The Keck Interferometer

KI is implemented as a classical Michelson interferometer, analogous to most radio interferometers, and is similar in approach to its predecessors, including PTI and the Mark III interferometer (Shao et al. 1988). AO systems on the two Keck telescopes, using the visible band for wavefront sensing, in conjunction with additional tip/tilt correction implemented by the interferometer via sensing at J or H band, provide corrected pupils from each telescopes. The interferometer field of view (FOV) is generally set by the diffraction pattern of the telescope aperture or subaperture,⁸ which is matched to the FOV of single-mode fibers for the fringe tracker cameras, and pinholes for the nuller. While fundamentally a single spatial mode system, because of the broad wavelength coverage, spanning a decade from J to N band, and requirements on visible-wavelength metrology and alignment as discussed later, spatial filtering was implemented only after beam combination.

KI supports two FOVs from each telescope, each propagated separately (although sharing some common transport optics) to the beam combiner. For nulling and the split-pupil V² modes, these two FOVs correspond to separate beams from the left and right halves of each Keck pupil in order to synthesize a four-telescope nulling interferometer, or to feed separate fringe trackers. While accommodated in the original design, but only implemented as part of the ASTRA upgrade (Woillez et al. 2010), the interferometer can also be configured to propagate separate FOVs from two stars separated by up to 30″. The two beams from each telescope are referred to as “primary” and “secondary”, where the primary beam propagates via a traditional on-axis coude train, and the secondary via an active off-axis coude train.

KI uses active fringe (phase) tracking in all modes to control the path length between the AO-corrected pupils. For ordinary (incoherent) V², temporal coherence is only required during the typical 5–10 ms frame time used to measure the fringe phasors, as only squared phasor quantities are summed between frames. Thus the key requirement for the fringe trackers in this mode is to stay within the coherence length of the spectral channel in the presence of atmospheric and other disturbances. However, several of KI’s modes require stable cophasing to achieve high sensitivity. In particular, for nulling, the target phase at K band is used to synthesize long coherent integration times for N-band observations, as the 10 μm signal-to-noise ratio (S/N) is insufficient for high-bandwidth control. Similarly, as part of the ASTRA implementation, the target phase measured on one camera using a portion of the K-band light is used to stabilize the fringes on a second K-band camera to enable long coherent integration times to allow high spectral resolution measurements. ASTRA also demonstrated true dual-star cophasing using an isoplanatic guide star to stabilize the fringes on a fainter science target (Woillez et al. 2012b).

One aspect of the use of large telescopes in an interferometer, besides the obvious increase in sensitivity, is that the areal dilution of the array is typically much less than for a long-baseline interferometer with small apertures. For KI, with baseline B = 85 m and aperture diameter D = 10 m, the array diffraction limit (λ/B) is 5 mas at K band, while the primary beam is 45 mas, so that there are ∼10 diffraction limited pixels across the FOV. Thus only modest spectral resolution R = λ/Δλ > B/D ∼ 10 is required to avoid bandwidth smearing in order to maintain coherence for imaging over the entire primary beam. Because of the general need to fringe track on the science object, the effective size of the object is limited in order to provide adequate fringe contrast and thus adequate S/N for tracking, and so the small relative FOV was seldom a significant constraint.

There are three tiers of path-length control relevant to KI’s implementation: (1) coarse delay, to correct for the mean geometrical delay for the selected region of sky—a range equal to twice the baseline length is required for full sky coverage; (2) fine delay, to correct for earth rotation of up to 73 μm/s/m of baseline, cophasing offsets, and small beam combiner geometric offsets—the required range depends on the desired length of a continuous track; and (3) high bandwidth vernier delay, to apply corrections from the fringe trackers and other sensors to correct atmospheric turbulence and other disturbances.

At KI, tiers (2) and (3) are implemented with eight fast delay lines (FDLs), each of which provides fully independent high bandwidth control over a delay range⁹ of 15 m. For nulling, the most complex operational mode, there is one delay line per telescope for the K-band primary beam, the N-band primary beam, and the K- and N-band secondary beams. The coarse delay tier (1) is implemented with two move and clamp delay lines (LDLs), one per telescope, each providing a total delay range of 72 m, which corrects for the mean external delay as well as a ∼40 m internal delay, as the beam combining laboratory is offset toward Keck II. While other path-length control implementations are possible, this one is efficient with respect to the number of reflections and the use of available space in the facility. Operationally, as the coarse delay line is not a cat’s eye, changing its position requires pausing observations to complete a move-and-realign operation.

KI uses extensive laser metrology to monitor the optical paths. An early design decision was to use visible wavelength HeNe metrology wherever possible for reasons of ready availability, ease of alignment, eye-safety, and the general insensitivity of the science arrays at this wavelength. In addition, boresight lasers for beam-train alignment were also at visible wavelengths. The later ASTRA implementation added an infrared metrology system, as required by S/N considerations, described in § 2.11.

While KI is an integrated system that includes the Keck telescopes and their AO systems, most of the hardware added to allow interferometry is at the basement level of the observatory, a.k.a., the interferometry basement. The potential of interferometry was considered early in the development of the Keck telescopes, and their implementations included provisions for a coude feed, a large coude tunnel, and with the completion of Keck II, a large beam combining laboratory.

Figure 1 presents an overview of the facility, illustrating light propagating from the telescopes at the front end, through the AO system, DSM, and coude train, to the interferometry basement, and within the interferometry basement through the delay lines into the beam-combiner area.

Zoom In Zoom Out Reset image size

While the interferometry control room has standard environmental controls, passive thermal control is relied upon for the beam combining lab. To attenuate environmental disturbances, the overall space was partitioned using modular Plexiglas cleanrooms into three rectangular spaces (see Fig. 1a): an LDL and beam transport area, as an extension of the coude tunnel between the two telescopes; an FDL area; and a beam combiner area. The latter two areas include panels above the spaces to provide further isolation from facility equipment (and from the small amount of make-up air provided near the back end), and also include access corridors outside of the cleanroom space. The coude tubes were also capped with optical windows, and various air leaks sealed, to minimize air flow within the interferometry basement. The LDL and FDL areas are largely free of electronic heat sources in order to limit induced turbulence; for example, the lasers and sensors used for the FDL metrology are fiber-coupled from the control room. There are more electronics along the periphery of the beam combiner area, but the small-diameter compressed beams (see § 2.1.4) and shorter propagation paths in that area (as well as the actual location of the electronics) limit their impact on beam-train seeing.

Figure 2 shows a plan view centered on the beam combiner area showing the final instrumentation complement. The outputs of the FDLs, after beam compression, are directed to the aft switchyard, where successively shorter wavelengths are extracted for the 10 μm nuller; the L-band or K/H-band fringe tracker(s); the H/J-band angle tracker; and the visible alignment system at the back end. The wavelength hierarchy isn’t strict due to the requirements for laser metrology and alignment. For example, the dichroic beamsplitter extracting light for the fringe tracker also needs to be partially reflective in the visible to support laser metrology and boresight functions. More detail on the overall optical configuration is given in § 2.5.

