[time-nuts] FLL errors [WAS: Chinese GPSDO...]

[Charles Steinmetz]

Tim wrote:

>Many hobbyist GPSDO's work, by counting OCXO cycles between some number of
>GPS PPS assertions.
>
>Software adjusts EFC based on frequency count. Often times the frequency
>count used as input to the software has not just random +/- 1 bobble in
>last digit, but also an extra count or two in last digit depending on the
>hobbyists' gating choices. The systematic extra count or two will result in
>a frequency offset between hobbyist GPSDO and PPS. Other times bugs in the
>software that drives EFC might result in a systematic frequency offset.

Good explanation.  The thing that hobbyists seem to miss is that 
carefully designed dither is necessary to randomize irreducible FLL 
errors after systematic errors are minimized.  To elaborate a bit, it 
is first necessary to chase down ALL of the systematic biases and 
minimize them.  Once that is done, one must add just enough dither to 
the process to make the remaining errors stochastic and 
gaussian.  [But understand -- dither is not a magic bullet.  Adding 
enough dither to obtain a stochastic result without minimizing the 
systematic biases only gets you a source that is on frequency on 
average, but has high phase noise and xDEV.]

Best regards,

Charles