The telescope collected light. What will we measure?

One camera shows a patch of a distant galaxy. Another instrument produces a band with dark and bright lines. A third finds that the signal depends on a polarizer’s rotation. The mirror could have been the same, yet the records answer different questions. Where did the familiar photograph go, and where did the extra information come from?

The telescope collects and directs radiation; the scientific instrument selects a property to separate or compare; the detector converts an interaction with radiation into a record. An image preserves spatial distribution, photometry measures brightness in a defined band, a spectrum separates wavelengths, and polarimetry compares polarization states. A larger pixel count does not by itself provide all these capabilities.

We will follow an absorbed photon to a digital count, examine different receiver structures, and see what a filter, spectrograph and polarization analyzer do before light reaches them. Useful prerequisites are signal, resolution and noise and focus, eyepiece and camera. The visible mathematical branches are optional: they check the examples without hiding the main explanation.

From the eye to plates and electronic arrays

The eye gave an immediate impression, while a drawing preserved the observer’s work. Photography made it possible to accumulate light’s effect during an exposure and revisit the material later. The development of dry gelatin plates in the 1870s simplified astronomical photography. Large glass-negative archives allowed comparisons across decades, including changes in stellar brightness. History of the Harvard Plate Stacks collection.

A plate did not output a charge table: after exposure and development, its photographic darkening had to be measured and related to brightness. Emulsion sensitivity and development conditions also affected the result. The DASCH authors on photographic plate processing. An electronic receiver changed the readout method itself: electronics could measure the accumulated signal and store it numerically.

In CCD history, conception and demonstration should be separated. The Nobel material reproduces Willard Boyle and George Smith’s notes dated September 8, 1969, containing the initial sketch, and dates the video-camera CCD demonstration to 1970. Controlled transfer of charge packets became the underlying mechanism. Nobel Committee, pp. 5–6.

Another direction was an active pixel with local charge-to-voltage conversion and integrated electronics. A JPL announcement dated June 20, 1995 describes Eric Fossum’s team’s CMOS Active Pixel Sensor for compact, low-power systems. This is a milestone in a particular development, rather than a claim that every CMOS device originated then or that CCDs immediately stopped being useful. JPL’s original announcement.

The same source can be described by different data

An image records where the signal appears on the receiver plane. Summing calibrated source counts in a selected spectral band and accounting for the background gives a photometric measurement. Repeating it over time gives a light curve. A simple filter passes a wavelength range; a pixel behind it generally does not report each visible photon’s wavelength separately.

A spectrum shows how signal is distributed over wavelength. Lines help investigate composition, physical conditions and motion. STIS’s long slit, for example, compares spectra from different parts of an extended object. Polarization adds information about the electromagnetic field’s state: signal changes after an analyzer help investigate scattering, dust geometry or, with appropriate models, magnetic fields. It is not simply another pixel color. NASA: the data STIS produces, STScI: polarimetric measurements.

Three views of one hypothetical source

Three views of one hypothetical sourceOur synthetic point source has an integrated flux of 200 photons/s, ideal QE=1 and no noise. Left: a specified normalized discrete 9×9 Gaussian spot, σ=1 pixel, sum 200, rather than a diffraction PSF. Middle: two infinitely narrow lines at 499.85/500.15 nm, 100 photons/s each, after a Gaussian response with FWHM=0.5 nm (R=1000 at 500 nm); this is spectral density, not the height of each input line. Right: I=200,Q=40,U=20,V=0 photons/s, the same polarization fraction in both lines, and an ideal linear analyzer. Circles mark angles 0°/45°/90°/135° and rates 120/110/80/90. Exposure and channel throughput are matched. This is neither a real star nor a random count realization, and differs from the single-pixel noise budget below.ImageFilter → arraySpectrumDispersion → arrayPolarizationAnalyzer → arrayx · px9 × 9 · Σ = 200 s⁻¹0200400499500501λ · nmR = 10000701400°90°180°α

More contrasting cells indicate a greater signal. Expected count rates: spot center 31.831 photons/(s·pixel), whole source 200 photons/s. Spectrum: photons/(s·nm); polarimeter: photons/s. Analyzer angles: 0°→120, 45°→110, 90°→80, 135°→90.

Our synthetic point source has an integrated flux of 200 photons/s, ideal QE=1 and no noise. Left: a specified normalized discrete 9×9 Gaussian spot, σ=1 pixel, sum 200, rather than a diffraction PSF. Middle: two infinitely narrow lines at 499.85/500.15 nm, 100 photons/s each, after a Gaussian response with FWHM=0.5 nm (R=1000 at 500 nm); this is spectral density, not the height of each input line. Right: I=200,Q=40,U=20,V=0 photons/s, the same polarization fraction in both lines, and an ideal linear analyzer. Circles mark angles 0°/45°/90°/135° and rates 120/110/80/90. Exposure and channel throughput are matched. This is neither a real star nor a random count realization, and differs from the single-pixel noise budget below. Source and limits.

