UHF oscillator

[T3sl4co1l]

Trying to make a VCO, say, 400-600MHz would be nice.  Actual frequency not critical, just trying to get something to work with a reasonably wide range.

Circuit:



The circuit isn't very specific, considering how important layout is...

The idea, at least, is a basic common-collector Hartley configuration.  I've added a detector output, and bias input for AGC.  This loop -- when the oscillator decides to start up -- is working fine.

The oscillator "coil" is a loop, rectangular, 4mm above the ground plane, 25mm long, made of 0.8mm wire, with the 'emitter' tap 6mm from the 'base' end.  The output loop is... whatever's good enough (too close and the loading quenches the oscillator; too far and there won't be enough to view the output).  The output is BNC, terminated into 50 ohms at the scope.

Pictures:





Kind of hard to see with solder everywhere, but it's "Manhattan style" with chip components where possible.

Detector diode: house marked (HP something?), point contact, seems to be low capacitance.  "Varactor" diode: BAT85 schottky.

Problem: I can't seem to get it to oscillate any higher than 425MHz.  The current form runs 305-425MHz.  There appears to be 5pF stray just laying around, of which at least 3pF I can't account for.  If I assume the loop is acting inductively, it should be around 19nH, which is reasonable, but it's also not, because I got exactly the same frequency range when the loop was 6mm high (which ought to be 25-30nH!).  Any attempt to shorten the loop only quenches oscillation.

Are there peculiar tricks to using UHF transistors that I'm missing?  I've seen microwave oscillators with larger tuning elements (stripline, etc.), I can't imagine it's the size that's screwing it up.

I also tried variations on this circuit with an MMBTH10, with similar results, except that I occasionally got something around 800-900MHz instead (with almost no adjustable range -- suggesting it was a parasitic loop hardly affected the 'varactor'), nothing inbetween (i.e., never something in the 400-800 range).  I think I can shrug this off as phase shift near the cutoff frequency causing a dead band (its fT ~ 800MHz, depending on voltage and bias), but with the slick-snot 2N5179 in here, I should have no problem going into the GHz.

I also tried a microstrip format (which should've been around 50-100 ohms Zo, maybe 40-50mm long depending on how you measure it), with much the same results (though with more squigglies: the AGC vs. Vadj response had a number of peculiar bumps in it, suggesting harmonic resonances or something like that).

Note: my scope is 350MHz, which isn't enough to see waveforms as such and measure the amplitude accurately, but there appears to be enough gain left to at least trigger and measure the frequency.  Even without the scope, the detector output at least tells if it's working or not.

Tim

[calexanian]

I am no expert in UHF land, but I think you are asking that device to go a bit high in frequency. Its gain will suffer above 200mhz. 425 may just be where the gain drops below oscillation capability. Also I might add another decoupeling cap right at the collector of the device. Those air wires are a touch long for that frequency. Just some thoughts from a person who does not know any better.  Additionally you need to have some proper coax out or the losses will be great. Possibly enough to swamp oscillation.

[Richard Head]

T3sl4co1l

I agree with the previous poster.
Also, what stops the RF from being pulled down by the agc line?
It seems that an RF choke is required on that line.

My 2c worth

Rich

[calexanian]

Also looking at the output transformer there. Your mutual inductance is probably not that great. I would use coax in a 4:1 balun type situation. Much wider band than parallel conductors.

[Richard Head]

T3sl4co1l

Also, what stops the RF from getting onto the Vadj line?
You must decouple and block the RF from going down those lines by means of chokes (or beads) and bypass (or feedthrough) caps.
This ensures that there is no RF where it isn't meant to be.
Ideally the oscillator should be enclosed in its own little metal box with feedthrough caps for all DC and control lines.
If you want to change the RF level it's probably better to use a PIN diode attentuator on the output than alter the operating point of the transistor.

Rich

[T3sl4co1l]

Should've put these on the schematic; the Vbias line has 1n to GND (visible in the first pic) and Vadj has 100p to GND.  Touching the wires has zero effect on the response (AGC bias or frequency).

As far as damping goes, the pair of 10kohm resistors should dampen things a little, but if I'm getting 10s of mV into 50 ohms from the output tap, it's got bigger loads to worry about.  Put another way, the reflected power from a 10kohm resistor should be very large indeed.  Not as large at the top of a resonator like this, but still not big.  I could use 100k resistors in the same places, but chose 10k for the faster RC response.

