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

Manfred Mornhinweg manfred at ludens.cl
Mon Dec 12 10:44:15 EST 2016


Several good points were brought up.

Some comments:

Indeed a diplexer filter looks very much like a dummy load to the 
amplifier, aiding stability, while a normal lowpass filter looks highly 
reactive at harmonic frequencies. A narrow-band resonant network looks 
highly reactive at even more frequencies, specially those lower than the 
operating frequency. So the stability situation is even worse in that 
case. But it should be quite possible to make stable amplifiers, that 
remain stable even when facing these loads.

When somebody develops an amplifier while using a dummy load without any 
filters, indeed a diplexer filter avoids some surprises, compared to a 
lowpass filter. What remains is the "surprise" of lower output power, 
since much of what was measured into the dummy load was in the 
harmonics. That's specially the case when experimenters try to get good 
figures and drive their test amps into saturation, getting great output 
and efficiency. But 30% of that output is in the harmonics, so after 
adding a low pass filter the numbers look less spectacular. And when 
measuring IMD and staying at an IMD-acceptable power level, the output 
and efficiency are both back to "normal", which means "poor".

Conceptually there is no problem with using a diplexer filter, when the 
amplifier is designed and operated such that its harmonic output is low 
anyway. Let's say, a clean class AB amplifier with lots of feedback 
might have the third harmonic down by 30dB. This means that 99.9% of its 
output power is on the desired frequency. It needs the low pass filter 
only to meet FCC rules, and dissipating the 0.1% of harmonic power in a 
diplexer filter's dummy load is perfectly fine. On the other hand, an 
amplifier designed for saturated operation, far example used in FM, or 
externally linearized, would be terrible with a diplexer filter! 
Naturally it generates a waveform that has a lot of harmonic contents, 
and dissipating all this in a diplexer filter's dummy load kills the 
efficiency advantage of such an amplifier. Such an amp MUST use a plain 
filter, and must be designed to cope with the reactive load at harmonic 
frequencies.

Consider the example of a switchmode amplifier: A half bridge of RF 
MOSFETs (one with source to ground, drain to the source of the second 
MOSFET, that one having the drain connected to +B), with its output (the 
connection of the two MOSFETs) connected to a filter that has high 
impedance at harmonics (an inductor-input low pass filter, or a 
series-resonant tank). Now drive this half bridge with an RF square 
wave, or simply overdrive it with a sine wave. The MOSFETs will 
basically switch on and off in opposing phase. The voltage waveform will 
be pretty much a square wave on the lower bands, degrading into a 
trapezoid on the higher bands. That waveform contains a whole lot of 
harmonics. But the output filter only allows a fundamental-frequency 
current to flow. So the current in the MOSFETs will be clean half sines, 
while the voltage is sqaure. The efficiency will be close to 100% - how 
close depends on how fast the FETs switch, how well they saturate, and 
how low their capacitances are. In this case the "reflection of harmonic 
energy", if anybody wants to analyze it that way, doesn't cause added 
dissipation in the FETs, because each FET is either on or off, and so 
cannot dissipate much power at all!

The situation is very different in a class AB linear amp that has 
moderate feedback. Due to this feedback, the FETs behave like signal 
generators with some internal resistance. So the reflected harmonics 
will indeed cause some dissipation, which depends on their phase, among 
other things. If the filter is on the other side of a broadband 
transformer, the situation is quite bad, because the transformer adds a 
phase shift that varies a lot according to the band.

This shows once again that diplexer filters can be a good idea in some 
situations, but not in others.

The phase distortion arising from voltage-variable MOSFET capacitances 
indeed is something that worries me time and again. I don't have yet a 
good way to simulate or calculate that, and when measuring IMD it's hard 
to tell what comes from amplitude distortion and what from phase 
distortion. The good news is that LDMOSFETs capable of VHF and even UHF 
have capacitances that are comparatively low at HF, so the problem is 
reduced. I would really like to know how much distortion comes from 
this...