The hypothetical source in the figure has the same total flux in all three views. Yet a two-dimensional spot, a spectral curve and an angle-dependent signal organize information differently. These are synthetic expected signals from an ideal instrument, rather than three real observations of a star.

A pixel collects charge carriers, not little pictures

An absorbed photon in a photodiode can create a mobile electron and hole. An electric field helps separate and collect the carriers before they recombine. For visible light in a silicon receiver, we often count collected photoelectrons. A hole means a missing electron behaving as a mobile positive carrier, rather than a physical opening in the chip.

Accumulated charge changes a node’s electrical state. The readout circuit converts this change into a voltage; an analog-to-digital converter, ADC, records a numerical count. Photon, electron and digital count ADU/DN are therefore different units at different stages. Equal ADU values from two cameras do not guarantee equal charge or sky flux. Conversion gain, exposure time and calibration are needed. STScI: photodiode, field and voltage.

In a hybrid infrared detector, the sensitive layer and silicon readout can be separate components. Individual indium contacts connect their pixels. The photon is absorbed in the active material; an electrical signal then reaches the electronics. Drawing that same photon traveling through every layer to the digital cable would be wrong. NASA: Webb’s hybrid detector layers.

A hybrid: absorber, contacts and readout

A hybrid: absorber, contacts and readoutOur functional hybrid photodiode diagram after NASA and WFC3/IR. HgCdTe absorbs photons; In connects pixels to the silicon CMOS circuit. Rust light arrows end in the absorber; the downward output is an electrical signal, and e⁻ in the chain denotes equivalent electron charge. Thicknesses, contact and pixel sizes are not to scale; internal layers, contacts and fields are incomplete. This is not a full MIRI Si:As IBC cross-section or a fabrication recipe. The lower chain separates physical units rather than tracing the same photon through electronics.LightHgCdTeAbsorberIn · ContactsCMOS · SiReadoutElectrical signal
  1. Incident photons · photons/s
  2. Collected charge · e⁻
  3. Node voltage · V
  4. ADC → digital count · ADU
Our functional hybrid photodiode diagram after NASA and WFC3/IR. HgCdTe absorbs photons; In connects pixels to the silicon CMOS circuit. Rust light arrows end in the absorber; the downward output is an electrical signal, and e⁻ in the chain denotes equivalent electron charge. Thicknesses, contact and pixel sizes are not to scale; internal layers, contacts and fields are incomplete. This is not a full MIRI Si:As IBC cross-section or a fabrication recipe. The lower chain separates physical units rather than tracing the same photon through electronics. Source and limits.

A CCD transfers charge; active CMOS reads pixels locally

CCD electrodes establish potential wells that collect charge. Changing the voltages moves packets toward an output: a row enters a register, and charges reach an amplifier from there. It resembles orderly transfer of portions, but charge carriers move inside the semiconductor, rather than the cells themselves. Real CCDs may have multiple outputs. Charge traps and incomplete transfer distort weak signals. The manufacturer: CCD structure and transfer.

In a typical active CMOS pixel, a photodiode is connected to a local charge-to-voltage circuit. Switches select pixels, while column amplifiers and ADCs can work in parallel. The entire frame need not travel through a long charge-transfer chain to a single output. Column-parallel readout does not, however, mean that every pixel in every CMOS camera starts and finishes its exposure simultaneously. Shutter behavior and readout mode must be checked separately. Hamamatsu: active pixels and column electronics.

CMOS primarily names an integrated-circuit technology; in cameras it commonly names the corresponding sensor architecture. It does not specify a unique absorber material. A HgCdTe hybrid, for instance, can have a CMOS readout circuit. Asking “infrared or CMOS?” therefore mixes separate characteristics. Compare actual quantum efficiency, noise, uniformity, speed and saturation in the required mode. STScI: HgCdTe combined with a CMOS MUX.

Material and fabrication determine what reaches the signal

A silicon device needs controlled semiconductor regions, insulating layers and ordered electrodes or transistors. Their geometry determines charge collection, storage and readout. Light’s route to the active region also matters: front-side electronics can obstruct it. Back illumination after suitable thinning opens another route; antireflection coatings reduce reflection, while active-layer thickness affects absorption. None of these methods provides identical efficiency at every wavelength. CCD structure, front and back illumination, the manufacturer on sensitivity and coatings.

