Jeff,
Interesting contribution, and I will comment on many of your points.
PC water cooling technology is able to get rid of 150W of CPU heat from a 1
cm die with only 10-20 degrees of temperature rise. For an RF amplifier,
allowing 70 degrees of rise, the same CPU thermal block should support 1000W
of heat takeaway per transistor. The best water cooling blocks have thermal
resistance below 0.03 deg C per Watt, for example see
http://skinneelabs.com/v1-swiftech-mcr-320-qp/4/.
In my planned, begun, but not nearly close to completed class AB legal limit
amplifier using $1 FETs, I'm using water cooling, but with homebuilt cooling
blocks rather than those from the PC industry. I did the detail calculation for
them, and found that while water cooling can extract significantly more heat
than air cooling, from a given transistor, it's still not easy to get a huge lot
of heat out of a small part. The reason is the finite thermal conductivity of
copper, along with the finite heat transfer from copper to water. Even with
water you need a certain surface to get a desired value of copper-to-water
thermal resistance, and this surface is much larger than the flange of a high
power transistor. So you need copper to bridge the distance from the flange to
all your water contact surface, and that becomes a bottleneck.
So, a high efficiency amplifier simplifies things a lot!
When checking data on thermal resistance of heatsinks or water cooling blocks,
you have to keep in mind that unless otherwise stated, the published values are
valid only when the heat is applied evenly over the whole surface. But with a
point source, the thermal resistance is much higher than published, because of
the thermal bottleneck formed in the small area under the transistor.
Hams need to think more about the difference between thermal mass and thermal
resistance. The solid state amp people often add a big heat sink with
copper and aluminum...that's thermal mass. The thermal resistance is set by
the fans blowing across the fins.
Don't forget that a thick base plate does reduce the thermal resistance from the
transistor to the fins, which is usually not small enough to ignore. And the
copper spreader is essential when using small devices that produce a lot of
heat, to bring the thermal resistance of the path just under teh transistor into
a usable range. When mounting a transistor on aluminium, and then asking it to
dissipate several hundred watts of heat, the thermal resistance in the cubic
centimeter of aluminium under the transistor is high enough to cause a burnout.
There's really no need for thermal mass when a very thin bit of copper with
water cooling is a much lower thermal resistance from Silicon die to water.
It's true that thermal mass is completely unnecessary, if the amplifier is rated
for CCS. You need low thermal resistance, and that's it. But with amplifiers
that have higher peak dissipation than average, such as ICAS rated ones, or ones
rated only for low duty cycle modes, you can get away cheaper by combining a
small dissipation capability with some thermal mass.
Modern RF transistors have enough margin to handle 100% VSWR for some
milliseconds (Freescale MRFE6VP1k25, NXP BLF-188) which is achieved by having
enough Voltage tolerance to handle peak voltage and enough current/thermal
tolerance to give the shutdown circuits enough time to act.
Old ones have that too, but maybe it's less clearly specified! ;-)
Many VHF and UHF RF power transistors (typically LDMOS) have good linearity
(-40 dB IMD) because the Cellular market demands it.
When operating at UHF, the gain is low due to the high internal feedback
capacitance. That _is_ built-in negative feedback! Some would say it's a way of
turning a bug into a feature. But when you use these same transistors at HF, the
capacitance becomes comparatively small, and the IMD gets worse while the gain
goes up.
Even if you use FETs that are barely able to operate at 30MHz, you might get
good IMD on 10 meters but very poor on 160, unless you add the proper external
negative feedback.
But perhaps some of the transistors you mean get good IMD specs from using lots
of internal source ballasting. That should hold true even at low frequencies,
and is a good thing to do in conventional linear amplifiers. For switchmode
amplifers instead, such ballasting means just a useless reduction of efficiency.
Then amplifier designers added additional linearity with digital pre-distort
and typically get a system IMD of -60 dB for 10 MHz of bandwidth. Practical
pre-distort algorithms can improve IMD by 15-20 dB, although lab experiments
can show 30 dB improvement in any one case.
But that's typically more in the realm of integrated transmitters, and not in
add-on amplifiers. I have thought a lot about such schemes, but the idea of
taking a 100W drive signal, dissecting it into phase and amplitude, digitize it,
digitally process and predistort it, and then output processed phase and
amplitude to put them back together in a 1500W stage, seems rather unreasonable.
I think digital predistortion of such a kind belongs in complete transmitters
and not in amplifiers.
On the other hand, maybe can implement it in such a way that despite its
complexity it's inexpensive enough to make it reasonable even in amplifiers.
