I've did some troubleshooting of a common-collector 3rd-hamornic crystal Colpitts oscillator this weekend, discovering that my emitter inductor (used to force harmonic operation) was too big, thereby sometimes allowing oscillation at the fundamental rather than the harmonic. In tracking this down, I re-read the relevant sections of Matthys's "Crystal Oscillator Circuits" and Hayward's "Radio Frequency Design." The two of them have somewhat different focuses, but I find it interesting that when it comes to the two capacitors in the "Colpitts network" Matthys says, "There is no requirement for any specific ratio of C1 [the base-emitter cap] to C2 [emitter-ground cap]." While arguably an entirely true statement, in contrast Hayward spends plenty of time discussing the ratio, even going so far as to create a graph of the oscillator's loop gain and phase as a function of one capacitor's value vs. the other, with the intent of letting you choose a gain that's not too much bigger than one so as try to minimize output waveform distortion.
Anyone have some thoughts on this? I have a suspicion that Hayward addresses the topic because he's trying hard to keep the transistor in small-signal (active) mode as much as possible, where Matthys right off-the-bat says that you can expect your transistor to be off ~80% of the time, saturated ~10% of the time, and operating actively ~10% of the time.
Matthys also suggests that the C2 (base-emitter) cap should generally be "between 40-70pF," whereas Hayward likes 100pF (and generally prefers larger caps... which I suspect reflects his emphasis on oscillators down at HF whereas Matthys often wanders into VHF territory).
I'm interested in whether those of you who design such oscillators tend to do it more from heuristic techniques (I have another book which suggests very heuristical approaches, e.g., "make the ratio of C2:C1 about 4:1") and expect to do a bit of tweaking on a bench vs. simulating the entire thing in SPICE first and expecting that it'll work just ducky on a bench. Also... when you're looking at a Colpitts oscillator, what do you tend to "see?" The most "natural" representation to *me* of the various configurations are:
Common base: Transistor thought of as a gain=1 emitter follower, and you're chasing loop gain. Hayward uses this point of analysis (and points out that it is, of course, an arbitrary choice.) I can see it readily as a negative resistance too. Common collector: Transistor thought of as a negative resistance, and with a few quick network transformations you're back to an RLC network where the transistor's negative R cancels our the load R. (Plastonek takes this approach. Matthys does loop gain analysis with just a bit of hand-waving thrown in. Hayward points out that all the topologies are really the same, you're just shifting the ground reference around...) Common emitter: Classic loop gain approach, with transistor base being a resistive load and the collector being a current source driving a pi network. Everyone seems to do it this way, since it's perhaps the most intuitively obvious topology and it's easy to analyze.
I also find it interestingly the Hayward's book -- aimed at hams -- seems to have one of the most comprehensive discussions of all this. The Art of Electronics only has a page on it!
---Joel