Zoom In Zoom Out Reset image size

Standard optical tables are used throughout the facility to support the interferometer hardware. Because of the need to maintain stable alignment over a large and distributed facility, limited isolation of the interferometer optics from man-made disturbances is possible. Rather, isolation and suppression on the source side was implemented, along with selective stiffening and damping of interferometer components, and active control and feedforward as discussed in § 2.11. The large machines needed to support the Keck telescopes produced significant vibrations, and major efforts were made early in the interferometer project to reduce the influence of these vibrations (especially from large induction motors with tones at ∼29 Hz and ∼58 Hz) on the interferometer hardware (Hess et al. 2003). These mitigations included the seismic isolation of machinery, air handlers, and fluid lines; the use of damping on the hydrostatic bearing oil system and on the transport optics; and operations policies on the status of equipment during observing. The vibration environment improved over time as old machines were replaced with new machines with specific attention paid to vibration and isolation issues. In conjunction with active measures, these improvements led to more stable visibilities, and allowed the use of longer integration times. More discussion on vibration mitigation and final performance is given in § 4.3.

As the KI science instruments are all single mode, for simplicity of implementation a collimated beam is propagated from the pupil at the deformable mirror in the AO system to the interferometer back end without any explicit pupil reimaging. This design affects the selection of the compressed beam diameter, the beam-train clear aperture, and the strategy for control of differential chromatic refraction (DCR).

The demagnification from the effective 10 m Keck aperture¹⁰ to the pupil on the deformable mirror is 80.4, so the starting beam diameter is 124 mm. A clear aperture (CA) of ∼150 mm is maintained for propagation at this diameter; the propagation distance in this space is typically 125 m, although it can vary by -45 m to +35 m depending on the LDL and FDL positions.

After 4:1 beam compression at the exit of the fast delay lines to 31 mm diameter (and for which a 44 mm CA is maintained), the propagation distances to the FATCAT fiber coupler, KAT, and the auto alignment system are 10.5, 14, and 16 m, respectively. Including this additional distance, the effective propagation distances, L, referred to the 124 mm beam diameter, d, are 290, 350, and 380 m, respectively.¹¹ Thus, for K band, the Fresnel number d²/(4λL) at the beam combiner is 6.0, which is still within the geometric-optics limit. For the nuller, the propagation distance for the compressed beam to the 20 mm cold pupil stop in the nuller camera is 12 m, for L = 320 m, and a Fresnel number of 1.2: smaller than desired, but a compromise that greatly simplified the implementation.

The other aspect of the pupil management strategy is control of DCR between the angle tracking and science wavelengths. The angular dispersion attributable to astronomical refraction at Mauna Kea at a 45° zenith angle (ZA) is given in Table 1. Refraction is accommodated with a Risley prism pair at the input to the angle tracker. It is configured for the wavelength difference between fringe tracking and angle tracking in order to maintain a fixed alignment on the fringe tracker at the expense of beam walk on the angle tracker. The Risley prism pair differential angle tracks the telescope elevation, while the common angle tracks the telescope azimuth to account for pupil rotation. For J-band angle tracking and K-band fringe tracking at a 45° ZA, the beam walk on KAT computed using the propagation distance above is 3 mm for the 31 mm beam, which is accommodated in the design CA. The reverse calculation, allowing for beam walk equal to ½ of the CA, gives the usable acquisition FOV at the angle tracker of approximately ± 0.65″.

Table 1 also includes the broadband coupling loss attributable to image elongation, which is < 2% at K band. For the angle tracker, the image elongation causes a small amount of S/N degradation at large zenith angles, although the effect is usually masked by image residuals attributable to imperfect wavefront correction; systematic aspects are accommodated by our per-star flux optimization procedure (see § 3.2). Finally, the table includes the magnitude of the broadband longitudinal phase dispersion computed for a near worst-case delay of 60 m (45° ZA in the plane of the baseline). The broadband V² loss is ∼25% and ∼6% at H and K band, respectively, and is quadratic with delay and with the number of spectral pixels. We absorb the small S/N impact, and account for the effect in the broadband channel through the usual target/calibrator observing sequence (see § 3.3.3). For the spectrometer channels at all of the provided resolutions, the effect is small.

As shown in Figure 8, a simple cat’s eye assembly is implemented using a 1077 mm focal length parabola and a 12.5 mm diameter flat mirror. Focus is maintained with invar spacers and shear panels, and protected silver coatings are used on the optics. The starlight beam input and output are arranged vertically, 176 mm center-to-center, and the parabola is cut to a rectangular shape, allowing close packing of 275 mm between delay lines. The flat mirror is mounted on a custom piezoelectric (PZT) actuator consisting of a preloaded series assembly of a 25 μm stroke PZT (-10 to +150 V) for servo control and an 8 μm stroke PZT for fringe demodulation (a.k.a. dither). Separate PZTs are used such that the fringe demodulation—programmed open-loop via a lookup table—remains unaffected by PZT nonlinearities as the servo PZT responds to the control system. A second PZT is mounted on the front of the FDL cart in opposition to the first for momentum compensation. A dust cover helps keep the PZT mirror clean, as the starlight and metrology light are focused on its surface, and flexible magnet panels cover the access holes in the invar shear plates.

Zoom In Zoom Out Reset image size

The cat’s eye assembly is supported on a three-wheeled optics cart using a four-bar linkage implemented with rotational flexures. The optics cart uses two 75 mm vee-wheels on one side riding on a 38 mm round rail for angular guidance, and a single 60 mm wheel on the other side riding on flat rail to complete the support. A voice coil motor drives the cat’s eye assembly with respect to the optics cart. The relative position is sensed with a linear variable differential transformer (LVDT) sensor, and a composite coil form is used in the voice coil motor to minimize eddy current damping. The optics cart is driven via a second voice coil motor from a separate tractor cart shown on the right in Figure 8; the relative position of the optics and tractor carts is measured with a second LVDT (this one with an internal bearing). The tractor drive uses a micro-stepped stepper motor with an inertia damper coupled through a 5:1 gearbox to a 25 mm vee-wheel which drives against a central 25 mm drive rail. FDL drive and sense power is via a ribbon cable which runs in a plastic tray adjacent to the drive rail. The cable is managed using an idler cart that is preloaded using a passive spring-reel cable tensioner. A precision magnetic home switch (Sony Magneswitch) attached to the cat’s eye assembly provides a micron-repeatable zero point. The track straightness (both lateral translation and tilt) limits beam shear to ∼2 mm over the range of delay line travel.

