[time-nuts] Adjusting HP 5065A frequency
magnus at rubidium.dyndns.org
Mon Oct 22 18:08:06 EDT 2012
On 10/22/2012 04:46 PM, EWKehren at aol.com wrote:
> A good place to start is with a manual of the Austron 2055 delay if there
> is real interest. Does any one have it in PDF form?
> As to the disciplining a Rb or specifically a HP5065A, the filter part is
> the challenge. Having worked on it for the last ten years always using
> Brooks loop and developing work around's it will do every thing except
> temperature control and barometric pressure. I have done temperature analog but rely
> mainly on holding the Rb temperature stable within a tenth degree C.
> Bob recently put things in perspective when there was talk about a 32 bit
> DAC. On one side you have GPS and time averaging is required to get full use
> of it. That is why a Rb is uniquely qualified for it. The original Shera
> input implemented on a 1 $ G/A is more than enough for time capture wether
> using a sawtooth corrected GPS receiver or a Tbolt. On the other end a
> LTC1655 is again more than enough in an Rb application. You can use it 16 bit
> direct or dither two or 4 bits. Filter is easy and do not forget that the
> Rb has an additional filter between cell and OCXO. Brooks V402NE will do the
> job, how ever others and I have not been able to buy some in the last nine
> month. Any one knows what is going on?
> What is needed is some one in the group be willing and able to step up to
> the plate and develop the filter on a PIC with all the things learned, and
> believe me, we are still learning as we adapt a M100/8600 via a Tbolt .
> I will be willing to help with every thing except the PIC.
> If you bring out the C field coil directly you have full isolation and do
> not have to worry about ground loops.
> A complete unit would cost less than $ 40 and more important be assembled
> by any one. I brought it up before but no one responded. If no response I
> will still be able to do every thing with the HP5065A RVFR that was given to
The needed loop filter isn't all that hard to achieve. Do read the
Stanford Research PRS-10 manual. The core PI-filter can be formulated as:
VI = VI + Vd * I
VF = VI + Vd * P
Where Vd is the phase detector value (may be in number of nanoseconds or
whatever), VF is the output frequency control (EFC) value. The I and P
values is the control parameters and VI is the integrator state.
Adding a pre-filter for Vd values can be done as in the PRS-10:
Vd = Vd + (Vi - Vd)* F
where Vi is the raw input and F is filter control parameter. F can be
set in power of 2 steps without too much loss of control, allowing for
shift-steps, which is what the PRS-10 uses. The PRS-10 runs it all in a
whopping 6805 if I recall things correctly, just your off the shelf 8
bit processor that felt right in the moment.
You can then allow for some front-end processing to cook up the I and P
values from more user-friendly parameters if you like.
Tossing in a FLL functionality on top of it for improved lock-in
performance isn't hard either.
There is a few scaling issues and most things is about getting a stable
timing and such, but it's not all that hard. If you want to do PPS there
is a little more attention to lock-in details naturally.
Cooking up an open "lock-it-all" isn't that hard thing to accomplish
with a bit of knowledge and experience. One might even have a bit of fun
and cook up a Kalman filter, which is essentially a self-tuning PI
PS. Sorry for not making you aware in advance that I was changing planes
in Miami this Friday. It was a busy week in Atlanta.
More information about the time-nuts