Amplitude Calibration

TIP Last Update: JeffMangum - 30 April 2014

Specification

  • Relative Calibration Accuracy: 1% for nu < 370 GHz; 3% for nu >= 370 GHz
  • Absolute Calibration Accuracy: 5% for all frequencies


Contents


The ALMA Amplitude Calibration Requirement

Document which attempts to clarify the various terms used to describe the amplitude calibration requirement and how they relate to the different pieces of this specification.

Amplitude Calibration Memos (ALMA, EVLA, etc.)


Flux Calibration Sources Discussion


Flux Calibration Source Measurements


Tsys Measurement


Calibration Device Technical Specifications and Reports

Calibration Device Technical Specifications

See Calibration Device Technical Specifications (FEND-40.06.00.00-009-A-SPE; Patt). The current version of the specs (2007/07/31) is FEND-40.06.00.00-009-B-SPE-20070731.pdf. Folder with all hardware development documentation is at 40.06 Calibration Systems on almaedm.

NOTE: Current (2007/07/31) version of the calibration device technical specs does not address following conflicts between device review and current specs.

Comments to the 2006-07-10 draft by JeffMangum:
  • Section 2.1 (Equipment Definition) states that "The simultaneous use of a calibration load and the solar filter is not possible". This is in direct conflict to the requirements laid-out in the amplitude calibration device review of 2005-09-07.
    • 2007/09/03: Ferdinand Patt responds: "As you have correctly stated in an earlier email, specification do not derive from review meeting, they derive from top level specification. To give some input on this anyway, the load and the solar filter are attached to the calibration wheel. Moving the loads on top of the solar filter could be possible but does not make sense. You have a 13 to 16 dB attenuator in between the receiver feed and two load that are ~ 50 K apart in temperature. Your Y-factor is about 1 something."
  • Section 4.1.2 (Hot RF Load Temperature Stability) states that "specified temperature stabilities shall be reached within 1 hour after the calibration device is going from its off-mode or transport-mode to operation-mode". This timescale should not be longer than the time required for the FE to reach operational mode following an antenna move. In other words, the amplitude calibration device must always be slaved to the operability of the frontend.
    • 2007/09/03: Ferdinand Patt responds: "As long as you have power to the receiver cabin the calibration device is operational. Transport or off-mode does not mean the unit is switched off. The calibration wheel is positioned in one of its park position and the backends are switched on. Once you loose power, or the power is switched off, the receiver also does not has power and is warming up. Getting the receiver operational again probably needs several hours (cool down of cryogenics). The 1 hour required is well within spec." Apparently, "off-mode" is commensurate with turning the frontend off.
  • The calibration timescales listed in Sections 4.5 (amplitude calibration device; 9.0 seconds), 5.3 (solar filter; 7.0 seconds), and 6.3 (QWP; 7.0 seconds) for each component of the amplitude calibration device is in conflict with information provided by Matt Carter at the amplitude calibration device review of 2005-09-07. During that review, Carter said that the move time for the amplitude calibration device will be less than 1 second. As only total calibration timescales are given in the current technical specification, we assume that:
    • The stow position to ambient load, hot load, solar filter, or QWP measurement (over commanded receiver band) movement time is less than 3.5 seconds.
    • The ambient load to hot load movement time is less than 2.0 seconds.
      • 2007/09/03: Ferdinand Patt responds: "As you have correctly stated in an earlier email, specification do not derive from review meeting, they derive from top level specification and sorry they certainly not derive from Matt Carter saying something. But again to give some input here and where this number is deriving from is the famous meeting in Grenoble. Section 5 (c) is asking for less than 1% for a cal cycle of 15 minutes. This is all I had and this number was also confirmed by Matt Carter."
  • Since the calibration system must be switched over the same time scale as any other device in the receiving system, the maximum time spent positioning a calibration subsystem (ambient load, hot load, solar filter, or QWP) must be comensurate with the maximum receiver band and fast switching time scale of 1.5 seconds.
    • 2007/09/03: Ferdinand Patt responds: "If this is still a requirement, we have to redesign the entire calibration system."
  • A document describing the proposed specification for the QWP is attached.
    • 2007/09/03: Ferdinand Patt responds: "Okay these are proposed QWP specs. They need to be reviewed, approved and put into the calibration spec and requirements. For your information the approved FE technical specification do not require a QWP anymore. Only the two prototypes will be equipped with the actuator for the QWP, all other units will only have the mechanical and electrical provision to put a QWP actuator in. We currently have no QWP build, we only have a design." Apparently, the production calibration systems will not include QWPs. Perhaps installation is delayed until later in project?.
  • Section 4.5: Why the limitation on how often the calibration cycle can be repeated? 5 minutes is a reasonable estimate, but should not be imposed as a limit. Can this be reduced to 1 minute?
    • 2007/09/03: Ferdinand Patt responds: "You could calibrate every minute, the calibration device will do this but there is a limited life time of the mechanics. The mechanical design must be based on some number to start with. I had a number to start with and this is out of the amplitude calibration device meeting section 5 (c) on page 8. There you talk about a calibration cycle every 15 minutes. On this see also my previous comment."
  • Section 9.1.3: The lower limit to the relative humidity operating range (20%) seems too high. I would expect that RH will quite often be much lower (near 0%) than this.
    • 2007/09/03: Ferdinand Patt responds: "Version B spec is updated and ask for 0 to 95%."

