Oscillators at GHz frequencies -- question

On Tuesday, December 11, 2018 at 11:16:34 PM UTC+11, snipped-for-privacy@gmail.com wrot e:

do I combine the two sub-circuits, while

My decidedly inexpert impression is that if you want to design an oscillato r to work at about 1.5GHz you design it around transmission lines.

You can't get all that much of a range of impedances - and the simple formu las you find in application notes for ECL logic are only good between about 50R and 75R. I've got a couple of microwave texts that thrown in a few hig her order terms that get you up to perhaps 120R (with very thin tracks on r ather thick low dielectric constant substrates) and down to perhaps 25R.

Transmission line transformers can give you a bigger range of impedances, a nd you can always use a resistive divider.terminator to give you a sensible voltage swings at the base/gate even if the collector/drain is giving you a useful output swing (power output).

This will probably irritate people who are rather more expert into telling you what you really need to know.

On Tuesday, December 11, 2018 at 11:16:34 PM UTC+11, snipped-for-privacy@gmail.com wrot e:

do I combine the two sub-circuits, while

My decidedly inexpert impression is that if you want to design an oscillato r to work at about 1.5GHz you design it around transmission lines.

You can't get all that much of a range of impedances - and the simple formu las you find in application notes for ECL logic are only good between about 50R and 75R. I've got a couple of microwave texts that thrown in a few hig her order terms that get you up to perhaps 120R (with very thin tracks on r ather thick low dielectric constant substrates) and down to perhaps 25R.

Transmission line transformers can give you a bigger range of impedances, a nd you can always use a resistive divider/terminator to give you a sensible voltage swings at the base/gate even if the collector/drain is giving you a useful output swing (power output).

This will probably irritate people who are rather more expert into telling you what you really need to know.

Impedance matching is not a requirement for oscillator design.

You _can_ use 50R gain blocks and use a bandpass element plus suitable transmission line lengths to achieve overall 0 degree phase shifts. Going this way with a discrete transistor will give you headaches: The frequency response of your matching networks combined with the frequency response of your bandpass network has to provide the narrow resonance you want (high Q) together with the 0 degree phase shift.

If you are able to find a suitable equivalent circuit for your active element in the chosen configuration, it may be easier to directly find the element values that give you either the response* you like without any matching elements.

• Negative resistance + suitable resonator or open loop transfer function > 1 (+ zero phase shift). Both are equivalent and it's your choice.

Pere

Pere

Please remember the velocity factor of the transmission line, if you are using a microstrip or semirigid coax section.

Right. That would probably kill the resonator Q.

What is a common-emitter Colpitts?

I like coaxial ceramic resonators, which are shorted transmission lines with huge dielectric constants. You can pretend that one of them is a parallel LC.

I did, but only 20 minutes after I'd posted what you have responded to.

I immediately deleted the original post, and reposted a corrected version and groups.google now shows 13.3cm and 6.7cm.

In fact the standard velocity factor for most printed circuit substrates is about 0.67 which I've known for years, but it got screwed up in my mind with the 1GHz to 1.5GHz translation. A definite oops.

On Wednesday, December 12, 2018 at 1:32:35 AM UTC+11, snipped-for-privacy@ieee.org wro te:

ote:

ow do I combine the two sub-circuits, while

tor to work at about 1.5GHz you design it around transmission lines.

n line on regular printed circuit board materials, though you might be bett er off going for a high frequency substrate.

mulas you find in application notes for ECL logic are only good between abo ut 50R and 75R. I've got a couple of microwave texts that thrown in a few h igher order terms that get you up to perhaps 120R (with very thin tracks on rather thick low dielectric constant substrates) and down to perhaps 25R.

and you can always use a resistive divider/terminator to give you a sensib le voltage swings at the base/gate even if the collector/drain is giving yo u a useful output swing (power output).

Afterthought.

Tapered transmission lines can be used for impedance transformation, at lea st over a narrow frequency range. They should be more practical and more fl exible than stacking transmission lines to make a broad-band transmission l ine transformer.

I've never seen it done, but apparently it works

What is the oscillator to be used for? What are the specifications? How accurate is need to be? Does it need to be voltage controlled? What about phase noise? What temperature stability does it need?

I second "o pere o"'s post in that forget all this matching nonsense.

Overall starter method is to get the fastest (Ft) transistor that you can can and a SPICE model. Forget S parameter nonsense. An oscillator is nonlinear, making S parameters, essentially, useless. S parameters are valid only at the specific bias current they were measured at.

I would start with a common emitter driving a pi, Colpitts oscillator configuration. The other option is a Butler. This uses another tuned circuit to improve frequency response. However, without knowing a lot more about what you are trying achieve, recommendations are pretty much all guesswork.

-- Kevin Aylward

Its the topology in fig.2

fig.3 is a colpitts Common collector/emittor follower.

Most seem pretty confused as what a colpitts actually is. There are 3 "technical" topologies, common emitter, common collector and common base. All of which however, are essentially, topologically identical.

