[evlatests] Temperature-sensitive gains at Ku, K, Ka, and Q bands
Bob Hayward
rhayward at nrao.edu
Tue Nov 1 19:41:05 EDT 2011
Rick asked about how we can determine if the noise diode calibration
level (i.e., T(Cal)) is changing with temperature on an antenna.
I ran several tests in December 2010 that gives us a handle on what is
happening but we will need to do further tests this winter to better
understand it.
The test that I did used the C-Band on Antenna 7 over a 30 hour period.
It was set to a frequency of 5 GHz, pointed at the zenith and not
allowed to move. Using the EVLA Monitor Archive I collected the Total
Power (TP) and Switched Power Ratio (SDR, sometimes referred to as SP)
values from the T304, along with the physical temperatures of the
Weather Station, the Vertex Cabin and the receiver's RF Box where the
noise diode is mounted.
This was in the early days of Ken Sowinski's new SDR algorithm which
determines the percentage power that the noise diode adds to the
receiver based on the variance of the T(off) & T(on) of the 10 Hz
switched noise diode waveform. It is essentially a phase-insensitive
detector scheme, which is necessary since the T304 does not know what
the noise diode switching waveform is doing. Ken's SDR is, essentially,
given by
TP(variance) P(Sys+On) - P(Sys+Off)
SDR = ------------ = ----------------------
TP P(Sys+Off)
which is the incremental change of (linear) power from the T(Cal). We
try to set the T(cal) levels on our receivers to be 5% of T(sys). Ken's
SDR usually comes up with a value of 0.05 for the typical receiver. The
SDR is then used to determine the System Temperature by
T(Sys) = T(Cal) / SDR
For this experiment, the SDR should stay flat with time unless there is
a change in T(Sys). The SDR value is not affected by changes in gain in
the signal path since the ratio eliminates this common gain factor. As
we are looking at Zenith with (hopefully) no RFI present in the band,
the SDR in a perfect system should be unchanging. Dew or frost could
affect things but this is unlikely since no rapid changes in SDR were
seen. Thus if we see a change in SDR, and it is not from T(Sys)
according to our assumptions, it must be from a change in the injected
T(Cal) level. Since we had a 30 hour run, we should see the effect of
the diurnal temperature change on the noise diode.
I have attached a number of plots of the analysis of the data:
Page 1 shows the temperature change over the run. Note that the outside
temperature changed by 30C, the Vertex Cabin by 4C while the receiver
temperature changed by 9C. Note that the receiver sensor is not actually
mounted on the noise diode itself, so its value is not completely accurate.
Page 2 is a normalized plot of the temperature changes. We were
surprised that the receiver was changing by this much. We were hoping
the temperature stabilization inside the Vertex Cabin would hold it to
just a few degrees.
Plot 3 shows the Total Power outputs seen on all 4 T304 IF channels. The
RCP (A & B) side had problems (it had an amp that was oscillating) and
its output power was all over the place. The LCP side (C & D) was, by
and large, slowly changing but did have a jump in gain a few times.
Plot 4 shows the Switched Power Ratio (SDR=SP) for all 4 channels. Both
sides are running about 5% with the LCP side slightly hotter than the
RCP (this is about typical for setting the T(Cals) for the two sides).
Note that the RCP side looks very similar in shape to the LCP even
though its TP was varying terribly over the run (all the glitches are
caused by the gain changing within Ken's integration period).
Plot 5 shows the TP and SDR of a single LCP and RCP channel, IF-A &
IF-C. This shows that Ken's SDR does ignore the changes in TP gain.
While the TP traces look completely different, the SDR traces are very
similar. From here on we ignore the RCP side and concentrate on the LCP.
Plot 6 isolates the IF-C channel and shows how the TP and SDR curves are
essentially independent. Infact the SDR is only showing us the change in
T(Cal).
Plot 7 plots the SDR of IF-C with the monitored temperature of the RF
Box. There is an obvious inverse relationship between the temperature of
the noise diodes and its output power.
Plot 8 shows a normalized plot of the SDR (i.e., the deviation from the
average value). We now see that the max to min change in SDR is about 6%
for a temperature change of 9C.
Plot 9 zooms in to the area where the SDR is changing the fastest. We
see about a 1% change over 30 minutes, which is supposedly the time
period when an observer should probably consider recalibrating the
system on his favorite calibrator source. This drift in T(Cal) is barely
meeting the Project Book requirement.
Note that there is about a 2C change in temperature over this 30 minute
period. According to the manufacturer of the noise diode, the
temperature coefficient is better than 0.01 dB/C, so we should only see
less than a 0.5% change in T(Cal), not 1%. Maybe the noise diode is
seeing a bigger temperature change that what the RF Box sensor is
monitoring (this might be the case as the sensor is sitting on a board
that has a bunch of warm voltage regulators on the other side).