Calibration Load Radiating Temperature Test Report

See:

Calibration Group Discussion of Prototype Calibration Load Test Results

The prototype load test report was discussed by the Calibration Group at its 2007/08/16 telecon. From that discussion:

  • Several issues from load report need resolution. Basic problem is that it is not possible to optimize the backscatter, total scatter, and thermal performance of a load system with a single type of absorber. Since the amplitude calibration sequence will involve comparisons of spectral measurements of the loads, standing waves are a problem.
  • Based on the analysis of the "chopper wheel" amplitude calibration technique presented in ALMA Memo 434, there isn't much room for relaxation in any of the load temperature regulation requirements. It would seem that adopting a more relaxed +-0.5 K (0.15% at 300 K) load temperature regulation spec would result in a missed 1% amplitude calibration spec at frequencies less than 370 GHz.
  • A revised version of the Calibration Device Technical Specs is available. Still several issues:
  • Richard Hills has produced a summary and recommendation document describing the current status of the amplitude calibration device testing and development. In summary, Richard points to four issues one must overcome when determining the effective temperature of the current hot and ambient loads:
    1. Determining the actual physical temperature, as determined by thermometers embedded in the structure of the load.
    2. Temperature gradients between the temperature sensors and the actual emitting surface. These will depend on IR cooling and convection (both that due to the gradients in the air temperature which in turn depend on the orientation, and that due to the forced air circulation produced by the HVAC system).
    3. The emissivity of the load.
    4. Most critically, the backscatter characteristics of the load.
  • Note that the specification for the required accuracy of the loads is 0.2 K (+-0.07%) for the 20 C (ambient) load and 0.6 K (+-0.2%) for the 60 C (hot) load. This specification refers to the effective black body temperature of the loads.
  • Richard recommends the following:
    1. Continue the procurement of two sets of loads to get us through the initial test phases. The Calibration Group concurs.
    2. Confirm the figures in the calibration design document as being the requirements for the knowledge of the effective black-body temperatures of the loads when used with the ALMA receivers at all frequencies of interest. The Calibration Group concurs.
    3. Suggests that it might be productive to slightly soften the load temperature stability requirement and set the specifications at +/-0.5K for the ambient load and +/-1K for the hot load, while maintaining the values of +/-0.3K and +/-0.6K as goals. The Calibration Group discussed this and opted not to follow this recommendation. Since the current 0.2/0.6 K spec only marginally allows one to meet the 1% amplitude calibration specification, there is no room for relaxation in the load stability specification.
    4. Start immediately on a program to develop an alternative design of load that does meet these requirements. The Calibration Group concurs. Richard is convinced that the cone design is sufficiently promising to make this worthwhile, but clearly we should not start on such a study without a clear understanding of what is to be done and on what timescale.
  • The Calibration Group further discussed the temperature stability limitations of the prototype systems with Patt and Murk. It appears that the need to provide a hot load which meets the temperature stability requirement from 30 to 900 GHz is the main limitation. According to Patt and Murk, if the hot calibration load were not needed for Bands 1 and 2, then it would be easier to meet the overall load temperature stability requirement. During the next month or so JeffMangum will look into whether we can get away with using just an ambient load for relative amplitude calibration for Bands 1 and 2.
