That's a useful paper, and Chris Trask is pretty thorough.
It shows that the 2N3866 is best kept below 50mA, which the mag loop amps on the go do.
One thing he misses saying is that on the plot of the family of curves, you should consider a resistive load line, representing what the collector of the transistor will see. This load line for the NPN polarity devices will be a straight line from a point towards upper left to a point towards lower right. So you can now see why the region where the saturation current starts to curve off to the right becomes important because 'bent' operation encroaches into a larger and larger fraction of the available load line length.
So the obvious way out is to accept that you need to bias-up the transistor to not go above, say, 50mA and to rig the load line impedance to match this. But there is a second escape... use more of the space to the right by using a higher supply voltage.
The more curved saturation line can be seen in those curve tracer plots of transistors with higher Ft and plenty of those with lower noise figures. It is mostly due to the doping recipes used to get those high Ft values.
Things like 2N3866 and 2N5109 were shaped by the needs of the cable TV industry for small distribution amplifiers. In this application, linearity is the crucial need. But in Trask's curves, these devices do not look amongst the best. The answer to this conundrum is that these amplifiers were designed around 24v power supplies.
Transistors give their worst linearity in their low voltage-saturation region so keeping the voltage up and out of that zone is the direct fix. And it's a fix that works well.
Some transistors on a curve tracer will start to show the 'flat' part of their curves bending upwards at the right extreme. Yup, this is another source of non linearity, but it tends to be most strong at higher currents, and the load line tends to keep the transistor away from the high voltage and high current simultaneously condition. So it's less of a threat and Trask doesn't bother showing it.
Now, there is another effect on the go. It doesn't show up on curve tracer plots, and it is significant.
Notice the spec snippets include Ft, the transition frequency and the Miller capacitance Ccb - the base-collector feedback capacitance. These things are a related.
That capacitance is that of the collector - base junction, essentially a reverse biased diode where you could think of transistor action as a big leakage problem (
) controlled by carrier injection from the nearby forwards biased base - emitter junction. But you should still see it as a reverse biased diode junction.... As Bruce Forsyth said, "What do points make? points make prizes!" So what do reverse biased diodes make? Reverse biased diodes make Varactors.
Oh b*gger! varactor diodes are variable capacitors, controlled by the reverse voltage across them. And they are FAST. They have a long history as VHF and microwave frequency multipliers as well as parametric amplifiers. This doesn't sound good. Frequency multiplication is harmonic creation and that suggests non-linearity.
In an amplifier there is usually voltage gain, so the collector, being the output, swings around more than the input. With a common emitter amplifier, the collector - base capacitance is multiplied by the Miller effect making it more significant still. And as the collector swings around, Ccb swings with it. So the high-frequency roll off point, and the phase shift associated with it ALL swing around with the signal voltage.
So if you have a strong signal, it will pump Ccb and this will amplitude and phase modulate all the other signals. Inter-Modulation is the word.
So there is an RF intermod mechanism which is off of the curve tracer charts. This effect gives an advantage to those devices whose Ft is comfortably far above the wanted frequency range. So the high Ft devices, with the bendier saturation region on the curve tracer plots actually work a lot better in active antennae than you would expect from the plots. Using higher supply voltages and higher collector voltages also puts you onto the lower capacitance and less bent part of the varactor curve of Ccb.
Next!
Just because you're only interested in receiving a chosen frequency range doesn't mean that signals intercepted by your loop stop there. Oh, no! In fact a small loop becomes a progressively more efficient interceptor of signals at growing frequencies. From the point of view of an amplifier attached straight to a loop, it will see everything on the go in the RF environment. It's remarkably difficult to arrange filtering between loop and amplifier without spoiling the bandwidth of the loop, so amplifiers screwed directly to loops it has to be, then.
So low bandwidth transistors get to really flaunt their HF and VHF non-linearities, pumped by signals that are likely outside you range of interest. They don't understand you're not interested in them, so they don't know you want them to swerve around and miss your loop!
This is where resonated loops with remotely controlled variable capacitors (Mechanical ones, please, for god's sake not varactors!) come to the fore. A remote servomotor or stepper will do the job. I got given a number of really nice small steppers intended for very long life.... would you believe for driving the spools in one-armed bandits? They were microstepped for smooth rotation, and even the joggle as the spool halted to show you a lemon was faked by software.
A way around the Miller multiplication of Ccb effects... the reduction of frequency response as well as the less well known modulation by big signals is the cascode circuit. It pays dividends here.
Some time ago I wass designing myself a hifi amplifier. I wanted to get intrmod products more than an hundred dB down even at the top of the audio band and at high (1dB below clipping) levels. Ccb modulation turned out to be a major factor and the eventual structure I developed was heavily cascoded. It did the job. Besides, A colleague and myself were having a bit of a laugh. He was designing himself a very minimalist amplifier and I decided to design one using as many transistors as possible as a feat of creative p***-artistry. Of course, I wanted every transistor to do something useful... something I could point to as giving an advantage.
Could I hear this? most probably not, not even at that age. But I was having fun.
Anyway, Trask's paper shows how a curve tracer makes a mockery of a simple Hfe tester. The same goes for valves. You see the crazy prices the audiophile toob-rolling world has pushed AVO VCMs up to, well consider how much better the Tek and telequipment curve tracers are at providing you with information. But remember the bit above, about RF non linearity not appearing on curve tracers.
The one really nice bit about valves is they lack the varactor effect. They don't have depletion regions whose thickness gets modulated. The bad bit is you don't get 'PNP' complements. Maybe I have the seeds of another April the first audio article?
Enough typing for now. Time to rustle up some breakfast.
David