You mention an L-network, but do you additionally have a
capacitor in series with the shunt-feeding arm? Or is your
L-network just one series-L from the coaxial feed point to
the arm, and one shunt-C from the arm to ground?
If you have just the two parts, (and no series-C feeding the
Shunt-arm), then you can get some really high voltages.
On 80 meters, and with a 15-m shunt arm, the input Z to the
Shunt-feeding-arm will look something like 50+j375. In parallel form,
that's 2860 ohms resistive in parallel with 382 ohms inductive.
That high-resistive number should ring alarm bells!
If the shunt-C of your tuning network cancels out a fair
amount of that parallel-L (forming an almost parallel resonant
circuit), then your network can have a really high voltage
point across the shunt-C but also across the series-L of your
tuning network. Thus you could have arcs either across the
shunt-C, or across the series-L. Especially so if there are tap
points on the inductor that go over to a switch (bringing those
tap points physically close to one another).
Using a series-C in the feed arm brings the impedance
down to a reasonable value, placing high-voltage stress just on that
one component - the series capacitor.
-- Tom, N5EG
TowerTalk mailing list