    • On 2007/09/07 JeffMangum made some calculations of the accuracy of the single- and dual-load calibration systems at Bands 1 (40 GHz) and 2 (80 GHz). Following the same formalism described in Load Calibration at Millimeter and Submillimeter Wavelengths: Mangum; October 18, 2002; ALMA Memo 434 the situation is directly analogous to the calculation done at 230 GHz. Assuming 50th-percentile tau(40)=0.02+-0.002 and tau(80)=0.03+-0.003 the single-load (traditional) chopper uncertainty is ~1.2% at both frequencies under "best" conditions (as laid out in Memo 434) assuming double sideband operation. The total uncertainty is almost completely dominated by the uncertainty in the sideband gain ratio and the uncertainty in the rear scattering and spillover plus radiative efficiencies (etal). Note that this analysis assumes that the frontends do not suffer affects due to saturation.
    • At our 2007/09/13 telecon it was pointed out by Richard that in fact Bands 1 and 2 will use HEMPT's and not SIS mixers, so they will be intrinsically SSB. One can then eliminate this element of the calibration error. Recalculating with sigma(Ri)=0 yields a single-load chopper uncertainty of 1% which is completely dominated by the uncertainty in etal.
  • Richard noted in an additional follow-up email that there appear to be three options to resolve the load temperature stability issue:
    1. Insist that both the ambient and hot loads have good performance at all bands (i.e. ~30 to 950GHz).
    2. Revise the system so that the ambient load is for 30 to 900GHz, while the hot load is used for 85 to 950GHz only (i.e. not Bands 1 and 2). NB B2 is formally 67-90 GHz. The advantage of this option is of course that the hot load can be made smaller (roughly a factor of two in the size of the aperture), which eases the design considerably. The argument is that the atmospheric opacity in Bands 1 and 2 is low enough that one does not need to use the two-load calibration method: one (ambient) load is good enough to meet ALMA's accuracy requirements. It seems that this approach was adopted in e.g. the Calibration Unit Design document, but it was not directly taken on board in the development work done so far.
    3. Make two (essentially identical) loads for 85 to 950GHz with only one of them heated, and then a third ambient-only load for 30 to 90GHz. This option is attractive in that the ambient load can be made with a smaller aperture as well and one could perhaps make saving by using the same design as the hot load. (Remember that the ambient load has the tighter requirements on the accuracy. Note also that even the ambient load is not really at thermal equilibrium - the air may be at one temperature but much of the radiation environment will be the cold sky or even the cold interior of the cryostat.) A point here is that, since there are no active plans to build Bands 1 and 2 at present, we can postpone actually making the third load until those bands are funded. We would of course need to be sure of the feasibility of making a low frequency ambient load that fits in the space available.
  • A point to appreciate about having only one calibration load for Bands 1 and 2 is that it does not directly give you a value for the receiver noise temperature - only the overall system temperature. Given that we will have other ways of estimating the atmospheric contribution (e.g. the WVR plus model), Richard believes that this is not a problem.

-- JeffMangum - 16 Aug 2007

Summary of Calibration Group Discussion of Prototype Calibration Device Design and Load Test Results

  • Total calibration cycle time of 9 seconds is too long. This was calculated based on an assumed 10% of 15 minute calibration cycle interval. A better value to use would be 5 seconds. Perhaps the alignment and positioning accuracy used to set this spec is too stringent?
  • Inability to use solar filter and loads is probably ok. MarkHoldaway has devised a scheme for calibrating solar observations with current prototype calibration system design.
  • Load temperature accuracy specification refers to the effective black body temperature.

Comments to CRE Proposing Some Changes to Calibration Device Specifications

Richard Hills and JeffMangum submitted a CRE which requests changes to the Specifications of the Calibration Device:
  1. Total amplitude calibration timescale upper limit.
  2. Use of Hot Load for amplitude calibration of Bands 1 and 2.

