[Amps] MOSFET amp filtering - was: auto-tune

[Manfred Mornhinweg]

Gary,

your post pretty much condenses this matter. I will just comment on a 
few of the points:

> Inter modulation is just what the term implies. Two or more signals modulate
> each other in the signal path. For this to happen there needs to be a non
> linear device in the path to cause modulation to occur.

Right. My doubt was what kind of products, and at what intensities, a 
certain sort of nonlinearity causes. That's why I made my tests and 
simulations, and found what others surely have found long before: That 
cutting off an entire half of a signal does cause strong harmonic 
distortion, as expected, but causes strong IMD only at low frequencies, 
while the IMD products close to the tones are weak.

Instead a nonlinear behaviour over the remaining half wave does cause 
strong IMD products close to the tones, and thus is not acceptable.

> Of course SSB contains more than one signal
> (frequency) at any given time as does a two tone signal.

I like to always look at these things both in the frequency domain and 
in the time domain. Right, if you look in the frequency domain, with a 
sufficiently long sampling time, then you get a mix of many frequencies 
and amplitudes within a 2.4kHz bandwidth, for a typical SSB signal. And 
they vary all the time. But if you look at this SSB signal in the time 
domain, what you see is a single RF signal, with a clean sine shape, 
that varies in amplitude and in frequency. And these amplitude and 
frequency variations are rather slow, when compared to the carrier 
frequency. Except for one point: When the signal reachs zero, it can 
start increasing again very quickly - a fast inflection that contains 
higher frequencies in the envelope. And at that same point, the phase 
generally jumps by 180 degrees.

It's interesting to consider what each sort of nonlinearity does to this 
signal, as seen in the time domain. Seen over a few full RF cycles, the 
signal stays almost constant, that is, it's basically a single-tone 
signal, and thus a nonlinearity causes harmonics but no intermodulation. 
Seen over a longer time span, like thousands of RF cycles, the 
modulation in amplitude and phase becomes obvious, and so any 
nonlinearity causes intermodulation too - but at lower levels than the 
harmonics. Instead with signals that are modulated at high frequency, 
such as at RF frequencies, the intermodulation would be basically as 
strong as the harmonics.

> If those signals are pre-distorted before they reach the amplifier that in
> itself will not cause IMD. But if the distortion causes new and more
> frequencies to be generated then the linear amplifier will have more signals
> to deal with and can more easily be driven into overload which will cause
> the amplifier to operate farther from linear and that will cause IMD.

I would like to see this in a simpler way: Take an amplifier that isn't 
fully linear, but make sure it does have some response at all amplitude 
levels - no dead bands allowed. Drive it from an exciter, compare its 
instant output amplitude to the instant drive amplitude, and either 
predistort the drive signal or control the gain of the amplifier or 
control the attenuation of an attenuator between the driver and the amp. 
If properly done, this will render the output amplitude a very precise 
copy of the undistorted drive signal amplitude. At this point, I would 
expect that the only significant distortion that can remain, within the 
bandwidth over which the correction system works, is that coming fro any 
phase distortion introduced by the amplifier. And I WOULD LOVE if 
anybody here could derive equations to calculate how much IMD of what 
order would be caused by what type and amount of phase modulation! This 
is one thing I'm missing dearly in my plans for a high efficiency linear 
amplifier, and it seems that I'm too mathematically challenged to derive 
these equations myself!

> A class B amplifier doesn't cause IMD because it cuts off one half of the
> cycle provided that it remains linear over that half cycle. It does however
> generate high second harmonic energy.

That's what my simulation shows too.

> Class B audio amplifiers are usually run in push pull to cancel the harmonic
> distortion. But if they generated IMD because they were a class B amplifier
> they wouldn't be very useful as an audio amplifier.

In audio work any nonlinearity causes both harmonic and intermodulation 
distortion in roughly the same proportion, since we are always dealing 
with complex waveforms, not clean sines. But in RF signals carrying 
audio modulation, things are a bit different. Usually at RF we get far 
more harmonics than IMD. This needs to be always considered when audio 
people start working in RF, or vice-versa.

> Although a distortion free class B amplifier is hard to obtain, in reality
> what is called class B is usually AB, which is not totally cutoff when the
> lower half of the signal is present. This increases the linearity and allows
> close to class B operation.

Yes. Basically the gain of all semiconductor devices drops when the 
current is low. The class AB bias and the resulting idling current 
reduce this effect, tending to linearize the response in the small 
signal area - but it never becomes fully linear. Negative feedback at RF 
further linearizes amplifiers up to a usable level. External envelope 
feedback through a high gain, medium frequency error amplifier would 
linearize them much better, but it seems that no commercial amplifier 
uses it. The only ones I have seen using it are homebuilt ones, and with 
good results.

> Maybe I am missing something here?

What you are missing is just the more complex extensions to this, such 
as the phase modulation caused by voltage-variable capacitances in 
semiconductors. We tend to think about linearity as the ratio between 
output amplitude and input amplitude only, but in SSB work we do need to 
also look at the phase: The output phase should closely follow the input 
phase, not affected by amplitude changes. That is, the amplifier should 
be able to reproduce all signal amplitudes from zero to saturation, 
without changing the phase. And I suspect that this is the hardest part 
with semiconductors, and that it's far easier with tubes. And I lack the 
knowledge about how important it is in practice, that is, how much 
amplitude-to-phase-modulation is acceptable for a given, required IMD 
performance. If anybody can provide input on this, with firm numbers or 
with a way to calculate them, that would be very interesting!

Manfred

========================
Visit my hobby homepage!
http://ludens.cl
========================