strange oscillator

We use Eclips Lite gates, 3 GHz stuff, and the fast ADCMP series comparators. Even a few GigaComm parts, roughly 7 GHz.

There is some really fast logic around, ballpark 10 ps, but it's like hundreds of dollars per gate.

Instant hostility to idiotic ideas is easy. Did you know you tried to make an oscillator with inductor Q's of 20,420?

Inductor reactance: 2*pi*2.5e-9*1.3e9 = 20.42 ohms

You do not realize the inductor ESR defaults to 1 milliohm if it is not specified. I have told you about this many times but you consider advice to be an insult and you ignore it. The resulting Q is

Where do you plan of getting inductors like that?

Next problem. Have you looked at the waveform at the source pin of your oscillator? How do you plan on getting useful output from that?

Again, you completely ignore helpful advice that could teach you how to go about designing a useful oscillator. The link I provided was a complete waste of time, but here it is again:

Don't be silly. What matters is the loaded Q. And this oscillator will run just fine even with an effective Q below 10. The bare coil Q is mostly irrelevant.

Many oscillator designs let the transistor current hit zero in order to limit the amplitude. Inevitably, the signal at the source ends up looking nasty, at least in the simulation. In a real circuit, it won't look nearly as bad. If you need a purer output, you can always take it from the top of the tank.

Jeroen Belleman

He doesn't like thinking about inductors - too complicated - and buys them off the shelf when he has to use them.

The business of designing a special purpose transformer and getting it wound strikes him as more trouble than it can possibly be worth (though he has wound a special purpose inductor around a pencil and made a lot of fuss about the result.)

He's here to garner flattery, not to listen to advice, nor humiliate himself by admitting that it was useful.

Your first inductor doesn't have any parallel capacitance which isn't all that physically realistic either.

John Larkin's approach to design looks like evolution in action. He does seem to come up with useful - or least marketable - circuits, but if there is a design element in the development process he doesn't seem to be willing to talk about it.

I don't consider your advice to be an insult. But real inductor Qs are high enough that I can ignore them here; the phemt has gobs of gain. The first step of circuit design is topology.

I doubt we'd see those squiggles in real life. But we don't truly trust Spice models for things like this; As Mike say, Spice is for training your instincts. After futzing with a lot of sims, we pick the best ones and build them.

I sell lots of products with oscillators. They always work. About the only rule I respect is Conservation of Energy. This biz is full of "Rules" that are wrong.

Of course you aren't all that picky about modelling your components carefully either.

A sim that has an inductor without any parallel capacitance may look better than one with a more realistic inductor.

If you understood the rules better, they'd look wrong a lot less often.

Changing the Q of the inductors to 10 has little effect.

However, adding 1pf stray capacitance across each inductor converts the squiggles into distorted sine waves suitable for driving a load.

Adding mutual coupling of 0.1 between the inductors will smooth the waveform even more.

However, these changes turn the circuit into a conventional Hartley, which nobody uses any more. Vastly improved perfomance is obtained with a conventional Colpitts. This requires fewer inductors which are expensive and adds one capacitor which is cheaper.

Congratulations. You have invented a Hartley oscillator.

Ignoring the Q is plain sloppy.

The waveform at the source pin is unusable. You should take a look for yourself.

Most oscillators operate in class C, where the conduction angle is less than 50%. The transistor is biased off most of the time.

Driving the oscillator into limiting means forward biasing the base- collector junction. This decreases the tank Q and generates unwanted harmonics. A much better option is to limit the energy fed into the tank so it equals the energy dissipated in the tank. This is normally achieved by reducing the current into the transistor.

The signal at the source should not look nasty. JL's circuit ignores component parasitics. This makes the sim irrelevant.

If the real circuit does not resemble the simulation, the simulation is bad. At the frequencies where SPICE is usable, a close correlation between actual and simulation should be possible, otherwise you are fooling yourself.

