[time-nuts] Short-Term Stability

[Poul-Henning Kamp]

In message <006b01c64922$1355e5b0$0201a8c0 at n4fs>, "Mike Feher" writes:

>Like I said I did a similar thing for the front end of an FFT spectrum
> analyzer back in 1974, but that was by adding analog noise.

No you didn't do the same thing.

You add analog noise to dither out imperfections in the transfer
function in the ADC.  In that case, the most important property of
the injected signal is that you can pull it out again afterwards.

This is quite different in that you don't actually add a new signal
in, you just use a "frequency spectrum optimized rounding function".

>Certainly not optimum as having
>an algorithm and doing it on the fly [...]

Again: the algorithm does _NOT_ run on the fly.

You precompute a bitstream that tells you when to round the opposite
way of the "normal" rounding algorithm, and all you have to do is
add the memory for the bitstream and two XOR gates (one for the
possible carry).

The reason you can precompute it is that the DDS produces a cyclic
and predictable signal.

Over and out.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.