The delay line positions are monitored using standard heterodyne metrology, as widely used in optical interferometers (see Shao et al. 1988). The laser source plates use stabilized (10 ppb) 633 nm HeNe lasers and acousto-optic modulators (AOMs) with a 2 MHz frequency difference. The source plates are located in the KI control room, and feed their s- and p-polarized outputs using separate single-mode polarization-maintaining fibers to 1:4 fiber splitters into the laboratory area. These fiber splitters feed fiber-coupled beam launchers as illustrated in Figure 9. The metrology launchers, and corresponding retroreflectors mounted in front of the delay lines in the vertical gaps between starlight beams, are used to measure the FDL optical paths in double pass; the metrology apertures on the front plate of the cat’s eye assembly are shown in Figure 8.

Zoom In Zoom Out Reset image size

The optical output of the launchers—reference and unknown signals—are returned to the computer room using multimode fibers where they feed fiber-coupled preamps followed by an automatic gain control (AGC) stage¹⁶ and comparator. The fringe phase is quantized in a digital laser-counter card to 7 bits, and as the metrology is double pass with respect to the starlight, provides a delay resolution of 1.2 nm.

The delay line is fully programmable in position and rate, and includes a target generator (500 Hz update) to generate smooth trajectories from low rate inputs such as sidereal targets.

In track mode, four nested control loops are implemented:

• 1.  
  PZT loop (4 kHz update): This loop implements PD (proportional-derivative) control of the PZT rate via a digital integrator connected to a digital-to-analog converter (DAC) which feeds the amplifier for the servo PZT. The servo error is computed from the current target trajectory and laser metrology feedback; delay line targets requiring low latency can bypass the target generator and feed the PZT loop directly.
• 2.  
  Optics voice coil loop (800 Hz update): This loop uses the PZT position read back from the digital integrator (and includes an additional feedforward from the servo error to maintain stability in the event of PZT saturation) and implements PID (proportional-integral-derivative) control of the optics voice coil to continuously offload the PZT. The optics voice coil force is compensated by adding a feedforward of its command to the tractor voice coil servo output to prevent reaction of the optics cart.
• 3.  
  Tractor voice coil loop (800 Hz update): This loop uses the optics voice coil position as measured by its LVDT sensor and implements PID control of the tractor voice coil.
• 4.  
  Motor loop (100 Hz update): This loop uses the tractor voice coil position as measured by its LVDT sensor and implements PD control to update the motor rate; the mean motor rate is provided as a feedforward from the target generator.

For large moves, the FDL controller switches to slew mode, which disables the PZT loop, stabilizes the two voice coils using their collocated sensors, and drives the motor using the servo error. The delay slew rate must be kept less than 0.6 m/s in order to keep the Doppler shift on the (double-pass) metrology to less than 2 MHz.

The controller also implements several calibration functions, including home set/check to clear/read the metrology position at the fiducial mark, and a dither-calibrate function. The latter creates a position lookup table that accounts for non-linearity and hysteresis of the dither PZT to generate the triangular waveform for fringe demodulation; high time resolution to provide good waveform fidelity is accomplished through varying the phase of the interrupt clock used by the FDL control system.

The two fastest loops are usually tuned for ∼800 Hz and ∼25 Hz command bandwidths, respectively. Based on measured error rejection versus frequency, the effective command latency is 0.7 ms, providing disturbance rejection below 250 Hz. For feedforward commands from external metrology and accelerometers, provided directly to the PZT loop, a lead filter is implemented to improve mid-frequency rejection at the expense of increased peaking above 300 Hz, where there are typically only limited disturbances, achieving 20 dB suppression at 90 Hz. See § 2.7 for discussion of the performance of the FDL as used with the fringe tracker.

Figure 10 shows the free-space combination optics, implementing a traditional Michelson beam combiner, similar to that used at PTI. It is fed from dichroic beamsplitters in the aft switchyard. It is optimized for the H and K bands, although the beamsplitter is usable in the visible to support metrology and alignment functions. The two beamsplitter outputs are directed into single-mode fibers with integral vacuum feedthroughs directly into an infrared dewar. No photometric channels are provided; calibration for flux imbalance is provided using interleaved single-aperture calibrations as discussed in § 3.2. The use of free-space combination, and post-combination fibers, allows more ready incorporation of end-to-end laser metrology.¹⁷

Zoom In Zoom Out Reset image size

The stimulus and alignment beams are introduced through a visible/IR dichroic fold mirror, which leaks slightly at H and K, to allow some infrared transmission. The stimulus includes a fiber-coupled tungsten-halogen white-light source, and a 658 nm single-mode fiber-pigtailed laser coupled into the combined beam with a rugate notch filter. A remote-actuated flip-in iris in the stimulus provides a pencil beam used for beam-train alignments. A second rugate notch filter in series with the white-light and laser stimuli can be used to inject the 633 nm constant-term metrology as discussed in § 2.11.

The fold mirrors directing light to the fibers are mounted on PZT tip/tilt stages (“raster mirrors”) and are used for optimizing coupling into the fibers. The fiber feeds use 81 mm focal length off-axis parabolas (OAPs) to couple light into custom fluoride-glass fiber assemblies (LVF IRguide-1: 6.5/125 μm; NA = 0.17; 1.45 μm cutoff) with integral vacuum feedthroughs. The mode-field diameter is 27.5 mm, which is close to optimal for a 31 mm beam, providing a theoretical coupling efficiency of ∼70% for the (obscured) Keck pupil. A separate, standalone white-light stimulus, which provides nearly uniform (vs. Gaussian) illumination over the starlight pupil, can be used to provide improved sensitivity for alignment of the fiber feeds.

The dewar optical schematic is shown in Figure 11. It uses a custom dewar with a 30 cm diameter cold plate, and two LN2 reservoirs connected to an autofill system, with an ion pump used to maintain good vacuum between observing runs. The vacuum segments of the four fibers are routed to fiber collimator doublets (19 mm diameter clear aperture; 70 mm focal length) which are mounted on PZT tip/tilt stages to adjust pointing onto the array. The four collimated beams are combined and reflected via a pyramidal mirror through a 5-position filter wheel and 5-position prism wheel, to a final camera relay (50 mm diameter; 84 mm focal length) onto the infrared array mounted directly behind the annular fold mirror. A cryomotor allows remote control of the focus of this final stage. The filter and prism wheels are remote controlled via external motors and mechanical feedthroughs.

Zoom In Zoom Out Reset image size

The filter wheel provides positions for H; K; K^′, continuous H-K, and a split filter which provides a K filter for the two of the beams and is open for the other two beams. This latter configuration was most commonly used, with the K side used for the broadband channels, the open side used for the spectrometer channels. For the typically fast readouts used, any background penalty from the lack of a filter in the open channel was made up for by the higher transmission. The prism wheel provides, for two of the four beams (corresponding to the open half of the split filter setting), narrow- and wide-deviation direct-view prisms, and for one of the four inputs, a medium-resolution grism. At K band, the prisms and grism provide 5, 10, or 42 pixels across the band, with the ASTRA version, which includes a high resolution grism, providing 330 pixels across the band for SPR observations. At H band, the wide deviation prism is used providing 4 pixels across the band. The wavelengths are determined from FTS measurements.

