|
There are a few issues with the generic types of LC oscillators (Collpits, Armstrong, Hartley, Seiler, Clapp), which this tries to solve:
1. Frequency stability and phase noise - in any of the standard topologies, the active device participates heavily in the tank circuit. The problem is that most junction capacitances are bias dependent. And bias is anything but steady at VHF. What usually happens is the following:
- Signal amplitude causes device transconductance to change > device capacitances change > tank frequency is altered.
Most standard topologies couple the feedback mechanism heavily into the tank circuit. Especially Collpits and Armstrong - the tank circuit is directly a part of the positive feedback. Seiler oscillators try to decouple the tank circuit as much as possible, which does have a positive effect to some extend, but that leads to other kind of issues.
2. Amplitude stability across a range - Most oscillators don't have a stable amplitude across a wide tuning range. Seiler oscillators are considered with the highest tuning range amongst the standards, and still struggle at anything higher than 1.5:1. This is somewhat expected, because the tank impedance varies with frequency. For instance, a resonant tank at 108MHz would have 68Ohm impedance (100nH/22pF). Any non-linear stray capacitance would significantly alter the frequency. Just an example, one more pF of capacitance (22pF > 23pF), and the resonant frequency is now 105MHz. That's 3MHz deviation per pF. And how that relates to the amplitude, you're thinking? Back to point 1 - Signal amplitude causes device transconductance to change > device capacitances change > tank frequency is altered.
What I attempted to do differently here:
1. The active device which presents negative feedback is common base. This isolates the Miller effect almost entirely, so you're dealing with predictable Ccb. Cbe becomes a non-factor, and Cce is usually very low with RF transistors in the first place, and somewhat constant.
2. The coupling to get positive feedback happens through another amplification stage - common emitter. This allows the coupling capacitor to be very small - 3.3pF in the simulation, but can probably be lower with the right transistor in reality. That creates a capacitance compression effect, so that even if Ccb and Cbe vary, the tank will see them in series with a very small capacitor, thus compressing their effect on the tank.
3. An Automatic Gain Control (AGC) is formed by sampling the voltage at the common emitter stage, rather than directly on the tank, thus reducing the loading effect, and transforming the voltage into current, injected as tail current. Amplitude rises > tail current reduces, eventually reaching equilibrium. This has three very important effects on the circuit:
- The non-linear active device behavior (and participation into the tank) becomes predictable across the tuning range, as signal amplitude is constant. No amplitude variations > no capacitance changes in the active device.
- The amplitude is limited softly, rather than clamped. Positive effect of harmonic distortion and phase noise.
- Finally, and probably most importantly if you're doing FM or tuning via varicaps - the amplitude is limited at a level that doesn't forward bias the varicaps. No forward bias, no RF rectification, no unpredictable frequency pulling.
Coupled with a PLL loop, and this oscillator should be able to tune reliably across a huge tuning range, even significantly outperforming Vackar oscillators. A major improvement could be made if you use self biasing active devices, like tubes or JFETs, rather than BJTs. Then, the tuning range becomes limited only by the available gain to start the oscillations, the rest is self-limiting. Ah, yeah - also needs a buffer stage
|
