aa r X i v : . [ a s t r o - ph . I M ] M a r Astronomy&Astrophysicsmanuscript no. final c (cid:13)
ESO 2018November 17, 2018
An integrated optics beam combiner for the second generationVLTI instruments
M. Benisty , J-P. Berger , L. Jocou , P. Labeye , F. Malbet , K. Perraut and P. Kern Laboratoire d’AstrOphysique de Grenoble (LAOG), 414 rue de la piscine, 38400 St Martin d’Heres, France CEA-LETI, Minatec, 17 rue des martyrs, 38054 Grenoble, FranceReceived 06 / / / / ABSTRACT
Context.
Recentely, an increasing number of scientific publications making use of images obtained with near-infrared long-baselineinterferometry have been produced. The technique has reached, at last, a technical maturity level that opens new avenues for numer-ous astrophysical topics requiring milli-arc-second model-independent imaging. The Very Large Telescope Interferometer (VLTI)will soon be equipped with instruments able to combine between four and six telescopes.
Aims.
In the framework of the VLTI second generation instruments Gravity and VSI, we propose a new beam combining conceptusing integrated optics (IO) technologies with a novel ABCD-like fringe encoding scheme. Our goal is to demonstrate that IO-basedcombinations bring considerable advantages in terms of instrumental design and performance. We therefore aim at giving a full char-acterization of an IO beam combiner in order to establish its performance and check its compliance with the specifications of animaging instrument.
Methods.
For this purpose, prototype IO beam combiners have been manufactured and laboratory measurements were made in theH band with a dedicated testbed, simulating a four-telescope interferometer. We studied the beam combiners through the analysis ofthroughput, instrumental visibilities, phases and closure phases in wide band as well as with spectral dispersion. Study of the polar-ization properties was also carried out.
Results.
We obtain competitive throughput (65%), high and stable instrumental contrasts (from 80% in wide band up to 100% ± e.g. ◦ ± ◦ ) which we attribute to internal optical path di ff erences(OPD) that can be calibrated. We validate a new static and an achromatic phase shifting IO function close to the nominal 90 ◦ value( e.g. ◦ ± ◦ ). All these observables show limited chromaticity over the H band range. Conclusions.
Our results demonstrate that such ABCD-like beam combiners are particularly well suited for interferometric combi-nation of multiple beams to achieve aperture synthesis imaging. This opens the way to extending this technique to all near infraredwavelengths and in particular, the K band.
Key words. optical interferometry – integrated optics
1. Introduction
Optical long baseline interferometry o ff ers a unique way to di-rectly probe astrophysical environments with milli-arcsecondresolution. The study of stellar surfaces, evolved stars, youngstars, our galactic center and the heart of active galactic nu-clei require access to direct imaging. Until now, a large frac-tion of observations in the near infrared (NIR) were obtainedwith 2 to 3-telescope arrays, with little spatial frequency cov-erage (so-called uv coverage), restricting the astrophysical in-terpretation to a parametric one in most of the cases. However,discriminating between di ff erent successful scenarios of com-plex or rapidly-changing objects raises the need for imagesas model-independent as possible. This translates into the re-quirement to use as many telescopes as possible in order tofill the uv plane and allow an unambiguous image reconstruc-tion. Until very recently, most of the images produced withoptical long baseline interferometers had moderate complex-ity and therefore did not bring additional information with re-spect to parametric modelling. In our opinion, the complex-ity barrier where the reconstructed image adds meaningful sci-entific value to the astrophysical interpretation was recently Send o ff print requests to : [email protected] passed by Monnier et al. (2007) and Zhao et al. (2008) using theMIRC instrument, an image plane 4-beam combiner using sin-gle mode fibers, at the CHARA interferometer (Monnier et al.2006b; ten Brummelaar et al. 2005).The Very Large Telescope Interferometer (VLTI, (Sch¨oller2007; Haguenauer et al. 2008)) will be equipped in 2012-2015 with two second-generation instruments: Gravity(Eisenhauer et al. 2008) and VSI (Malbet et al. 2006, 2008)which will be capable of exploiting the imaging capability ofthe array by combining four beams for the first and six for thesecond. The stringent requirements for these two instrumentshave triggered the interest in using integrated optics (IO)as a core technology for the beam combining function. Theability to integrate a singlemode circuit on a substrate, ableto interfere all the beams o ff ers numerous advantages both interms of performance and ease of operation. Single-mode beamcombiners provide natural modal filtering, which associatedwith proper photometric calibration has been shown to lead toaccurate visibility measurements. The compactness of the chipallows the instrument footprint to be minimized and the thermalcontrol to be optimized (further enhancing the calibrationaccuracy). No alignment is required, other than the injectionin the input guides, even though the combination scheme is Benisty et al.: A 4-beam IO beam combiner with ABCD encoding
Fig. 1.
Upper panel: theoretical design of the integrated optics 4-way beam combiner allowing pairwise combination and usingphase-shifting devices to produce 4 outputs in quadrature. We refer to each output using the index m,l,k . ml k is the k th output out of4, resulting from the combination of the beam m and l . The lower panel is a picture of a prototype that is 80mm long and 8mm wide.complex. Finally, this technology o ff ers the flexibility to easilyswitch beam combiners to adapt to a particular situation ( e.g. target, number of telescopes).Since the initial proposition by Kern et al. (1996), LAOGand its industrial partner LETI / CEA have been developing theuse of IO technology to interferometrically combine light beamsin optical waveguides lying on a solid substrate of a few cen-timeter (Kern et al. 1996; Malbet et al. 1999; Berger et al. 2000).This instrumental research program has consisted in designing,fabricating and characterizing all the IO building blocks requiredto build an astronomical interferometric beam combiner. Severalbeam combining schemes have been implemented and tested.Some of them have led to successful on-sky demonstrations suchas the VINCI / VLTI (2 telescopes) and IONIC3 / IOTA (3 tele-scopes) instruments (Berger et al. 2003; LeBouquin et al. 2004;Kraus et al. 2005; Monnier et al. 2006a).In the context of VLTI second-generation instrument studies,LeBouquin (2005) have studied the global e ffi ciency of a greatvariety of IO beam combiners. This study has concluded that oneof the most e ffi cient ways to combine four beams ( e.g.
2. The beam combiner: technology and design
Prior to fabrication, the IO circuit was designed and numeri-cal computation simulating the propagation of an electromag-netic signal was carried out to determine the expected proper-ties in terms of flux routing. Each IO function was checked andits throughput and flux distribution were optimized numerically.
Fig. 2.