FATCAT uses one quadrant (512 × 512) of a Rockwell HAWAII HgCdTe 1024 × 1024 infrared array with a 2.5 μm cutoff. Using the PZT adjusters on the collimating lenses, each spot or dispersed spectrum is arranged onto one 18.5 μm pixel, or onto one row of pixels, to minimize the required number of pixels that must be read; multiple nondestructive reads of each pixel are used to further reduce noise. The PZTs on the collimating lenses are adjusted using the FATCAT stimulus seen in retroflection through a narrowband filter to provide repeatable spectral alignment (see § 2.13). The usable flux per frame per pixel is ∼60,000 e^- in order to keep V² errors attributable to array saturation to < 0.3%.

The array electronics are commercial (Leach & Low 2000), using a pre-amp attached to the dewar feeding an external post-amp and 1 MHz/16 bit digitizer (0.26 DN/e^-). The output is connected via a custom fiber link to a custom clock control and interface card in the FATCAT VME chassis, which includes a downloadable waveform memory for the clock pattern, as well as an interface to the digitizer. A coadder provides hardware support for multiple nondestructive reads, and CDS read noise of 5 e^- to 10 e^- can be obtained for the low resolution modes where only a small number of pixel are read. While they may share a single dewar, from a controls perspective each beam combiner is logically separate.

In retrospect, the single-piece fiber assemblies generally worked well, providing a low-background dewar without the need for baffles and stops, avoiding losses from window transmission, and allowing for easy multiplexing of multiple inputs. The fiber assemblies survived multiple thermal cycles over the life of the instrument, and the few issues we encountered occurred only on the first cool-down. The tradeoffs for this approach are that pointing and focus inside the dewar required internal actuators, rather than allowing compensation from an external feed, and that the far end of the fibers were not directly accessible, so that fiber input alignment and optimization could only be done via readout of the array.

In conjunction with path-length modulation on the delay line, FATCAT uses an ABCD algorithm to measure the fringe phasors. The fringe phase is used to command the delay lines to stabilize the optical path, and the fringe amplitude and phase are used for science measurements. When fringe tracking, FATCAT operates at frame rates of 4, 5, 8, or 10 ms, depending on the mode and seeing; when phase referenced for the ASTRA SPR or DFPR mode, much longer integration times are possible (see § 3.1). The fringe tracker has strong heritage to PTI, and algorithmic details and performance of the KI implementation are provided in a separate paper (Colavita et al. 2010a).

In low resolution modes, the white light (WL) and coadded spectrometer channels are coherently combined to increase the S/N for phase estimation via the usual ABCD algorithm, but implemented using a sliding integration window to decrease control latency. The phasors are corrected for mismatches between the dither wavelength and actual wavelength (‘‘dewarping”). In all modes, the spectrometer data are phase referenced to the WL phase—typically for 12 frames—to improve S/N before estimation of group delay via a coherent DFT algorithm.

Phase and group delay can be combined in different ways to generate the FDL control signal. For V² operations, a simple fast inner phase loop with a slow outer group delay loop is employed. With a zero target to the outer loop, the servo asymptotically tracks to zero group delay while correcting fast phase fluctuations. However, to reduce V² systematics that depend on phase (e.g., from a nonlinear dither), the target for the outer loop is typically configured as a low frequency triangle wave, one wavelength peak-to-peak, so that visibility measurements are averaged over 2π of phase.

A different servo loop, which tracks to a fixed phase point within the coherence envelope, is used for nuller and ASTRA cophasing. The system is optimized for low control latency in the cophasing path. Both the feedback and phase referencing paths include active control of narrowband disturbances (as does the feedback path for the V² mode). In addition, a robust algorithm to estimate the slowly varying water vapor dispersion as required for nuller cophasing—essentially synthesizing a long coherent integration time on the dispersion only—was also implemented.

Transitions among search, semi-lock, and lock states use the fringe S/N estimated using a modified ABCD algorithm optimized for rejection of scintillation. The fringe search uses a geometric spiral about the last lock position. The transition to semi-lock is based on the instantaneous S/N exceeding a threshold, while the transition to and from lock depends on the average S/N. The threshold includes an additive term to set a minimum V² floor in addition to a scalar S/N threshold.¹⁸

As shown in Figure 12, four dichroics reflect two primary and two secondary beams to fold mirrors on the KAT foreoptics table. These fold mirrors direct the four beams, on an approximate 75 × 150 mm grid, into a Cassegrain telescope with a parabolic primary (305 mm diameter; 914 mm focal length) and a convex spherical secondary (51 mm diameter; -125 mm focal length). The beams are deviated from the telescope’s optical axis such that their projections overlap ∼20 m front of the telescope; the telescope images this pseudo pupil close to the detector lens at a demagnification of 1.8:1. An apochromatic triplet (38 mm diameter; 101 mm focal length) produces four images on one quadrant of the detector array on an approximate 35 × 70 pixel grid for the array’s 18.5 μm pixels; the images are centered near a corner of a quadrant to enable rapid array readout. The quasi-pupil approach employed is scalable to a larger number of input beams. Remote-controlled shutters in front of the KAT fold mirrors and the dewar are used as part of the calibration process.

Zoom In Zoom Out Reset image size

A custom dewar with a 25 cm diameter cold plate houses an input window, baffles, a filter wheel including J-, and H-band filters, the detector lens, and the infrared array with its mount and fanout card. The cryogenics and electronics are similar to FATCAT.

A pair of Risley prisms is located prior to the dewar window; although the beam there is not precisely collimated, it’s at a slow focal ratio (f/50). The Risley prisms are controlled based on the science and KAT wavelengths as a function of telescope elevation and azimuth as discussed in § 2.1.4. The deviation was calibrated empirically for zero point and scale factor based on sky tests.

Fundamentally, KAT senses tip/tilt errors from image motion on the array and drives fast tip/tilt mirrors to minimize the errors.

Each KAT image is allocated a 16 × 16 pixel track box, read at up to 80 Hz. KAT implements a Fourier-domain matched-filter tracker using, typically, a 4 pixel FWHM Gaussian image template. It provides improved performance versus our original windowed centroid tracker. The matched-filter tracker, even with a fixed template, was found to be robust to varying image sizes, and it’s also fundamentally insensitive to a constant image background. While a quad cell (with appropriate optics changes to provide a finer image scale) would be more sensitive, the flexibility and diagnostics provided by the imaging mode are significant benefits. KAT can also be run at lower frame rates to improve sensitivity.²⁰

The KAT servo uses integral control with conservative tuning. Since the AO system has removed most of the fast seeing, including tip/tilt, KAT addresses primarily seeing and drifts in the beam train: the tuning procedure results in 3–5 Hz closed-loop (command) bandwidth for 80 Hz sampling, which is adequate for the typical power spectra of the residual image motion.

