[time-nuts] Phase Noise and ADCs

[Magnus Danielson]

Hi John,

On 2020-09-26 17:10, John Ackermann N8UR wrote:
> We know that phase noise scales with frequency, so if you multiply
> frequency by 10 you get a 20 dB increase in noise.
>
> What I don't fully understand is how that relationship works with
> other than simple multiplication/division.
>
> For example (and my real life concern), if I have an analog to digital
> converter that is clocked at 122.88 MHz and know the phase noise of
> that clock signal, what do I know about the effective phase noise when
> the ADC is receiving a signal at, e.g., 12.288 MHz?
>
> In other words, if I were to measure the phase noise at the output of
> the ADC when fed a high-enough quality 12.288 MHz signal, would I see
> something like the 122.88 MHz phase noise, or something better due to
> the scaling by 10?

In this case, your 12.288 MHz phase-noise will be augmented with the
scaled-down version of the 122,88 MHz phase-noise. The trick being used
is to actually let the sampling clock of the ADC be a transfer clock
such that it samples a reference also, at which time one subtracts the
phase-data from the reference, most of the transfer clocks noise cancels
out, and it does so fairly well as it have common integration time
between the channels, so you avoid the decorrelation that DMTD normally
suffers from. The second trick being used is to use a second pair of
ADCs to make ADC deficiensies cancel out too, as the DUT and REF is
common mode and the individual ADC noise contributors can be averaged out.

Now, as you decimate data etc. you still have the issue of ADC
resolution, as that has a tendency to loose weak side-bands. The
non-linearity is kind of peculiar actually.

I recommend you to dig up Sam Stein's papers on his phase noise
development. I think you will find them a good read and help you with
your understanding.

Cheers,
Magnus