Details of the beam combining function: for each inter-ferometric pair ( e.g. [12]), one arm is shifted by 90 ◦ leading tofour outputs in quadrature (with phases written as ϕ to ϕ ).Combinations of beams occur in couplers that present two out-puts in phase opposition to maintain energy conservation. Byrecording the four phase states (ABCD-like, see the right fig-ure), one can retrieve the interferometric observables (amplitudeand phase of the fringes).This step done, the simulation parameters were turned into tech-nological parameters and a photolithographic mask was fabri-cated.LETI uses a silica-on-silicon technology to fabricate IO cir-cuits. This technological process requires several photolitho-graphic steps to etch di ff erent layers. The beam combiners aremade by depositing alternatively 3 doped silica layers on asilicon substrate. The second layer is etched to define chan-nel waveguides and the other two layers constitute the opticalcladding. For the first time, the etching technology allows us tocompletely isolate each waveguide from the others (Labeye et al.2006). The produced beam combiners have been designed to op-erate in the atmospheric H band and more recently in the K band.The so-called “pairwise static ABCD” beam combiner canbe described as follows. Each beam combiner is designed tohave 4 inputs and 24 outputs, allowing 6 interferometric pair-wise combinations, each one producing 4 phase-shifted outputswith a phase di ff erence of 90 ◦ . For each injected beam, the lightpropagates through waveguides and is split in three in a tricou-pler (item (a) in Fig. 1) to enter the combining function (consti- enisty et al.: A 4-beam IO beam combiner with ABCD encoding 3 tuted of Y-junctions and couplers). The light is then divided intwo in a Y junction that acts like a classical beamsplitter (itemb), each beam later being combined in a coupler (item c) with abeam coming from another telescope. A coupler allows a con-trolled power transfer between one waveguide and another. As aconsequence of energy conservation, each coupler has two out-puts in phase opposition. In only one of these four arms, thereis a phase-shifting device designed to change the phase of thepropagating beam by 90 ◦ (Fig. 2). This leads to four outputbeams, two of them being in phase opposition with an additionalphase-shift of 90 ◦ with respect to the other two ( e.g. ϕ and ϕ + π ; ϕ + π/ ϕ + π + π/
2, following Fig. 2). The phase-shifting function is based on the variation of the e ff ective index( i.e. index seen by the fundamental mode propagating into thewaveguide) with the waveguide diameter (Labeye 2008). To cre-ate a phase shift, enlarging one of the two waveguides createsa di ff erence in the e ff ective index and leads to an optical pathdi ff erence between two parallel waveguides of the same phys-ical length. In order to achieve an achromatic phase shift, thewavelength dependence is compensated by concatenating a fewwaveguide segments of di ff erent diameters separated by tapers( i.e. adiabatic functions) to avoid any loss due to discontinuities(Fig. 3). Since the photometry is extracted from a linear com-bination of the interferometric signal itself, the beam combinershave no dedicated photometric channels. This allows us to e ffi -ciently use all photons for the interferometric combinations. Bydesign, each interferometric pair simultaneously gives access tofour phase states in quadrature (ABCD-like but without tempo-ral modulation). These 4 measurements allow the visibility am-plitude and phase to be retrieved using the ABCD method de-scribed in Colavita (1999). In practice, the departure from idealquadrature forbids the use of simple algorithms and leads us toconsider a generalized algorithm capable of handling a realisticdescription of the beam combiner properties.Throughout the paper, the outputs are identified with the in-dex m , l , k , such as ml k , where m , l are the interfering beams, and k = [1..4], the output for this combination (similarly, the A-C-B-D measurements of Fig. 2). The same nomenclature applies tofunctions. In the case of Y-junctions, we denote them using theindex m , l to specify the beam combination to which they are re-lated, with m corresponding to the actual beam that enters theY-junction. The index k designates its two outputs. For exam-ple, Y and Y are the two outputs of the Y-junction that splitsbeam 1 in signals that will interfere with beam 2. Similarly, Y and Y are the outputs of the Y-junction that splits beam 2 intosignals that will combine with beam 1. We use the same notationfor the couplers, e.g. C and C are the outputs of the couplercorresponding to the combination of beam 2 and 4.With this notation, the intensity recorded at the outputs ofthe combination of beams m , l can be written : i kml = N m t kml + N l t klm + V ob jml V kml q N m t kml N l t klm cos ( ϕ kml + ϕ pml + ϕ ob jml )(1)where N m is the number of photons in the m beam and t kml thetotal transmission of the k output for the ml beam pair. V kml is theinstrumental contrast, ϕ kml is the instrumental phase introducedby the IO beam combiner between the two interfering beams. ϕ pml is the residual atmospheric phase due to piston e ff ects. V ob jml is the object visibility and ϕ ob jml is its phase. Fig. 3.
Principle of the phase-shifting function: a variation of theoptical path is induced by a di ff erential change of the waveguidee ff ective index due to a change in their width. The concatenationof carefully optimized portions of waveguides with controllede ff ective index allows to flatten the wavelength response.
3. Laboratory set up
In this section, we present the aims of the experiments, ourtestbed as well as our operating mode for the data acquisitionand processing.