As discussed in § 2.5, the primary tip/tilt actuators for KAT are the fast tip/tilt mirrors shown in Figure 7. In nulling mode, the fast tip/tilt mirrors of the nuller (not seen by KAT) are slaved to the V² tit/tilt mirrors.²¹ To minimize the stroke of the tip/tilt mirrors, there is an option of using a crossover network to send low bandwidth (< 1 Hz) control commands to the M14 mirrors, which are common to both N and K band, and high bandwidth commands (>1 Hz) to the fast tip/tilt mirrors: This is referred to as split-actuation mode. Whether or not split actuation is enabled, an outer (integral) control loop with a 1 Hz sample rate is implemented back to AO (via a CORBA/Keyword bridge [see § 2.12]) to continuously offload the commanded tip/tilt. For the secondary beams, there is additional offloading of the difference between the secondary and primary tip/tilt commands to the secondary M7 control.

Similar to FATCAT, there is the option to implement active suppression of narrow-band vibrations. This is implemented on a separate processor running at a 2500 Hz update rate which provides the actual control of the fast tip/tilt mirrors. Originally, this processor also supported use of a laser beacon back-propagated to a sensor at the DSM to provide very high bandwidth tilt sensing to deal with environmental disturbances. This mode, including the narrow-band suppression, was deprecated as the environmental disturbances became better controlled over time. Recently, as part of the ASTRA project, preliminary experiments were conducted using a beacon from the DSM shear sensor, and sensors in the beam-combining lab, in order to provide high bandwidth control of the secondary beam train for faint stars that would allow only very slow KAT frame rates (Woillez et al. 2012b).

Finally, to support chopping implemented by the AO system as used for nuller alignment, the KAT servo control can be gated during the off-target phases of the chop pattern. AO chopping is implemented synchronously, so no handshaking between KAT and AO is required, except for knowledge of the overall AO mode.

KAT track mode includes a state machine which transitions through the image acquisition process to the lock state and automatically accommodates image loss and other common fault conditions. The tracker zero point is initialized using autocollimation from corner cubes illuminated by the FATCAT stimulus (with the Risley prisms at null). Fine adjustment of the KAT input mirrors is used to set the image offsets to zero. However, during a typical V² run, the zero point of the controller is updated for each star based on feedback from FATCAT as discussed in § 3.2. Array calibration makes use of the remote-controlled shutters to measure background and read noise. The numerical value of the read noise is not needed in real-time, but is measured to ensure proper operation of the camera. Typical KAT read noise is 13 to 19 e^-, depending on whether two (using 4 subreads) or four (using 2 subreads) images are read out. The camera also supports an alignment mode using windows larger than 16 × 16 with reduced frame rates.

Light from the two “primary” and the two “secondary” beams enters at the lower right of Figures 13 and 14. Longitudinal atmospheric dispersion compensators (ADCs), consisting of motorized translating wedges of ZnSe, correct for systematic and variable dispersion. The primary and secondary light is then directed into the primary and secondary MMZs. For each MMZ, there is an outer and inner path through the interferometer, and the outer mirrors include PZT (and manual) actuators to allow for vernier path-length adjustment between the two paths. The inner and outer dark outputs of the primary and secondary MMZs are interfered at cross-combiner beamsplitters. Path-length modulators in series with the outputs of the primary MMZ provide a 20 Hz triangular OPD modulation for the cross combiner, and there are additional PZT actuators for vernier pathlength control between the primary and secondary inputs. The four cross-combiner fold mirrors are mounted on fast angular choppers, used during gated-mode fringe acquisition. Also shown are metrology beam launchers injected via the unused bright outputs of the MMZs. These launchers were ultimately not used in operation, as commissioning tests with the launchers showed that cophasing errors attributable to unmonitored paths between the nuller and FATCAT were primarily small drifts that could be accommodated within the (low) bandwidth of the nuller servo.

The nuller source place, shown in the center of Figure 13, is used for nuller alignment and calibration. The source plate includes a fiber-fed HeNe boresight laser collimator and a ceramic-element thermal source. The thermal source is collimated, spatially filtered, combined with the laser boresight, and injected back toward the nuller breadboard. The source plate includes a chopper wheel located near the pinhole which is synchronized to the real-time system to allow coherent detection of the white-light source during nuller breadboard alignment. Retroreflected light from slide-in corner cubes either on the nuller source plate (for coalignment of the nuller laser and white-light sources), at the entrance to the nuller breadboard (for coalignment of the nuller camera ports), or in the transport optics (for coalignment of the nuller and FATCAT), is directed to the nuller alignment CCD camera. A Risley prism in series with the boresight laser is used to account for the dispersion in the wedges of the nuller optics so that the laser boresight correctly traces the 10 μm path after exiting the nuller breadboard. Note that the stimulus-injection beamsplitter remains in place during observations, enabling the laser boresight, stopped down in diameter with an iris, to be sensed at the DSM using its shear sensor for real-time beam-train shear adjustment.

The four outputs of the nuller breadboard are directed to the nuller camera via a periscope to a fold mirror beneath the dewar, as shown at the top left Figure 13; the camera is described below.

The nuller camera, KALI (“Keck Aperture Nulling Interferometer Camera”; Creech-Eakman et al. 2003), uses a large (40 cm × 120 cm) LN2/LHe dewar, accepting four beams through a ZnSe window via a fold mirror beneath the dewar. The dewar includes motorized mechanisms (via ferrofluidic feedthroughs) for aperture, pinhole, filter, and prism sets (along with manual focus mechanisms for the first two lens sets); only the typical settings are given below. The camera internal optical path is illustrated in Figure 15. Light enters through a pupil mask (typically 20 mm diameter) and is focused by a Ge singlet (172 mm focal length) onto a pinhole mask (typically 160 μm diameter) at f/8 with respect to the pupil mask. Note that the geometric beam size at the pupil mask is a 31 mm diameter half-circle. The selection of aperture mask diameter and pinhole size was a compromise between the degree of spatial filtering and the system throughput. Following the pinhole is a Ge singlet recollimator (56 mm focal length), filter (typically a 7.7–13.5 μm bandpass), and prism (typically a “high-resolution” direct-view ZnSe-ZnS pair). The four beams, at this point arranged in a square grid, are multiplexed onto a common pyramidal mirror. While the prior optics are cooled to 77 K, the remaining elements are in a LHe “coldbox”, including a long wave block filter and a final biconvex Ge relay (60 mm focal length) that focuses light onto the array. The clocking of the prisms and the angles of the fold mirrors result in four dispersed spectra, each 3 × 20 pixels (coadded in software to 1 × 10), arranged in a square pattern on the array.