In a follow-up, Ferdinand Patt suggested some additional changes be included. These changes resulted from a simulation done by Luitjens Popken, have been discussed among many, and are the following:
  1. Effective emissivity (or reflectivity): From Luitjens simulation. Suggest this should be 0.999 or higher, i.e. total scatter below 1x10^-3. This implies a maximum error of ~0.2 K for a change in the "surroundings" of 200K which would be a pretty extreme case - e.g. from seeing the cold sky at zenith reflected from the top of the cryostat to seeing the warm sky near the horizon. As far as I can see from the test report this is should be achievable with a good design. We could live with 0.998 if that was the best that could be done. Ferdinand proposes to write 0.998 and 0.999 as a design goal. The test report shows scatter of ~ 3*10-3 at 90 GHz and ~ 7*10-4 at 323 GHz for the pyramids (probably used as hot load) but also 2.4 * 10-3 for the cone at 90 GHz (probably used for ambient load).
  2. Variation of the brightness temperature across the aperture (TBC, TBD): This is a bit tricky. The measurement will presumably give us an average weighted by intensity pattern of the receiver at this location. This pattern is of course strongly peaked in the middle so we could actually tolerate a significant gradient in temperature towards the outer edge. You could argue that any errors would be included in (4) but in practice we would like to avoid a situation where there are strong variations since it would make the results sensitive to alignment, etc. As a pretty arbitrary number, how about +/-1K for the hot load and +/- 0.5K for the ambient?
  3. Total calibration uncertainty (difference between "measured" brightness temperature and the load's brightness temperature calculated from thermistors (PTRs) and knowledge of the emissivity); this includes implicitly the required accuracy and repeatability of the physical temperature measurement (present Spec sect. 4.2) that would not be specified any longer.
    • Hot: +/- 0.5 K at 90 deg C
    • Ambient: +/- 0.3 K

Comments?

-- JeffMangum - 03 Dec 2007

Semi-Transparent Vane

Multi-Load Amplitude Calibration Device

Short Timescale Total Power Fluctuation Discussion

Email discussion regarding how to deal with short-timescale total power fluctuations.

Mark Holdaway response to Ryohei Kawabe

Ryohei Kawabe wrote:

> 2) Specific  comments & questions
>     - Amplitude calibration is planned to be calibrated by the "Chopper Wheel Calibration".
>       But, short-timescale (visibility) amplitude fluctuation due to the atmospheric opacity
>       fluctuation cannot be calibrated by this and might cause the error of radio source flux. 
>       Is there a plan to calibrate this.

These fluctuations will have a random part and a systematic part. We
actually DO need to worry a bit about the random part -- if we performed
an important flux calibration measurement during an opacity "blip", we 
would have a systematic error.  Systematic trends between measurements 
can be removed by interpolation.

It seems to me that the best hope for dealing with this is the WVR -- 
we can, through atmospheric models, convert the WVR data into
opacity-at-observing-frequency, which we can use to correct the 
visibilities on an integration-by-integration basis.

I have been worrying about a related problem: fluctuating decorrelation.
Of course, our great hope for that problem is also WVR.

Dave Woody chimes-in...

There actually is a method to measure the fast atmospheric
opacity fluctuations simply using the total power monitors
in the receiver IF chain.  If the gain and noise temperature
of the receiver system is stable enough, then you can interpret
the total power variations as arising from opacity/emission
variations from the sky.  We do this on the OVRO/CARMA
system.  Essentially it is a continuous version of the ambient/sky
load calibration technique where you measure the power from
the ambient load periodically but measure the sky continuously.
Cheers
David

Koh-Ichiro points to the fact that NMA uses such a system too...

David Woody wrote:
> There actually is a method to measure the fast atmospheric
> opacity fluctuations simply using the total power monitors
> in the receiver IF chain.  If the gain and noise temperature
> of the receiver system is stable enough, then you can interpret
> the total power variations as arising from opacity/emission
> variations from the sky.  We do this on the OVRO/CARMA
> system.  Essentially it is a continuous version of the ambient/sky
> load calibration technique where you measure the power from
> the ambient load periodically but measure the sky continuously.