The signal at the source does not have to look ugly, even in the simulation. JL's oscillator can be made to look usable by adding 1pf caps across each inductor, and adding a bit of mutual coupling to smooth it even more. However, this turns it into a Hartley, which nobody uses any more.

Inductors are expensive. Changing the circut into a Colpitts gives vastly better performance and eliminates one inductor.

Taking the output from the top of the tank destroys the Q.

We prefer to use parts that we have or can buy easily, so tapped inductors are not preferred. Possibly one could plop two 0603 Ls next to one another and get some coupling... that would depend on placement.

Little surface-mount baluns can be fun, but the inductances are too high for a 1.5 GHz oscillator. A balun could do our phase splitting function if we use a single-ended oscillator.

But you said it's not a Hartley! I knew that.

I might do the oscillator with an ECL comparator, which gets me the differential output. That could be RC, or LC.

Some of the fast comparators have resistor programmable hysteresis, which might be useful in an RC oscillator.

It would also depend on the way the 0603 inductors had realised. Presumably there is going to be some kind of heical current path, but you have to know where the axis of the helix lies before you can place the parts in way that will let them couple.

Making the inductor yourself, or getting somebody to make what you need takes more work, but at least you know what you have got.

At GHz frequencies you can make transmission line transformers on a printed circuit board. Buried stripline is less dispersive than microstrip on the surface of the board, tracked above a buried ground plane, and you'd probably need to pay for some kind of Rogers high-frequency substrate, rather than relying on FR4-epoxy resin bonded glass fibre.

Probably not. Hysteresis is useful when the comparator is fast enough to switch a couple times as the input goes through the switching threshold. If you want a GHz oscillator the comparator probably isn't fast enough to generate problematic hash at cross-over.

On a sunny day (Thu, 03 Feb 2022 15:58:32 -0800) it happened John Larkin snipped-for-privacy@highlandtechnology.com wrote in snipped-for-privacy@4ax.com:

I have used the twisted wire oscillator, very stable too, tunable with a potmeter causing Vce changes that result in Ccb Cce changes etc:

Chip specs may change. add that ECL to make differential output. There are probably a million types of oscillators, cavities come to mind too.

We connected an MC10EP11 as a pecl gate, with pulldowns to ground. One diff outout pair was cross-connected to the input, and it oscillates at 1.5 GHz with a fairly square wave out the 2nd diff pair.

Tempco is mediocre. An LC would probably help. Given that I currently need a 1.5 GHz clock, doing this might be risky.

We'll try a comparator next.

I could buy a commercial VCO, but they are expensive.

Huh? What is wrong with temperature coefficient in this application?

Logic isn't intended to oscillate without a network (because oscillation forces the inputs into the logic margin, not a logically valid state). The comparator DOES specify behavior at the threshold, so that's 'safe' in a sense.

So, now you want to make, not an oscillator, but a clock locked to a frequency standard? Every PC has something like that, like maybe this

On a sunny day (Sun, 06 Feb 2022 09:54:32 -0800) it happened John Larkin snipped-for-privacy@highlandtechnology.com wrote in snipped-for-privacy@4ax.com:

Yes I have some Sirenza VCOs vco190-1572t.pdf

There is a cheaper one at about 1.5 GHz on ebay, dont know how good it is:

And I have a 960 MHz vco190-964t.pdf

The twisted wire is cheaper :-)

Huh?

Sirenza advertises GHz VCOs, but doesn't seem to want to sell tham in smallish quantities. MiniCircuits does, but expensive.

Varil seems to be gone. I can't find them online.

A PCB transmission line oscillator would be OK, I guess.

Anything that oscillates that high probably has enough slew rate, and '1.5 GHz' usually means between 1.45 and 1.55 is good enough. What, if any, criterion is in danger of thermal drift out of compliance? Do you know that dependence isn't desirable? The target of this clock surely has some temperature coefficients too.

Oops; I just read the fine print; it only does 500 MHz with direct crystal input. On the other hand, it has four outputs; you can combine them so the fundamental is cancelled but the third harmonic is enhanced. Kinda like a tripler, but no tank required, just resistors.