[time-nuts] DC Voltage Ramp?
cfmd at bredband.net
Sun Sep 4 16:15:30 EDT 2005
From: "Tom Van Baak" <tvb at leapsecond.com>
Subject: Re: [time-nuts] DC Voltage Ramp?
Date: Sun, 4 Sep 2005 12:26:46 -0700
Message-ID: <001601c5b186$9810d0a0$9415f204 at computer>
> > You need to come up with the ramp/parabolic which will compensate your
> > oscillators linear frequency error and frequency drift and the only way to
> > that properly is to compare it to something which is better. So, you need
> > compare against a Cesium or GPS, and then calculate the parameters, and
> Even if you compare to something better you seem
> to imply that a given quartz oscillator has "parameters"
> as if they were some fixed immutable value valid for
> all times and scaleable up or down for any tau. This
> is not the case.
No, if you read what I wrote not long ago, you would know I was not implying
that. I was just discussing the problem at hand at a particular moment in time.
> Drift rate specifications can be made per day or month
> or year. These are not specs of how the oscillator must
> perform or will perform; they are often guaranteed limits
> on how bad the oscillator will perform.
> You can measure the actual daily drift rate by taking
> a month of data and dividing by 30. But if you do this
> my guess is you will be amazed how much the daily
> drift rate varies from day to day, or hour to hour.
Indeed. There are many sources of frequency drift. Some may be predicted and
thus compensated where as others will not allow themselfs to be very well
modeled. If you can observe a particular noise source, then you could
potentially use that to compensate for the effect, but only when you have
observed the transfer-function for that noise source. However, there are a
number of noise-sources we can't observe directly and thus we need to either
live with them or compensate them by some sort of control loop.
> > redo this. In order to handle other perturbations such as random phase
> > and result of mechanical chock, you need to do this on a regular basis.
> > What you now have is a control-loop of some sort, wherein a PLL is really
> > one perticular class.
> E.g., GPSDO.
Yes. But the point is that you end up doing GPSDO as a consequence. Also, what
does GPSDO actually mean and why do you have to do it. Just concluding you have
to doesn't suffice until you have understood the reasons why.
More information about the time-nuts