With such beam combiners, all the information about the coher-ence of the object is included in the way the 4 pixels are relatedto each other, including the instrumental contribution. This con-tribution has therefore to be known, i.e. fully calibrated.The relationship between the measured fluxes on the pix-els and the visibility amplitudes and phases of the object canbe expressed with a matrix representing the behavior of the in-strument. With an unresolved internal source ( i.e. V ob jml = ϕ ob jml =
0) and without piston ( i.e. centered at zero OPD), Eq. (1)becomes : i kml = N m t kml + N l t klm + V kml p N m N l X kml (2)with X kml = q t kml t klm cos ( ϕ kml ), a coe ffi cient di ff erent from 1, thatcorresponds to the level at which the beam combiner conservesthe coherence and that depends on ϕ kml , the internal IO phasespecific to the output ml k . If one isolates a combination cell [ml] ,the relation between the output intensity and input number ofphotons (Eq. (2)) can be written as : i ml i ml i ml i ml = t ml t lm V ml X ml ( ϕ ml ) t ml t lm V ml X ml ( ϕ ml ) t ml t lm V ml X ml ( ϕ ml ) t ml t lm V ml X ml ( ϕ ml ) ∗ N m N l √ N m N l (3) ml and ml correspond to two outputs of the same coupler (thesame is valid for outputs ml and ml ). Therefore, ideally, be-cause energy is conserved at the output of a coupler the follow-ing relations should apply: ϕ ml = ϕ ml + π and ϕ ml = ϕ ml + π .Similarly, the beam combiner is ideally designed to introduce aphase quadrature between the outputs therefore: ϕ ml = ϕ ml + π/ ϕ ml = ϕ ml + π/ ×
3] ma-trix with zero elsewhere. In reality, crossing terms appear bothas incoherent and coherent contributions, and the actual outgo-ing intensities should be described using a general matrix of24 ×
10 terms :
Benisty et al.: A 4-beam IO beam combiner with ABCD encoding t t V X : : : :: : : : t t V X ∗ N : N √ N N : √ N N (4)This matrix, called the V2PM (visibility to pixel matrix),in accordance to previous work on multiaxial interferometers(Tatulli & LeBouquin 2006), completely characterizes the in-strumental behavior of the beam combiner e.g. the transmission,the visibility, the phase relations and the parasite flux. The ul-timate goal of such a study would be to precisely estimate andcalibrate it (Lacour et al. 2008). However, it is out of the scope ofsuch a paper to present and discuss a full characterization of theglobal beam combiner matrix. We prefer to focus on character-izing the individual tricoupler functions, Y-junctions, couplersand phase-shifting devices described by Eq. (3) as well as theglobal routing of the incoherent flux (so-called crosstalk) insidethe beam combiner. As it will be seen later, the crosstalk terms(related to the crossing terms in Eq. (4)) are su ffi ciently small,which justifies this approach.To reach this goal, we set up a laboratory testbed and testedthe IO beam combiners through photometric and interferometricmeasurements to calibrate this instrumental matrix. The follow-ing quantities, as well as their dependence on wavelength, havebeen measured:1. the so called “normalized kappa matrix” κ kml = t kml P m P l m t kml ;2. instrumental contrast V kml ;3. instrumental phase shift between outputs, ϕ kml - ϕ k + ml , with ϕ kml being the individual phase of the output signal ml k ;4. instrumental closure phase Φ kml j = ϕ kml + ϕ kl j + ϕ kjm ; We designed a dedicated interferometric testbed capable of sim-ulating an 8 telescope interferometer (Jocou 2007). Figure 4 de-scribes individual functions of the setup. The bench includes var-ious items :(a) an object simulator that can reproduce a single star or abinary star with an adjustable flux ratio(b) up to 8 optical devices simulating telescopes and cou-pling the light into single-mode polarization-maintainingfibers(c) optical path compensation and modulation devices (delaylines of a few mm long)(d) an IO beam combiner(e) a spectrometer(f) a Wollaston prism to split the linear polarizations(g) an infrared detector.All laboratory tests were carried out in the H band with lightsources of di ff erent coherence lengths. The object simulator canreproduce a single star as well as a binary star. In the latter case,its design is based on an optical setup that mimics a Michelsoninterferometer, but with a tilted mirror in one of its arm and withan unbalanced pathlength between the two arms. It producestwo non-coherent luminous spots simulating a binary star, whoseseparation can be adjusted by tilting the mirrors. This setup willbe used to characterize the dynamics of the testbed. The image isplaced at the focal plane of a F / Fig. 4.
Schematic view of the laboratory testbed simulating theVLTI (see text for details).
Fig. 5.
A detector image of the 24 outputs, obtained when usingthe spectrograph and a Wollaston prism (splitting the two linearpolarizations P1 and P2). The patterns are due to various nonzero OPD for the di ff erent beam combinations.diameter collimated beam. The wavefront is sampled by up to 8telescopes that can be set to reproduce a replica of the VLTI en-trance pupil. These telescopes are made up of an F / ff er-ential e ff ects ( ∆ L = ff racting grating provides 15spectral channels through the H band, while the Wollaston prismsplits the linear polarization states to improve the instrumentaltransfer function. These two last elements, that can be placedand removed easily depending on the need, are located in anafocal mount with a magnification of 1. The detector is a near-infrared InGaAs PICNIC chip, with 40 µ m-large pixels. In thecase of wide band measurements, each waveguide output is im-aged on one single pixel. Figure 5 is a detector image obtainedwhen both the spectrograph and the Wollaston prism are used. Itshows the 24 beam combiner outputs spectrally dispersed alongthe vertical direction. P1 and P2 correspond to the two linearpolarization states. enisty et al.: A 4-beam IO beam combiner with ABCD encoding 5Step Sh1 Sh2 Sh3 Sh4 Measurement1 X X X X B g P P P P I Sh = shutter; O = open; X = closed Table 1.
Experimental protocol including photometric and in-terferometric measurements.
The sequence of acquisitions performed to characterize the beamcombiners consists of 6 steps that can be done with or withoutspectral dispersion (see Table 1). Step 1 is a background mea-surement with all shutters closed to prevent any light from prop-agating through the instrument. The 4 consecutive steps are mea-surements with only one beam at the time that give access tothe flux splitting ratios in the couplers and tricouplers. Finally,Step 6 is the interferometric combination of all input beams. Allmeasurements are repeated 1024 times.The first 5 steps are used to validate the design in terms ofphotometry (light routing, transmission and splitting ratios, un-desired flux, together with their wavelength dependence). Step 6leads to the determination of the value, stability and chromatic-ity of instrumental contrasts and closure phases as well as ofthe phase relations between phase-shifted outputs, supposedly inquadrature. To get a complete and independent laboratory char-acterization of each output of the tested beam combiners, the in-terferometric measurements presented in this paper are obtainedwith OPD modulation and polarization splitting on a point-likesource.
The data processing derives four quantities : the kappa matrix( i.e. the photometric contribution of each beam to the interfero-gram); the instrumental contrast; the phase-shift induced by thedevices and the closure phases.The kappa matrix is extracted from each individual set ofdata where only one input is illuminated (Table 1, Steps 2-5).The instrumental contrast is computed using a classical visi-bility estimator on the interferograms to evaluate the coherence.This consists of estimating the envelope amplitude and calibrat-ing for the photometric inbalance between the interfering beams.Since our experimental data are obtained with a SNR of ∼ ml and ml , i.e. outputs k = k = ϕ ml − ϕ ml = atan ℑ [ F ml ( ν ml , ) F ∗ ml ( ν ml , )] ℜ [ F ml ( ν ml , ) F ∗ ml ( ν ml , )] where ℑ , ℜ stand for the imaginary and real parts respectively.F kml ( ν ml , k ) (here with k = ,
3) is the intensity of the Fourier spec-trum of the signal ml k , taken at the maximum value (to which ν ml , k corresponds). The closure phase is measured using a triplet of telescopes i.e. from three pairwise combinations. It is calculated as thephase of the bispectrum, that is the complex product of the cor-responding three Fourier spectra. Since each beam combinationproduces 4 phase-shifted output signals, there are 4 closure-phase signals per telescope triangle. For the telescope triangle[mln], the closure-phase derived from the output k can be writ-ten as: Φ kml j = atan ℑ [ F kml ( ν ml , k ) F kl j ( ν l j , k ) F k ∗ m j ( ν m j , k )] ℜ [ F kml ( ν ml , k ) F kl j ( ν l j , k ) F k ∗ m j ( ν m j , k ) The closure phase is computed with the constraint that thefrequencies respect the closure relation ( e.g. ν m j , k = ν ml , k + ν l j , k ).Definitions of F kml ( ν ml , k ), F kl j ( ν l j , k ) and F km j ( ν m j , k ) are identical tothe phase-shift case.The methodology for data reduction with spectral dispersionis identical. From all illuminated pixels of the detector (in a casesuch as in Fig. 5), we measure interferograms from which wederive the chromatic behavior of instrumental quantities.
