PILOT: FIR Balloon-Borne Polarimeter

• PILOT is a balloon-borne far-infrared polarimeter designed to measure polarized thermal dust emission and map the Galactic magnetic field.
• It uses a rotating half-wave plate and multiplexed bolometer arrays to achieve high-sensitivity polarimetry, crucial for foreground cleaning in CMB studies.
• Advanced calibration techniques correct cross-talk, readout latency, and intensity-to-polarization leakage, enabling reliable, science-ready polarization data.

PILOT is a balloon-borne far-infrared polarimeter developed to measure the polarized thermal emission of interstellar dust, primarily at $240\,\mu\mathrm{m}$, corresponding to about $1.2$–$1.25\,\mathrm{THz}$. In the literature, the acronym is expanded as both “Polarized Instrument for Long-wavelength Observation of the Tenuous interstellar medium” and “Polarized Instrument for the Long-wavelength Observations of the Tenuous ISM.” Its scientific role is to probe dust-grain alignment, Galactic magnetic-field structure, and the spectral behavior of polarized dust emission at frequencies above Planck’s polarized dust channel, while its instrumental role is to test multiplexed bolometer arrays for precision polarimetry in a stratospheric environment (Misawa et al., 2014, Mangilli et al., 2018, Bernard et al., 2022).

1. Scientific purpose and observational niche

PILOT was conceived as a high-frequency dust-polarization experiment for the diffuse interstellar medium, molecular clouds, and nearby galaxies. The main astrophysical motivation is that polarized thermal dust emission is simultaneously a tracer of aligned grains and magnetic-field geometry and a dominant foreground for cosmic-microwave-background polarization studies. The $240\,\mu\mathrm{m}$ band lies near the peak of the modified-blackbody spectrum of Galactic dust at about $17$ K, so it complements lower-frequency polarized measurements such as Planck $353\,\mathrm{GHz}$ by constraining both the dust spectral energy distribution and the wavelength dependence of polarization (Mangilli et al., 2018, Misawa et al., 2014).

A central question motivating PILOT is whether the dust polarization fraction varies with wavelength. This matters because different dust models predict different FIR-to-submillimeter behavior, and because any mismatch between high-frequency dust polarization and the frequencies targeted by CMB experiments propagates directly into foreground-model uncertainty. The experiment was therefore designed to characterize polarized dust emission in regions that range from the Galactic plane to diffuse high-latitude fields, with explicit relevance to CMB foreground removal as well as to ISM physics (Misawa et al., 2014).

The project also had a demonstrator function. Future CMB polarization missions require very large detector counts and stringent intercalibration, and PILOT was explicitly framed as a test-bed for using filled, multiplexed bolometer arrays in polarimetric mode. This technological objective is inseparable from the science case, because the same instrumental systematics that limit precision FIR polarimetry also limit the credibility of foreground templates for cosmology (Misawa et al., 2014).

2. Instrument architecture and polarimetric measurement principle

PILOT is built around an off-axis Gregorian telescope. The in-flight description gives an off-axis parabolic primary and off-axis ellipsoidal secondary satisfying the Mizuguchi–Dragone condition, with a used primary aperture of about $0.73\,\mathrm{m}$, effective focal length $1790\,\mathrm{mm}$, and a field of view of about $1.0^\circ \times 0.8^\circ$ (Mangilli et al., 2018). A rotating half-wave plate (HWP) near a Lyot stop modulates polarization, and a fixed wire-grid analyzer at $45^\circ$ splits orthogonal linear polarizations onto two focal planes, TRANS and REFLEX. Each focal plane contains multiplexed bolometer arrays, for a total of $1.2$0 detectors, cooled to about $1.2$1 by a closed-cycle $1.2$2He refrigerator; the cold optics are at $1.2$3, and the detectors derive from Herschel/PACS technology (Mangilli et al., 2018).

The detector model used for polarimetry is written, in instrument coordinates, as

$1.2$4

where $1.2$5 is detector response, $1.2$6 optical transmission, $1.2$7 an electronic offset, and the sign is $1.2$8 for REFLEX and $1.2$9 for TRANS. In sky coordinates the same model becomes

$1.25\,\mathrm{THz}$0

with $1.25\,\mathrm{THz}$1, where $1.25\,\mathrm{THz}$2 is the parallactic angle (Mangilli et al., 2018). This formulation makes explicit that the experiment measures linear polarization through a known HWP modulation law plus detector-dependent response and offset terms.