Zoom In Zoom Out Reset image size

The array is a DRS HF-128 128 × 128 high-flux silicon BIB array with 75 μm pixels, operated at 10 K, slightly above the cold-box temperature, for best sensitivity. Similar to FATCAT, the array is controlled by a combination of commercial front-end electronics (Wallace Instruments) that include clock buffers, bias control, preamps, and 14 bit digitizers, combined with custom electronics for clock-pattern generation and data interface to the real-time system. The 16 digitized camera outputs are multiplexed into four data streams to four camera interface cards in the nuller VME rack; two additional camera cards are used for clock generation and bias control. The 2.5 ms array readout is synchronized with the 20 Hz modulation of the rapid ramp mirrors on the nuller breadboard, which uses five 5 ms ZABCD bins on each of the up and down strokes. Unlike the shorter wavelength arrays, the reads for this array are destructive, and the Z bin data are not used. In operation, the system is mostly background limited, with read-noise contributing about 1/3 of the total variance in the measurement of a broadband fringe quadrature.

The nuller is phase stabilized at high bandwidth using FATCAT at K band on the two long baselines; details of that system are described in § 2.7. One important aspect of the stabilization is accounting for the effects of atmospheric water vapor. While the dry atmosphere is not very dispersive in the near- and mid-infrared, the water vapor contributions are very dispersive, so that even though water vapor is typically a small contributor to total OPD, the dispersion must be accommodated to allow high accuracy cophasing (Colavita et al. 2004b; Colavita 2010). In practice this means that the K-band system must separately estimate the contributions of dry air and water vapor, and account for them in the feedforward targets generated for the nuller FDLs, which cannot simply be slaved to the K-band FDLs. In addition, a separate feedforward target must also be provided to the nuller ADCs.

The nuller ADCs correct variable atmospheric dispersion, quasistatic dispersion from the basement OPD differences, and residual static dispersion from optical elements in the beam train. Using only group and phase delay measurements, the ADC zero point has a 2π ambiguity; this was calibrated using the spectral phase curvature measured from internal fringe tests. In operation, a quasistatic initial target for the ADC—before enabling the starlight servo—is determined based on the LDL position, the basement temperature and humidity, and a dispersion model, to enable convergence to the proper lobe.²³

While the cophasing system controls most mid- and high-frequency phase and group delay errors, the nuller itself, at lower bandwidth, must adjust the FDLs to acquire and track the fringe phase, control the nuller ADCs to minimize group delay, and adjust the three nuller vernier delays between the inner and outer paths of the MMZs and the primary and secondary cross combiner paths so that the fringe quadratures measured at the four outputs of the breadboard can be coherently combined. These control loops are illustrated in Figure 16.

Zoom In Zoom Out Reset image size

Because the single nuller camera is shared among the MMZs and the cross combiner, nuller control is time-multiplexed using two different micro-sequences (Colavita et al. 2010b, 2009, 2008, 2006; Booth et al. 2006; Garcia et al. 2006). The gated-mode micro-sequence is used for fringe acquisition, employing conventional fringe scanning implemented by the FDLs and the rapid ramp mirrors on the nuller breadboard to measure the MMZ and cross combiner fringe quadratures. These measurements are used to control the four nuller delay lines as indicated in Figure 16. Differences in the quadratures between the inner and outer paths are used, on bright stars, to adjust the three vernier actuators on the nuller breadboard. Finally, the difference between the group and phase delays is used to control the primary and secondary ADCs. In this gated-mode micro-sequence, the four angular chopper mirrors on the nuller breadboard are used to point away different beams during different sequence phases to allow disentanglement of the observed fringe quadratures.

When the OPDs have converged to less than ∼1 radian, nuller control switches into Null-Peak (NP) mode. In this mode, the angular choppers are disabled, the FDL scanning modulation is replaced by 0 or 180° phase steps, and the FDL and ADC commands are derived from coherent combination of the signals from the four camera ports. NP mode uses a 400 ms long micro-sequence divided into eight 50-ms slots. Two of those slots are used to implement low-bandwidth control on the long baselines; five are used to measure the coherent leakage at null; and the final slot is used to implement low-bandwidth control on the short baseline, and also to measure the coherent leakage at peak (i.e., at anti-null) to serve as an in situ photometric calibration. This is our most sensitive mode—NP mode has a 2 × higher S/N in our application than fringe scanning for control of the FDLs for the MMZs—and all science data are taken in this configuration. The NP mode also has the fundamental advantage of directly servoing on the key observable, the null leakage, rather than indirectly controlling the null leakage by maintaining a particular phase offset.

The interferometry basement from the fringe trackers through the FDLs exhibits low levels of vibration: fringe jitter (rms first difference) in double pass from FTS corner cubes in front of the FDLs is typically < 50 nm, and thus is a small contributor to the error budgets. On the other hand, without compensation for the path-length fluctuations from the FDLs to the telescopes, it is not possible to maintain reliable fringe lock, and metrology of that path is required. The solution employed for most interferometer modes was to add metrology, launched from just outside of the starlight clear aperture in front of the FDLs, as shown at the bottom of Figure 7, toward the M14 mirrors, and terminating at the M3 corner cubes in the DSM (Fig. 5). This optical configuration (“coude metrology”) avoided the use of dichroics, and also avoided interference with centered alignment and metrology beams, at the expense of some tilt-to-piston coupling in the sensing. Operationally, PZT fine adjusters on the beam launchers were used to optimize the pointing of this metrology to the DSM corner cubes for each star.

The beam launchers used for this system were identical, except for the beam expander, to those used for monitoring the FDLs described in § 2.6, and were fed from remote source plates in the computer room. The electronics were identical, too, except for the use of APD detectors to provide higher S/N. However, during earlier testing it was discovered that the polarization properties of the individual beam trains led to signal dropouts (with respect to our S/N margins) of this polarization-sensitive metrology system at certain azimuth angles.²⁴ The solution employed was a polarization rotator at the output of the launchers. The launchers output circular polarization, and the rotator just used a quarter-wave plate, converting the polarization back to linear, which tracked the telescope azimuth. With an appropriate zero point for the rotator, this system provided consistent performance at all telescope orientations. As this metrology was used for vibration control only, it was high-pass filtered at ∼1 Hz, so that it wasn’t necessary to worry about its zero point or momentary dropouts, and was used as a feedforward signal to the FDLs via the low-latency path described in § 2.6.

To monitor the rest of the optical path from the DSM corner cubes to the Keck primary, an accelerometer-based feedforward system was implemented. Figure 17 shows the deployment of accelerometers on the system, including six accelerometers, total, symmetrically mounted to three primary mirror segments, and three accelerometers, each, on the secondary and tertiary mirrors. There are also two accelerometers on the AO bench, aligned and adjacent to the input and output beams, and a final accelerometer on the DSM that serves as the tie point to the coude metrology (see also Fig. 5). To within the limitations of the sensors, this system allows correction for rigid-body motions of the three primary telescope optics, AO bench, and DSM, as they project onto the interferometer’s optical path.

