SPS: A software simulator for the Herschel-SPIRE photometer
AAstronomy & Astrophysics manuscript no. sps˙paper˙accepted c (cid:13)
ESO 2018October 19, 2018
SPS: A software simulator for the Herschel-SPIRE photometer
B. Sibthorpe , P. Chanial , and M. J. Gri ffi n UK Astronomy Technology Centre, Royal Observatory Edinburgh, Blackford Hill, Edinburgh, EH9 3HJe-mail: [email protected] Astrophysics Group, Blackett Laboratory, Imperial College, Prince Consort Road, London SW7 2AZ, UKe-mail: [email protected] Cardi ff School of Physics and Astronomy, Cardi ff University, Queens Buildings, The Parade, Cardi ff , CF24 3AAe-mail: [email protected] Received XXXX XX, 2008; accepted XXXX XX, 2008
ABSTRACT
Aims.
Instrument simulators are becoming ever more useful for planning and analysing large astronomy survey data. In this paper wepresent a simulator for the Herschel-SPIRE photometer. We describe the models it uses and the form of the input and output data.
Methods.
The SPIRE photometer simulator is a software package which uses theoretical models, along with flight model test data, toperform numerical simulations of the output time-lines from the instrument in operation on board the Herschel space observatory.
Results.
A description of the types of uses of the simulator are given, along with information on its past uses. These include examplesimulations performed in preparation for a high redshift galaxy survey, and a debris disc survey. These are presented as a demonstrationof the sort of outputs the simulator is capable of producing.
Key words.
Methods: numerical, Instrumentation: photometers, Space vehicles: instruments
1. Introduction
Software simulators are now used in a wide range of astron-omy instrument projects. They have many uses, from guidingthe instrument design, through observation planning, and finallyhelping to understand the astronomical data. This paper de-scribes the software package developed to simulate the Spectraland Photometric Imaging Receiver, SPIRE (Gri ffi n et al. 2008b)photometer instrument on board the European Space Agency’sHerschel space observatory (Pilbratt 2008), scheduled for launchin 2009. This software is designed to simulate the behavior ofthe instrument and spacecraft, and return realistic data in a for-mat compatible with the Herschel data pipeline (Gri ffi n et al.2008a).SPIRE is a dual instrument, comprising a three-band submil-limetre camera and an imaging Fourier transform spectrometer(FTS). In this paper we will only be considering the photome-ter; for information on the spectrometer simulator see Lindneret al. (2004). The SPIRE photometer uses arrays of hexagonallypacked feedhorn-coupled bolometric detectors, and has a fieldof view (FoV) of 4 x 8 arcminutes. The FoV is observed si-multaneously in three bands centred approximately at 250, 350and 500 µ m, giving di ff raction limited full-width-half-maximum(FWHM) beam widths of approximately 18, 25 and 36 arcsec-onds respectively. There are 139, 88 and 43 detectors in the250 µ m, 350 µ m and 500 µ m arrays respectively, along withtwo ‘dark pixels’ – bolometers positioned outside the instrumentFoV – and two thermistors located on each array.Maximising the aperture e ffi ciency of the feedhorns requiresan aperture corresponding to an angle of 2 λ/ D on the sky,where λ is the wavelength and D is the telescope diameter.Consequently, the detector beams have an angular separation ofapproximately twice the FWHM beam size on the sky. As a re-sult, specific observing patterns – either jiggling or scanning – must be employed to achieve Nyquist sampling of the sky bright-ness distribution (Gri ffi n et al. 2002). Jiggling is achieved us-ing a dual axis internal beam steering mirror (BSM), while fullysampled scanned observations are obtained by scanning the tele-scope at the specific scanning angle of 42.4 ◦ with respect to shortaxis of the bolometer array (Sibthorpe et al. 2006; Waskett et al.2007).The bolometer signal time-lines are filtered on-board by a 5-Hz low-pass filter. After multiplexing, an appropriate 4-bit o ff setis removed from each data time-line before the final gain stage,after which the data are digitized and telemetered to the ground.The o ff set removal allows the data to be sampled with 20 bitaccuracy using a 16-bit ADC. The timelines are reconstructedto 20-bit accuracy on the ground using the telemtered o ff set anddigitized data time-lines. The photometer on-board signal chainis described in detail by Gri ffi n et al. (2008a).In Section 2 of this paper we describe the simulation soft-ware itself, and its operation. Details of the format of the simula-tor input and output data are also given. Section 3 presents someexamples of how the SPS has been used so far. This demon-strates the kind of outputs the SPS can produce, and the typesof investigations that it can be used to perform. The usage per-missions and availability of the SPS are explained in Section 3,and Section 4 describes the future development plans for the SPSpost-launch. Finally, Section 6 reviews the main conclusions ofthis paper.