The observing strategy uses a stepped HWP rather than fast continuous rotation. The pre-flight rationale was that the instrumental background can be as much as $1.25\,\mathrm{THz}$3 times larger than the polarized astrophysical signal, so HWP-position-dependent background terms are better handled as scan-to-scan offsets than as rapidly modulated signals vulnerable to minute HWP transmission anisotropies. This choice also matches the slower multiplexed bolometer readout. A dedicated stellar sensor provides pointing reconstruction accurate to a few arcseconds even at scan rates up to $1.25\,\mathrm{THz}$4/s, and an internal calibration source (ICS) behind mirror M3 illuminates all detectors simultaneously to monitor responsivity in flight (Misawa et al., 2014, Mangilli et al., 2018).

Pre-flight tests established several instrument properties that remained important for later data analysis. The HWP was found to be consistent with an ideal half-wave plate, with no significant differential transmission between fast and slow axes and phase-shift parameters $1.25\,\mathrm{THz}$5 on the transmission side and $1.25\,\mathrm{THz}$6 on the reflection side. The optics also induce a focal-plane-dependent polarization-angle rotation of about $1.25\,\mathrm{THz}$7, small but not negligible, and therefore requiring calibration in science analysis (Misawa et al., 2014).

3. Flights, observing campaigns, and in-flight characterization

PILOT flew from Timmins, Ontario, Canada, in September 2015, from Alice Springs, Australia, in April 2017, and again from Timmins in 2019. The in-flight performance paper notes an internal inconsistency for flight #2: the main text states April 2018, whereas the abstract and conclusions state April 2017; later PILOT analyses use April 2017 (Mangilli et al., 2018, Bernard et al., 2022). Flight #1 lasted 24 h and yielded about 15 h of science data but suffered a major front-baffle straylight problem after sunlight exposure. Between flights, the instrument underwent cryogenic, baffling, and scan-strategy improvements, including a variable-elevation/arbitrary scan-angle mode (Mangilli et al., 2018).

The second flight provided the main in-flight performance characterization. Detector time constants were estimated from both cosmic-ray glitches and ICS OFF decays. Most arrays had $1.25\,\mathrm{THz}$8, while array #6 was slower at about $1.25\,\mathrm{THz}$9. The inferred intrinsic ICS time constant was $240\,\mu\mathrm{m}$0, and final detector time constants averaged about $240\,\mu\mathrm{m}$1, again with array #6 slower at $240\,\mu\mathrm{m}$2. These constants were used to deconvolve timelines, and planetary beam measurements showed that the resulting point-spread function was not scan-smeared (Mangilli et al., 2018).

The measured beam at $240\,\mu\mathrm{m}$3 had an average FWHM of $240\,\mu\mathrm{m}$4, larger than the nominal optical beam of $240\,\mu\mathrm{m}$5 but consistent with simulations once convolution by the square $240\,\mu\mathrm{m}$6 detector pixels was included; the simulated value was $240\,\mu\mathrm{m}$7. The beam was described as characteristically boxy rather than temporally smeared, with somewhat stronger wings than the optical simulation, an effect already seen on the ground (Mangilli et al., 2018).

The same flight also quantified instrumental background, responsivity, and noise. The in-flight background near field center was estimated at $240\,\mu\mathrm{m}$8–$240\,\mu\mathrm{m}$9, higher than the pre-flight photometric prediction of $17$0, and rose by about a factor of two from center to corners. During HWP stepping the background showed a polarized component of about $17$1 ADU, or roughly $17$2, corresponding to about $17$3 of the total background, with a polarization angle peaked near $17$4 in the instrument frame. ICS-based responsivity varied by about $17$5 over time in flight #2; a linear model based on housekeeping parameters reproduced these variations with median residuals around $17$6. The spatial flat field derived from atmospheric loading was stable at the $17$7 level or better, with $17$8 relative dispersion between skydip-based and full-flight maps (Mangilli et al., 2018).

Noise performance was judged satisfactory. The focal-plane median high-frequency noise level in flight #2 was $17$9, with array #6 reaching $353\,\mathrm{GHz}$0. On that basis, the sensitivity study predicted a polarization-fraction signal-to-noise ratio of about 10 in the weakly polarized Galactic-center region at $353\,\mathrm{GHz}$1 resolution, about 6 for bright $353\,\mathrm{GHz}$2 Ophiuchi sources at $353\,\mathrm{GHz}$3 resolution, and about 16 integrated over the whole diffuse BICEP field assuming $353\,\mathrm{GHz}$4 polarization (Mangilli et al., 2018).