Recent temperature coefficient tests that I did on the Solar L-Band
receiver in the lab suggest that its noise diode is about 0.003 dB/C,
which is 3 times better than the manufacturer's 0.01 dB/C spec. The new
Solar Mode design requires an amplifier after the noise diode, along
with a 62 dB digital attenuator, to achieve the wide levels of T(Cal)
and S(Cal) values to cover the huge range of possible system
temperatures that the receiver could experience (i.e., looking at cold
sky (Tsys = 30 deg) all the way up to a 25 million degree solar flare).
I found that this unit had a temperature coefficient of about 0.01 dB/C,
so it was no worse than the worst case manufacturer's spec for a noise
diode all by itself. But, we've already seen the C-Band noise diode on
Antenna 7 is varying a couple of times worse than we can (currently)
account for.
As you might guess, we will be doing more tests over the next couple of
months. Wayne Koski has built us a nice 8-channel temperature box that
will allow us to measure temperatures anywhere we want inside the Vertex
Cabin and dump the data in to the EVLA Monitor Archive. Thus we can map
the true temperature profile of a receiver from the top of the feed all
the way down to the noise diode and hopefully determine what is the true
culprit affecting the temperature of the noise diode. Is it due to a
change in the ambient temperature of the Cabin or from the feed acting
as a "heat pipe".
The limitation of this type of SDR test is that it works fine for when
the telescope is doing an imitation of a birdbath and looking straight
up. If we move in elevation, the T(Sys) will no longer remain constant
as atmospheric and spillover contributions will change. We can try
covering the feed with an absorber but that will make the measured
switched power minuscule (we would have to crank the T(Cal) level up)
and we will need to monitor the physical temperature of the load. I'd
also like to try placing a metal plate over the feed - the amps will
essentially see their own physical temperature - but the standing waves
could be pretty awful which might make the receiver's total power
stability problematic. And remember, moving in elevation will disturb
the temperature layers inside the Vertex Cabin, thus making the ambient
temperature profile much more complex to sort out.
-Bob
Rick Perley wrote:
> So far as we can tell, the application of PDif nicely removes these
> temperature-induced gain changes. So we're not too concerned about
> this. However -- the tests so far cannot easily discern if the noise
> diode output is varying significantly (defined here as more than ~1%).
> The 'flux density' run of last year provided some evidence that noise
> diode power was varying by about this much ... we'll need to think of a
> good test to give a more definitive judgment.
>
>
>
> Bob Hayward wrote:
>> A few words about gain change versus temperature on the EVLA. There
>> seems to be some surprise that we might be seeing gain variation in
>> the signal path in the Vertex Cabin. I'm not surprised at all.
>>
>> The rule of thumb for a typical amplifier is that it will have a
>> temperature coefficient of -0.012 db / deg C / stage. Most of our amps
>> are 3-stage devices so a 1 deg C change in temperature will cause a
>> -0.036 db change in gain. Thus a 10 deg change (to use a nice round
>> number) from 15 to 25 C will cause a drop in gain of -0.36 dB (i.e., 8%).
>>
>> Looking at the data from last winter of the temperature sensors built
>> into each of the room temperature RF Boxes on our receivers, we saw up
>> to a 10C temperature between day to night. We also saw a 20 C change
>> from the hot of summer to the cold of winter. We suspect that not all
>> of this change is due to the ambient temperature inside the Vertex
>> Cabin. Much of it may be due to the fact that the receivers are
>> hanging off a big metal feed (or feed tower) that sticks up through
>> the roof, thus forming a nice heat pipe.
>>
>> The amplifiers in the LO/IF rack will see much less of a temperature
>> change - perhaps a degree or two at most - thanks to conditioned air
>> inside the enclosure. The UX Converter, on the other hand, is much
>> less regulated, and it is hanging off of the wall of the hut.
>>
>> So a 10 deg change in the receiver's warm RF path would give us a 0.36
>> dB gain change from its post-amps. On the UX Converter, there are 1 or
>> 2 amplifiers depending whether the signal is going through the direct
>> of converted path. If it only saw half of the 10 deg temperature
>> change that the receiver experiences, there would be another 0.18 to
>> 0.36 dB change in gain. Thanks to Murphy's Law, these will add, so
>> overall we would see a total change in gain of 0.54 to 0.72 dB, or 13
>> to 18%.
>>
>> All 8 of the receivers will have roughly a similar gain change but the
>> high frequency receivers will be slightly worse because of the added
>> contribution from the UX Converter.
>>
>> It should be noted that this performance is no different than what we
>> had on the old VLA system and it is exactly why we use switched power
>> to compensate for gain changes. Our real problem is that the
>> temperature change that we are seeing in the receivers is bigger than
>> we had expected (thanks to the heat pipe effect) so the output power
>> of our noise diodes may be changing more that we would like.