WFC3/IR provides a concrete hybridization example. Its HgCdTe was grown by molecular-beam epitaxy on ZnCdTe; implanted regions and contacts were made after growth. A separate CMOS circuit was joined to the detector through one indium contact per pixel; the substrate was removed after hybridization. This is a documented sequence for that product, rather than a recipe with arbitrary doping doses and temperatures for any array. WFC3: materials, growth and hybrid assembly.

Integration must be followed by signal checks: wavelength-dependent sensitivity, differences between pixels, exposure dependence, dark current and read noise in a specified mode. A faulty contact, carrier trap and nonuniform conversion gain are different causes of a poor exposure. Adding cells cannot replace fabrication of the active layer and subsequent calibration.

IR arrays, microchannels and scintillators solve different problems

Near-infrared detectors include HgCdTe: changing the mercury-to-cadmium ratio adjusts its long-wavelength sensitivity limit. Webb uses such near-IR hybrids, whereas MIRI uses arsenic-doped silicon, Si:As. These are impurity-band-conduction devices with sensitive and blocking layers, different from near-IR HgCdTe photodiodes. A field directs freed carriers toward readout, while the structure limits unwanted dark signal. Cooling is also part of the instrument. Hybrids may allow repeated nondestructive reads before reset, following signal accumulation and sudden cosmic-ray jumps. Nonlinearity and persistence after strong illumination still require separate checks. NASA: Webb’s different absorbers, STScI: MIRI’s Si:As IBC. The band is explored further in the infrared telescope article.

In the ultraviolet, individual events can be amplified separately. A STIS MAMA photocathode releases a photoelectron; a microchannel plate multiplies electrons, and an anode system locates the pulse. FUV uses CsI on the MCP entrance face, while NUV uses Cs₂Te on the inside of the entrance window. This is event counting, with its own rate and brightness limits, rather than an unlimited CCD charge well. Events can accumulate into an image or be recorded with coordinates and times. STScI: MAMA structure and modes.

In a scintillator, deposited energy first creates a light flash, read by another receiver. Fermi LAT’s CsI(Tl) calorimeter crystals are read by photodiodes; together with the tracker, they help estimate the original gamma photon’s energy and direction. Optical photons inside the crystal are secondary signals. NASA: LAT detectors.

These examples do not cover every wavelength receiver. ALMA, for example, has cooled radio-frequency assemblies, conversion and digitization before the correlator. Its measurement chain differs from direct optical imaging. The operator: ALMA Front End and Back End. Selection therefore starts with the radiation and information that must be preserved.

Absorption does not guarantee a useful count

Quantum efficiency, QE, describes the conversion of incident photons into registered signal. In our simple optical example, QE=0.8 means an 80% probability of obtaining a collected photoelectron from an incident photon. High-energy photons or a device with internal amplification require separate definitions of charge per event and quantum yield. Sensitivity depends on wavelength; a high peak QE does not describe the entire band. The manufacturer: QE and spectral sensitivity.

Sky background accompanies source light. Additional carriers may arise without illumination: dark current. Readout adds electronics noise. Even a perfectly steady mean flux has statistical fluctuations in the number of registered events. Subtracting mean background or mean dark signal helps, but cannot cancel random fluctuations in the exposure already taken.

Cooling can reduce dark current; it does not remove the source’s photon noise or guarantee the best result in every mode. Temperature, time, readout and spectral band matter together. Hamamatsu: dark current and noise types, pp. 17–18.

Branch · With equationsOptional detail; the main explanation continues below.

Count electrons and noise separately

Our example concerns one pixel: it receives Φ_(s) = 200 source photons/s and Φ_(b) = 50 sky photons/s; specified parameters are η = 0.8, dark current d = 0.1 electron/s and one-read noise r = 3 electrons rms. For exposure t, expected electron counts are:

S = ηΦ_(s)t,; B = ηΦ_(b)t,; D = dt.

Assume independent Poisson events and independent read noise, an exactly known mean background, and a linear unsaturated chain before the ADC. Then:

SNR = (S)/(√(S + B + D + (r)^(2))).

At t = 5 s, we obtain S = 800, B = 200 and D = 0.5; variance is 1009.5 electron², noise 31.773 electrons rms and SNR≈25.179. Fractional D is an average over possible exposures, not half an electron in an individual measurement. Subtracting B and D reduces the mean count but leaves their variance contributions.

Dividing the same total time into n separate frames replaces (r)^(2) with n(r)^(2). Splitting can avoid saturation or allow a damaged frame to be rejected, at the cost of repeated readout. A finite background measurement, calibration errors and ADC quantization are omitted here. This separate-frame formula is not a universal noise model for complex nondestructive readout. The three noiseless source views in the earlier figure are a different example: their 200 photons/s are spread across all spatial pixels. Noise model and components.