We are also using such a system for Nobeyama Millimeter Array,
although the system is a bit old and the time resolution
is a problem.  I believe the performance of ALMA BE system
will be good enough.

Cheers,
Koh-Ichiro

-- JeffMangum - 07 Sep 2004

Well - may be it's a bit late, but I'll just mention that we've been doing this too at Plateau de Bure, for several years. I'm sure that the emission changes are still the best way to monitor the opacity changes, with no need to rely on the atmosphere model. Hovever this assumes the receiver stability is good enough.

-- RobertLucas - 10 Feb 2005


Decorrelation Correction Change

Concerning Section 2.4 Amplitude Decorrelation Correction

I recommend we AX that long and tedious section (I wrote it, I can insult the writing), and replace it with this short text:

2.4 Amplitude Decorrelation Correction

Fast switching phase compensation alone results in significant residual phase fluctuations which lead to variable decorrelation. This variable decorrelation must be accounted for both to maintain an accurate flux scale and to make high quality images. However, there is strong indication that a phase compensation scheme using both fast switching and water vapor radiometry will be able to reduce these residual phase errors to the point that they will not result in significant decorrelation. Additionally, the residual phase errors of a combined FS/WVR scheme will probably be dominated by Gaussian noise from the WVR, so if a decorrelation correction is required to achieve an accurate flux scale (ie, at the highest frequencies), that correction will be nearly independent of time and baseline, ie, a simple scaling of the interferometric data which can be accomplished with an interferometric/total power cross calibration or by dead reckoning based on the known noise properties of the WVR units. Variations in the noise level of different WVR units will naturally be calibrated out as antenna-based gains in the flux calibration process.

-- MarkHoldaway - 08 Sep 2004


Berkeley Absolute Flux Calibration System (Gibson/Welch)

Frontend Compression and the ACD

On 2009-01-25 Pavel Yagoubov provided the following note to Richard Hills and
JeffMangum.

A purpose of this note is to clarify how the top-level absolute amplitude calibration requirement flows down to a Front-End saturation requirement and the Amplitude Calibration Device (ACD) absolute accuracy. Being still relatively new in the project, I apologize if this topic has been sufficiently addressed in the past and there are conclusive answers to my questions summarized at the end of this note, which I might have failed to find, - in this case please indicate relevant references.

The relevant top-level specifications are:

A few ALMA memos address amplitude calibration with a goal to obtain 1% absolute accuracy: #321 (Plambeck), #401 (Kerr), #434 (Mangum), #460 (Kerr), and #461 (Guilloteau). A few of these Memos also deal with an effect of SIS mixer and IF amplifiers saturation on the instrument calibration uncertainty.

However, there seems to be no obvious link from the top level amplitude calibration requirements to the Front-end saturation specification. FEND-40.00.00.00-00130-00 / AT specifies “The large signal gain compression resulting from an RF load of 100°C shall be less than 5%” and is referred to the system requirement 227 to have <1 dB compression in 11 dB dynamic range. The 11 dB comes apparently from the most sensitive receiver dynamic range, ~30-400 K. This is obviously not a very clear link and by far doesn't fulfill the absolute calibration requirement. I assume that a substantial part of the 1 dB compression is expected to happen in a 3-level correlator and that there are developed algorithms for non-linearity correction of that sub-system. Apparently this link is not much relevant and the Front-end compression is better to be specified based on amplitude calibration requirements because the latter is more demanding.

However, the 5% Front-end compression does not appear to be correct flow down from the calibration accuracy requirements, definitely not for the lower frequency bands, because the calibration is done in a non-linear region and far away from Tsys. Figures below show how the accuracy depends on the input temperatures for the receiver noise temperature Trx = 40 K (relevant for bands 3-6) and Trx = 100 K (higher bands). I used Tsat=10.000 K for the Trx=40 K and Tsat=12.500 K for the Trx=100K.

PYFig1.pngPYFig2.png

Figure. Tmeasured/Ttrue as a function of input temperature. 5% compression between 77 K and 373K as specified in Front-end technical specifications. Trx=40 K (left) and 100K (right).