4. Results
In this section, we present the results of the laboratory experi-ments obtained on a point-like source, in terms of flux through-put and routing, instrumental contrasts, phase-shifts, and closurephases. The results correspond to two di ff erent IO chips manu-factured in the same wafer (called Chip1 and Chip2 in all tables)in broad band, as well as with spectral dispersion for the secondchip only.For the latter experiment, we report in all tables the aver-age values as well as the amplitude of the variation over thewavelength range ( i.e. | X max ,λ − X min ,λ | , for the instrumentalquantity X ). We refer to the latter as chromaticity in the text.Detailed studies that relate the performance to the IO designand simulations will be given in a following paper (Labeye etal. 2009, in prep.). All results are commented on in Sect. 5. Thenotations used in the tables and figures are the same as definedin Sect. 2. Transmission:
When injecting 100 photons in one input, thetransmission is the total number of photons detected at all out-puts. The overall transmission budget includes the coupling ef-ficiency from the telescope point spread function to the fiber, aswell as the propagation losses inside the fibers and the IO chiptransmission. The latter quantity is determined in broad band byusing a fiber at each input and at each output, and the measuredflux is normalized by a fiber-to-fiber transmission. The measure-ment gives about 65% in the H band for the transmission of theIO chip itself. The telescope-to-fiber coupling is ideally ≈ ≤ dB / km ),consequently, for the 2m fiber lengths that we are using, the cor-responding transmission is ≈ + combiner’ is therefore ≈ Flux routing and individual IO functions:
For the followingparagraphs, we use Step 1 to Step 5 (Table 1). Figure 6, left,gives the broad band photometric coe ffi cients for the 24 beamcombiner outputs obtained when shutters prevent three telescope Benisty et al.: A 4-beam IO beam combiner with ABCD encoding F F Day 1Day 30.060.080.10 F F F F
0 10 200.060.080.10 F F
0 10 20
Pixel κ c oe ff i c i en t κ c oe ff i c i en t F F F F F F F F F λ (nm) κ c oe ff i c i en t κ c oe ff i c i en t κ c oe ff i c i en t κ c oe ff i c i en t Fig. 6.
Left: kappa matrix photometric coe ffi cients obtained when the light is injected in one input at a time (F .. ) for Chip1. Thestars and squares show results obtained on two di ff erent days of experiments (called Day 1 and Day 3). Pixels from 1 to 24 are the24 outputs of the beam combiner, identified as 12 to 34 in Table 2. Right: variation of the kappa matrix photometric coe ffi cientswith wavelength over the H band range, measured on Chip2. The four panels correspond to the light injection in one input at thetime ( e.g. F for the injection in input 1). Inside each of them, among the 12 outputs illuminated, only 3 are plotted (in full, dashed,and dotted lines), corresponding to each combination cell. Table 2.
Kappa matrix photometric coe ffi cients obtained when the light is injected only in the 4th input (F ). The first two linescorrespond to the wide band experiments while the third gives the average value over wavelength, as well as the peak-to-valleyamplitude over the wavelength range (chromaticity). Values are divided by 10 . Bold numbers, preceded by a star, indicate theilluminated outputs. Output 12 ∗ ∗ ∗ Chip 1 0.2 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± / ∆ λ / / / / / / / / / /
11 621 /
344 926 /
357 624 / ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ Chip 1 1078 ±
10 800 ±
10 1055 ±
10 811 ±
10 1037 ± ±
10 10 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± / ∆ λ /
184 794 /
270 1032 /
150 820 /
171 1055 /
124 8 / / /
194 865 /
412 766 /
360 913 / beams from propagating through the chip. These coe ffi cients aredefined, for each pixel, by the ratio between the flux detected onone pixel and the sum of the flux on all outputs. As expected insuch a case, 12 out of 24 pixels are illuminated. Two measure-ments taken on 2 di ff erent days are compared (stars and squaredsymbols) showing very small time variability. Table 2 gives anexample of the averaged photometric coe ffi cients for all 24 out-puts, obtained when the light is injected in the input.With the same experiments using spectral dispersion, onecan derive their dependence on wavelength. Figure 6, right, givestheir variation with wavelength over the spectral range. Table 2also gives the chromaticity of photometric coe ffi cients obtained when injecting in the input.To be more general, Table 3 gives measurements in all cases, i.e. when the light is injected in all 4 inputs, one at a time. Weonly provide the values corresponding to the outputs presentingthe minimum and maximum chromaticity as well as the averagechromaticity over the 12 signals. Because there is an importantspread across the photometric coe ffi cients values (see Table 2),the chromaticity is given with respect to the coe ffi cient value ob-tained when averaging over the spectral band ( i.e. divided by thisvalue).From these coe ffi cients, in both wide band and spectrally dis-persed experiments, we determine the splitting ratio of the dif- enisty et al.: A 4-beam IO beam combiner with ABCD encoding 7 T T T T λ (nm) T r i c oup l e r s − S p li tt i ng r a t i o Y Y Y λ (nm) Y − S p li tt i ng r a t i o Y Y Y λ (nm) Y − S p li tt i ng r a t i o Fig. 7.
Variation with wavelength of the tricoupler splitting ratio (left) and of the Y-junctions (middle, right). Middle: three Y-junction splitting ratios corresponding to the illumination in the first input (these Y-junctions are related to the [12], [13] and [14]combinations); Right: three Y-junctions splitting ratios obtained when injecting the light in the second input. For clarity, only oneflux ratio coe ffi cient (among 2 or 3) is plotted for each function. The error for each spectral channel is estimated from the dispersionover 1024 measurements and is smaller than the symbol sizes. The theoretical values of 33% for the tricoupler, and 50% for theY-junctions, are given by the horizontal dashed line. Values for all functions are given in Table 4 and 6. Table 4.
Tricoupler splitting ratio measured in wide band. The dispersion (rms) over 1024 measurements is 0.1%. The two first linescorrespond to the wide band experiments while the third gives the average value and the variation amplitude over the wavelengthrange.