Achieving good linearity of the transistor requires precise control of the
impedance seen by the part around the peaks of the modulation cycle...which
probably can't be achieved with ham antennas that blow in the wind and get
wet.
Sorry, I don't get you here. Do you mean that ham antennas dynamically vary
their impedance according to the signal's envelope? I don't think they do
that...
What is true is that they change impedance with heat, rain, snow, ice,
frequency, time, and sometimes with the wind. But that's independent of the
modulation cycle! And it happens with all antennas, not only ham's!
So the solution is to use active pre-distort which measures distortion in
real time and controls the input to the amplifier.
Yes. It can control either the input, or the amplifier proper, for example by
bias modulation.
I've been working on the design of a system which samples the RF output of
the amplifier (or transceiver) and adds pre-distort to the microphone audio
to cancel out distortion. The whole SSB signal chain is linear and the
output is just the frequency shifted input, so it should work.
How do you intend to handle the delay through the radio? My radio's delay, from
mic input to RF output, is 1.95ms in SSB. This is a conventional radio. I
understand that DSP-based radios have even longer delays. So, I fear that by
using feedback all around the radio and amplifier, you won't be able to
linearize the amplifer, but instead you will create a mighty oscillator!
There are some issues: for example a pre-distort system like this would work
very hard to cancel out the audio compression of the transceiver...so audio
compression and frequency response adjustment would have to occur in the MIC
path before the pre-distort.
That would be just a minor problem. You could actually include the audio
compression into the linearization network, which would then not be a true
linearizer, but a circuit intended to create a desired amount of envelope
distortion, in addition to ironing out the unwanted distortions. But the problem
with the delay in the radio seems to require a time machine to fix! Basically
when you sense your output amplitude, you have to compare it to the audio signal
that was there about 2ms before, and inject your correction 2ms before you can
sense the output! That's not possible.
So you would have to make a feedforward correcting system, that learns about the
radio/amplifier chain's performance from the sensed signals, and then applies
the learned curve to the feedforward section. That might work, but requires DSP.
I don't see how you could do it with a purely analog circuit. With an analog
circuit, you can work only if the total delay through the corrected circuitry is
short, compared to the maximum frequency you want to correct at.
Protection circuits are not really that complex any more because a $30
Arduino (or equivalent) can sample the voltages, currents, and RF levels well
enough to protect the amp.
In fact analog protection circuits are simpler, and MUCH faster, than involving
an Arduino!
Moore's Law has done it's work. Once a designer commits to digitizing and
processing many points in the amplifier design for protection, it is a
relatively short step to increasing the sampling speed and implementing
digital predistortion.
Yes, that's indeed often true. I have found that when I decide to include a
microcontroller to do "just a few simple functions" in one of my designs, I end
up putting most of the functionality into the microcontroller, and the circuit
gets very simple and inexpensive! But some things simply require response times
too fast to rely on a microcontroller. Like shutting down a shorted amplifier in
a few RF cyles.
As an aside, what's the problem with a SWR sensor between LPF and antenna?
Several:
- If for any reason drive is applied on a band higher than the selected low pass
filter, there is near-infinite SWR and no protection.
- Any high SWR due to bad low pass filter elements, or relays, is not detected,
and can kill the transistors.
- The harmonics and IMD created by the SWR sensor's diodes are sent to the
antenna, unattenuated. Putting the sensor before the low pass filter would at
least attenuate the harmonics, but not the IMD. Eliminating the SWR sensor
eliminates that IMD. Anyway with a good design the level of these distortions
might be small enough to ignore. But there are a lot of SWR sensors and meters
out there that create a lot of IMD and harmonics!
- SWR sensors just sense SWR. But any specific value of SWR, except an exact
1:1, can be caused by any of an infinite combinations of resistance and
reactance. Of these, some may be dangerous to the amp, and others not. So,
instead of sensing SWR, it would be better to sense the actual conditions that
burn out transistors. That is, voltage, current, and power. SWR sensors are only
used because it's simpler and cheaper to use an SWR sensor, than to sense those
three important parameters.
- SWR sensors inserted before the low pass filter aren't very useful, because
there is high SWR there anyway, because of the harmonics created by the
amplifier. So, if an SWR before the filter is useless, and an SWR sensor after
the filter is only partially useful, but also problematic, my opinion is to
forget the SWR sensor, and work just with voltage, current and power
protections. And in a bridge amplifier we don't need the voltage sensing, and in
a switching amplifier having high efficiency we probably don't need any way of
measuring dissipated power. So the protection circuit would need just a current
sensor and fast shutdown of either the drive, the bias or the power supply. Bias
is fastest to shut down, normally. Just don't shut it down so fast that it
creates a voltage spike in any inductance associated with the power supply!