-- Kevin Aylward

I agree that S parameters are listed in manufacturer's data sheets for specified Vce and Ic and some frequency, say 600 MHz etc., However, the reason I am a confused is I have already tried "SPICE"-ing a common emitter Colpitts oscillator with a HFA3134(ft - 8 GHz) with a target frequency of 1.5 GHz and the output did not contain the target frequency -- I checked the output Fourier transform. The transistor was biased in the standard way for Vc = 0.5Vcc and Ve = 0.1Vcc. It also has a Dc blocking capacitor at the output, a emitter bypass capacitor.

My Pozar text, as well as those by Ludwig-Bretchenko and Grebennikov, spend a lot of time-effort on the S-parameter model, and datasheets from NXP specifically mention the biasing conditions when listing the S parameters.So immediately the S, Z matrix etc., come into play.

So, how do I resolve this mess.

dakupot wrote

Forget about S parameters, in my school days I had to do those calculations, never used since. And forget about simulations for anything above a GHz or even 100 MHz :-)

Get a real transistor, piece of copper board, build the oscillator, use scope, spectrum analyzer. And remember if you think it is now OK, any mechanical change / housing change / caps / trace length and even individual transistors will change the frequency and everything else. So here comes trimmers, varicaps, trimpots, PLL loops, crystals TCXO etc etc to get anywhere near.

Maybe at that point you get fed up, type ebay.com in your browser and for example order an ADF4350 development board for 60 $ or so:

Well, I did anyways, but this also works:

You can use a rtl_sdr USB DVB-T stick and my spectrum analyzer to check the output:

or increase range by adding a down mixer:

Simulations are, especially for RF, a wet dream, so are calculations with S-parameters. After you spend the days and hours and switch off that 'puter your have: Absolutely Nothing.

Build some test circuits and you are good to go.

The ADF4350 output:

RF is easy. :-)

Giggle Hertz

You should be able to find the oscillation frequency without Fourier transforms. Are you getting an oscillating signal? What's the frequency?

Pere

dakupot wrote

Agreed.

Yes. Modern design rarely needs more than dv/dt=i/c. Solving equations is what computers are for.

For an ASIC, design, simple not a problem to simulate and have a 1st time pass, whatever the frequency. For a simple *one transistor* discrete circuit is is straight forward to include pretty much all the rational parasitics of a real board into the simulation. Doesn't matter if you don't know them exactly. You have 4 or 5 inductances in leads, 4 or 5 capacitances at various nodes, maybe throw in a few spice transmission lines to boot. Its not rocket science and its all handled in the wash in spice. You can run

The deal is, it appears many just throw their hands up look at a circuits and say can't simulate it, and even try. If you don't know enough to add in reasonable estimates of the parasitics, you probably shouldn't be designing rf stuff.

Sure, and all routinely and easily included in the spice schematic.

Fortunately, in my line of ASIC design work, the chips have banks of selectable caps and varacters to digitally tune with no hands near the pots shifting the frequency. :-)

-- Kevin Aylward

A key point is that an oscillator runs to cut-off or hard limiting. The instantaneous values are all over the place. You need a real, time domain analysis. S parameters are only a small signal approximation. S parameters buy you absolutely nothing, today, with modern tools.

er.... wow.... measuring the time between consecutive waveform points is the usual way!

My own SuperSpice, although it does have FFT, has an option that automatically looks at the xing points and calculates the average frequency. It just prints it out after the run is finished.

Modern design needs to be highly optimised to succeed in the market. There is way too much competition trying to eat your lunch. As for the above example, this means, that one is going to iterate out the optimum bias current for starters. The likelihood is if you use the same application note from the vendor as your design basis, you will have another "me too" product.

Understand that today, all the billions and billions of products with billions and billions of analog transistors in them, are, essentially,

A nominal starting point in oscillator design, (after a basic dc bias setup) is to do small signal plots of loop gain. Small signal loop gain plots run much faster than trans, so you can quickly see if the thing is going to even oscillate near the target frequency at all.

There are examples as to how to do loop gain in LTSpice. Vary a bunch of components to check that you can get say, at least 6dB of gain at the net 0 deg phase point, and that the 0 deg phase point is where it needs to be. If this looks ok run trans.

Understand the *principles*. Ignore the big equations, they are useless for design. Spice will solve it, whatever it is. Leave big equations to the likes of Steven Hawkins (RIP).

For an oscillator, you need to get a loop gain > unity, and the zero frequency point. The non-linear transient runs will get the exat frequency, noting that the capacitances are nonlinear and saturation causes extra nominal phase shifts, which all move the frequency from its small signal value.

-- Kevin Aylward

Post your .asc file! And I'll post my CCR Colpitts sim.

Please post your sim

Oh, all right.

That one works in real life, but the tempco was awful. Pulling the base a bit more negative helped a lot.

Adding some NTC caps made the tempco do this:

More: