillation shift in frequency when this happens?

This is getting desperate but I'm not sure what else to suggest other than to try finding another cavity and swap things between the two to see what causes the shift to the 1800MHz oscillation.