ALMA Frequency Switching
Larry D'Addario describes the state of frequency switching capabilities in ALMA as follows:
Frequency switching is certainly possible, but we need to consider the
parameters: maximum switching speed and maximum frequency change
between signal and reference. An old specification requires that a
change of .03% of the sky frequency be accomplished within 10 msec (indeed, this was item
#431 in 'System Technical Requirements' 80.04.00.00-004-D, now
obsolete), but it applies only in single dish mode, so that phase
tracking across the frequency step is not necessary. We expect to be
able to achieve this, and the actual time should be much less than 10
msec. Originally it was assumed that this must be done in the 1st LO,
but that becomes difficult in the high bands (.03% of the sky frequency reaches 284 MHz in
band 10 (sky frequency = 950 GHz)) because the 1st LO cannot re-acquire lock that fast.
However, it can still be done in the 2nd LO. It needs to be
understood that because of tuning resolution constraints, the
frequency step size cannot be freely choosen (except at the lower
bands, where it can be done in the FTS). The spec is understood to
require
at least .03% of the sky frequency. The actual step size may be rounded up from
that requested to the next larger available value, so it might be more
much more than .03% of the sky frequency. (The tuning of the 2nd LO is not continuous, but
has gaps of up to 40 MHz.)
Outstanding Questions
- Is 0.03% of the sky frequency the frequency throw or the total switch range (i.e. is it +-0.03% or a total range of 0.03%)? Answer: It is the throw. -- JeffMangum - 09 Aug 2004
--
JeffMangum - 25 Jul 2004
Chip Scott's comments:
I agree with what Larry is saying with some minor clarifications.
The YIG tuning speed is roughly 2ms per 1 GHz step. This will be the
slowest tuning. The FTS tunes even faster but the frequency change takes
place on a timing event. The overall tuning speed will depend on whether
phase lock is lost or not.
I think the subtle point everyone misses is that there is no way the
system can verify successful tuning for anything under 48 ms or one
timing event. It is possible the tuning is taking more than 10 ms, all
the way up to close to 48 ms. On the next timing event, the PLL will
show lock but when was lock achieved or when did the tuning event settle
out?
It might help if I understood this particular requirement better.
Chip
--
JeffMangum - 01 Aug 2004
System Requirements Summary
Based on previous single dish (140ft, MWO, 12m) frequency switch capabilities, FS on ALMA should:
- Switch by at least +-0.03% of the sky frequency
- Switch frequency and lock LO on a timescale not more than 10ms
- Integrate at a given frequency switch state for not less than 1TE (48ms)
Dick Sramek suggested the following elaboration on these general specs:
- The 2nd LO shall be able to switch frequency by at least 285 MHz (0.03% of sky frequency at 950 GHz) and establish phase lock in 10 ms;
- The change in frequency should only be initiated at a timing event, TE (48 msec). This implies that the duration in any frequency state will be a multiple of 48 msec;
- Since frequency switching will only be used in total power mode, the requirement that the 2nd LO be phase continuous after a frequency switching cycle, is not needed while frequency switching;
- Continuous selection of LO frequency within the 285 MHz range is not required; 35% of the band must be available for use, but gaps up to 41 MHz are permitted;
- After a frequency change, a monitor bit should be set if lock is not achieved within 10 msec. This monitor point would be read after each command for a frequency change.
--
JeffMangum - 07 Sep 2004
Discussion from Darrel
Here are some thoughts on the frequency switching issue.
There are separate things: what we'd really like, what compromises
we've already agreed to, and what we'll actually get.
First what we'd like: obviously this is only relevant to
single dish spectroscopy, not to any form of interferometric
observing.
IN AN IDEAL WORLD:
1. Frequency switching is the most efficient single dish spectroscopic
observing mode, since the "Off" position that all observing modes
require is simultaneously looking "On" source. So, we would prefer
not to compromise this important observing mode too much.
