[Amps] Price per Watt Conversation

[Manfred Mornhinweg]

Cathy,

> You guys should publish a construction article(s) in QST and other 
> nations' equivalent magazines. A lot of us would love to build such 
> an amp, but we're not competent to design it.

Such articles have been published already. I'm mostly aware about
publications in Germany, both in magazines and online. For example, in
1997 (20 years ago!) DL9AH produced a design using 32 inexpensive IRF710
MOSFETs, to produce 750W output over the entire HF spectrum. He reported
an IMD3 of 35dB below each tone (41dB below PEP) and an efficiency
around 60%. One year later he published it in the magazine Funk, in two
installments, in february and march 1998. DL2GAE published an
improvement on this design in the september 2002 issue of Funk. DL4JAL
published a digital power/SWR meter for this amplifer, in 2010. In that
same year, DO2AS published a detailed construction article for this
amplifier. So this is really a well published and well documented
design, replicated by a good number of hams. That's why I took it as
basis for my own experimentations with that kind of amplifier, but I
have not yet built my final implementation, which would use 48 slightly
better MOSFETs for 1200W output.

DK6AE published a small amplifier using six IRF820 MOSFETs, that
produces 250W. He combined four of those modules to produce an amplifier
delivering the German legal limit power.

My friend Rodrigo, CE7MCK, built an amplifier much like DL9AH's, using
50 IRF510 MOSFETs. Yesterday I had a QSO with him on 40 meters, after he
transmitted a bulletin using that amplifier. He was booming in at 
S9+25dB, with an impeccable clean signal. Rodrigo is the same age as I, 
and when we both were young students and new hams we competed building 
the best transceivers and accessories. Rodrigo is the only guy I know 
who made his own ferrite cores! Back when importing components was 
difficult, his toroids saved the day for us homebrewers.
Nowadays he is the president of Radio Club Llanquihue, and lives 400km
away from me, so I don't see him often, but we stay in contact. As far
as I know he hasn't published his amplifier. There is currently no ham
magazine in Chile.

> Being able to work with off-the-shelf, widely-available BJT or FET 
> components in current production, and off-the-shelf heat sinks that 
> don't require custom machining, should bring the cost and skills 
> required down significantly.

Yes. But be careful with "current production". When I did all the tests
to select the best small MOSFET for HF amplifiers, I ended up with the
IXTP1R6N50P. It performed significantly better than the IRF710 or any
other I tried. But since then this MOSFET has been discontinued! Which
means choosing the second best, or checking what's new on the market and
doing all the tests again...


Roger,

> Join the SSPA (Solid State Power Amplifier) group on FaceBook.

Maybe I'm crazy, but I dislike Facebook. I registered there several
years ago, hated it, and left. Maybe it's better nowadays, I don't know.


Bill,

> At that voltage it is tempting to directly rectify the AC line, 
> greatly simplifying power supplies. Of course proper ground isolation
>  practices must be taken.

Yes! In fact DL9AH's amplifier uses direct line rectification from
220VAC, using a half wave rectifier, phase-angle control for voltage
regulation, and a big filter capacitor bank. He relies on the neutral
being at ground potential. I definitely don't like that approach, as it
has a horrible input power factor which doesn't suit my power source.
Instead my approach is to use an off-line buck regulator of the exact
same type as the one I built to power the DC motor of my lathe:

http://ludens.cl/Electron/latsup/latsup.html

That one is for 160V and 3A, and the design can be easily modified for
110V 20A or whatever the amplifier would require. In fact I did that 
redesign, and bought the parts needed to make my amp supply using that 
basic circuit, but haven't yet assembled it.

This design definitely puts the amplifier at dangerous voltages, tied to
the line, so that good insulation needs to be provided in the input and
output RF transformers, and reasonable care needs to be used when
working on it. On the other hand, of course it's far less dangerous than
working on tube amps, and many of us here have done that often enough
and are still alive.

For use in the USA, it's tempting to simply rectify the 127VAC line and
use that. But that would give you an unregulated voltage with
significant ripple, requiring huge capacitors to do just as bad as
common tube amplifiers do in that regard, and suffering from a poor
power factor. So I think that it's better to use a buck regulator like
mine, given its simplicity and low cost. The benefits are voltage 
regulation, no significant ripple, and a somewhat improved power factor.

If desired, a real and true power factor correction circuit could be
inserted before the filter cap. Specially when operating in the field
from a portable generator this would be an advantage.


Joe,

> Ok say you did have a unit that could do that voltage, For a legal 
> Limit Amp, what would be the current draw?
> 
> I'm assuming all of Ohms Laws still apply don't they?