I would expect that additional phase distortion coming from MOSFET 
capacitances and crooked waveforms when not using a diplexer, should be 
low. The basic phase distortion coming from variable amplitudes is 
probably much larger, and anyway, during operation on a given frequency, 
the phase of the reflected harmonics stays constant. And putting a real 
resonant tank circuit directly on the drains, without a transformer in 
between, assuring a clean voltage waveform while the transistor conducts 
distorted current, should be best in reducing this form of distortion.

It's wrong to think that transistors are naturally broadbanded and tubes 
are naturally narrow-banded. The truth instead is that any device can be 
used broadbanded only over the range where its internal capacitances can 
still be absorbed into low Q lowpass networks. With transistors that's 
anywhere between a few MHz and probably 100MHz or so. With tubes, the 
upper limit is far lower. When operating above this capacitance-dictated 
limit, it's necessary to use a higher Q load circuit to absorb this 
capacitance, and it's here where things start getting narrow banded. In 
a typical HF tube amplifier, on 10 meters the tube's capacitance is so 
high that the Pi tank operates almost without any added tuning 
capacitance, and often needs a Q of 14 or even higher to be able to 
absorb the comparatively very high capacitance of the tube! This 
prevents making broadband HF tube amps, while it's easy to make 
broadband transistor amps - but not mandatory! In fact there are quite a 
few advantages to making narrow-band solid state amps.

There is another reason that makes tube broadband amps impractical: 
Broadband transformers usually need magnetic cores. The best material 
currently available for them are certain ferrite mixes. But the 
characteristics of this material are such that to work at the high 
voltage tubes need, the transformers either need very large cores, or a 
lot of turns. Both approaches result in excessive winding length, so 
that the transformer has a much too low frequency limit, given simply by 
the outphasing along those long wires!

Pure transmission-line transformers get around this restriction, but 
winding transformers from high impedance transmission lines isn't very 
practical either. And a pure transmission line transformer matching 
roughly 3000 ohm to 50 ohm needs eight separate, big ferrite cores wound 
with 400 ohm line. Or a three-stage transmission line transformer, using 
six cores, the first stage being wound with 1500 ohm transmission line...

Pi tanks are indeed impractical at low impedances, but there are enough 
other tank configurations that work fine at low impedances. So that's 
not a problem. Take a Pi tank, remove the tuning cap, and add it in 
series either with the input or the output. There you have two tank 
configurations that result in practical values when designed for low 
impedance transistors and 50 ohm loads.

Transistor amps using a resonant tank circuit directly from drain to 
antenna work very well, but only on one band. The problem is that it's 
very hard to do any bandswitching at impedance levels of a few ohm. So 
practical solid state amps first use a broadband transformer to raise 
the impedance to 50 ohm or anything in that order, and only then do the 
bandswitching - be it switching low pass filters, or bandpass ones. But 
this broadband transformer is the biggest factor causing poor efficiency 
and linearity! It's enlightening to use a circuit simulator to try a 
basic push pull RF amplifier using either a life-like transformer having 
  a coupling factor of 0.97 or so, and then a perfect transformer having 
a coupling factor of 1. This simple change often makes the efficiency 
change from 45% to 68%, the drain waveform to change from a horrible and 
funny thing to a clean sine wave, and the IMD go down by 10 or 15dB - at 
the same power level. Too bad that we don't have perfect transformers 
available, to build real amplifiers! In practice the performance of a 
real-world push pull solid state amp stands and falls with the quality 
of the coupling between the two drains. The usual bifiliar chokes are 
very often not good enough on the higher bands, the usual transformers 
with two brass tubes as primary do not provide any coupling between the 
drains (!), and the amplifiers often built nowadays that don't even have 
bifiliar chokes are even worse. No wonder many of them barely make 45% 
efficiency at full power, and have poor IMD.