Zoom In Zoom Out Reset image size

Quartz-flexure accelerometers were employed, with local, custom gain-strapped preamplifiers to accommodate the long signal runs required to cable the telescopes. The data acquisition systems installed in auxiliary crates in the AO computer rooms on the telescopes’ Nasmyth platforms ran synchronously with the FDLs, triggered at the 4 kHz rate of their fastest loop. The processing steps included simultaneous data acquisition, unit conversion, deconvolution of the antialiasing filter in the preamplifier, double integration from acceleration to position, and high-pass filtering to limit low-frequency noise propagation from the double integration. The high-pass corner frequency of 0.7 Hz was chosen so that the accelerometer position noise spectrum remained below the uncorrected atmospheric OPD power spectrum. Finally, the 15 sensor signals were combined as shown in Figure 17 to yield the feedforward signal used by the delay lines.²⁵ However, in practice, the secondary mirror was usually excluded from the sum, as its mechanical response to focus offloads from the AO system exceeded the dynamic range of the high-gain signal chain.

The achieved end-to-end rejection of the accelerometers in conjunction with the FDLs is 5 × or greater from 7–60 Hz, limited by latency at the high end and the noise-limiting high pass filter at the low end. Between runs, the accelerometer system captures 10 s long, 1 kHz data bursts every 1000 s, useful for long-term trending of the disturbance environment. It’s worth noting that while the coude metrology system remained essential to system operation, the accelerometers played a less critical role over time as the disturbance environment was better controlled, and the fringe tracker was augmented with narrow-band suppression capability: see § 4.3 for a more detailed discussion.

Prior to integration and test of the nuller, we were uncertain if the relative optical paths between the fringe tracker and nuller to the K/N splitters in front of the FDLs (after which the K- and N-band beams are common mode) were adequately stable at frequencies above the nuller starlight control bandwidth. While subsequent testing verified that the optical paths were indeed adequately stable without additional control, we discuss the constant-term (CT) metrology system for completeness, as it had implications on the system architecture, and also because it was restored and upgraded for the ASTRA project.

A version of the beam launcher design shown in Figure 9 was used, modified as described in the caption, to launch both s and p polarizations. The beam launcher injection was via a rugate beamsplitter in the FATCAT stimulus as shown in Figure 10. The injected light was split by the FATCAT starlight beamsplitter, and both polarizations were propagated toward the telescopes along each arm. A similar implementation was employed on the nuller, in this case, launching via the unused port (bright output) of the MMZs as shown in Figure 13. For both FATCAT and the nuller, the metrology was injected parallel to, and just below, the geometric starlight beam, but within the system clear aperture, toward corner cubes on the telescope side of the K/N splitters.²⁶ Polarizers in front of the corner cubes selected s for the K1 arms and p for the K2 arms; crosstalk between nuller metrology retroreflected from the corner cubes back to the fringe tracker, and vice versa, was accommodated through the use of different absolute and relative AOM frequencies.

The signal processing for the CT metrology was generally similar to that used elsewhere, although more care was required to maintain adequate S/N. In particular, since both the s and p polarizations propagate, the final S/N is lower than for a conventional launcher for which one polarization remains internal to the launcher, and not attenuated, serving as a local oscillator. Thus, in addition to the use of low noise preamplifiers, a bandpass filter was added to the electronics centered at the heterodyne frequency.²⁷ A consequence of the use of this filter was that the metrology would drop out as its frequency Doppler-shifted out of band as the delay lines slewed between targets. However, this was only an issue for sequencing, as continuity between scans was not important in this application.

The system performed acceptably, although care was required to align the beams to the optical axis to avoid shear-induced signal fluctuations (doubled by the corner cubes) as the delay line tracked,²⁸ and to align the polarization axes to avoid leakage. To minimize beam shear caused by motion of the fast tip/tilt mirrors, KAT was usually operated in split actuation mode (see § 2.9) when the CT metrology was used.

For dual-star use, end-to-end metrology—more than just the path monitored above—is required to enable accurate cophasing. In addition, and as was borne out in tests, the metrology needs to be coaxial if it’s to be used at frequencies below 1 Hz. Finally, while there was a deliberate design decision for KI to use HeNe metrology whenever possible, the ASTRA metrology system (Woillez et al. 2012b) incorporated a stabilized infrared Nd:YAG laser in order to provide adequate power to monitor the entire beam train to a corner cube in front of the deformable mirror in the AO system. This metrology system was launched coaxially from the aft switchyard between the tip/tilt mirrors and FATCAT,²⁹ and the CT system described above was now used to monitor the quasistatic path from the beam combiner to the Nd:YAG launcher via a common corner cube on the far side of the ASTRA metrology injection dichroic.

Despite its complexity, KI achieves very high instrument uptime, typically 90%, aided by good practices with respect to configuration control, maintenance, and documentation. KI observations were scheduled into observing campaigns during bright time, and significant instrument alignment and checkout occurs before going on sky. Ideally, all steps to bring the instrument into an operational state would be fully automated, in order to prevent human error and provide repeatability. In practice, a number of manual steps are involved, and a simple, but invaluable, tool is a continually-updated instrument checklist including week-before, day-before, day-of-the-run, and daytime and nighttime activities required to achieve successful operations. The checklist also includes threshold values of quantities such as measured array read noise and FDL tracking jitter, which if exceeded, require additional actions. The use of a rigorous checklist helps guard against more insidious errors where misconfiguration of, say, an OPD feedforward system, could be misconstrued during observations as simply poor seeing. The checklist also includes certain quarterly or biannual actions, such as beam-train collimation checks and FDL tracking jitter tests. In steady state, the effort to operate the interferometer, and maintain the interferometer hardware and software infrastructure as a facility instrument for observers at all levels of interferometry familiarity, was approximately 4 FTEs; it was slightly higher when the nuller mode was routinely used.

The degree of advanced setup leading up to an observing run depends on the mode. For all modes, the DSMs must be reinstalled if they had been previously removed to allow access to AO by other instruments, configured for either single or split pupil mode, and coaligned to AO. The internal beam train must be properly configured, and the various instruments configured. The RTC database includes a number of tagged sets which enable proper configuration of the real-time software; the high-level configuration tool IFSys helps to automate this process to minimize errors. For all modes, internal fringes are tracked to confirm proper operation and the expected delay offsets with respect to the metrology zero points. For the nuller, while the dewar is always maintained at 77 K via an LN2 autofill system, during pre-run setup the camera is manually cooled to LHe, and an extensive alignment and checkout procedure begins. This procedure concludes with a full-up daytime nuller sequence using the nuller’s internal source and the observing sequencer, with the results processed through the nuller data pipeline to confirm proper operation. This procedure is also used to select the four 3 × 20 pixel regions of the array used by the nuller, as well as to determine channel wavelengths using internal fringes. An abbreviated version of the nuller sequence is also used on the first day of the run to manually adjust the nuller vernier stages to be in the center of their PZT actuator ranges.