Yes, they sure apply. My multi-FET amplifier should achieve 1200W output
at around 60% efficiency. That means 2000W input. At the planned 108V
supply voltage that's 18.52A.

That current is low enough to be handled with a single switching MOSFET
and a single freewheeling diode in a buck regulator.

Given the small loss of this power supply, the AC input power would be
only slightly larger, say 2050W. At my 220V line voltage, that would be
  9.32A, but due to the mediocre input power factor of such an
uncompensated power supply, the actual RMS input current would be more
like 16A when a full power carrier is being transmitted. Not much 
different from a typical tube power amp of the same power. That's why a
power factor correction circuit would be useful. With that PFC, the 
input current would stay below 10A.


Roger,

> I hope I didn't make any of my usual math mistakes.

Don't worry - making mistakes is part of human nature! ;-)

I wonder how many mistakes I have made myself in this thread...

> Yes there is a limit on how many you can parallel, but it's dependent
>  on the frequency and circuit design.

What I had in mind when I wrote that I think there is no limit, was
properly combining them. Sure, if a lot of transistors are simply
paralleled, phasing differences between them set a limit that depends on
frequency. But I see no limit on how many can be combined when using
equal lengths of cable into splitters/combiners.

> Temperature.   Although ratings are given at 25 C, typically air 
> cooled devices run between 40 C and 70 C (based on experience with 
> computer design and cooling)

There you have a mistake, although it's not a maths mistake. Computer
CPUs and high power LDMOSFETs have roughly the same footprint. But a CPU
will dissipate 10 to 30W, an extremely powerful one might dissipate 100W
(although I have never seen that). Instead with LDMOSFETs we are talking
about 600W dissipation, and some people even think they can do more!
While it's quite possible to run computer CPUs between 40 and 70°C,
depending on their power dissipation, I don't think that you can achieve
that with an LDMOSFET dissipating 600W. Certainly not with air cooling,
and probably not even with water cooling.

> The MRFX1K80H is derated at 11.11W / deg C, so at 40 C (40-25 = 15 X
>  11.11 = 166.5) or 1800-166.5, or a MAX rating of 1633.5,

Another mistake: The max dissipation rating of this device at 25°C is
2222W (preliminary data), so at 40°C it's 2055.5W. 1800W is the rated
maximum RF output power, not the dissipation rating. But of course you
can't keep it at 40°C, unless you use an extremely good cryogenic
cooling system...

> but for 70C we have (70-25= 45 45 X 11.11= 499. 95, or 500W  1800 - 
> 500 = 1300W.

1722W, really. But you can't keep it at 70°C either, while dissipating
that much power...

> Again, this is compared to class C,

LDMOSFETs are rarely used in class C. Even for nonlinear, saturated,
high efficiency service they are usually biased into class AB. When you
do a detailed circuit design, you soon notice why. For class C
amplifiers, VDMOSFETs are better suited.

> but 4 devices should loaf along at 375W output ea in linear service 
> at 40C.

That's true. While they would run much hotter than 40°C in a practical
design, they would still be loafing along.

> With water cooling and chilled water, the case temp really can be 
> close to 25 C (77F)

Yes, that can be done, but is totally impractical for a ham amplifier.
And it's probably not cost-effective for a commercial amp either. It's
cheaper, smaller, lighter, more efficient and more reliable to use
enough devices/pallets with a simple cooling scheme, than fewer
devices/pallets cooled by chilled water (which requires accurate
temperature control to prevent condensation).

> and the much lower voltages used in SS amps makes water cooling much
>  easier than in tube amps.

That part is completely trivial, since the cooling surface of LDMOSFETs
is usually at ground potential.

> Running linear at 60% of 1300W is 780W gets us into only half the 
> derated power which "should be OK for SSB, but digital?

I think that 1300W output in linear broadband service will exceed the
dissipation that can be handled properly. You must keep in mind the
whole equation: The real-world dissipating capability depends on the sum
of all thermal resistances. This device has 0.01 K/W less internal
thermal resistance than the BLF188XR. That's 0.09 instead of 0.10 K/W.
Whatever cooling system you use will have the same thermal resistance
for either device. So, if a realistic cooling system achieves 0.2 K/W,
the total thermal resistance will be 0.3 K/W for the BLF188XR, and 0.29
K/W for the MRFX1K80H. Meaning that the real-world dissipation
capability in this case would only be 3% better for the new and bigger
device. So, if the new device costs significantly more, or has
significantly higher capacitances, it would be ill suited to linear
applications, in comparison to the smaller device.