There is an example of these kinds of "amplifiers" from the induction heater
industry. The designs operate IGFETs at line voltage and use pulse width
modulation to generate "RF" at about 50 KHz to drive the heater coil. You
can get a 1.8 KW hot plate for less than $100.
Yes. And many other systems use this configuration too.
IGFETs don't have the switching speed to operate at HF frequencies, but we
can probably learn from how they design their systems.
Don't be too sure! I have successfully ran $1 FETs in switchmode, with
reasonable efficiency, up to about 10MHz. The efficiency is not the 98 to 99%
typical of 50kHz operation, but more like 90%. However this is still quite
acceptable. One pair of those $1 FETs can give 200 watts of RF output in this
mode. In those tests I didn't use any antiparallel diodes, and so I had to be
VERY careful with tuning, because any slight detuning causes conduction of the
body diodes during a small part of the cycle, and that makes the loss shoot
through the roof, and kills the FETs in just a few milliseconds, unless the
protection is faster.
The problem is that when I search for inexpensive FETs that can work at 400V or
above, and can switch with acceptable efficiency at up to 30MHz, so far I have
given up in frustration. If anybody has a good hint in this regard, I'm very
interested! Such FETs need to have extremely fast slew rate, low capacitances,
and low gate resistance. Instead the current rating, power dissipation, RdsON,
and transconductance, don't need to be anything spectacular. SMDs are a good
option, because their lead inductance should generally be lower than that of
through-hole devices, and the power dissipation required for a switching
amplifier can be handled easily by SMDs.
A SS high power amplifier consists of a three high power stages; Power factor
correction, Power Voltage generation, and RF amplification.
Sometimes you can skip the second stage, and/or the first stage! The simplest
configuration possible is just an off-line rectifier/filter, followed by the RF
stage. But that has a poor voltage regulation, barely a little better than tube
amps, and a poor power factor, just like tube amps. We should do better than
that! So, the minimal practical _good_ configuration is the power factor
corrector stage plus the RF amp.
Instead of using PWM for controlling RF output in the RF stage, the amplifier
can the use the existing PWM in the power supply stage to control the main
power voltage to the RF amplifier stage.
Yes. And I have done it, but I have found quite a few problems. These problems
are described in the literature, but one gets a real appreciation of them only
when actually trying it hands-on!
The RF amplifier only sees enough supply voltage to meet the power output
needs of modulation cycle and so operates at maximum efficiency at all times
with effectively near zero idling current.
In an ideal situation, yes. But it's hard to operate that way.
Modulating the supply voltage has a hard to predict effect on linearity and
IMD...but we let the digital pre-distort measure and compensate for that in
real time.
Yes. The problem is in two areas: One is unwanted phase modulation when the
internal FET capacitances change due to varying supply voltage. This creates the
much commented "AM to PM effect". The other problem I have encountered, and that
is less often described, is drive bleedthrough to the output, when the supply is
at low voltages. At some point the RF output no longer follows the supply
voltage, and the phase at the output inverts! So, any supply modulation scheme
needs to be complemented by some other system that controls the drive at low
power times, which MIGHT be possible to do just by bias modulation. This is an
area I would like to experiment in. Supply modulation for the mid and upper
parts of the envelope, but freezing the supply at maybe 20% nominal, and
modulating the bias for correction, at low levels.
And I fear that I will encounter a huge new can of worms when trying that!
Mostly in the area of phase modulation.
Two other high efficiency techniques from cellular industry should be
considered. Doherty amplfiers and out-phasing. A Doherty amplifier uses a
linear stage and a separate peaking amplifier running in class C.
I shudder when thinking about the difficulties of making such a Doherty pair
work well together over the whole 1.8 to 30MHz range! On a single band it would
be easy...
Outphasing amplifiers modulate the current draw to control RF output (see
"Ampliphase"). The amplifier consists of two output stages in parallel
operating out of phase. When perfectly out of phase the power cancels and
each amplifier sees a very high impedance load, as the phase shifts to in
phase, the power output add and the amplifiers see a match load.
I made a test amplifier one time, using a full bridge, each side of the bridge
driven by a full amplitude square wave, and modulating the output amplitude by
outphasing the two sides of the bridge. It's not a linear process, so it needs
linearization, but the big problem is that it was never possible with my
amplifier to produce a clean zero output! Some power would always leak through.
