[time-nuts] Effects of filter delay

[SAIDJACK at aol.com]

In a message dated 12/23/2006 15:17:33 Pacific Standard Time,  
bruce.griffiths at xtra.co.nz writes:

The FIR  filter used in some GPSDOCXOs is not the optimum prefilter for 
the control  loop.
An exponential averaging (IIR) filter is better.
However an FIR  filter has the advantage of being easy to do and FIR 
filters are not  subject to limit cycle oscillations due to roundoff 
error that plague some  IIR filters.
There is a maximum prefilter delay that can be used (for a  particular 
loop time constant) before the control loop becomes  unstable.
The prefilter output is typically further processed (in a second  order 
loop) by a Proportional + Integral feedback stage whose output  controls 
the OCXO frequency via a ADC connected to the OCXO efc input.  Even 
better performance is achievable with a third order loop containing 2  
cascaded integrators, however stabilising the loop can be more  difficult.

Bruce


Hi Bruce,
 
surely you mean "DAC" driving the OCXO EFC :)
 
There are several problems with PI(D) controllers: the Temperature  
coefficients of the OCXO's are not corrected well, they are only corrected after  the 
fact (after generating an error in the output frequency that can be sensed  and 
corrected by the PI loop). Also, aging must be taken into account,  
especially for hold-over periods. So by the time the PI loop senses some offset,  the 
frequency may have, and is in the process of drifting quite  significantly.
 
The best temperature correction is done by using the sensor inside the  OCXO 
that controls the oven, using this to store the required DAC settings  for a 
given temperature over long periods of time, correcting for aging, then  
calculating the DAC offset versus temperature by linear interpolation etc. There  
are ways to sense the OCXO internal temperature very accurately without having  
access to the actual heater circuitry :)
 
Appearantly Kallmann filters are the best predictors for Temperature  
coefficients etc.
 
Does anybody have any information of the performance of PID controlers  
compared to simpler PI controllers (not doing the differential part)?
 
bye,
Said