Routine daytime setup for all modes includes setting the LDL position for the first observation sequence of the night, beam-train alignment, beam combiner and camera alignment, array read-noise calibration, metrology alignment, and coalignment of the various systems; details are given in § 2.13. The final steps of daytime beam-train alignment include telescope azimuth rotations to thread the beam correctly through the telescope coude train to the DSM. Beam-train alignment is facilitated by the alignment sequencers described in § 2.13.

The final pre-observing checkout includes a detailed configuration check, final calibration and coalignment of the various systems, and end-to-end checks including a beam-train vignetting check, and acquisition of the AO stimulus on the angle tracker.

All observing modes employ various sequencers to ensure consistent and efficient data collection as described in § 2.12. In addition to the primary V² and nuller sequencers, dedicated sequencers are used at different parts of the night, for example, to enable rapid LDL repositions, to adjust the beam-train shear for nuller targets, and to optimize the nuller alignment between clusters, as described in § 2.13.

Several software tools are used to monitor system performance throughout the night. An alarm-handler system is used to monitor the state of the various systems, particularly motion control, that use the EPICS framework. IFSys monitors the health of the RTC components. Finally, the engineering GUIs are configured with threshold values on key telemetry to allow quick-look views of system state and to aid debugging. With respect to the science data quality, quick look analysis of the V² and nuller data is initiated at the conclusion of each scan to further aid in performance monitoring, as well as to optimize science observations, and if appropriate, to make changes to the scheduled observing program. The top panel of Figure 20 gives an example of phase jitter (rms first-difference of fringe-tracker phase) versus time—one of the most useful metrics for assessing instrument stability and performance. High levels of jitter cause both a reduction in fringe visibility, and at higher levels, prevent reliable phase unwrapping in the fringe tracker. The plot shown is representative in exhibiting an instrumental noise floor, a slight dependence on target magnitude, and a slow variation over time attributable to changing atmospheric seeing.

Zoom In Zoom Out Reset image size

Figure 21 illustrates a typical V² observing sequence, emphasizing the parallelism of the various parts of the process. As noted previously, KI does not include separate photometric channels, and we instead rely on the fact that if intensity fluctuations are uncorrelated between apertures, knowledge of the average fluxes of the individual apertures over the data-collection interval is adequate for full correction of the data (Shaklan et al. 1992). At PTI, only a single ratio measurement of the separate fluxes was taken at the end of a scan, and primarily served as a diagnostic, as application of the ratio correction was not shown to improve the data quality. However, at KI the observed intensity fluctuations are larger, and in particular, display a pink noise power spectrum. Subject to these statistics, interleaving the ratio measurements as shown in Figure 21 provides a more accurate measure of the average flux over the interval, and the ratio correction computed from these five sets of 5 s single-aperture fluxes generally improves the data quality of the 125 s of fringe data. The V² sequence also includes a single dark measurement at the end, implemented with a shutter. While this approach is adequate for photometric calibration of H/K data, for L-band observations, an additional telescope nodding step is required to get a useful dark measurement for calibration (Ragland et al. 2009).

Zoom In Zoom Out Reset image size

One other aspect of the sequence is worth noting: the flux auto-peak step prior to data collection (which happens in parallel with optimization of the alignment of the coude metrology). This step uses the raster mirrors on the FATCAT breadboard (Fig. 10) to determine the pointing offset which maximizes coupling into the single-mode fibers. In V² mode, an offset to the KAT zero-point is used to apply the pointing correction. In nulling mode, where KAT controls pointing to both the nuller and FATCAT, the raster mirror zero points are changed in order to not affect the nuller pointing.

The V² sequence is very efficient, and for H- and K-band targets, when the seeing is average or better, six or more stars per hour can be observed, as illustrated in Figure 20.

Finally, for the case of the nuller, Figure 22 illustrates the sequencing involved in getting to the point of collecting science data. The first three minutes includes the time for telescope slew, AO acquisition, and acquisition of the four beams on KAT. After a shear adjustment, K-band fringes are acquired, phase referencing is enabled, and background calibrations are acquired on the nuller in gated and NP mode. The next three minutes are spent acquiring and stabilizing nuller fringes using the low bandwidth nuller control system in gated mode. At eight minutes into the sequence, the nuller switches to the NP micro-sequence for data collection, enables the vernier control loops, and after settling, begins science data collection, typically for 12 minutes, for a cadence of about 25 minutes per scan. Addition information on the nuller science sequencing is in Colavita et al. (2009).

Zoom In Zoom Out Reset image size

All KI science observations were executed in service observing mode. Science observing was performed by staff at WMKO with support from NExScI. Interferometer runs were generally scheduled to take place two to four times during a six month semester, and science and engineering observing both took place during a given run. PI’s and Co-I’s could participate in the observing either from the interferometer observing room at WMKO headquarters in Waimea, or remotely. Remote participation was either data monitoring only, from any computer via a password-protected website showing the near-real-time data reduction, or monitoring of all telescope parameters at one of the Keck Remote Operations sites. Each project was assigned a NExScI contact scientist to help with observation planning and data analysis.

KI schedule files used a pre-defined ASCII format to describe the sources to be observed along with their properties. The interferometer files were read by the KI real-time software and also translated into the format used by the telescopes. These files could be produced by hand, but generally the getCal program was used.

getCal³⁰ is an interferometric observation planning software suite. Its tools can be used to resolve common star names into standard catalog designations and astrometry; extract visibility calibrators from the Hipparcos catalog according to a variety of geometric, brightness, and astrophysical criteria; compute target zenith and delay accessibility; and interface with the Simbad astronomical database, the PTI and KI sequencer GUIs, visibility calibration codes, and sky visualization software. It can be installed locally on the user’s computer or used via a web interface (gcWeb).

To further assist with investigation planning, the KI observables can be simulated using the Visibility Modeling Tool, VMT.³¹ VMT allows the user to specify an observation configuration (object coordinates, time coverage), instrument configuration (V² or nulling mode, spectral band, and dispersion), and a model of the object to be observed that can be specified either as a combination of elementary geometrical morphologies (with user-specified relative fluxes and sky offsets) or as an uploaded FITS image. VMT then computes and plots the resulting KI observables, and is therefore a valuable tool in assessing the viability of a given investigation.

3.3.3.1. Visibility Data

3.3.3.2. Nulling Data

The KI raw data were processed automatically at WMKO immediately following each observing night. Following successful processing the resulting data files were automatically sent to and ingested in an archive located at NExScI. All the KI data are now public and available to users via the Keck Observatory Archive (KOA).³⁵