JC wrote:
Thanks for putting this all in perspective for me. Your description of
the high Q, and resultant heating of the loading capacitors, and detuning
makes sense. Users have reported much higher peak power than RMS with these
amps. I always attributed that to the voltage doubler, and B+ sag under
load. That is probably the dominate factor, but apparently RF losses are
significant as well after a sustained key down at 1 KW input.

First, power you read on RF power meters is RMS power. Peak envelope
power is the peak value of short term RMS power, average power is the
short term average of RMS power. This is why for CW PEP and average
power is the same, while for SSB and AM it is not.

The RF losses are the same between the two.

A weak power supply can cause a disparity between peak and average
power, but that has little to do with the supply being a doubler. It
mostly has to do with the size and quality of components. Remember
when Dentron's were made the legal limit was 1000 watts DC plate input.


So what is your opinion about the tank coil being wound of stranded and
insulated wire where the wire strands are silver plated, and Teflon
insulated?

That's OK. The big problem is the inductance is too low.

They are also close wound where the turns touch each other? I
have never seen that done before.

Cost cutting. Efficiency is determined by the ratio of unloaded Q to
loaded Q, and certainly stranded wire, insulated wire, and close
spacing reduces unloaded Q. Losses in the tank are reduced by half if
you double overall unloaded Q, or if you simply cut operating Q in
half.

It would be difficult to double unloaded Q, but it is easy to cut
loaded or operating Q in half.

It sounds like I need to hit the charts for pi-networks, and pick the Q,
and I/O impedances, and then make it happen. The inductance of that orange
160-80-40m coil with Teflon wire will be hard to model. I will try to
measure it with my RF bridge, and see how it comes out (uh / 10 turns).

You need to measure it in place with the lid on. The chassis and lid in
the Dentron, since they are right across the coil axis, greatly reduce
inductance.

Whatever you measure outside the amp, you will have a lot less with the
coil inside and the cover on.

I have 2 large 400pf 2500v silver micas, and a strange 4 section variable
mica variable where each section is 300pf. This item has a common stud, and
each section is about 1" diameter, and 1/4" thick. I might be wrong, but I
recall someone saying these were used in the ART-13. These capacitors should
handle lots of RF current, but I'm not sure of the voltage rating of that 4
section mica variable capacitor. The Dentron 1000pf loading capacitors are
CRL type 858S, 5KV, X5U. These are best for bypassing use, and have high
dissipation factor, and high capacitance drift as it heats up.

You will find some people complain about current rating, but in truth
current rating is the least of your worries. The real problem with the
caps is temperature coefficient and the fact Dentron's have terrible
tank Q. The operating Q is far too high on many bands.

I suggest you get all the components correct, and get a tank operating
Q between 10 and 15 on bands where you can. Ten meters will be higher
than that because of stray C, so just use the lowest possible value
there.

Then you can get NP0 (N P zero) caps of smaller value and parallel them
for loading. Ameritron uses some ceramic chips that are high current
temperature stable NP0, or you could use multiple N750 or less ceramic
caps in parallel. You will not find high value doorknobs in stable
temperature coefficients.

For the anode padding caps, you will probably need to use small value
HV ceramics in parallel. You'll see Dentron, in MOST amps, has far too
much capacitance and too little inductance.

Remember to measure that tank coil with the metal in place at the coil
axis. You also must measure the coil at radio frequencies. One cheap
common tool is a grid dip meter with a few fixed caps of known value...
although a small light bulb, some fixed capacitors, and a transceiver
will work every bit as well.

73 Tom