RF choke, maybe; a mere ferrite bead isn't going to have more than a hundred or so ohms (I don't think I even have any good for this frequency range), and lossy anyway.  But a proper coil would have reactance rather than loss, so even though its impedance might be 100s of ohms (as a plus... the parallel reactance would help raise the frequency a little), it needn't load the circuit as much as 10k.  (As always...assuming anything still makes sense at all.  Standard RF disclaimer?)

As for cutoff, like I said, the MMBTH10 was doing 400MHz or so, out of an 800MHz fT.  The 2N5179 should be good out to a gig even, by the same reasoning.  So, Idunno.

Tim

[Richard Head]

Tim

I would try another approach to the AGC design.
I reckon the diode and 100pF cap will de-Q the resonator.
In the interest of best phase noise and stability I would be inclined to put a buffer after the oscillator and use the buffered output for the AGC.
You don't want any unneccesary load on the oscillator tuned circuit at all.
Also, I would ground the metal case of the transistor if possible to reduce the "hand"effect.
For the output couping, why don't you rather just tap the inductor a couple of mm from the cold side? This will eliminate the extra coupling loop.
You can also try a BFR92 or similar. It has a much higher Ft and is available in SMD.

Rich

[G0HZU]

I have to say that the VCO design and physical layout looks pretty grim to me.

I've designed quite a few wideband VCOs in my time but nothing recently. Sadly, these days (at my place of work) it makes more sense to simply buy VCOs from Zcomm or Mini Circuits for design work because it saves so much money on NRE costs etc

But...

You really need to approach this again and draw up some basic design requirements for the VCO and then use some basic design rules to try and hit your targets.

I would suggest you try for the following (very loose) spec.

Tuning range:      400 - 650MHz
Tuning voltage      3 to 18V   (is this OK?)
Phase noise      better than -95dBc/Hz at 10kHz offset
Output level      >+3dBm
Layout         smaller than 30mm x 30mm with a layout and components that give lowish microphony


Normally, you design the VCO for a certain loaded Q and resonator power level to meet the phase noise requirements whilst making sure the varactor and its steering resistance don't degrade the phase noise too much.

Also, the traditional method I used to use was to analyse the circuit in open loop on a linear simulator and make sure the loaded Q is achieved and also that there is sufficient gain margin at the 0/360degree phase point across the whole tuning range. Otherwise, it won't start up reliably.

The good news is that it's pretty easy to meet the above spec and I would probably use a simple Hartley design with a printed inductor. This should give reasonable resistance to microphony. I'd be tempted to use all SMD devices but you could use something like a BFR91. I wouldn't use a big old 2N5179 here although it should be possible to get one to oscillate up at UHF.

I've designed VCOs to cover 650-1200MHz in the past and managed to get -100dBc/Hz phase noise at 10kHz offset but this was many years ago. These days you can use modern tools like Microwave Office to do the whole design on a PC including the layout and it will tell you everything including the phase noise performance.

But if you use a few rule of thumb equations (and a simple linear simulator) you can predict the performance pretty well anyway

[Richard Head]

G0HZU

I've never tried a printed inductor for a VCO before. I wonder what the unloaded Q would be.
Another option may be an SMD 0805 inductor (coil type only). Wouldn't work with a Hartley but definately with a Colpitts.
Microphonics can be a huge problem with certain products so an inductor hanging in the air is asking for trouble!
I agree with your suggestion to compile a spec beforehand but I suppose for someone new to the game it can be a little overwhelming at first.

Dick

[T3sl4co1l]

Thanks for the comments.

This is a learning project so it's going to be big, ugly and crude.  I've moved things around a few times already on the board pictured, hence the ugliness.

For sure, if I needed to do this right the first time, and have some chance of success, I'd buy off-the-shelf blocks, like Mini-Circuits' VCOs, amps, mixers...  Un/fortunately, this isn't one of those projects.

My intention is to play with radio in the VHF-UHF range.  I don't have any particular band in mind, hence the wide range.  Maybe more of a spectrum analyzer thing, just put a log detector on the output.  So this will be the LO, and 300-500MHz would cover a 0-200MHz band with a 300MHz IF.  That's certainly enough to see, for example, the local FM radio stations -- like the guy in the video, the homebrew spec.  Other ranges are possible of course.  So I need enough output to drive a mixer (probably a DBM, with a DC coupled input port to go all the way down to 0MHz), and low enough phase noise to be useful (-95dB/10kHz means good nearby-channel rejection, probably as good or better than any crystal filter I can find?).  As a mere proof of concept, things like microphonics and size aren't important (though I appreciate the mention, and come to think of it, those X7R bypass caps will pick up a bit... not the best thing to use around a varactor, say).