2. If there were no other constraints, we would insist on the
1st LO being switched. The whole idea of frequency switching is
to keep everything in the receiver constant between the two switch
cycles, so all the slope and ripples in the rx passband, due to the
rx itself, cancel as perfectly as possible. By switching the 2nd LO
instead, we are effectively moving the IF filter response of the
first IF back and forth on the sky. This has two bad effects:
(a) the IF filters and other circuitry will not have a perfectly
flat frequency response. When the two phases of the 2nd LO switching
cycle are subtracted, we'll be subtracting a shifted version of the
1st IF response from itself. As it's shifted, it won't cancel
completely. In other words, by switching the 2nd LO rather than
the first LO, we'll get worse baselines, even in the total absence
of atmospheric effects.
(b) the total usable spectrum width is reduced by the switching range;
the edge of the spectrum is going to fall outside the IF filters in
one phase of the switching cycle. If we're trying to look at a spectrum
that's more than 2 GHz wide, we're going to have to arrange the different
2-GHz 2nd IF bands to overlap by more than the switching distance.
3. Switching offset precision. What doesn't cancel in frequency
switching is the frequency response of the antenna itself. This is
usually dominated by a standing wave ripple, typically coming from
coherent noise being radiated out of the receiver feed horn and
reflecting back into the feed after bouncing off the subreflector,
feed struts or whatever. This causes a variation in the rx input
noise, typically dominated by a sinusoid corresponding to the path
length involved in the strongest reflection. With frequency switching,
if the precise switching frequency interval is made to match an integer
number of periods of this dominant sine wave, the sinusoidal ripple
is effectively filtered out, and the spectral baseline is very much
better. So, in an ideal world, we'd like to match the frequency
switching interval to the standing wave. The period is likely to
be around 25 MHz (say, a 12-meter round trip reflection from
the rx to the subreflector), and we'd probably like to match
a multiple of that within, say, 1(?) per cent - that means the
switching distance should ideally be settable to within 0.25 MHz.
4. Frequency switching cycle time:
If we want to do frequency switching OTF observing, the cycle time
obviously needs to be very fast - less than a millisecond probably.
However, I don't think we're planning to do rapid-scanning-OTF
observing of frequency switched mode (although I don't know why not).
This is probably just one of those things we compromised on years
ago. At the Kitt Peak 12 Meter, we typically used a 1-second
switch cycle (but not with rapid OTF scanning).
WHAT WILL ALMA ACTUALLY PROVIDE?
I remember these discussions taking place years ago, and we had
to accept a compromise. That battle has been lost, so it probably
shouldn't be brought up again. However, we should just try to approach
the ideal (i.e. first LO switching, with complete choice of frequency
switching interval, and a reasonable fast switching cycle) so far
as the existing hardware plans and agreements allow it.
This doesn't answer Clint's questions, but it might nevertheless
be worth pointing out what we'd really like if there were no constraints.
--
JeffMangum - 07 Sep 2004
Comments from Al Wootten to Clint Janes regarding a document written by Chip Scott describing his understanding of the frequency switching science requirements.
You asked me to comment on a document you and Scott wrote.
I think your document makes sense to me. The switch of .03% is
always about 90 km/s. For frequencies above 3mm that will always
exceed 25 MHz which will make matching the FS interval to the
antenna period of 25 MHz achievable under the 'at least 0.03%'
understanding until perhaps when one gets to the lower bands.
Your summary:
- 1st LO can't do it, but that needs to be verified.
- 2nd LO can do it without too much grief if the minimum integration time (observing time on the sky for a single frequency) is at least 48 ms.
- 2nd LO can still do it for shorter integration times, but with a ~$30k redesign of the 2nd LO. BE requests a CRE in this case in order to document the funding request.
You asked three questions. Would you please verify:
- Our understanding of the frequency switch,
- Advise if the minimum integration time on a single frequency is 48ms, and
- Verify that 1st LO cannot reasonably perform the frequency switch?
My answers are:
- I think your understanding is good. Scientifically one would want to use FS when one wanted to observe an atmospheric line (position switching would cancel out the signal from the large source) or a bright continuum source (position switching switches between two signals with a large signal difference while frequency switching does not, making this preferable for some experiments).