This is just a minor imperfection for an AM transmitter, which will then be
capable of just 98% modulation instead of 100%. But in SSB, it's unacceptable. I
haven't found a fix for this.
The last thing to consider is whether to implement an amplifier or to
implement a complete transmitter.
Yes. YES! It's the same point I have been considering for a long time! Making an
add-on amplifier is one thing, and should be done with a relatively simple
system, making best possible use of the available 100W drive signal. Instead a
legal limit power stage that synthesizes the signal from its phase and amplitude
components, is better applied in a DSP-based transceiver, and not in an add-on
amplifier for existing transceivers.
Power supply modulation and out-phasing techniques lend themselves to
separating the modulation into amplitude and frequency components and it's
easier to perform that separation at audio than at RF. The predistort system
works best on MIC audio so we have control of that part of the system.
Lastly, the RF power amplifier has 30 dB of gain and doesn't need a 100W
transceiver driving it. We could include something like a Softrock for the
transmitter (and use the receiver for the RF sampling of the output)
We have been thinking in the same line. I could very well imagine a DSP-based
transceiver using the OpenDSR modules, or perhaps something simpler (reduced to
a single board), parametrically driving a 1.5kW swutching output stage. All
nicely built into a single, and quite compact, box. With power supply and
everything included. A 1.5kW, all mode, high performance, high efficiency,
digital transceiver.
Now you know my ultimate dream, folks.
It doesn't absolutely need to have a direct sampling receiver, though. I would
be satisfied with a DDS-controlled traditional front end, moving over to DSP at
the IF level. The transmit side could use the DDS's phase modulation functions,
under the control of the DSP, along with envelope modulation. That would be a
practical, high performance implementation, and less expensive than direct
sampling, that still allows the whole range of digital tricks to be done, such
as predistortion.
I have been playing with an advanced DDS, but I still haven't started any more
involved DSP programming.
Anyone actually developing this concept to market maturity can put all
existing ham amplifier manufacturers out of business. A scaring thought -
for them!
no, the old farts will resist ;-)
Right! ;-)
- The transistors needed are very much smaller and cheaper than those
needed for class AB, due to low dissipation requirements.
A niggle, but the die will be the same size, and peak thermal density will
be the same.
I'm not a specialist in semiconductor manufacturing, but I think that the die
size required is smaller, because the current rating we need is small, and the
power rating too.
The package can be cheap epoxy, but it would be really good if it had very low
source terminal inductance. Some such packages exist, and the more innovative
semiconductor companies are pioneering them.
So, cheaper only if produced in very high quantity because they are used in
power suppliers or other applications.
Yes, I'm thinking 100% in the line of fast switchmode FETs, not FETs purposely
made for RF. Because anything labelled for RF is automatically at least 20 times
as expensive as a similar device labelled for switchmode use.
In switching amplifiers its easier to run FETs in parallel, than in linear ones.
So the option of using a large number of very small and very cheap FETs is
attractive. That method solves the source lead inductance problem, too.
Another aside. In crowded band conditions, the - 30 dB IMD from many
transmitters across the band is the most significant limitation on
communication. Buying an $8000 transceiver with 100 dB of dynamic range
doesn't do any good when the splatter from 10 nearby stations raises the
noise floor. I would be in favor of requiring improved IMD in transmitters
and power amplifiers...say -40 dB for 100W transceivers and - 50 dB for 1000W
transmitters. This would drive the state of the art, improve the bands
dramatically.
The problem is that it wouldn't fix old radios and amps. You cannot outlaw them,
and many of us hams keep using old stuff for many decades! And another problem
is that the best radio design is futile, if the radio falls into the hands of
some ham who disables the ALC to get some more power. I know loads of hams who
always, as a matter of fact, take the cover off any new radio and turn down the
ALC pot to zero, to "unleash the true power of the radio". I hear them often
talking about their idea that most ham transceivers are really 150 or 160 watt
radios "intentionally limited" by the evil manufacturer to just 100 watt! So
they "liberate" their poor radios, and happily splatter around on the bands...
Three mic preamps in a chain help in "getting the ultimate power out of that radio".
I guess you all have heard such hams on the bands... In my part of the world
they are the rule more than the exception.
And the difficulties involved in this approach:
- Finding ways to get around the limitations of present-day RF power
transistors, in terms of voltage-dependent internal capacitances, slew rate
limitations, and high voltage handling.
Solved at VHF with unmatched LDMOS FET's from Freescale and NXP. Will those
work at HF?