Things I don't know... (could fill a library?)

- Where do I find a linear simulator for this sort of thing?

I'd normally throw things into SPICE and have a go, but I already know that's ludicrous at best; even if I can find models for the transistors (I think I actually have models for 2N5179, BFT92, etc., at least), and do my best to express the circuit as an approximate mesh of parasitic components, that still doesn't account for the striking nonlinearities of a mere oscillator (like the hysteretic gain response of an oscillator that's going into blocking behavior).  Only a transient simulation can produce those data, a timestep at a time; horribly inefficient.  A linear simulation is enough to run the sums and say, "yep, looks like it'll oscillate, go get the soldering iron", but tells nothing about the nonlinearities of the thing (which ultimately are best tested empirically, I guess).

- I'm a little familiar with S parameters, but I haven't used them quantitatively.  I think there are two main issues with using them: 1. they are ratiometric (you have to figure against system impedance and reflection coefficients and stuff to get ohms out), and 2. they only apply to the circuit specified in the datasheet/appnote (if specified at all), at the specified conditions.  Not, like, intrinsic properties at the device pins (unless it is, and I've missed that?).  Which would suggest a "correct" design process would require building the amplifier, then to make an oscillator for example, one would loop the input to the output and insert an absorptive (or diplexing) narrowband filter (or if there's enough s21, a lone resonator on the input might suffice).  A normal (reflective) filter could be used by taking into account gain, reflectivity and cable length (one example being a simple 1/4 wave stub).  But really, this doesn't sound anything like the design process of an oscillator, at least not by a sane engineer.  So there must be another way.

- Circuits are still circuits, but I don't have much experience with lead inductance and junction capacitance (as AC steady state phase shift, rather than how it appears in a switching circuit, which I am good at).  I've dabbled in HF tech before, but down there, a 2N3904 is still a 2N3904.  All the usual amplifier circuits are easily realized: common emitter, common base, differential, cascode, etc.  At worst, a resistor or ferrite bead might be used to clean up parasitic oscillations.  Up here, a 2N3904 is a dull lump, and that's not a parasitic oscillation, that's the desired signal!

- Resonator impedance: sure, I can estimate the inductance of a loop of wire, but how physically relevant is it at this frequency?  (Partial answer: a one inch loop should be a damned good inductor under, say, 1GHz or so.  I should still be well within the 'lumped constant' regime here.)
- Tapping, coupling, matching: transformers aren't transformers up here.  Or rather, they can't be made in whole numbers of turns, with asymptotically large inductances and near-unity coupling factors -- transformers as we like to think of them, not just mushy inductors sharing a few lines of flux.

This is one good reason I went for the secondary loop: it's adjustable, from about zero (pushed up against the divider wall) to, oh, maybe 30% equivalent (nearly touching the primary loop).  I observe the AGC bias rises when the loop is near, since it's coupling more power out, which needs more gain to keep going.  Makes sense.  But how would I interpret its impedance?  A shorted turn is surely not a resistive 50 ohm source.  I would be inclined to call it a rather low impedance (I am taking a fairly low ratio, after all), but it's got considerable stray (self and leakage) inductances as well.  Which could be matched, but that requires tuning, yadda yadda.

- I do have some mix #61 ferrite cores: dual aperture (hopefully should be good for mixer baluns, RFCs and the like?) and 0.38" toroids (same idea, or general purpose).  I think these should still be okay in the 100s of MHz, achieving enough impedance/inductance on one or a few turns to do the job, while using as few turns as possible in total (parasitic capacitances should add up quickly, no?).

- Coils are coils, except when they are not.  I can wind a nice little, oh, ten turn, 6 x 6 mm solenoid with a useful reactance in the band, but it will exhibit higher modes as well, which may affect harmonics and thus cause unexpected pulling or trapping or peaking or losses.  I don't know how to model or estimate this.

- Wires are wires, except when they are not.  Again, should be lumped constant regime, but even just that black wire leading over to the BNC connector I wonder about.  (I took a series of data the other day with the output cable apparently un-terminated... no wonder the AGC was so spiky!)



Anyway, after more poking around, I've got it up to 357-516MHz (0.5-32V), with stable AGC over the range (no dips, just a gradual change in bias, well behaved).  It seems changing the length did finally change the frequency, some.  I have no idea what was going on before.