These are the figures we should expect with ALMA receivers. I looked through different ALMA cartridge test reports; compression is typically measured between 1-3 % with 300 K using different techniques, corresponding to 4-5% at 373 K. All these numbers are also in line with theoretical predictions. Compression is a function of SIS mixer architecture (number of junctions, their size, type of mixer), but also is a function of LO frequency and especially dc bias as the SIS mixers are usually saturated at the output.

As can be seen from the Figures above, the 5% compression requirement at 100°C set on the Front-end is by far not sufficient to meet the top level requirement on amplitude calibration accuracy, definitely not at the lower bands, where an error of up to 30% is expected even if this would be the only contributor to the total budget. A 1% accuracy is obtained with Tin>200K for Trx=40 K and T>230 K for Trx=100K.

A few ALMA Memos address SIS saturation effect in details, conclusion of #461 was that to meet the 1% calibration requirement one needs 2 calibration loads (ambient and hot), as well as a possibility to measure combinations of loads and sky. It appears that the ACD specifications, +-0.3 K for the ambient and +-0.5 K for the hot load, are copied from this Memo conclusions.

The 5 calibration step measurements configuration has not been accepted for ALMA, and it is not obvious now whether the top level absolute amplitude calibration specification can be met with current 2 load configuration and also whether the requirements for ACD are still optimally defined, especially for the hot load and for the lower frequency bands.

To summarize, a list of questions and comments:
  1. Is there a plan to characterize Front-ends non-linearity? If yes, I guess we need to have <0.2 % characterization accuracy to stay within 1 % error for Tsys=50K. Important to note that such characterization should be required for all receiver settings.
  2. Front-end non-linearity can be improved by tuning the SIS mixer away from the optimum dc bias on the expense of sensitivity degradation. Are there plans for doing this? If yes, we need to investigate to what level the non-linearity can be improved and check saturation of the other components which might start influence, as IF amplifiers.
  3. In case Front-end non-linearity will not be characterized, what is the ACD uncertainties budget in the final error? It appears that in this case the cold sky calibration based on atmospheric model and opacity measurements by WVR and sky dip will determine the final accuracy at low frequencies and the need for the hot load gets marginal. In any case, it does not look reasonable to specify the ACD to +-0.1% absolute accuracy (+-0.5K for the hot load) on top of 5% non-linearity uncertainty. I do realize that importance of the hot load increases with frequency and that the high end bands will benefit from it, but the ACD uncertainty impact gets fractionally smaller there because Tsys gets closer to the ambient load temperature.

P. Yagoubov, 25-01-2009


Augmenting this discussion, Kamaljeet Saini and Pavel provided the following references to several ALMA receiver gain compression measurements:

Compression values provided below are (min, max) of compression data as a function of (test frequency, sideband tested, article number) based on the data for the first eight pre-production units, measured as described in the accompanying procedure(s).

Pavel notes the following regarding these gain compression measurements:
  • To strengthen an importance of this issue, please note that this original note is based on the FE specification for the compression, "5% large signal compression @ 373 K".
  • Cartridge specifications have adopted a different definition: "The large signal gain compression caused by the changing the RF load temperatures from 77 K to 373 K shall be less than 5 %". Because of this different definition, the large signal compression @ 373K translates to about 8%. And the calibration error of the Front-ends will become even larger.
  • In practice the cartridge groups verify the compression using the 77 K and the 300 K loads (except for the Band 9 who are using the 373 K load) and even with this smaller range get close to the relative compression specified.
  • In order to synchronize the specs we either need to relax the FE specs on compression @ 373 K to 8% or change the definition to a "5% large signal compression caused by the change between the 77 K and 373 K loads".

-- JeffMangum - 2009-06-12

FE internal CRE to stop measuring Band 3 gain compression in steady-state Production

Proposed CRE to relax the requirement to 6% (posted on Nov 16)

Comment From Bernard Lazareff Regarding Gain Compression Measurement Techniques

Following the ACD CDR in December 2009 Bernard Lazareff provided a summary analysis of the Band 3 gain compression measurements quoted in FEND-40.02.03.00-179-A-REP comparing the sector and tone methods used. Bernard points out: My conclusions from this exercise are that the results from tone injection are not compatible with the results from the sector method, and are compatible with the absence of gain compression (within the tbc accuracy). Furthermore, Bernard also notes that the report quoted above concludes "Although the two measurement techniques show quite a difference in results, they both show that the Band 3 cartridge gain compression is well below the 5% specification".

