[Elecraft] K2 key clicks

[Masleid, Michael A.]

Gary wrote:

>If you want to understand exactly how speed and keying waveform affect =
the
>keying bandwidth of a transmitter, refer to the analysis on the =
following
>website (the math is at the level of advanced calculus but the graphics =
and
>narrative explain the conclusions very clearly).
>http://fermi.la.asu.edu/w9cf/articles/click/index.html

Kevin, W9CF, suggests an optimum wave shape for the keying waveform.  =
The
trick is to generate the waveform.  Let's set S(t) to be a square pulse
with a duration of 0.004 seconds.  If you convolve S(t) with itself =
several
times you will get a very nice Gaussian pulse, with T =3D 0.004 seconds.

With the first convolution, you get a triangle wave form.  Not too much
like a Gaussian, but it gets better.  One way to make a triangle wave =
form
is to integrate one cycle of a square wave, or S(t+0.002)+S(t-0.002).

With the second convolution, the edges get smoothed out.  It starts to =
look
more like a Gaussian pulse.  To get this waveform, we need a second
integrator.  We feed S(t+0.004)-2S(t)+S(t-0.004) into the first =
integrator,
and then feed that into a second integrator.  So, were integrating the
sequence 1, -2, 1 two times.

The third convolution looks really nice.  In this case
we need the integral of the integral of the integral of
S(t+0.006)-3S(t+0.002)+3S(t-0.002)-S(t-0.006)
which is a good place to stop.  We are integrating the sequence
1, -3, 3, +1 three times.  We could go for 1, -4, 6, -4, 1, but we
won't.  Instead we can now produce the erf(t) function by integrating
one more time!  Of course, to make a "dit", we have to turn it all
off again.  Turn off is done 20 milliseconds latter by the reverse
sequence -1, 3, -3, 1.

This is a good place to stop because it uses 4 integrations, and
quad op amplifiers can do 4 integrations.  Also, the waveform takes
16 milliseconds to complete.  It would be nice to leave a little time
for break in keying.

I believe that the bandwidth envelope breaks downward at about 80 Hz,
and drops at 24 dB per octave, with nulls at multiples of 250 Hz.

This might lend itself well to a variable transmitting band width.
Figure that the 1,-3,3,1 wave shape can be generated by VPWR feeding
the quad integrator, which then drives pin 3 U10A.  The amplitude
on VPWR still controls VALC, but VPWR gets clocked 5 times faster
than the keying rate.

Of course, the devil is in the details.  You probably need FET switches
to act as charge dumps on the integrator chain.  You have to figure
out where to put the stuff, and VPWR may not have enough resolution
to step bandwidth and power both at the same time.

Good things about the approach:  1>  Provides better attenuation than =
the
Gaussian envelope at small frequency offsets.  2>  No tails.  The =
Gaussian
and RC envelopes have tails that go on forever.  The approximation
waveform is over and done with at T*4.

Bad things about the approach:  1>  Needs more parts.  2>  Needs more =
room.
3>  I suspect that you can probably do as well or better by modifying
the existing RC filters.  4>  I may have the math all wrong.

Thanks for the bandwidth!

73, de Michael, AB9GV