A couple tricks that are misleading about the Wikipedia image:
It shows an amplifier, but it doesn't say what kind. Note I distinguished between CC/CV amps. An amplifier, in a very general sense, simply amplifies; it doesn't matter what characteristic it has. Your LT1007 SPICE model, however, has a CV characteristic. Which means the tank is in parallel with a very low impedance (the output impedance of the amplifier), which reduces its Q considerably -- or to put it another way, to allow the tank to "do its own thing", it would need to be a much lower resonant impedance (Z = sqrt(L/C)). But that would also be undesirable, because too low of a load forces the amplifier into a "short circuit" operating range, which causes higher power dissipation.
It's the same sort of problem as impedance matching a speaker. Yes, the speaker is 8 ohms (..more or less..). Yes, the amplifier has a [small signal] output impedance of fractional ohms (often, easily 0.01 ohms or thereabouts). Yes, you will get the most power out of it, for a given signal level (that level being perhaps a few millivolts!), when the load is matched (i.e., also milliohms -- basically a dead short..). No, you won't get full power (or necessarily even have the amplifier survive such treatment), because ultimately, the amplifier is not a linear system, and violates that premise of the power transfer theorem.
But you *need* aspects related to the power transfer theorem. Real filters (bandpass, in this case) depend upon power transfer -- energy sloshing between reactances -- to do their job correctly. A poorly matched filter has a whacked out response. Which is exactly the observation here -- you get a crappy oscillator under those conditions.
Most trivial solution: make it match. Put in a dumb resistor.
Now, rather than a Hartley oscillator (which looks weird with an op-amp, anyway), you have an op-amp, driving a parallel resonant tank through a series resistor. And the feedback comes off the tank. No need for a tapped inductor anymore, because we have inverting and noninverting inputs on the amplifier (it should use noninverting, because the tank's phase will be in-phase at resonance, so that the loop sums to, well, 0 degrees, but that's the same as 360 degrees, isn't it?). The op-amp output will be a square wave (give or take its response time and such), and the tank voltage will be somewhat lower than the amp's peak output. If you do an AC analysis, you'll see the inductor and capacitor reactances cancel in parallel, leaving whatever loss component remains (BTW, check that the default series and parallel resistances of LTSpice "passives" are realistic!), which acts as a simple voltage divider against that series resistor. So you get a smooth sine wave (almost), the tank is able to do its job, and the frequency is reasonably accurate.
And about that tapped inductor. Yeah, it works with independent inductors (just as the Colpitts works with a "tapped" capacitor), but it's best with coupled inductors. Do this in LTSpice by adding a coupling coefficient (K) statement addressing the two (or more) coupled inductors. Typically, you'd build a Hartley oscillator with anything from an air cored solenoid with the tap near one end, to a cored transformer with everything tightly coupled; k from 0.8 to 1.0 will be reasonably representative of anything in this range.
The thing about transistors is this: the output is constant current. The current varies with input voltage. A transconductance device -- conductance is amps per volt, so the gain of such an amplifier is output/input, or amps/volts, a conductance.. but it's not the conductance of a single port, so it's trans. (Tran-sistors don't really trans-resist, so maybe they aren't the most aptly named. But ohm = 1/mho, so maybe it's close enough.) The other thing is, simplified block diagrams like to indicate bulk things, like "oh by the way, here's some gain", and boom there's your triangle symbol. But it leaves all the magic to your imagination -- including how the signal goes in and out (coupling capacitors??), where the DC power is introduced (resistors, inductors??), and what impedances are like, in and out (is it constant current like a transistor, or constant voltage like an op-amp?). So I think this is where that image was coming from: an abstract block diagram, not an instructional schematic. Textbooks love to lie like that, too...
Anyways, doing it with transistors is a fine way, of course. The main downside you'll find -- with any simple oscillator circuit like this -- is the many ways in which it fails, often dramatically so.
Since you say you're interested in power conversion... that opens up still other concerns. Like... did you know diodes aren't perfectly quiet, either? PN junction diodes exhibit reverse recovery (essentially, it keeps conducting for a little time after being reversed -- using the water analogy, compare with the mass of the moving piece in a check valve), and even those which do not (Schottky diodes), still exhibit similar effects (nonlinear capacitance), and even in the absence of that (which you can't very well test in reality, but you can easily prepare such a device in... the SPICE simulator!), the simple fact that you are performing a rectification function necessarily creates discontinuities in the waveform -- harmonic distortion.
So, naturally -- although you can create a clean oscillator, it's a small part of the total performance. Which I suppose should be easy enough to expect.
What might not be as expected is, ease of filtering. You kind of have two options there:
- Make the operating frequency fairly high, so you can use LC filtering, and maybe an active filtering or cancelling circuit, without it being overly bulky and expensive.
- But not so high, or so noisy, that you get hairy stuff in the 20MHz+ range. That shit's hard to clean up. It doesn't like to stay in wires, so you tend to need a lot of shielding and ferrite beads and arcane radio-frequency magic like that. About your only advantage is that, because the frequency is high, the components required to filter it are also small -- but you may end up needing a lot of space (for the shielding) to deal with it.
- Or, make the operating frequency relatively low, so that, alright, you have to use awfully large capacitors to filter the ripple, but, you can tolerate much more ripple because you can regulate it down much more effectively with an LDO or whatever.
Regulators use feedback loops, which have a dominant-pole characteristic to their feedback: at frequencies 100 times less than the internal cutoff frequency, the feedback has a gain of 100, causing errors to be reduced by a factor of 100, so that 1V ripple at the input becomes 0.01V at the output. LDOs inherently have poor high-frequency performance, but they aren't bad at lower frequencies. HDOs (for lack of a better term; conventional regulators -- like the 7805 and such) have modest performance at most frequencies, but also get much better at lower frequencies, for the same reason. The increase is not unlimited as frequency goes down, because the DC gain is finite (it levels off at some point). But reductions in the 60dB (for LDOs) to 120dB+ (for precision op-amps -- PSRR and CMRR) range are absolutely achievable.
A quick alternative to a proper LDO, you can skip the voltage regulation (or place it before or after!) and use something to filter the noise. One such circuit is the BJT "capacitor multiplier" (an emitter follower with bypassed base). This can reduce noise, at any frequency the transistor can handle, by roughly the ratio of emitter resistance (very small, r_e = 26mV / Ic) to Early effect resistance (very large, usually in the range of 100V / Ic). Two or even three stages can eliminate just about anything -- to the point where, even at fairly modest frequencies, you *really need* shielding to be able to tell how quiet it really is! And, rather than directly cascading these stages, but using a chain of RCRCRC filtering on the bases instead, won't even cost you too much voltage drop.
Tim
Tim
Home
Help
Search
Login
Register