Now the capacitances of the new device do look good. Specially the 
reverse transfer capacitance, the most important of the three, is much 
lower! That makes it attractive. But first we must see if the final 
production devices meet the specifications of the preliminary datasheet, 
and then we must check the actual street prices.

> This brings up the question of duty cycle. Just what are we 
> referencing against.  Deep class C could be 50% or even less.  If so,
>  then SSB at 20% should be OK, BUT digital at 100% in linear (AB1) is
>  likely to overheat the devices.

If you look at the transient thermal impedance specs of such
transistors, you will see that the internal heat capacity of the device
is far too small to be of any value in averaging out the thermal load in 
SSB. Instead the spreader and the heatsink proper can do that to some 
extent, but the problem is that these devices are so tiny, relative to 
their power dissipation in linear use! You get a thermal bottleneck in 
the small amount of copper just below the device. So you need to extend 
the transient thermal impedance calculation to include the spreader and 
heatsink, and that's pretty hard to do. I wonder if the professional 
manufacturers of transmitting equipment do that, or if they just rely on 
trial and smoke assessment! ;-)

> then these devices are nowhere near the advancement they appear to 
> be.

Exactly. They are an advance in their intended application: High
efficiency, saturated, nonlinear amplifiers. But in linear service they
won't do much better than lower spec'ed devices, simply because the
power limit is dominated by how much heat one can extract from their
tiny footprint. And that's an external limitation, independent of the
device.

What we need is either devices with a larger footprint to match the
higher dissipation, or (much better) using several devices, spatially
separated. And that means using pallets of moderate power each,
combining them to achieve whatever power is desired.

Or else, develop true high efficiency linear amplifiers, so that we
don't have to extract such huge amounts of heat. And that's the correct
thing to do, in my humble opinion, but the hardest too. Brute-force
solutions are simpler.

> I'd be more comfortable with 8 of the 150Gs in PP / parallel in a 
> manner similar to the 8 standard 151s in the Quadra as 8 X a real 300
>  W = 2400.  Not cheap, but at $107 ea they look pretty good compared 
> to the newer MRFX1K80HR5. Actually they look mighty good!

It's because they are made for linear service. Instead for an FM
broadcast transmitter, the big LDMOSFETs are clearly more cost-effective.


Karl-Arne,

> The way to go here may be the abandoning the concept of conventional
>  linear amplifiers altogether, as the AM broadcast industry already 
> has done.

That's what I'm promoting. With SSB it's harder than on AM, but entirely 
possible.

> Using SDR based carrier phase modulation synthesis, switching 
> techniques and PWM envelope restoration it would be possible to 
> create the output RF waveform using SDR methods, thus incorporating 
> the adaptive predistorsion in the RF waveform synthesis process from
>  the very beginning.

> Some of the newer MF and HF broadcast transmitters have these 
> techniques already incorporated.

Yes, but not only broadcast transmitters do. The OpenHPSDR project,
developed by and for hams, implements it too. If you buy a Hermes board
and download the free PowerSDR software, you get the entire SDR stuff
ready made. In EER mode the Hermes board will deliver 0.5W of
phase-modulated, constant-amplitude RF drive power, and a 200kHz PWM
signal that has the envelope modulated on it. In the PowerSDR software
you can adjust a delay between the envelope and phase signals, to
compensate for the delay of your modulator's low pass filter. All you 
need to do is build a high efficiency class C, E, F, or even saturated 
AB power amplifier chain, and a simple modulator using two MOSFETs, 
driver and low pass filter. And there you are, working on all ham bands 
through 6m, on all modes, in high efficiency. All along with an 
excellent receiver.

This system also incorporates adaptive predistortion. To use it, you
only need to feed back a sample of the transmitted signal into the
receiver, and set up the software.

The biggest problem is that the Hermes board costs around 1000 dollars,
which I find too expensive. It was developed for top notch performance,
hence the price. I recently bought a Red Pitaya board, which largely
replaces the Hermes at slightly lower performance (equivalent to the
IC7300) and one quarter the price. But there is a big drawback: The
currently available firmware for the Red Pitaya does not pass through
the envelope signal! That's an unexpected showstopper for me, as it 
prevents high quality EER transmission. As the software is now, the Red 
Pitaya can only be used as a conventional exciter. Maybe the author of 
this software might add the necessary code later. Otherwise someone else 
would have to step in, and either modify his software, or write new 
software for the Red Pitaya. The board does have the necessary hardware 
on it.

Manfred


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