2. The SPIRE photometer simulator
The SPIRE photometer simulator (SPS) is a software packagewhich simulates the operation of the SPIRE photometer instru-ment and associated spacecraft functions. The data are output astime-lines in engineering units, which corresponds to the inputdata level (Level-0) for the Herschel data pipeline. a r X i v : . [ a s t r o - ph . I M ] J un B. Sibthorpe et al.: SPS: A software simulator for the Herschel-SPIRE photometer
The SPS allows users to input their planned observations,along with representative maps of the sky (either synthetic orderived from existing data), and be returned data which can bereduced and analysed using the o ffi cial Herschel data pipeline,or any other suitable data reduction software the user wishes touse.It is important to note that the SPS should not be used asan observation planning tool as it does not take into accountinstrument and spacecraft overheads. All observation planningshould be carried out using the Herschel observation planningtool HSpot (Herschel-Spot Users’ Guide 2007). The simulateddata are intended to be a faithful representation of what in-flightdata will be like, but while every e ff ort has been made to makethe data realistic, it is inevitable that as yet unknown instrumentsystematics will be present in the post-launch data which are notcurrently present in the simulated data.The SPS is controlled via a graphical user interface (GUI).Observations are set up using a window which contains the sameparameters as those in HSpot. A simulated observation can thenbe performed using any of the SPIRE astronomical observingmodes, including parallel mode (only SPIRE output is returnedby a parallel mode simulation). The SPS is a modular software package written in the InteractiveData Language (IDL), with each module being controlled by acore program. This design was chosen to allow various modulesto be developed independently based on a defined set of moduleinput and output parameters. With the exception of the core pro-gram, each module contains four di ff erent phases of operation:an input , initialisation , temporal , and afterrun phase. The core module contains only the input phase operation.The input phase loads all of the user defined inputs and storesthem in the appropriate data structures in preparation for thesimulation run. The initialisation phase uses these input datato compute a variety of standard parameters, including opticaltransmissions, detector array layout etc. These are parameterswhich will not change throughout the simulations, and thus canbe computed at the start of the simulation. Parameters whosevalues might change as a result of feedback from other systemcomponents are computed in the step phase. Most commonlythese will be sets of parameters which are mutually dependent.In most cases the majority of a module’s operation can be per-formed in the initalisation phase, with the step phase allowingfor flexibility in the system for future development. Finally, the afterrun phase is used to compute any parameters which requireentire data time-lines to be complete before their operation canbe performed. For example, the 1 / f noise in the SPS is com-puted using Fourier methods, and cannot be computed in a stepby step method. Likewise the simulated on-board filtering of thedata must be done on entire data time-lines.The processing sequence of the SPS is shown in Figure 1 as aflow diagram, with each phase of operation being displayed. Thisdiagram also shows how multiple simulation runs are performed. The SPS contains 11 modules, with each module representingthe operation of a specific system component. For example, the detectors module contains all code associated with simulatingthe detectors, including the generation of noise contributionsarising from the detector itself, and the associated load resistor.
Begin simulation Assess progress of stepping phase Assess completion status of simulation set Exit simulator Input module parameters and set up structures for the given Perform non-time step (static) module operations and initialise modules for time step Perform module time step processes Perform module after run processes e.g. filtering Format and output data to a specified file BEGIN run = 0INPUT PHASE
INITIALISATION PHASETEMPROAL PHASEAFTERRUN PHASESIMULATION COMPLETEt = t total run = run total run = run + 1 t = t + 1
DATA OUTPUT
Fig. 1.
Block diagram showing the operational architecture im-plemented in the SPS.The modules are named according to the subsystem or functionthat they simulate and are described in detail below.
