[time-nuts] Square to sine wave symmetrical conversion (part 2)

[Charles Steinmetz]

skipp  wrote:

>[the 10MHz output is] not even close to being symmetrical.  The waveform
>on-portion (duty cycle) appears (surprising to me) to be much less than 20%
>
>Now I'm under the assumption that proper rounding or conversion of the non
>symmetrical 10 MHz square to a sine wave will be a bit more involved.

That means there are significant even harmonics present, including 
the second harmonic, which is a lot closer to the fundamental and, 
therefore, harder to remove by filtering (a perfect square wave 
contains only the fundamental and its odd harmonics).

>Before I launch toward part two of this latest saga, I'd really be 
>interested in reading
>suggestions and comments regarding methods to improve/fix the 10 MHz waveform
>symmetry.

You could do a lot better than 20/80 by simply using the 10MHz pulses 
to trigger a 50nS one-shot (astable multivibrator).  However, that is 
a quick-and-dirty solution, not a precision solution -- the duty 
cycle would wander around with temperature, power supply voltage, 
noise, and other factors.

Alternatively, you could use very aggressive filtering, but that 
could degrade the phase noise of the output sine wave due to the 
temperature coefficient of the filter cutoff frequency.

One sure way to get a symmetrical output would be to use a Dflop 
frequency divider to generate a symmetrical 5MHz square wave, 
followed by a frequency doubler to get back to 10MHz.  There would be 
a phase noise penalty, but it could be less than the phase noise 
penalty of a sufficiently aggressive filter.

Finally, you could use the 10MHz pulse train as the reference for a 
1:1 PLL.  I suspect this would be the best way to go about it.

Best regards,

Charles