Changes:
- Tuning loop 17mm length
- Base coupling capacitor 10pF
- Removed emitter bypass cap
- Emitter tap moved to 4mm from base end
- 10dB pad added between pickup loop and BNC output

Reflections and mismatch in the output seemed to contribute to dips and bumps in the AGC response.  Reducing loop gain (and the coupling cap value) seem to have alleviated most of the AGC hysteresis problems.  Haven't got the AGC quite stable under step changes in tuning, but for the most part, it seems to settle in under 100us.  But more things to optimize first!

Tim

[G0HZU]

Hi Tim
Your words above appear to be making this task appear a lot harder than it really is.. you only need it to work at 600MHz... Even with the crudest analysis you should be able to make a 600MHz VCO oscillate up near 600MHz first time

-95dBc/Hz phase noise at 10kHz offset is the baseline performance level you should be aiming for. It's better than the Rigol DS815TG LO phase noise spec but the Rigol has a very noisy LO spec for a spectrum analyser of something like -80dBc/Hz at 10kHz offset. -95dBc/Hz should be easy to get across the whole tuning range.

I'd normally throw things into SPICE and have a go, but I already know that's ludicrous at best

It wouldn't be my first tool of choice but I don't think it's ludicrous to use a SPICE simulator. It's only a 600MHz VCO so you can get away with quite simple component models and still get reasonable results. I would expect to be able to redesign your VCO and simulate it in SPICE and then build it and get very similar results. It's only 600MHz!

But I would prefer to start off with a linear simulation and look at open loop gain and the loaded Q and the phase response but for a basic low spec oscillator you can't really fail to get reliable  performance at 600MHz.

However, if you want to use it as a first LO in a spectrum analyser then you will need to make a really good design because you need low phase noise, low drift (assuming it's a free running VCO) and fairly linear tuning. If it's going to be a free running VCO then I think you need to aim at something like 5kHz/min maximum drift rate.

Note: I would have used a BBY31 SMD varactor diode (or a BB405B leaded type) for this type of VCO design. I don't think the diode you have chosen will give very good performance as a tuning diode. Also, I think the 10k steering resistance you have chosen will add noise and this would need to be changed to an LR in series to reduce the noise contribution.
I wouldn't have bothered with the AGC detector system and I would have used much lighter coupling to the active device and the varactor diode and I'd have used a fairly crude/fixed biasing method.

I've never tried a printed inductor for a VCO before. I wonder what the unloaded Q would be.

To meet the -95dBc/Hz phase noise spec the loaded Q for the system only needs to be about 5 or 6. However, now that I know it's
for a spectrum analyser LO I'd be thinking along the lines of using a VCO with a big fat printed resonator to get a much higher loaded Q.

But I think the baseline spec you can get with a very basic (non optimised) Hartley design would be about -98dBc/Hz phase noise @10kHz offset, 5kHz/min drift rate and maybe -3dBm output power (reduced RF power to minimise drift?)

The VCO would run from maybe 12V and need a tuning voltage of 2V to 20V to cover a 200MHz tuning range. eg 400-600MHz. A better (more complicated) design would improve on all of this, especially the phase noise.

[T3sl4co1l]

I don't like the idea of fixed bias for three reasons:
1. Distortion (I certainly can't see any harmonics coming from this thing, until they end up as spurs later on, anyway)
2. What's all that AC swing doing to my varactor?  If I want the widest range, I need the lowest voltages, which means spending the most time at low voltages.  A large swing, say 10Vpp, will necessarily run at a higher frequency (since it's 5V average*) than a small swing (1Vpp can comfortably run around 0.5V bias).
3. I want a reasonably constant voltage from this too, since that would be handy for crude signal generator purposes.  Like, if I know the V-F curve, I can do crude filter tuning by sweeping the varactor voltage.  Which would be handy for tweaking the 1st IF filter, and perhaps with some conversion, other things as well (a 350MHz BFO would get a solid 0-150MHz range, of course subject to further variation in signal level, distortion, filtering and so on..).

*Of course, the capacitance being nonlinear means the waveform is fat-bottomed (which goes back to 1., also), so the average will be closer to 0, maybe more like 2.5V or 3V in that case.

Varactor datasheets seem to suggest small signal amplitudes.  I don't know that they're always used that way.  Oscillators running flat out tend to saturate at some multiple of the supply, lots of voltage.  Yeah, it'll still tune, but it's trading nonlinearity for convenience.  I'd rather have something more linear, and... less convenient... I guess?  Maybe I shouldn't.

The tuning performance doesn't seem to be ridiculously terrible, at least among the diodes I have at hand right now.  Others kinds -- that have been claimed as usable by other resources -- seem to just short out the oscillator (low Q?).