Core
This module does not represent any physical part of theHerschel-SPIRE system. Instead it manages and runs all oper-ations specific to operating the simulation software, such as call-ing modules in the correct order, performing the required num-ber of simulations, and setting the simulation output file names.It also controls the time step resolution used within the simula-tor, which directly relates to the simulation run time.The nominal time step used in the SPS is 21 ms, whichequates to 48 Hz, approximately 3 times the nominal instrumentdetector sampling rate. The internal SPS clock must run fasterthan the instrument sampling rate to account for short time-scaletransient systematics, such as high frequency noise fluctuations.The 3:1 ratio of instrument to internal clock sampling rate wasfound to give provide an accurate simulated output, with no im-provement found when operating at higher ratios.
Sky
Like the core , this module does not correspond to a particularinstrument system. Instead it represents the astronomical skyviewable by the telescope. The module reads in three user de-fined input sky files, one for each waveband in a Flexible ImageTransport System (FITS) file format. The units of the input mapsare Jy / pixel, and the pixels must have the same integer arcsec-ond size for all three bands. It is recommended that a pixel sizeof 2 arcseconds is used ( ∼ /
10 of the 250 µ m beam). Pixel sizeslarger than this can result in digitisation artefacts from the in-put map being present in the output data. The source spectrumin each pixel is treated uniform across the SPIRE band, there-fore any spectral slope information must be included in the inputmodel. This input will, however, be multiplied with the SPIREfilter response function. . Sibthorpe et al.: SPS: A software simulator for the Herschel-SPIRE photometer 3 The astrometric data associated with the input images arealso read and stored for use in the observatory function module.The astrometic information contained within the FITS headermust therefore be correct and contain the CDELT, CRPIX,CTYPE, CRVAL, CROT and EQUINOX keywords.
Optics
This module represents the main optical properties and param-eters of the telescope and the SPIRE photometer (including thefilters), and the positional mapping of the detectors on the sky.These parameters are used to derive the optical transmission andemissivity profiles for each optical element, as well as the finalinstrument spectral response function in each band. This spec-tral response function is then applied to the input sky model as-suming a flat source spectrum across the SPIRE bands. The truesource spectrum must be taken in to account in the generation ofthe input sky model (see Section 2.2.2).The telescope beam is also taken into account in this module,and is used to produced a beam convolved version of the user’sinput map. A single indentical beam model is used for each de-tector within a single array, and three beam models are used intotal, one for each band. The beams used are based on opticalmodels and include the telescope side-lobes. The beam model iscontained within a library file and loaded from memory when re-quired. Each beam is symmetric and is contained within a squaregrid of size ∼ × The observatory function module specifies the photometer ob-serving mode to be simulated in terms of the appropriate ob-serving function and its associated parameters. Observing pa-rameters are supplied via either a simple HSpot style interface,or an advanced interface option. The HSpot option is the nom-inal input method, and provides input parameters akin to thosefound in the HSpot software. This does not mean however thatthe software can read in HSpot output files, rather the input pa-rameters are defined in the same way as those in HSpot. The ad-vanced screen allows the user to select di ff erent options in orderto investigate the e ff ect of di ff erent observing strategies whichare not normally permitted by Herschel.The user input information supplied is then used to gener-ate both the telescope pointing information, including pointingerrors, as well as the input parameters for the BSM module.
BSM
This module simulates the movement of the BSM. It generatespointing time-lines in terms of the fixed array spacecraft axes.These time-lines are generated originally in arcseconds, and thenconverted to analog-to-digital units (ADU) via a look-up table.Both outputs are provided to the user.