-- JeffMangum - 2009-12-09

Saturation Measurement Using the Pizza Slice Method

Bernard Lazareff has provided a report entitled "Saturation by the Pizza Slice Method: Experimental Investigation" by B. Lazareff and Patrice Serres. In this report the authors have conducted an investigation of the "pizza slice" method used to characterize the linearity of frontend Band 3 cartridges. The authors find, as expected, a spurious excess of non-linearity, strongest when the measurements are performed at a place where the beam is narrowest.

-- JeffMangum - 2010-01-15

Discussion of gain compression and future measurement strategy

ToddHunter: After speaking with Saini on Friday, I can clarify some of the points raised at last week's Science IPT meeting regarding front end gain compression.

I. Specification, and the proposed CRE

Richard asked whether I thought Pavel's CRE was effectively requesting a factor of 2 loosening of the current spec (in addition to the loosening from 5% to 6% in Bands 3 and 4). This question relates to the fact that the ratio between the incremental gain compression and the "large signal" gain compression is ~2 (see Figure 4 of Tony Kerr's Memo 460). If one simply notes that both the current and proposed specs use the term "large signal gain compression", then the answer is "no". But I don't know if the term "large signal" in the original spec was intended in the same way as Memo 460 defines it. If not, then I suppose the answer can be argued as "yes".

II. Status of measurements at cartridge manufacturers

At present, only Bands 7 and 9 are measuring the gain compression for every cold cartridge. Band 3 measured the first 12 cartridges using 77K and 300K loads, but using the partial absorber method ("pizza slice") in front a cold load, which was recently demonstrated as flawed. Since after the first 8 cartridges, all measurements were below their assumed requirement (5%, large signal), this test was removed from the Production Phase test plan in a FE-internal CRE. However, it has since been realized that when one corrects the data to the proper hotload temperature (373K), 14% of the measurements on the first 12 cartridges become out-of-spec. Note that the revised Test Plan associated with this CRE is not yet signed.

Following Bands 3's change to an abbreviated schedule, Band 6 has also removed the gain compression test from later cartridges. Band 7 is measuring every cartridge and obtaining good performance (using the small-signal method with a broadband noise source). Band 9 is measuring every cartridge with the small signal CW method. They divide their result by 2 to obtain the equivalent large-signal method. (Initially they were not, as they were meeting their 3% requirement without doing so, but then had one unit slightly above 3%). Regarding Band 4, after initial Band 4 measurements showed non-compliance, the Band 4 group has been advised to measure it on at least the first 8 (or so) units using the small-signal method. I don't know of any measurements on Band 8 or 10, although the first Band 8 cartridge has arrived at the NA FEIC.

III. How to proceed?

It had been my vague impression that there was a plan for the calibration "pipeline" to receive laboratory compression measurements (i.e. unique to each serial number) that could be used to remove the effect of compression in such that the residual error was <1%. However, looking closer, there is no mention of this in the Amplitude Calibration Steps document. Indeed, as things currently stand, this cannot possibly happen since measurements are no longer being made for two of the bands (and the two likely to have the highest compression!). One could imagine using the average of the known measurements at a given frequency within each band but it is not at all clear that this would be better than doing no correction at all. In fact, the level of compression changes across the RF band (see for example a recent Band 9 report where the value varies from 2.6% at 614GHz to 0.6% at 710GHz: report).

At this point, it is also important to review Memo 331, which clearly demonstrates that the gain compression of (at least some) SIS mixers is quite sensitive to the bias point. In this case, one could detune the bias to improve linearity at the cost of sensitivity. (This is not surprising, because the concept of "detuning" a mixer for better stability is another familiar technique.) One may elect to operate this way for planetary observations, for example.

