How to select transistor for oscillator ?

1. Equal capacitors?? But C = C1*C2 / (C1+C2)?
2. Note this ignores component losses: inductor, capacitors, and transistor too.
3. To follow on from that, the reflected impedance of the base (assuming a common emitter configuration) needs to be similarly high, otherwise the tank goes thud, or insufficient signal gets to the base.

Mind that, in addition to being a resonant tank and phase shift network, the CLC is also an impedance matching network. The capacitor ratio gives the voltage ratio, and therefore the impedance ratio squared, but that impedance is only valid if the capacitor impedance itself is low enough. Which is also to say, the tank Q is greater than the voltage ratio as well.

1. I would start with Zo = sqrt(L/C) and Fo = 1 / (2*pi*sqrt(L*C)) instead, where Zo is the resonant impedance, which will be Q times the equivalent total load impedance.
2. Yup, for a BJT of course; and for anything else, whatever it is. This is minimum, of course, and probably 2-3 times more is a good idea. (The average Gm drops at large signal levels, as amplitude saturates. In this way, you could design an oscillator to continue to operate, at relatively low amplitudes, without hitting saturation in the output (collector/etc.) side.)
3. Collector limiting?? Did you mean emitter? But that wouldn't use the whole VCC, unsure.

Anyway, a normal common emitter amplifier can be used here, setting, say: Vb ~ 0.2 Vcc, Vc ~ 0.6 Vcc and therefore Ve = Vb - Vbe.

That's Vc if a collector resistor is used, otherwise with a choke load (which would be a tapped coil in a Hartley oscillator). Obviously, don't forget to factor Rc into the total load resistance!

Note that gm = 1 / (Ic/Vth + 1/R_E), that is, r_e + R_E in the hybrid-pi model. R_E can be bypassed with a cap if desired.

1. Maximum base bias is Ic / hFE(min). Set base divider current (i.e., Vcc / (Rb1 + Rb2)) to about 10x that, to ensure the base voltage doesn't droop much. In practice, it may turn out that the base divider current is about Ic/10, but this isn't a saturated switch we're making here. :)

Finally, transistor only needs fT high enough. fT is not the definition of Gm rolloff -- that requires knowing h_ie at fT -- but it is related. Typically you'll have fT > 10*Fo.

Tim

in most cases it just needs enough gain at the chosen frequency. How much gain do you need? Other parameters sometimes matter of course.

NT

My common NPN gumdrops are BCX70 and BFS17 for faster stuff.

If you care about noise, take a look at Win's table, page 501 of AOE3. It pains me that he selected BCX70 as the first/worse transistor in the list.

Actually, I rarely use bipolar transistors these days.

They're just too "unstable" HEH

This kind of detailed thinking is what I might go thru to build a 500 MHz oscillator, not a 5 MHz oscillator. For 5 MHz just find Colpitts schematic online it will probably use a 2N3904. select components in tank for frequency and simulate in LTSpice. Tweak values if needed. Build circuit on small slab of copper clad it will work fine 95% chance. Can be all done in the time one takes to discuss it here...

Unless one really need low phase noise or distortion or very high frequency or some other specific parameter from an oscillator I think detailed mathematical analysis of low-frequency oscillators is kinda a waste of time. Yeah you can do it but the circuits are nonlinear and the results are often not particularly enlightening. This is what SPICE was invented for.

Amplifiers like to oscillate very much in the proper configuration and if they're in that configuration the tolerances can be pretty broad they don't need math to cajole them.

Bipolars are not unstable. But mosfets are usually easier to use.

The last bipolar transistor circuit that I designed was in fact a Colpitts oscillator. It's a 600 MHz VCO that uses a coaxial ceramic resonator and a BFT25A. The time consuming bit was temperature compensating it; I wound up buying a reel of custom-brewed N4700 caps.

John Larkin wrote

I use bipolars all the time, Here a 2.4 Giggle Hz oscillator:

Think those were BFR91. These are both to some degree frequency controlled by the base current, that changes Vce and CVce and thus frequency.

Anyways that was locked to the FPGA board that was locked to the Rubidium reference on the left.

For lower frequencies I of course use unijunction transistors.

For even higher frequencies there are 'sjips', like this I am working on:

Removed the crystal from the board, here testing if it still works with crystal on thin wires on the side. Thing is small... bad PCB, tracks fall off... A tunable external reference will be connected via a connector on the bottom.

Will attempt that locking thing with this LNB too.

For even higher frequencies I have some various color lasers.

:-)

But anyways bipolars are great, the naming here in Elektor magazine was TUP and TUN for Transistor Universal NPN and Transistor Universal PNP, BC547 BC557, also known in a smaller package as BC847 BC857 have only a Ft of 100 MHz or so, good enough for most things.

I also still have an old germanium OC76 and an OC140 germanium transistor in black painted glass envelope.

The "gimmick" there makes a like 3-5pF cap in that configuration, yeah?

I have some CK722's!

and some high-voltage opamps

Here's my oscillator:

The CCROs have TCs, Qs, and initial accuracies over 10x better than an LC. The higher-frequency parts have Qs in the thousands.

I look at it more as a transmission line, or LC cicuit. It will oscillate if serial impedance is low and phase shift close to 180 degrees. Basically it is an LC tuned circuit and phase reversing transformer.

. . -||- ) ( ) ( ) ( | | caps all over the length, like a transmission line, inductors coupled 180 degrees phase shift :-)

There are probably other ways to look at it, but it works great,

John Larkin wrote

What is the Ft?

hehe, that is funny.

Ceramics are not better than quartz crystals I think?

The TCXOs are cool, there is one in here:

They are available up into the GHz. As oscillators, they are a lot more pullable than crystals. But most CCROs are probably used as bandpass filters.

Can Spice do useful noise analysis of oscillators?

Dunno! Can't say I've ever needed to design such a high-performance oscillator myself. If I should need to I have a couple books on my shelf that go thru the math...

The "problem" such as it is, is that OP asked "what the BEST transistor" is and as usual the answer is "it depends." He didn't specify a noise requirement just a 5MHz Colpitts. so why not just a 2N3904? They work fine for the Colpitts at that frequency and there are many examples online.

It's kinda a reinvent-the-wheel situation usually when you go to reinvent-the-wheel you a) don't really learn too much and b) end up with a crappy wheel.

Asking what the best thing is is always the wrong approach. But the OP said 'What is the best way to select a transistor for an oscillator ?'

NT

No he didn't. He asked "what is the best way to select a transistor".