Which is something that's always bothered me a little bit about diodes in general: sure there's junction capacitance, but how lossy is it?  I have a sneaking suspicion that, sure, you can put a schottky diode into a switching supply and eliminate reverse recovery losses as such.  But, you trade that for a steep C/V curve, which is almost as bad as reverse recovery (in some cases, it can cause more dissipation than a junction diode!), and because the charge stored by that capacitance is larger, you're putting in more reactive power, which means the Q of that capacitance matters more.  And I have a sneaking suspicion that it's generally pretty crappy, in whatever kind of (diffused, epitaxial, interdigitated, ..?) structures are normally used in those diodes.  It's tempting to use an MBR3040 to tune a circuit in the 100s, even 10s of kHz, but good luck getting it to behave.

Which then begs the question of, how do they manage to get high Q factors in varactors: what's so damned special about them?  Hyperabrupt doping, yes okay, but that's not all, they simply aren't normal junction diodes if they're that good.  (But whatever... excuse my being a physicist, semiconductors are interesting...)

Regarding drift, I'm pretty much expecting that I'll sit this thing in a temp controlled oven (I've made a nice one before, little more than a thermistor, op-amp, and a BJT soldered to a hearty copper plate on which the PCB is mounted), along with some also-probably-temp-sensitive diode clamp circuitry to try to get a linear V-F transfer function if I can calibrate it nice.  So, yeah, keeping drift down is good (only makes things better!), but more of a second-order priority I guess.  (And hey, I can always put thermistors on the thing for tweaking AGC and CV tempcos... oh, what a royal pain calibrating that would be...)

Tim

[KJDS]

Varactors have very low Q at low reverse voltages, hence 2V is a typical minimum to aim at without suffering horrible phase noise.

Always put a buffer on the output of the oscillator. The only exception to this rule is if you're making several million of them and want to save the cost of the extra transistor.

For a wide tuning bandwidth, i'd have gone with a single ended common base configuration, made sure I had decent negative resistance on the emitter using RFSim99 and then hung the variable tuned circuit on there in a way that I can move it about without too much hassle.*

*that's not actually the method I'd use, I'd go the Z-comm route first, or a similar configuration but with a harmonic balance analysis in Microwave Office before building it but the first approach is no fun and the second approach requires a good harmonic balance simulator which isn't readily available.

[G0HZU]

2. What's all that AC swing doing to my varactor?  If I want the widest range, I need the lowest voltages, which
means spending the most time at low voltages.  A large swing, say 10Vpp, will necessarily run at a higher frequency
(since it's 5V average*) than a small swing (1Vpp can comfortably run around 0.5V bias).

The BAT85 seems a curious choice for a tuning diode. I don't know its ESR up at UHF and there doesn't seem to be much capacitance swing except at very low voltages. Have you thought this through properly?

Because you have coupled the varactor direct to the top of the resonator it means you can't tolerate much in the way of RF voltage across the main resonator at low tuning voltages or it will upset the diode action.

Also, at low tuning voltages the pF/V of your tuning diode is huge. So I would guess you could see 65MHz/V VCO gain at the bottom end of your tuning range.

So just applying some basic physics tells me your VCO will be very noisy because the kTR noise from the 10 kohm steering resistor will 'noise modulate' the varactor. It is possible to predict the phase noise at a 10kHz offset based on the above data. It will be in the ballpark of -85dBc/Hz at a 10kHz offset if you are getting 65MHz/V VCO gain.

But you are also coupling the resonator direct to the active device (bad) and I think your loaded Q could be down near 3 at a guess. So this also gives very poor phase noise even without the noise from the tuning section.

It's many years since I did any detailed wideband VCO design so I'm a bit rusty on this stuff but it's fairly easy to approximate the performance of a VCO using a few basic sums. Your design looks really bad to me both electrically and also the layout looks bad too.

But I guess it all depends on what your requirements are

[G0HZU]

A few suggestions:

Try using proper tuning diodes and arrange to use two in a back to back configuration.

Try using a series LR network to reduce the kTR noise from the steering resistance.

Also, try to use lighter coupling to the active device. This will be better for noise and also drift.

I would suggest using a SMD transistor or maybe a BFR91.

Don't get too hung up on harmonic distortion. You can filter the harmonics and often they aren't that significant anyway because the VCO is usually buffered in a couple of amplifier stages before being fed to a mixer. So the signal can be limited and filtered after the VCO.

Simulate the circuit and learn about the physics of all the electronic components in the VCO and how they contribute noise and can degrade stability etc. Based on this and a few simple sums it is possible to predict the noise performance of a VCO quite quickly.