>>
>> More tests are planned for this winter...
>>
>> -Bob
>>
>>
>> Jim Jackson wrote:
>>> I just looked - the temperature inside one UX converter cycled over
>>> only about 5C over a 24 hr period. I'll have to talk to the guys
>>> tomorrow about whether that could explain anything like this - I'm
>>> skeptical that it could.
>>>
>>> Jim
>>>
>>>
>>> At 02:04 PM 10/31/2011, Rick Perley wrote:
>>>> The elevation-gain test taken early Saturday morning strongly
>>>> supports Michiel's evidence for temperature-sensitive gains -- and at
>>>> all the high frequency bands (Ku, K, Ka, and Q).
>>>>
>>>> A strong change in PDif was noted in the tests taken the preceding
>>>> week -- of a source setting from 82 degrees elevation to elevation 8.
>>>> All high frequency PDif values *increased*, by typically 10%, during
>>>> this period. It was difficult to be sure that the visibilities changed
>>>> by a similar amount, since the test was of a single source setting
>>>> during the period.
>>>>
>>>> The data taken Saturday morning were of a source which was
>>>> rising at
>>>> midnight, and transited at dawn. Temperature effects were minimized,
>>>> while the elevation vs time was reversed (i.e., the source was rising,
>>>> rather than setting). The PDif values for this run rise during the run
>>>> -- but by a much smaller amount than the previous week's test.
>>>>
>>>> We can rule out an elevation dependency of the PDif. And we can
>>>> strongly favor a temperature-dependent gain, since the vertex room
>>>> temperatures should considerably lag the external air temperature.
>>>> Both
>>>> of my runs were taken when the air temperature was declining (the first
>>>> one strongly so, the second one only slightly), while Michiel's test --
>>>> which showed the opposite variation in PDif -- was taken with strong
>>>> heating.
>>>>
>>>> Application of the PDif values gives an elevation-dependence of the
>>>> gains which appears to be the same for both of my runs -- this
>>>> strengthens an explanation that the PDif changes are due to real gain
>>>> variations, and not due to a changing value of the noise diode power.
>>>>
>>>> Rick
>>>>
>>>> -------- Original Message --------
>>>> Subject: [evlatests] Q-band holography d.d. 2011-10-29
>>>> Date: Mon, 31 Oct 2011 12:37:02 -0600
>>>> From: Michiel Brentjens <mbrentje at nrao.edu>
>>>> To: EVLA Tests <evlatests at nrao.edu>
>>>>
>>>>
>>>>
>>>> ** [2011-10-31 Mon 12:11] Q-band holography d.d. 2011-10-29
>>>>
>>>> Temperature sensitive gains?
>>>> ----------------------------
>>>>
>>>> A 3C273 Q-band holography observation from past Saturday shows a
>>>> systematic decrease of PDIF over the course of six hours in virtually
>>>> all antennas. This *decrease* in accompanied by a similar slow,
>>>> systematic *decrease* in visibility amplitude. Typically, PDIF
>>>> decreases by 20%. The visibility amplitudes also show a 20% decrease,
>>>> although they are much noisier than the PDIFs.
>>>>
>>>> The source had a fairly constant elevation between 36 and 57
>>>> degrees. The observation began at 08:00 MDT and ended at 14:00
>>>> MDT. Are we seeing a temperature dependent Q-band gain here? Notable
>>>> exception is IF 3C, which shows nicely constant PDIFs, yet also
>>>> decreases its amplitudes along with all others.
>>>>
>>>>
>>>> Other phenomena
>>>> ---------------
>>>>
>>>> For some reason, 11R lost all signal around 16:50 UTC. After 17:30
>>>> UTC, the PDIFs of 20L become increasingly more noisy, until they
>>>> suddenly snap together at a higher value around 19:05 UTC. 12A PDIFs
>>>> are consistent with noise. 4L PDIFs behave nicely until 17:00, at
>>>> which point they increase by 50% and begin behaving erratically.
>>>>
>>>> - Michiel
>>>> _______________________________________________
>>>> evlatests mailing list
>>>> evlatests at listmgr.cv.nrao.edu
>>>> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
>>>>
>>>> _______________________________________________
>>>> evlatests mailing list
>>>> evlatests at listmgr.cv.nrao.edu
>>>> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
>>> _______________________________________________
>>> evlatests mailing list
>>> evlatests at listmgr.cv.nrao.edu
>>> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
> _______________________________________________
> evlatests mailing list
> evlatests at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
-------------- next part --------------
A non-text attachment was scrubbed...
Name: EA07-C25_SP-vs-Temp_Plots.pdf
Type: application/pdf
Size: 2400428 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20111101/01d5c02b/attachment.pdf>
More information about the evlatests
mailing list