Thermal
This module simulates all information pertaining to the tempera-tures of the instrument and the telescope, and their temporal fluc-tuations. However, since the instrument system, and telescopecan be considered to be stable on time scales of a single obser-vation, the model used in the SPS assumes that the only sourceof thermal fluctuations is the He cooler. The cooler fluctuationsare simulated using a simple 1 / f α noise time-line generator, with a user specified noise spectral density, knee frequency, and spec-tral index ( α ).The cooler fluctuation time-line is low-pass filtered assum-ing an RC filter profile. This represents the suppression of highfrequency thermal variations by the various thermal linkages be-tween the cooler tip and the array, and the thermal mass of thedetector arrays themselves. A di ff erent user defined time con-stant is used for each array, representing the di ff erent distancesand linkages between the cooler tip and each specific array. Thisresults in the three arrays having a similar, although not identicalthermal fluctuation time-line. A demonstration of the impact ofthermal drifts on the output timeline can be seen in Figure 2(a),along with the reference astronomical power timeline.At present there are no thermal gradients within a bolometerarray, meaning that all bolometers vary in temperature simulta-neously. These thermal variations represent the dominant sourceof common mode 1 / f noise in SPIRE, and as with all systemat-ics simulated, can be turned on and o ff depending on the usersneeds. Astronomical Power
This module uses both the telescope and BSM pointing infor-mation to generate absorbed power time-lines for each detector.The observing strategy is projected onto the beam convolved in-put sky, and the flux from each pointing is registered. Additionaltelescope e ffi ciency factors and filter transmission profiles arethen used to compute the astronomical power falling on a givenbolometer during each simulation time step. A sample outputtimeline is given by the dashed line in Figure 2. Background Power
The background power falling on each bolometer throughouta simulation is generated in this module. Each element in theoptical chain is assumed to emit as a blackbody, with a wave-length dependent emissivity of (1 − t ), where t is the opticaltransmission. The primary mirror is modelled with a user de-fined thermal gradient (with temperature decreasing with dis-tance from the spacecraft’s sun shield). The background powerof the Herschel / SPIRE system is expected to be stable on theperiod of a single observation, therefore no background powervariations are implemented. This will be reassessed once post-launch data are available.
Detectors
This module produces an output voltage time-line for eachdetector channel based on inputs from the
Astronomical and
Background power modules. The bolometer thermal model pre-sented by Sudiwala et al. (1992) and Woodcraft et al. (2008) isused to calculate the correct voltage output based on the powerabsorbed by the detector. The thermal model provides an accu-rate characterisation of non-linear bolometer response to highlevels of incident power, and detector output fluctuations due tovariations in the bath temperature. Each detector is modelled in-dependently using its own set of bolometer parameters measuredduring instrument testing.In addition to the detector voltage time-lines, the theoreticalphoton, Johnson, phonon and amplifier noise contributions areall calculated here. These data are then used to generate uniquewhite and / or 1 / f noise time-lines, which are in turn summed B. Sibthorpe et al.: SPS: A software simulator for the Herschel-SPIRE photometer
Fig. 2.
Sample data timeline for the central bolometer in the250 µ m array under various noise conditions: (a) contains onlythermal drifts (correlated 1 / f noise) with no other noise e ff ects;(b) shows the timeline with uncorrelated 1 / f and white noise ap-plied; (c) contains both of the previous two outputs combined,and (d) shows the timeline from (c) once it has passed throughthe read-out electronics, i.e. low-pass filtered, sampled, and digi-tised. All signals have been calibrated to Jy / beam from their na-tive unit to allow direct comparison for direct comparison of thedi ff erent noise e ff ect. The original noiseless astronomical powertimeline is also given in each panel by the dashed line.with the detector output to produce the final detector voltagetime-line (see Figure 2(b)). Sampling
The sampling module simulates the processing of the outputdetector time-lines by the on-board read-out electronics. Eachdetector time-line is filtered by the on-board 5-Hz low-pass fil-ter, digitized, and output at the requested sampling frequency. Avoltage o ff set is also removed from the time-lines, and output asa separate single value for each detector.Sample input and output time-lines to this module are givenin Figures 2(c) and (d) respectively. The decreased noise levelresulting from the low-pass filtering, along with the digitisationof the signal are evident in these panels. Housekeeping
The housekeeping module outputs various instrument monitor-ing information. These data include flags which identify the ob-serving mode being used, and the building blocks used to makeup that mode.
The SPS output is a standard FITS file with 11 extensions.Each extension contains the data structure from one of the SPSmodules. All input parameters are contained within the outputfile, however for file size reasons not all derived time-lines arestored as standard. Nominally, only the time-lines returned bythe physical system are output, i.e. telescope pointing and digi-tised bolometer outputs. It is possible, however, to choose to out-put the astronomical power, background power, noiseless volt-age, and noisy voltage time-lines if requested. These time-linesare sampled at a higher frequency, and hence can significantlyincrease the output file size.The standard FITS output can be loaded into the o ffi cialHerschel pipeline processing software via a separate piece ofconverter software, whereupon it can then be reduced in thesame way as real SPIRE data. Alternatively the standard outputcan be converted from digitized units (level-0) to calibrated data(level-1) time-lines using a conversion routine supplied with theSPS. This new file can then either be loaded into the Herschelpipeline, or read in using another piece of custom software. The SPS will run on any modern computer (CPU ∼ ∼ ∼
635 MB of memory during the simulation,and generate a ∼
44 MB output file. The memory required duringa simulation, and the size of the output file are both dependent onthe size of the input file, and the selection of output parameters.In an optimal configuration for specific investigations the sys-tem has been shown to run up to 10 times faster than real time.Simulated observations can be performed for the full 18 hoursavailable in a single Herschel astronomical observing request.A lack of detail in the spacecraft pointing model means thatSPS cannot be used to calculate observing times. Many of therequired overheads such as slewing and pointing calibrations arenot included in the SPS. The pointing model used has been de-signed to provide realistic data for on-source regions only.The SPS contains all of the major known instrument sys-tematics, and represents the best current estimate of the SPIREphotometer’s in-flight performance.