T T T T T Chip 1 33.7 - 33.3 - 32.8 33.2 - 32.6 - 34.1 34.5 - 35.0 - 30.3 34.2 - 35.9 - 29.8Chip 2 31.9 - 35.3 - 32.7 33.8 - 34.0 - 32.1 32.4 - 36.4 - 31.1 34.3 - 34.0 - 31.6avg /∆ λ / / / / / / / / / / / / Table 3.
Minimum, maximum and average of the photometriccoe ffi cients chromaticity among the 12 illuminated outputs fromeach injection. Injection in input
Table 5.
Minimum, maximum and average chromaticity of tri-coupler splitting ratio (among the three outputs of each tricou-pler). T is the tricoupler corresponding to input 1. Tricoupler Average Minimum MaximumT
8% 3% 12%T
13% 6% 19%T
13% 3% 19%T
9% 5% 14% ferent optical functions (tricouplers, Y-junctions and couplers),under the assumption that all functions are ideal ( i.e no photonloss; P i x i = x i a splitting ratio coe ffi cient). For thesake of clarity, for each Y-junction and coupler, only one valueout of the two splitting ratio coe ffi cients is given in the tables,since the second output is obviously its complementary to 100%. Tricouplers:
Table 4 gives such values for the 4 tricouplershowing flux splitting ratios close to 33% for both chips, in wideband. The best flux separation is 33.7 / / / / ± Y-junctions:
Table 6 gives the splitting ratio for the 12 Y-junctions. The values, measured in broad band, are close to 50%,with a 0.1%-dispersion over 1024 points. Variations are smallfrom one day to another (2.9% maximum).With spectral dispersion, the splitting ratio are similar to thebroad band measurements. Figure 7, middle and right, gives theirwavelength-dependence. For clarity, only extreme behaviors areshown in the figure, with the smallest (middle plot) and greatest(right plot) variations over the H band. Out of the 12 Y-junctions,10 show a maximum variation inferior to 9% over the spannedrange of the H band, while 2 show a strong variation of about26 and 28%. These two Y-junctions are the closest to the inputs(combination [23]). Over all the Y-junctions, the average chro-maticity is 9.2%.
Benisty et al.: A 4-beam IO beam combiner with ABCD encoding
Table 6.
Y-junction splitting ratio. The dispersion (rms) over 1024 measurements is 0.1%. The two first lines correspond to thewide band experiments while the third gives the average value and the variation amplitude over the wavelength range.
Y Y Y Y Y Y Y Y Y Y Y Y Y Chip 1 52.5 47.4 47.0 44.4 50.2 50.2 52.6 50.0 51.5 52.6 54.7 51.0Chip 2 48.7 45.7 52.2 49.8 51.2 50.5 54.1 50.4 53.0 53.2 52.2 50.3avg / ∆ λ / / / / / / / / / / / / C C C λ (nm) C oup l e r s − S p li tt i ng r a t i o C C C λ (nm) C oup l e r s − S p li tt i ng r a t i o Fig. 8.
Variation of coupler splitting ratio with wavelength. Only one of the flux ratio coe ffi cients is plotted, and extreme behaviorsare given : left, when the light is injected in the first input; right, for an injection in the second input. The theoretical value of 50%is given by the horizontal dashed line. Table 7.
Coupler splitting ratio. The dispersion (rms) over 1024 measurements is 0.2%. The two first lines correspond to the wideband experiments while the third gives the average value and the variation amplitude over the wavelength range.
C C C C C C C C C C C C C Chip 1 54.1 59.6 50.5 58.4 59.7 61.5 57.5 58.5 48.3 50.4 53.9 55.2Chip 2 52.0 58.9 55.2 55.7 58.3 56.3 56.3 51.2 52.6 55.0 55.0 55.4avg / ∆ λ / / / / / / / / / / / / Couplers:
Table 7 gives the flux splitting ratio given by the12 couplers, in broad band, showing asymmetric splitting, up to61.5% / ff erences are up to 1.7%, 2.7% and 4%, forthe tricouplers, Y-junctions and couplers, respectively, while forChip2, they are of 0.9%, 2.3%, and 2.6% for the same functions. Cross-talk:
On the pixels where one should not detect anyflux, the measured intensity gives the amount of undesired flux, i.e. cross-talk flux. It can be due to direct propagation into thesubstrate and to light leak at the X-junctions level, where waveg-uides are crossing. We estimate the cross-talk flux to be less than1.2% ± Instrumental contrasts :
Figure 9, left, shows an example ofthe interferograms obtained with Chip 1 in broad band. Fromthese interferograms, we derive the instrumental contrasts af-ter calibrating for the photometric inbalance between interferingbeams. Table 9 gives the measured values, showing high con-trast values. The minimum and maximum values are respectivelyof 95% ±
1% and 98% ±
1% for the first beam combiner, and of82% ±
6% and 94% ±
1% for the second one.These results show a maximum variation of about 5% over aday timescale and 10% from one day to another. The measuredcontrasts on the four phase shifted outputs are only slightly dif-ferent (on average 2%, up to 9%). The non-perfect linearity ofour detector can lead to a bias of up to 5% in visibility. In thecase of our experiments, this e ff ect could not be reproduced andcalibrated. Therefore, although the statistical errors can be very enisty et al.: A 4-beam IO beam combiner with ABCD encoding 9 OPD (mm) N o r m a li z ed f l u x λ (nm) C on t r a s t C on t r a s t C on t r a s t C on t r a s t C on t r a s t C on t r a s t Fig. 9.
Left: an interferogram obtained with temporal OPD modulation with wide-band measurements. Right: the variation ofinstrumental contrasts with wavelength, for all four outputs of each beam pair. 100% contrasts are given by the horizontal dashedline.
Table 8.
Cross-talk flux (in %) determined when the light is injected in the 4 outputs successively. The two first lines report resultsobtained in broad band while the third line corresponds to the spectral dispersion case. Error bars are estimated from the dispersion(rms).
Day Injection in 1 Injection in 2 Injection in 3 Injection in 4Chip 1 0.7 ± ± ± ± ± ± ± ± / ∆ λ / / / / Table 9.
Instrumental contrasts (in %) obtained with wide band experiments (first two lines) and with spectral dispersion (thirdline) with a statistical error of 1%.