Those can't handle high voltages. So you have to deal with impedances of a few
ohm. And also they are expensive. Remember that I'm a cheapskate! ;-)
As for working on HF, sure they will, but it might be hard to get acceptable
stability with some of them. While a switching amplifier is safe from
oscillation while it is in eitehr state, it can still oscillate during the
transitions between states.
The most complex part might be the linearity management system. But that a
separable component with it's own value.
Yes. It's perfectly reasonable to first built a highly efficient, single band,
non-linear CW amplifier, then add linearization, and then see how to develop
this into a multiband design.
> We need a box which samples the RF
on any HF band with 2KW passing through it.
That's trivially simple.
A box which generates predistort
on the mic audio. And the software to process it.
That is far more difficult, but as it goes with software, one can add one line
at a time and test it, and try many differnet approaches, without having to
order new parts for every test! That makes software development easier than
hardware.
Second most complex part, building an 90% efficient RF stage based on switch
mode (class C) transistors and a modulated power supply voltage.
Those are two parts.
This would
be coupled with a separation of amplitude and phase in either RF or audio.
For an amplifier, it would have to be RF.
The DC power supply would put out 0 to 50V (or 0 to 400V) with a bandwidth of
3 KHz.
No. It needs to be higher. While the envelope of an AM signal is spectrally
limited to the same frequency range as the audio transmitted, the envelope of an
SSB signal has a much larger bandwidth. Theoretically it's an infinite
bandwidth. In practice it has been said that 30kHz bandwidth of the power supply
modulation is good enough.
The modulated power supply would have to run at 1 MHz or greater
switching frequency to achieve this bandwidth.
It should be possible to apply the law of tens. To get 30kHz bandwidth, a
switching frequency of 300kHz might just be enough. But then the filter must be
really good, or we will get 300kHz sidebands on our signal!
> The input to the modulated
power supply would be 400 VDC from power factor corrected direct line
rectification.
Yes.
Where to put the line isolation transformer? The 400 VDC from line
rectification and PFC is not isolated. The modulated power supply is simpler
if it doesn't have a transformer (would need 2 KW transformer for 1 MHz). Is
it cheaper to put the transformer in the RF path?
Certainly. A 1:1, 50 ohm RF isolation transformer is easy, small and
inexpensive. And at the drive side we have a transformer anyway.
Radical but scary thought, what about just capacitively coupling the RF
output and skip the whole line isolation completely?
Perfectly possible, but you need small caps that are safe at a few kilovolts to
meet safety standards, and they also must be able to carry the full RF current,
around 5A in this case. And the amplifier _must_ be grounded whenever used, or
it will give the operator a mighty (although not dangerous) tickle. Worse,
ungrounded it could damage the radio when plugging in the PL-259! So I think
that RF transformer isolation is the way to go.
The water cooling loops
would have to isolate the RF transistor case from safety ground, that used to
work for plate voltages and this would only be a few hundred volts.
Water? If the amplifier is high efficency, we won't need water cooling! Power
dissipation might be 150W during a brick-on-key test. That's easy to handle with
normal heatsinks and a small thermostatically controlled fan. During normal SSB
operation, the fan would never come on, if the heatsink gets reasonable natural
air flow.
Maybe we should start a collaborative open project, developing this thing!
The final goal: A solid state amplifier no larger nor heavier than a
typical HF radio, that can produce solid legal limit output in all modes,
with no time limit, with good IMD performance and high reliability, a total
parts cost around $500, and selling to those who are too lazy to build it,
for around $1000.
I'm willing to help. I've wanted to do something like this for a while now.
Good! Project is started, then! This is a historical moment! ;-)
And finally, a quick comment to Isaac:
As you have seen above, EER and class D are both part of the game, but when
playing with EER I ran into trouble with AM to PM conversion, and drive
feedthrough. Perhaps I shouldn't publish this, but the worst case IMD I got was
-18dB. And the best case IMD wasn't much better! Instead when trying an AM
signal with 50% modulation, the IMD produced by the EER approach was excellent.
So, EER with MOSFETs apparently can't be used all the way down to zero power.
A few years ago Saulo Quaggio published a nice article in QEX about his EER tube
amplifier. That was very inspiring. But he too used EER only from a certain
amplitude up, and from there down he ran class AB, and tried to splice the two
operating areas. And with FETs it's worse than with tubes, because of the
voltage-variable capacitances in FETs.
I would love to learn tricks to get EER to perform well all the way down to zero
amplitude!
Manfred
========================
Visit my hobby homepage!
http://ludens.cl
========================
_______________________________________________
Amps mailing list
Amps@contesting.com
http://lists.contesting.com/mailman/listinfo/amps
|