Given that detailed measurements of all Band 3 and 6 cartridges at a grid of frequencies seems unlikely (due to unacceptable schedule delays), what course would be most useful to Science & Operations? I think we would benefit from performing enough compression measurements of one or two cartridges in each Band to get a good idea of how much it varies with respect to mixer tuning parameters, and from unit to unit. In other words, measure the compression with the small-signal method (preferably using a broadband noise source to avoid standing waves) at the optimal level of LO and mixer bias. (The signal can be injected through the LO port -- see section "IV.b Linearity Calibration" of Tong et al. from April 2009 THz meeting). Then perturb these settings (individually) and measure the change in compression. Repeat this process at many frequencies in the Band. Repeat everything with a second cartridge. This would tell us which bands are most sensitive in their compression to the tuning.

Where can this be done? Proper measurements require calibrated loads at (a minimum of) three different temperatures (in order to fit the curvature of the gain curve). The only place where this capability is planned to exist in ALMA is at the EU FEIC, starting in March when they will receive a production ACD. An added benefit to making such measurements were there on the next FE is that after it ships to the telescope, one could attempt to develop a measurement method using the sky and the ACD that successfully replicates the lab results. This method could then be used on all FEs.

Comments?

-- ToddHunter (2009-01-25)

Amplitude Calibration Saturation Discussion (ALMA Memo 461)

Note from Mark Holdaway to Stephane and Aurore asking about some of their assumptions:


Dear Stephane & Aurore,

I think you are being overly pessemistic about sub-millimeter observing.
I see that you take the median value of sigma_{w} -- ie, the rms 
pathlength fluctuation as seen at an elevation angle of 36 deg
with the 300m interferometer.

Now, it is true that the good phase conditions and the good opacity 
conditions are not highly correlated, but Larry & I did look
at the joint probability distribution of TAU & PHI, and it is
written up in LAMA Memo 801:

http://www.tuc.nrao.edu/lamaMemos/lamaMemo801.pdf
(note the plot is in femtoseconds of delay, corrected to zenith, and
1 fs is 0.03 microns).

Or, looking at the statistics of the fluctuations only
(actually, these are based on statistics of hourly medians):

percentile      rms path @ 36 deg        improvement
                  [microns]

50%              251                     1.0
25%              122                     0.49
10%              68                      0.27


SO, I am guessing that we can certainly cut your T sky fluctuations in 
half for 650 GHz.


Another fact which you allude to:  the fluctuations get highly blended
if you are able to switch faster than 1 s (ie, the crossing time of the
dish) -- is that a possibility, to switch at 2 Hz instead of 1 Hz?



    -Mark

Stephane's response:

A 15:00 08/02/2005 -0700, Mark Holdaway a écrit :


> Dear Stephane & Aurore,
>
> I think you are being overly pessemistic about sub-millimeter observing.
> I see that you take the median value of sigma_{w} -- ie, the rms
> pathlength fluctuation as seen at an elevation angle of 36 deg
> with the 300m interferometer.


Yes, that is correct, but since it appears that receiver stability is the
limiting factor, this does not affect the conclusion: it re-inforce it.


> Now, it is true that the good phase conditions and the good opacity
> conditions are not highly correlated, but Larry & I did look
> at the joint probability distribution of TAU & PHI, and it is
> written up in LAMA Memo 801:
>
> http://www.tuc.nrao.edu/lamaMemos/lamaMemo801.pdf
> (note the plot is in femtoseconds of delay, corrected to zenith, and
> 1 fs is 0.03 microns).
>
> Or, looking at the statistics of the fluctuations only
> (actually, these are based on statistics of hourly medians):
>
> percentile      rms path @ 36 deg        improvement
>                   [microns]
>
> 50%              251                     1.0
> 25%              122                     0.49
> 10%              68                      0.27
>
>
> SO, I am guessing that we can certainly cut your T sky fluctuations in
> half for 650 GHz.
>
>
> Another fact which you allude to:  the fluctuations get highly blended
> if you are able to switch faster than 1 s (ie, the crossing time of the
> dish) -- is that a possibility, to switch at 2 Hz instead of 1 Hz?
>

It would always be better, but it may be difficult. The proposed device is fairly heavy
and large, and moving it faster than 1Hz may require too much power
(and power dissipation, which you want to avoid too).

        Thanks,

                Stéphane

-- JeffMangum - 19 Sep 2005
Topic revision: r66 - 2014-04-30, JeffMangum
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding NRAO Public Wiki? Send feedback