Output 12 Chip 1 95 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± / ∆ λ / / / / / / / / / / / / Chip 1 97 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± / ∆ λ / / / / / / / / / / / / small ( ∼ ff ects the contrast val-ues.With spectral dispersion, the measured contrasts are very high(up to 100%) showing very small variations with wavelength(see Figure 9, right). The maximum and minimum chromatic-ities, among all 24 outputs, are of 4.6% and 0.8% respectively,with an average of 2.3%. Phase shifts:
Figure 10 shows 4 phase-shifted interferogramsin each panel, that correspond to the intensity of the 4 phase-shifted outputs for the interferometric couple [34]. These 4 out-puts are in di ff erent phase states, as it can be seen, and in thesespecific examples, in the left panel, the phase-shift is close to theexpected value of quadrature. Table 12.
Examples of phase-shifts (in degrees) measured in thetwo polarization states (P1, P2), for the 6 beam pairs. Statisticalerrors are of 1 ◦ . Beam Pair [12] [13] [14] [24] [23] [34]P1 62 87 34 67 77 88P2 55 80 37 62 70 77
Table 10 gives the values of the phase-shifts obtained for theoutputs designed to be in quadrature. For the first chip, on 5 ofthe 6 interferometric combinations, the phase shifts are close tothe quadrature (from 78 ◦ ± ◦ to 82 ◦ ± ◦ ). For the sixth phase- OPD (mm) N o r m a li z ed f l u x Combination [34]
OPD (mm) N o r m a li z ed f l u x Combination [14]
Fig. 10.
The phase-shifted interferograms recorded for the 4 outputs (full, dashed, dotted, dot-dashed lines) of the [34] beam pairare close to quadrature (left). On the contrary, the 4 outputs of the central combination [14] produce interferograms only slightlyphase-shifted ( ∼ ◦ ) (right). Beam pair [ml]:121314 242334 1560 1580 1600 1620 1640 1660 0 20 40 60 80 100 λ (nm) P ha s e − s h i ft s ( deg ) Triangle 134 Φ Φ Φ Φ λ (nm) C l o s u r e P ha s e ( deg ) Fig. 11.
Left: wavelength dependence of the phase-shifts for the outputs designed to be in quadrature, for all 6 pairwise combinations([ml]). Error bars on each spectral point are of 1 ◦ . The theoretical value of 90 ◦ is given by the horizontal dashed line. Right: variationof the closure phase over the H band, for the triangle [134]. The four symbols correspond to the four outputs in quadrature. Table 10.
Phase-shifts obtained with the wide band experiments (two first lines) with errors of 1.0 ◦ (the dispersion (rms) over halfa day); The third line gives the average values over the wavelength range. Phase-Shifts Φ ( ◦ ) Φ ( ◦ ) Φ ( ◦ ) Φ ( ◦ ) Φ ( ◦ ) Φ ( ◦ )Chip 1 79.0 81.5 26.1 79.4 81.6 77.9Chip 2 62.5 87.3 32.9 67.5 77.5 87.7avg / ∆ λ / / / / / / Table 11.
Closure-phase measurements for one independent triangle, for Chip 2 with spectral dispersion, as well as amplitude ofthe variation over the H band range. Statistical errors are of 2.5 ◦ . Triangle [134] Φ ( ◦ ) Φ ( ◦ ) Φ ( ◦ ) Φ ( ◦ )avg / ∆ λ / / / / Table 13.
Contrasts and phase-shifts obtained when the split-ting of polarizations is done before or after the combination, andwhen no splitting is achieved. Errors are statistical.
Pixel 12 < V > before 83.7 ± ± ± ± < V > after 79.6 ± ± ± ± < V > ∅ ± ± ± ± ∅ Phase-shifts 78.98 ± ± ± shifting function, corresponding to the central [14] combination,the measurement gives about 26 ◦ ± ◦ . The right panel of Fig. 10corresponds to such a combination. The second chip gives dif-ferent results, with phase shifts spanning a larger range of values( e.g ◦ ± ◦ ; 88 ◦ ± ◦ ). The function corresponding to the [14]combination still shows a much smaller value of about 33 ◦ ± ◦ .Over a timescale of half a day, measured phase shifts showvariations of 1 ◦ at most, and from one day to another, a maxi-mum variation of 3 ◦ .With spectral dispersion, the obtained values for the phaseshifts are similar. Figure 11 (left) and Table 10 show the wave-length dependence of the measured phase-shifts for the outputsexpected to be in quadrature. The maximum chromaticity is of13 ◦ (for the [23] combination, as for the Y-junctions) and theminimum is of 3.1 ◦ . Over all outputs, the average chromaticityis 5.6 ◦ . Closure phases:
Table 11 gives a set of independent instru-mental closure phases measured for the 4 phase shifted outputsof one telescope triangle, with spectral dispersion. The measuredclosure phases have non-zero values which means that the beamcombiner functions themselves contribute to the phase budget.These terms can result from additional OPD originating in smalllength di ff erences between waveguides or from the delay in read-ing the detector pixels. The phase relation between the vari-ous beam combiner outputs can be found again in the closure-phase values. In fact, the closure-phases di ff er by about 180 ◦ forcouplers outputs in phase opposition and similarly, the closurephases measured at outputs theoretically in quadrature are dif-ferent to the corresponding phase sum between the telescopes( i.e. ϕ ml + ϕ l j − ϕ m j ).Figure 11 (right) and Table 11 give the variation of the in-strumental closure phases with wavelength, in the case of onetriangle of telescopes ([134]). The minimum variation over thewavelength range is of 18 ◦ while the maximum is of 30 ◦ . Polarization:
The two linear polarization directions are per-pendicular to each other following the symmetry axis of the chipitself ( i.e. within the beam combiner plane (horizontal) and par-allel to the light wavefront (vertical)). Such orientations are de-fined at the time of manufacturing the chip, and were confirmedby laboratory measurements.In our study, disparities between measurements obtained onboth linear polarization states were noticed. Contrasts and phase-shifts can be up to, respectively, 13% and 10 ◦ di ff erent, meaningthat the two polarizations propagate di ff erently. Also, for onepolarization state, instrumental contrasts are higher than for thesecond one for all beam combiner outputs except for the onesrelated to the central phase shifting function ([14] combination).Similarly, phase shifts are closer to the quadrature for one po-larization than the other for all outputs except the central ones (Table 12). Therefore, with respect to polarization, one phaseshifting function has a di ff erent behavior than the 5 others.In order to identify possible problems with instrumental po-larization and to determine the best instrumental set-up to reducethe visibility loss due to birefringent fibers, a detailed study on aday timescale was done with three optical set-ups: the first onehad polarizers before injection into the fibers and beam com-bination ( i.e. before the telescope mounts); the second one hada Wollaston prism before imaging on the detector, after beamcombination; and the third one had no polarization splitting.We found that the instrumental contrasts can drop by 10 to15% when no polarization splitting is done (Table 13). Slightlybetter contrasts ( ∼ ff erence was found.