3. Uses of the SPS
The SPS has been used for a variety of purposes throughout itsdevelopment. It was used to optimise and characterise the SPIREphotometer astronomical observing modes, and identify the bestobserving strategy possible with SPIRE (Sibthorpe et al. 2006;Waskett et al. 2007; Sibthorpe et al. 2008a). It has also pro-vided input data for a variety of map-making algorithms duringa map-making selection exercise. The di ff erent algorithms werecompared, and the best chosen as the standard map-making al- . Sibthorpe et al.: SPS: A software simulator for the Herschel-SPIRE photometer 5 gorithm for use in the Herschel pipeline (Chanial, submitted toPASP)In addition to mission planning, the SPS is currently be-ing used in the preparation and planning of various Herschelkey programmes, both in guaranteed and open time. These pro-grammes span a very wide range of targets, from high redshiftgalaxy surveys through to observations of di ff use galactic struc-ture and star forming regions. Investigation of saturation in sur-veys containing high dynamic range, uniformity of noise in out-put maps, and development and testing of specific algorithms foruse with survey data are all examples of recent uses of the SPS.An example output produced during planning for one of thehigh redshift galaxy surveys is presented in Figure 3. The inputto this simulation is a map derived from galaxy catalogues whichwere scaled to the appropriate flux densities. Three input imageswere created, one for each band, but only the 250 µ m band im-age is shown here. Two orthogonal simulated ‘large map’ scanswere performed with 1 / f noise included. The time-line data fromthese two observations were then made into a na¨ıve map and co-added to produce the final image seen in Figure 3a. This typeof image can be used to investigate the kind of problems thistype of survey might have with source extraction, P(D) analysis,and various other data analysis techniques. A coverage map, alsooutput by the map-maker (Figure 3b), provides additional infor-mation on the uniformity of integration time, and hence depthacross the map. With these two maps the noise statistics of thedata can be investigated, and any biases characterised. With aknown input sky model, methods which attempt to remove thevarious sources of noise can also be tested and compared.An example of a ‘small map’ observation is given inFigure 4. Here the observation of a debris disc, whose brightnessdistribution is based on that measured for the Epsilon Eridanidisc (Greaves et al. 1998), has been simulated without instru-mental noise. The input sky model is shown in Figure 4a, andconsists of a simple torus shaped object placed on top of a real-istic extra-galactic background. The goal of this simulation wasto provide a way to study the influence of extra-galactic confu-sion on the derived properties and structure of the debris discwhen performing a chopped SPIRE observation. This investi-gation found that a ‘small map’ observation results in a ∼ ff -source field. Consequently, methods to iden-tify and remove the contaminating sources are significantly moredi ffi cult to employ.The information from these types of simulations allows fora better informed choice of observing mode. A mode can be se-lected which fits the type of source, and confusion noise environ-ment. Simulations can also be used to develop, test, and charac-terise a series of analysis routines so that they are ready once thedata are taken.
4. Usage and availability
The SPS software is publicly available from the SPS website( ), along with the required conver-sion software which enables the SPS output to be used withHIPE. All relevant information on how to obtain and use the SPScan be found at on the SPS website, along with further details on (a)(b)
Fig. 3.