Given that transformers with perfect coupling are impossible in 
practice, and at the low impedances resulting from 50V high power amps 
we cannot built transformers that have even decent coupling, it would be 
  best to avoid those transformers altogether.

To make a transformer-less, band-switched solid state amplifier, it 
would be really great to have transistors that can operate at a load 
impedance that falls in a switch-friendly and relay-friendly range. 
Something around 50 ohm is fine. But for a legal limit amp this requires 
a supply voltage around 300V. Yes, I'm aware of the ARF1505, but its 
capacitances are very high, much higher than acceptable for broadband 
use to 30MHz, and still pretty high for use with tuned tanks. This is 
also bad news regarding phase distortion from the voltage-induced 
variability of these capacitances. And I have read that this transistor 
isn't really suited to linear applications at 300V, due to hotspotting 
issues. It's only intended for CW applications in the industry, like 
certain welding machines and such. If anybody here has tried that 
transistor in linear mode, I would be very interested in the results.

At this time my best bet to obtain high-voltage RF transistors is to use 
many small cheap switchmode MOSFETs in parallel, with individual source 
and gate resistors added. But even this approach works well only to 
about 150V supply voltage. Above that the capacitances kill the 
performance. In my experience, at this voltage level one can make legal 
limit amplifiers in which the transformer issues are reduced to an 
acceptable level, but not really transformerless ones. But the 
switchmode transistors, even if optimally selected and applied, are 
always a bit marginal in an RF power application. They weren't made for 
it, of course. The best side about them is that their low cost makes 
them very suitable for experimenters: 50 dollars will buy you the active 
devices for a legal limit amp! And 150V power supplies are easier to 
make than either 50V ones or 3000V ones, at the 3kW power level.

About LDMOSFET reliability: The new "extra rugged" parts should be 
impossible to destroy from load mismatch, as long as their maximum safe 
temperature isn't exceeded. Since heating them up takes time, at least 
several milliseconds, the combination of these LDMOSFETs with 
comprehensive protection circuitry should result in a pretty 
indestructible amplifier. Instead older technology MOSFETs were prone to 
failure from drain voltage spikes. If an antenna gets disconnected 
during transmission, for example due to a worn coax connector or a 
faulty relay, the voltage can go up far too quickly for any protection 
circuit to respond. Old solid state devices were prone to fail under 
such conditions, no matter how much protections were built in. Modern 
extra rugged parts survive this. That's their main ruggedness advantage, 
in my opinion.

But these modern LDMOSFETs have "don't-touch-me" gates. They are very 
prone to failure from gate overvoltage. It's essential in the design of 
any amp using them to make sure that the gate voltage will never be 
exceeded - neither from overdriving, nor from self-oscillation, nor from 
switch-on or switch-off phenomena.

Going through the reports I have from other experimenters who blew 
LDMOSFETs, it's obvious that the prime cause of death in the hands of 
experimenters is overheating, when they didn't understand how to 
properly calculate the temperature rise of the junctions under the 
selected operating conditions, and the second one is gate overvoltage, 
often arising from self-oscillation. The gates need to be resistively 
loaded down over a wide frequency spectrum, wider than the one over 
which the amplifier has gain! And those loads must go from gate to 
source (ground), not from gate to gate! A push pull amplifier with 
gate-to-gate load is unloaded in common-mode, and will happily oscillate.

And even with proper gate loads and strictly limited drive power it's a 
good idea to add voltage clamps to the gates, just in case...

So, solid state amp design does have its pitfalls. But so does tube amp 
design! Only that tube amps have been used for so long, that all these 
pitfalls were worked out long ago, and properly published in a time when 
mainstream ham literature was still of a technical nature... Just think 
about glitch resistors, plate choke resonance isssues, hum modulation 
from AC in the filaments, crooked unstable drive impedance of grounded 
grid amps, and so on. Today we know them all, and the solutions to them. 
With high performance solid state amps, there is still a thing or two to 
learn.

Manfred

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


More information about the Amps mailing list