4. Calibration logic, mapping strategy, and the role of Jupiter

PILOT’s science analysis depends on calibrating detector response, atmospheric loading, time constants, and instrumental polarization in a consistent timeline-based framework. Before the later pipeline additions, the processing already corrected detector responses, atmospheric residuals, internal calibration source variations, and detector time constants. What remained problematic were distortions of the measured point-spread function and false polarization signals produced by electronic effects and by leakage from total intensity into polarization (Bernard et al., 2022).

Jupiter became the central calibrator for these remaining effects because, at PILOT’s beam size of about $353\,\mathrm{GHz}$5, it is effectively point-like and can be treated as unpolarized for this purpose. During flight #3, PILOT observed Jupiter near maximum elevation of about $353\,\mathrm{GHz}$6 for roughly 30 minutes, using eight HWP positions and two scan angles. The data were mapped with the polarization version of Scanamorphos, and for systematics studies the maps were projected in instrument coordinates rather than sky coordinates so that the beam would not be smeared by sky rotation. High-resolution maps with $353\,\mathrm{GHz}$7 pixels were produced, including per-array and per-observation PSFs (Bernard et al., 2022).

This Jupiter strategy is methodologically important. It allows instrumental polarization to be measured empirically as non-zero $353\,\mathrm{GHz}$8 and $353\,\mathrm{GHz}$9 around an intrinsically unpolarized source, and it supports PSF-based modeling of leakage. It also isolates electronic artifacts that are tied to the readout geometry rather than to astrophysical structure. A plausible implication is that PILOT’s science-ready polarization fidelity is limited less by the abstract modulation formalism than by the quality with which these PSF-level and timeline-level systematics are removed (Bernard et al., 2022).

5. Science-ready pipeline: cross-talk, readout latency, and intensity-to-polarization leakage

The major pipeline maturation described for PILOT consists of three added corrections: detector cross-talk, readout latency or memory effects, and intensity-to-polarization leakage. These were demonstrated on Jupiter from flight #3, but the correction strategy was stated to apply to all PILOT data (Bernard et al., 2022).

The cross-talk effect appears in Jupiter intensity maps as a diagonal linear extension of the beam along the detector-line direction. It was interpreted as electrical cross-talk between pixels read simultaneously into a common buffer unit. The adopted model transfers a fraction $0.73\,\mathrm{m}$0 of the signal from pixel $0.73\,\mathrm{m}$1 to pixel $0.73\,\mathrm{m}$2 along a line, with $0.73\,\mathrm{m}$3, and corrects detector data through

$0.73\,\mathrm{m}$4

A single scalar $0.73\,\mathrm{m}$5 was solved for per array from Jupiter maps. The fitted values were all just below $0.73\,\mathrm{m}$6: $0.73\,\mathrm{m}$7, $0.73\,\mathrm{m}$8, $0.73\,\mathrm{m}$9, $1790\,\mathrm{mm}$0, and $1790\,\mathrm{mm}$1 for the operational arrays. In diagnostic regions the effect amplitude fell from $1790\,\mathrm{mm}$2 of the PSF peak to about $1790\,\mathrm{mm}$3 after correction (Bernard et al., 2022).

The readout-latency correction addresses a memory effect in the time-multiplexed electronics. Without glitch masking, Jupiter maps show a ghost source one array away along the column direction because signal from a bright pixel on line 16 leaks into line 1 when the multiplexing cycle resets. The correction is applied in timeline order,

$1790\,\mathrm{mm}$4

with a common coefficient for adjacent reads and a separate $1790\,\mathrm{mm}$5 for the reset from line 16 to line 1. Most $1790\,\mathrm{mm}$6 were found to be negative, while $1790\,\mathrm{mm}$7 was usually smaller in amplitude and mostly positive. An operationally important point is that this correction shifts the apparent planet position enough that sample coordinates must be recomputed afterward from the stellar-sensor pointing solution. Although the paper introduces cross-talk before latency, it explicitly states that latency occurs later in the chain and therefore must be corrected first (Bernard et al., 2022).