Simulated ‘large map’ observation of a high redshiftextra-galactic field. (a) This image is from the 250 µ m band, andshows an image of size 45 arcmin ×
45 arcmin. A na¨ıve map-maker was used to construct the image from the output time-line data. (b) coverage map showing uniformity of the observingtime.the models used within the SPS. This site will also contain in-formation on updates to the SPS as they are released . The SPS may be used free of charge only for non-commercial re-search purposes. If you make modifications to this software, you mustclearly mark the software as having been changed and you must also re-tain the copyright and disclaimer. In the interests of avoiding the prop-agation of modified versions of the SPS, we would request that you donot edit or redistribute this software without first contacting the author,or a member of the SPS team. All uses of this software must acknowl-edge it’s use, and reference this paper. B. Sibthorpe et al.: SPS: A software simulator for the Herschel-SPIRE photometer(a) (b)
Fig. 4.
Noiseless simulated ‘small map’ observation of a debrisdisc – (a) input sky model, (b) output map.
5. Future development
Following the launch of Herschel a new version of the SPS willbe released. This version will be updated to include in-flight per-formance data, and will aim to replicate the true characteristicsof the operating system. Feedback from users is also expected toguide the development of additional features in the SPS.
6. Conclusions
In this paper we have described the SPIRE photometer simulatorsoftware and given examples of its capabilities and outputs. TheSPS mimics the output of the SPIRE photometer such that it canbe loaded into the Herschel pipeline environment, and reducedas if it were real data. The software represents the primary in-strumental characteristics, and allows the user to investigate howthese characteristics might influence the quality of astronomicaldata for individual science cases. It also provides an opportunityto prepare, and characterise, science case specific data analysisroutines.We have presented several examples of preparation work car-ried out with the SPS, and provided information on how newusers can access and use the SPS in preparation for their ownHerschel observations. The SPIRE photometer simulator is aversatile tool for the investigation of a wide range of instru-mental influences on astronomical data obtained with Herschel-SPIRE.
Acknowledgements.
The authors wish to acknowledge significant contribu-tions to the SPIRE simulator project from Pierr´e Chanial, Trevor Fulton, Ren´eGastaud, Michael Pohlen, Emma Rigby, Richard Savage, Rupert Ward, TimWaskett, Steven Watkin, and Adam Woodcraft. We also acknowledge the Scienceand Technology Facilities Council for post-doctoral funding of this work.
References
Greaves, J. S., Holland, W. S., Moriarty-Schieven, G., et al. 1998, ApJ, 506,L133Gri ffi n, M., Dowell, C. D., Lim, T., et al. 2008a, in Society of Photo-OpticalInstrumentation Engineers (SPIE) Conference Series, Vol. 7010, Society ofPhoto-Optical Instrumentation Engineers (SPIE) Conference SeriesGri ffi n, M., Swinyard, B., Vigroux, L., et al. 2008b, in Society of Photo-OpticalInstrumentation Engineers (SPIE) Conference Series, Vol. 7010, Society ofPhoto-Optical Instrumentation Engineers (SPIE) Conference SeriesGri ffi n, M. J., Bock, J. J., & Gear, W. K. 2002, Applied Optics, 41, 6543Herschel-Spot Users’ Guide. 2007, Herschel-Spot (HSpot)Users’ Guide: Herschel Observation Planning Tool,http: // herschel.esac.esa.int / Docs / HSPOT / html / hspot om.htmlLindner, J. V., Naylor, D. A., & Swinyard, B. M. 2004, in Proc. of the SPIE,Volume 5487, pp. 469-480 (2004)., ed. J. C. Mather, Vol. 5487, 469–480 Pilbratt, G. L. 2008, in Society of Photo-Optical Instrumentation Engineers(SPIE) Conference Series, Vol. 7010, Society of Photo-OpticalInstrumentation Engineers (SPIE) Conference SeriesSibthorpe, B., Chanial, P., Waskett, T. J., & Gri ffi n, M. J. 2008a, MNRAS, 388,1787Sibthorpe, B., Phillips, N., Holland, W., & W., D. 2008b, in New Light on YoungStars: Spitzer’s View of Circumstellar Disks, 5th Spitzer ConferenceSibthorpe, B., Waskett, T. J., & Gri ffi n, M. J. 2006, in Presented at theSociety of Photo-Optical Instrumentation Engineers (SPIE) Conference, Vol.6270, Society of Photo-Optical Instrumentation Engineers (SPIE) ConferenceSeriesSudiwala, R. V., Gri ffi n, M. J., & Woodcraft, A. L. 1992, International journal ofInfrared and Millimeter WavesWaskett, T. J., Sibthorpe, B., Gri ffiffi