5. Discussion
The characterizations presented here shows that global proper-ties of the designed beam combiners are very satisfactory for afirst prototype. In this section, we discuss how departures fromthe ideal case might a ff ect the performance. Photometry:
The transmission of the beam combiners di-rectly impacts the instrument sensitivity and therefore, con-strains the limiting magnitude. For these longest and most com-plex IO beam combiners tested today, a 65% transmission is sat-isfactory. An improved technology has allowed to reduce thelosses with respect to previous beam combiners. For compar-ison, the IONIC3T / IOTA H band beam combiners that weremade with the first technology and had only 3 Y-junctions and3 couplers, presented a transmission of 60%. The gain comesessentially from improvement in propagation losses associatedwith a reduced beam combiner length.We show evidence for small crosstalk photon leaks that leadto unwanted flux in the combining cells. These potentially a ff ectthe photometry estimation and might introduce a small coher-ent contribution not revealed in this study. However the mea-sured values for both chips are mostly inferior to 1% of the totalflux and have contributions smaller than the typical error barsof the measurements. The origin of this e ff ect has been identi-fied as coming from imperfect fiber / IO coupling (in the exper-iment the fibers are not glued unlike in an actual instrument)and from flux coupling in the substrate that is partially guided.However, the latter contribution has been dramatically reducedthanks to new etching technology that allows each waveguideto be completely isolated from the others and to not waste flux(Labeye et al. 2006). As a matter of fact, our experiments usingthe old technology showed flux crosstalk of up to 8%.The flux routing of the interfering beams, for each pair, needsto be as equal as possible, although this is not a strong require-ment. In fact, the instrumental visibility (and therefore the SNR)decreases as the beam fluxes are unbalanced. For these beamcombiners, the tricouplers equally split the flux in three (withinthe error bars), and the flux splitting ratios of the Y-junctions aresatisfactory. On the contrary, the couplers can be as unbalancedas 60% /
40% leading to a maximum contrast loss of 2%. Thesesplitting ratios were estimated with the assumption of ideal func-tions, which actually present a small loss of a few % (Labeye2008). These unbalanced ratios are due to an error in the de-sign, and new prototypes are being made with improved cou-plers. Globally, the agreement of the simulated data with the ex-periments is very satisfactory.
Finally, the stability and repeatability of all observables arekey elements to reach high dynamics. Photometric quantities arestable over a day timescale at the level of a percent, and are re-producible from one day to another with a maximum variationof 3%. We suspect that the absence of a glued interface betweenthe fibers and IO beam combiners contributes to the small errors(except for the closure phases for which this e ff ect is canceledout). However, for a first demonstration, the performances aregood and the general flux routing is validated. The competitivetransmission together with the capacity to use all photons forcoherent combination lead to an overall high sensitivity for such4-beam phase-shifting beam combiners. Interferometry:
The instrumental contrast directly impactsthe sensitivity of the instrument and should therefore be as highas possible. Here the beam combiners produce instrumental con-trasts with values always above 80% in wide band, and reaching100% with spectral dispersion. In broad band operation, the ef-fects of di ff erential dispersion due to unequal length of the fiberscan lead to a contrast loss of up to 20% which explains the veryhigh contrasts obtained with spectral dispersion. Statistically, thestability of the contrasts over a day timescale is good (from 1%to 4%) but the reproducibility of such measurements onto an-other day could not be validated due to the non-linearity of ourdetector. This a ff ects the value with a bias of 5%, making ourresults only upper limits.The phase shifting functions were initially designed to sam-ple the coherence in four phase states in quadrature (the so calledABCD sampling). For the two tested beam combiners, the phase-shifting functions lead to 5 phase-shifts out of 6 in agreementwith our expectations, while one of them (the central one corre-sponding to the [14] combination) is far from 90 ◦ . We suspect aninhomogeneity in the constitution of the silicon substrate at theposition of this phase shifting function. Of all 6 functions, thisone is the most distant from the center of the beam combinerand is located next to the edge of the chip. This e ff ect could ex-plain the measurements obtained on both chips since they werelocated next to each other on the same wafer. This could alsoresult from a relaxation of the stresses on the silica during thechip cut. An error in the design has been ruled out. This is thefirst time that these functions have been tested and while still notcompletely controled, these results are encouraging and showthat the use of phase shifting functions in IO beam combiners ispromising. In addition, phase shifts are stable on a timescale ofa day, within 1 ◦ and vary up to 3 ◦ from one day to another.The incidence of this departure from phase quadrature onthe final complex visibility SNR estimation cannot be quantifiedwithout a proper calculation. It should be seen as a reduction inthe instrumental response. In the limiting case where the phaseshift is 0, one cannot retrieve unambiguously the complex vis-ibility information. However since all but one phase shift havevalues close to quadrature, we believe this validates the concept.Finally, the measured closure phases are not equal to zero.Specific phase contributions of each function, that result fromnon-perfectly symmetric pathways, are unknown. When observ-ing a scientific target, calibrating with a point source (that iscentro-symmetric and supposedly leads to a zero closure-phase)should allow us to remove the instrumental contribution. Itshould be expected that given the remarkable stability of suchbeam combiners (Berger et al. 2000) the instrumental responseshould be very well calibrated. Chromaticity:
The importance of the chromaticity of thefunctions directly depends on the spectral resolution used in theinstrument. On all photometric coe ffi cients, the average chromaticity cango from about 12% to 80%. Individual functions show varietiesof chromaticity, with variations for tricouplers and Y-junctionsfrom 3% to 28% while couplers can be considered as achro-matic. The Y-junctions and couplers for the [23] combinationare more chromatic than the others. More details will be given ina following paper (Labeye et al. 2009, in prep.). However photo-metric calibration in the presence of dispersion should solve thisissue. With correction for photometric e ff ects, the obtained con-trasts show very little chromaticity, with an average maximumvariation of 4.6% over the range of H band. Phase shifts showchromatic variations of 5.6 ◦ on average, the maximum variationcorresponding to the [23] combination as for individual func-tions. The closure phases show strong variation with wavelength,up to 30 ◦ , which results in the sum of all chromatic e ff ects forthe three telescopes in the closure relation.We can anticipate that in cases where spectral resolution islow the e ff ective wavelength might be a ff ected and a properwavelength calibration should be considered together with aproper stellar calibrator choice. In addition, in order to limitpotential biases, particular care should be taken to specify thealignment accuracy between the beam combiner, spectrographand detector pixels. Polarization:
Birefringence control is a critical part of aguided optics instrument. In our experiment, we used highlybirefringent fibers which have a well defined polarization axisbut, in turn, require specific care on how the polarization state ismodified along the propagation.Our study first shows that all estimated quantities (photome-try, instrumental contrast, phase shift, closure phase) can be dif-ferent depending on the polarization state. We have shown inparticular that the behavior of the central [14] combination is dif-ferent from the other phase-shifting functions. This confirms thesuspected marginal behavior of this combination, maybe origi-nating from an inhomogeneity in the substrate for both chips.Besides, our work showed that it is necessary to split the polar-ization states before or after the beam combinations, or to ac-tively control the phase shift between the two linear states toavoid contrast loss. The use of birefringent fibers and waveg-uides forces the phase velocity of the two polarization statesto be di ff erent and results in two shifted interferograms. Whenthese two are superimposed, it leads to a single low contrast in-terferogram. In the case of a sensitive imaging instrument, split-ting after the beam combination is recommended since the useof polarizers before the combination would lose half of the use-ful photons. As far as stability is concerned, without splitting,the di ff erential phase shifting of the polarization states inside thebeam combiner would lead to varying results. The new proto-types are being designed with special care given to this problem. On-sky operating modes:
By allowing all the complex coher-ent factors to be measured in one single detector frame readout,these beam combiners o ff er two observational modes, depend-ing on the stability of the fringes ( i.e. on the atmospheric con-ditions or on the availability of a fringe tracker and its perfor-mance). If the fringes are stabilized to better than a fraction ofthe wavelength, a long coherent integration of the flux on eachpixel is possible ( i.e coherencing mode), highly increasing theSNR compared to temporal encoding. Otherwise, by varying theOPD, one will access 4 phase-shifted interferograms on whichto estimate the interferometric observables. The latter mode isalso well suited for laboratory measurements and calibration. Precision interferometry and data reduction: enisty et al.: A 4-beam IO beam combiner with ABCD encoding 13
We have shown that the described beam combiners presentperformance well suited for astronomical beam combination ina four telescope imaging interferometer. Our extensive labora-tory characterization shows that on-sky performance, in termsof precision, should be comparable to what has been achievedwith IONIC-VINCI / VLTI, IONIC3 / IOTA or FLUOR / IOTA-CHARA. However, an interesting number of astrophysical prob-lems will soon be more demanding (e.g. debris disk, hot Jupiterdetection). In that case it is important to better characterizecalibration issues and tackle all imperfections that could beintroduced by the beam combiner. Such work is justified bythe tremendous stability of the beam combiner that has beenrevealed by numerous industrial developments. Therefore, aproper calibration of the beam combiner should allow system-atics biases to be removed. What we propose is to use the so-called ’visibility to pixel matrix’ calibration in order to carry outa global inversion of the matrix that links the measured intensi-ties with the properties of the scientific object. Such a method,that will be detailed in a coming paper allows all instrumentalcontributions to be extracted. We briefly discuss here the philos-ophy of this data reduction. By developing the cosine, Eq. (1)can be written in such a way that instrumental terms are sepa-rated from the object contribution (V ob jml , ϕ ob jml , or similarly thecomplex visibility V ), leading to a system of linear equationsthat includes the instrumental matrix (V2PM, see Sect. 2): I = V2PM ∗ V The complex visibility of the scientific source will be ob-tained by inverting the system. A full characterization of theV2PM matrix should be first done with internal calibration pro-cedures. This method allows all instrumental e ff ects to be in-cluded in a single matrix taking into account all crossing termsdue to unideal behavior .
6. Conclusion
We have presented a laboratory characterization, in the H band,of an integrated optics beam combiner dedicated to the combi-nation of a four telescope interferometer. It uses a novel ’pair-wise static ABCD’ combination scheme which optimizes the ex-traction of coherence information: visibility amplitudes, phasesand closure phases. Our measurements show that, although com-plex, the flux routing inside the beam combiner is e ffi cient andthat the global throughput is competitive (65%). In particularthe comparison with simulated performance of several buildingblock functions is very satisfactory. The instrumental contrastsare high ( ≥ ff er-ent from zero but stable and probably caused by small internaloptical path di ff erences. The specificity of these beam combin-ers, which is to produce four phase shifted outputs to samplethe coherence information (for each baseline) in one integra-tion, has been validated. Finally, the global chromaticity of thebeam combiner has required specific optimizations that are non-standard with respect to standard telecommunication IO func-tions. IO chips used in telecommunications are usually requiredto show a flat response on short bandpasses ( ∆ λ ≈ ≈ nm around 1 . µ m(Labeye 2008). These results on a first prototype validate thefeasibility of a ’pairwise static ABCD’ combination scheme andits suitability for an interferometric imaging instrument.Most of technological building blocks are now defined. Weare working on the definitive version that will be included inside Gravity and VSI. This will require a number of technologicalimprovements and innovations. In particular, the phase-shiftingfunction will be ameliorated to get closer to the nominal 90 ◦ phase shift. The throughput will be improved with an optimizedrouting that will reduce the global propagation losses. We arecurrently extending the demonstration, using the same silica onsilicon technology, to the K band, as required by Gravity andVSI. It is expected that, while less transmissive in this band, theshort propagation distances inside the combiner will lead to ac-ceptable global losses. Finally, we will explore how this combi-nation concept can be extended to a six-way beam combiner andfit VSI requirements. Acknowledgements.
This work was financially supported by CNES, CNRS,ASHRA, Universit´e Joseph Fourier and Agence Nationale de la Recherche grantANR-06-BLAN-0421. We thank J-B. Le Bouquin, and S. Lacour for fruitful dis-cussions. We acknowledge the referee, Markus Schoeller, for his careful readingof the manuscript and for thoughtful suggestions that improved its clarity.
References
Berger, J.-P., Benech, P., Schanen-Duport, I., et al. 2000, in Proc. SPIE Vol. 4006,p. 986-995, Interferometry in Optical Astronomy.Berger, J.-P., Haguenauer, P., Kern, P., et al. 2008, in The Power of Optical / IRInterferometry, ed. A. Richichi, F. Delplancke, F. Paresce, & A. Chelli, 589– + Berger, J.-P., Haguenauer, P., Kern, P. Y., et al. 2003, in Interferometry forOptical Astronomy II. Proc. SPIE, Vol. 4838, p. 1099-1106Colavita, M. M. 1999, PASP, 111, 111Eisenhauer, F., Perrin, G., Brandner, W., et al. 2008, in Proc. SPIE, Vol. 7013,pp. 70132A–70132A–13Haguenauer, P., Abuter, R., Alonso, J., et al. 2008, in Proc. SPIE, Vol. 7013, pp.70130C–70130C–9Jocou, L. 2007, PhD thesis, CNAMKern, P., Malbet, F., Schanen-Duport, I., & Benech, P. 1996, in