The most critical residual effect is intensity-to-polarization leakage, also called instrumental polarization. Even after electronic corrections, Jupiter $1790\,\mathrm{mm}$8 and $1790\,\mathrm{mm}$9 maps remain non-zero. PILOT models the spurious polarized signal as the convolution of the true sky total intensity with a leakage PSF measured on an unpolarized point source, following the approach introduced by Ritacco et al. for NIKA. In effective form,

$1.0^\circ \times 0.8^\circ$0

where $1.0^\circ \times 0.8^\circ$1 and $1.0^\circ \times 0.8^\circ$2 are array-dependent leakage kernels measured from Jupiter and rotated by the parallactic angle. For astrophysical targets the external intensity template is a Herschel $1.0^\circ \times 0.8^\circ$3 map, reprojected and rescaled by linear correlation with the observed PILOT intensity; for Jupiter, the reference is a synthetic Gaussian source with FWHM $1.0^\circ \times 0.8^\circ$4, corresponding to the Herschel $1.0^\circ \times 0.8^\circ$5 beam. The leakage PSFs are evaluated at discrete parallactic-angle values with a step of about $1.0^\circ \times 0.8^\circ$6, and the timeline prediction is obtained by interpolation, deprojection, and subtraction before remapping with Scanamorphos (Bernard et al., 2022).

The sequence of corrections shows that electronic cleanup is necessary but not sufficient. The Jupiter polarization fraction $1.0^\circ \times 0.8^\circ$7, derived from cumulative $1.0^\circ \times 0.8^\circ$8 and $1.0^\circ \times 0.8^\circ$9 profiles, changes as follows:

Processing stage	$45^\circ$0	$45^\circ$1
Uncorrected	$45^\circ$2	$45^\circ$3
Readout latency only	$45^\circ$4	$45^\circ$5
Readout latency + cross-talk	$45^\circ$6	$45^\circ$7
All three corrections	$45^\circ$8	$45^\circ$9

This sequence shows that the dominant false polarization is not removed by electronic corrections alone; the leakage model derived from Jupiter PSFs is the decisive step for making FIR polarization maps science-ready (Bernard et al., 2022).

6. Validation, residual limitations, and scientific significance

The headline performance result of the systematics paper is that the polarization-leakage correction is accurate to better than $1.2$00 on Jupiter during flight #3. More precisely, the residual false polarization after all corrections is $1.2$01 when integrated within the beam FWHM and $1.2$02 when integrated over the whole PSF. The often-quoted $1.2$03 is therefore best interpreted as a conservative upper bound on the residual intensity-to-polarization leakage systematic, not as a generic calibration uncertainty (Bernard et al., 2022).

The paper also states several limitations. Residual structures in corrected Jupiter polarization maps likely arise from imperfect pointing reconstruction, imperfect PSF knowledge, discretization in parallactic angle, and numerical inaccuracies in rotation, convolution, and deprojection. Ordinary reprojection routines were found insufficient; the implementation required drizzling methods and $1.2$04 pixels, about 22 times smaller than the Herschel beam, to keep numerical errors subdominant. A further caveat is that Jupiter is a point source. Because bolometric observations filter large-scale emission, leakage on extended sources should be lower than what is measured on Jupiter, so the residuals derived from Jupiter are a conservative indicator (Bernard et al., 2022).

This has direct scientific consequences. Before leakage correction, the residual instrumental polarization in Jupiter maps was of order $1.2$05–$1.2$06 over the whole PSF and $1.2$07–$1.2$08 within the beam FWHM, levels comparable to or larger than typical diffuse dust polarization fractions. Such leakage would seriously bias $1.2$09, $1.2$10, polarization fraction, and polarization-angle maps. After correction, the residual false polarization falls below roughly $1.2$11–$1.2$12, low enough to support quantitative FIR polarization studies, especially on extended interstellar structures where the effective leakage should be smaller still (Bernard et al., 2022).

In that sense, PILOT evolved from a successful balloon polarimeter into a mature science pipeline only once the remaining PSF and instrumental-polarization systematics were explicitly modeled and removed. The pre-flight characterization established optical and polarimetric viability, the in-flight studies quantified beam, noise, and calibration performance, and the later processing work converted those capabilities into science-ready polarization data products. The experiment’s significance therefore lies not only in its $1.2$13 view of Galactic dust, but also in its demonstration that high-frequency dust polarimetry with multiplexed bolometers can be made quantitatively reliable under balloon-borne observing conditions (Misawa et al., 2014, Mangilli et al., 2018, Bernard et al., 2022).