- I don't think there are compelling reasons to switch on time intervals less than 48 ms. At the 12m the time was 1s; at MWO we used a 5 Hz switch. 20 Hz will be fine I think.
- Well it would be best for the first LO to do it. I think at one point Shillue told me what you wrote in your memo--that the laser couldn't switch and grab lock in 10ms for the 0.03%. But is there a frequency interval for which it CAN grab and lock in 10ms? Or is there a time interval for which it CAN grab and lock at 300 MHz shifts? It seems to me this is what your conclusion is driving at, and it would be good to know the answer.
Clear skies,
Al
--
JeffMangum - 07 Sep 2004
--
Since frequency switching came up in our discussion yesterday,
here are a few comments on the observing technique.
Although the need to support frequency switching for single dish
spectral line observations is in the ALMA requirements, I'm not
sure that the importance of frequency switching being carried
out only in the FIRST local oscillator has been appreciated.
First, to reiterate the importance of frequency switching as
a single dish observing mode: all SD observations require
switching against something, whether that's blank sky, a cold
load or whatever. Frequency switching, where the frequency throw
is small compared to the total rx passband, has the advantage
that both the "OFF-souce reference" and the "ON-source" signal
are observed simultaneously for 100% of the time. It's like
an increase of a factor of 2 in observing time, making
frequency switching one of the most efficient SD observing
modes there is. Further, for observing terrestrial atmospheric
lines - which are of interest for atmospheric physics, but
also for some calibration issues - frequency switching is the
best way to go; you can't position-switch against blank sky,
because the atmosphere is more or less the same in all directions.
Why does the switching have to be in the first LO? When you
switch the first LO, everything in the rx chain beyond the
first mixer stays constant, so will cancel nicely when you
subtract 2 phases of LO frequencies, "ON-source" and "OFF-source."
The only thing that doesn't cancel is the noise spectrum
before the mixer, which includes antenna spillover noise,
atmospheric noise, and the desired astronomical signal.
This allows you to observe the astronomical signal, while
canceling out the strong ripples, slopes and drifts occuring in
anything in the receiver system after the first mixer.
If, instead of switching in the first LO, you chose to switch
just the second LO, the results will be very poor. The first IF
amplifiers and filters will inevitably have slopes and ripples
across the passband. By switching the 2nd LO it won't be possible
to distinguish between spectral features in the first IF amplifiers and
filters, and astronomical spectral features. Remember, these IF slopes
and ripples are multiplicative factors on the entire system noise.
If you're trying to observe a milliKelvin astronomical signal, which
will often be the case, it will have to be done in the presence of
ripples which may be effectively hundreds of degrees in amplitude - say,
40 dB to 50 dB stronger than the signal you're trying to see.
Results will be very poor indeed.
The ripples in the frequency response of the feed horn, and the
antenna itself, of course do not cancel with any form of frequency
switching. However, they will normally be very small indeed, being just
a multiplicative factor on the background (cosmic and atmospheric)
and spillover noise received at the feed. This compares to the
ripples in the first IF, which are multiplicative factors applied
to the entire system noise.
So, frequency switching should be supported in the first LO.
Frequency switching applied to the 2nd LO doesn't make much sense.
Cheers,
Darrel.
--
The 12 M at Kitt Peak used to do lots of detection experiments in
beam-switched mode. With only 4 nutators, that's not going to be
a very sensitive mode on ALMA. However, for a detection experiment
on extended objects, frequency switching may be the only mode possible.
If any source is sufficiently extended (i.e. more than a primary
beam width) than it won't be visible on any interferometric baseline.
The most sensitive prospect of detecting a weak signal under these
circumstances is to average the power spectra from 64 single
dishes. You have to be switching something for observations like that,
and frequency switching is going to be the only thing that may be
fast enough. Position switching, slewing the entire antenna back
and forth, may simply not be fast enough to take out the atmosphere
and still allow good spectral baselines. Frequency switching, since
we can't nutate the S/R on 64 antennas, is likely to be the only
way to go.
So, weak detections of extended sources will require frequency
switching.
Cheers,
Darrel.
--
AlWootten - 29 Apr 2005