[time-nuts] 10811 crystal orientation

bg at lysator.liu.se bg at lysator.liu.se
Sun Jul 12 22:40:23 UTC 2009


Hi Said,

> Hi Magnus,
>
> Agree with all your points.
>
> On the added noise due to an accelerometer, my thoughts are that this
> needs
>  to be carefully designed so as not to add more error than we are actually
> removing (due to phase shifts between crystal sensitivity and the
> accelerometer  response for example, and due to random noise).
>
> For example, if we do the compensation in the digital domain, and use a 10
> bit ADC on the accelerometer, and a full-scale compensation is say 2E-09,
> then  the quantization noise itself (1/2 of 1 LSB on average, or about
> 4E-013
> just due  to the quantization noise) may already be considerable. So we
> need to use a good  ADC, and very low noise accelerometer :)

Some parameters for a "random" state-of-the-art analog accelerometer...

    measurement range         60g
    bandwidth                 >300Hz
    resolution/threshold      < 1ug
    1 year bias repeatability  < 40ug   (over an "industrial" temp range)

Doing some hand waving here... gives more like a 24bit ADC. Noise (DC to
10Hz) can possibly be kept below 10ug. If we are going to compensate at
10Hz bandwidth, there is opportunity to oversample ten times or more.
Which also serves to reduce noise. Putting the accelerometer inside an
oven will improve its performance greatly, working to our benefit.

Building a system where accelerometer errors affect oscillator frequency
at the 1E-16 level should be possible.

--

    Björn





More information about the time-nuts mailing list