[evlatests] Round-trip phase meters.

[vdhawan at aoc.nrao.edu]

Prodded by Peter Napier and Steve Durand, I looked at some round-trip
phase monitor data. I see nothing to contradict the September VLA test
minutes quoted below, but I'm writing down some details for when we
revisit this.

" B. Clark reports that the L352 phase meter measurements are well
behaved and changes are slow.  The main effect is the diurnal thermal
term, which need not be corrected for many types of observing, as it
is removed in the calibration process.  At a time when the antenna was
observing a baseline file, the worst case jump seen due to antenna
motion was about 800 fs, or about 15d at Q band.  From this he
concludes: 1.) the L352 averaging time can be changed from 1s to 10s,
to decrease the output data rate; and 2.)  the phase is behaving well
enough that the priority of implementing the round-trip correction
decreases below many other software tasks. "

I looked at two data sets both from EA-16.

(1) 24 hours of data where two L352 modules measured the same fiber on
    17th Sept; antenna not moving.

(2) 4 hours during a baselines run, 11th Oct. Antenna all over sky.

The specs are mostly met. A few fixes may make it a factor ~2 better.
The action items (perhaps low priority for Barry and Ken, but the
engineers can work on them meantime) are:

a. Fix the worse L352 meter, #52.

b. Do a controlled test of fiber delay vs. elevation.

c. Retest one fiber with two L352 meters to check consistency.

d. Sweep phase on meter(s) to check for dead zones, nonlinearities,
   offsets, gain drift in analog electronics.

e. Fix temperature sensors / decide we don't need them / identify a
   surrogate temp sensor for the exposed part of fiber.

------------------------------------------------------------------------

Verbose version:


			Data set (1).

Scale factor / repeatability of measurement by different units:

	The two round-trip meters did not report identical fiber length, 
	one read 95% of the other on (24-hr) average, with variations
	both ways on 30min timescales. 

	Action: fix voltage regulator and any other known problems on
	unit #52.

	A 5% scale error in reading a 40ps (diurnal) change is 2ps.
	If this number is unit-specific it may be calibratable (but a
	nuisance to track). If the scale factor varies with time/
	temperature, as it seems to, then it needs to be fixed. 

	Question: what was the length of the test fiber compared to the max
	in A-array?  NM array?

	Action: re-do test with a new or fixed unit. 

	Action: Sweep phase of 512MHz reference (e.g. at a rate of cycle/min) 
	to check for dead zones, offsets, drifts, or changes in linearity of 
	the phase meters. (None of these effects are expected in the digital 
	part, but the analog bits ...?)

Temperature:

	Data from the other, presumably good, unit (#3d) showed a fiber delay
	change (one-way) of 40 picosec over 13 hrs or 3 ps/hr.  This is due to
	temperature, and the temperature coeff is the about the same as
	measured in Ken's run (see part 2) about -4.5ps / 1deg C.
	I found the Antenna 16 ACU/FRM mount temperature in the monitor data 
	and it correlates reasonably well with the round-trip phase change. 
	It may not exactly follow the fiber temperature; in the later part 
	of the day the temp coeff was halved (fiber in shade maybe?)
 
	Rumour has it that the temperature monitors on the L352 are
	unreliable; Also those in other modules?

	Action: Fix L352 temp monitors or decide we dont need them.
	
	Identify a temperature monitor near the antenna fiber.

The short-term variations over ~1s have an rms 70-100fs, and
	can be reduced further by averaging over 10s. The spec
	for the VLA as a whole is 0.5ps jitter on timescales <=1s
	(9deg rms phase @ 50GHz). 

Variations over 30minutes: 

	To look at these, I sampled the data every 30min; interpolated
	linearly between samples and subtracted from the original data.

	The residuals over 30min have peak-peak variation in one-way 
	delay <1.2ps over the whole day. 

	The spec is 1.4ps pk-pk. 
	
	Even the 'worse' module, #52, had deviations <1.8ps 

Delay slope with time:

	The slopes were under 0.1ps/min for the good module.	

	The spec is 0.2ps/min.

 	The other unit had 2 incidents of 0.28ps/min for 5min, and 
	0.55ps/min for 2min -causes unclear- otherwise it met spec.



			Data set (2).

Temperature: There is a change of 2.8C over 4hrs.  (Ken: was this measured 
	at the VLA weather station?). The temperature coeff of 1-way delay 
	change is again about -4ps / 1 degree C.

Time variation: After removing the trend with temperature, the 1-way fiber 
	delay varies by ~1.4ps pk-pk over 4hrs, and <= 1ps pk-pk in
	30min, within spec of 1.4ps for time variation. (The elevation is 
	also contributing variability.)

Elevation: The fiber delay cycles by <=1ps pk-pk as the elevation cycles 
	from 10 to 120 back to 10 degrees. But the fiber change is not 
	proportional to elevation - it is shifted by a quarter-cycle, 
	i.e. varies with the rate of change of elevation or, more likely, 
	is due to hysteresis, which might also explain a delay step at 
	110deg elevation (next item) as something flops over the top.

	There is a step in delay of upto 0.5ps, nearly every time the
	antenna goes UP over 100 degrees. Coming down in elevation,
	variation is more gradual.  During the step, the rate of change
	delay with elevation is 70-100 fs/degree of elevation, mildly
	out of spec.

	The spec is 0.7ps change for any two points in the sky and
	70fs/degree change of direction upto 10degrees.

Azimuth: There is no obvious change with azimuth, even during 2 unwraps.

Action: Monitor the fiber while El/Az are changed one at a time.

---------------------------------------------------------------------------