A bright pixel can hit two different limits

A physical storage node has finite charge capacity, its full well. Separately, the ADC has a finite code range. Conversion gain g is often specified as electrons per ADU step: fewer electrons per code make the code rise faster for the same charge. The ADC can clip the signal before the pixel physically overflows. Saturation destroys the relationship between flux and stored value: one maximum number can conceal different brightnesses.

Dynamic range compares the upper and lower useful measurement limits. In an idealized mode with little dark signal, it is sometimes estimated as full well divided by read noise. Background, nonlinearity, electronics mode and required precision also matter in a real task. File or ADC bit depth alone does not equal useful physical dynamic range. The manufacturer: saturation charge and dynamic range, pp. 15 and 18.

Pixel charge and the digital limit

Time · sg · e⁻/ADUCharge · e⁻Code · ADUSNRState
521000.5100025.179Linear
100220010.010505113.083Linear
100120010.016383—ADC clipped
Our single-pixel example: source 200 and sky 50 photons/s, QE=0.8, dark=0.1 electron/s, read=3 electrons rms/frame, well=30000 electrons, 14-bit ADC and bias=500 ADU. g is in electrons/ADU. SNR is the linear analog budget before quantization and clipping, with an exactly known mean background; after saturation it is not presented as an estimate for the stored exposure. These are expected values, not camera specifications, measurements or safe exposure settings. Source and limits.

Branch · With equationsOptional detail; the main explanation continues below.

Why 14 bits do not mean 16384 electrons

In our model, let full well be W = 30000 electrons, ADC depth b = 14 bits, conversion gain g = 2 electrons/ADU and fixed digital offset C_(0) = 500 ADU. Before clipping and rounding, the expected code for charge N is:

C = C_(0) + (N)/(g).

The largest code is 16383, so the permitted charge is limited by the smaller of two quantities:

N_(lim) = min⁡[W, g((2)^(b) − 1 − C_(0))].

With these parameters, W is limiting. Over 100 s, the previous example’s combined source, sky and dark charge is 20010 electrons, giving 10505 ADU. Changing only g to 1 electron/ADU leaves the well unfilled, but the expected code of 20510 is already clipped to 16383. The formal analog SNR after this clipping does not characterize the stored image.

At the constant mean rate of 200.1 electrons/s, W is reached after about 149.925 s; for g=1, the ADC limit is reached after 79.375 s. These are mean teaching-model thresholds, not guaranteed safe exposure settings: random departures and real-device nonlinearity require headroom. With r=3, the rough read-limited range estimate is 10000 in the first mode, and about 5294 after including the ADC in the second.

A pixel samples an already formed image

A small cell provides a finer grid on the image, but cannot correct diffraction, aberration, atmospheric blur or mount shake. If optics smeared a point, extra pixels describe the spot in greater detail. Overly coarse cells, on the other hand, can lose information the optics still transmitted.

Reducing pixel size spreads the same source’s light over more cells. They can be combined for total photometry, with noise and readout accounted for. Megapixel count, field size and optical resolution are related but different properties.

Branch · With equationsOptional detail; the main explanation continues below.

Convert pixel micrometers into arcseconds

For a central pixel of width p on a flat focal plane in an ideal central projection with focal length f, the angle between directions to its edges is:

θ_(px) = 2arctan⁡(p)/(2f) ≈ (p)/(f).

The angle is in radians; one radian contains about 206265 arcseconds. Our p = 5 μm and f = 1 m give approximately 1.031 arcsec/pixel. A 10 µm pixel at the same f gives about 2.063 arcsec/pixel; the mirror has not changed. Elsewhere in the field, distortion and projection can alter local scale. This angular step should not be confused with a spectral line’s width in nanometers.

A spectrograph organizes wavelengths before absorption

In an ordinary slit spectrograph, the telescope forms an image on the entrance slit, which selects a narrow strip of the object. A collimator turns the beam from a selected point into approximately parallel light; a dispersing element, such as a grating, directs wavelengths differently. The spectrograph camera, meaning focusing optics here, brings the separated beams onto a detector. The electronic array absorbs light already separated in space. ESO: spectrograph construction and geometry, pp. 12 and 19.

A fiber can deliver light to the instrument entrance, where its output serves as a source for the downstream optics. HARPS, for example, receives light through an object and a reference fiber. ESO: the HARPS fiber feed. A single simple slit or fiber does not automatically preserve the entire two-dimensional object. A long slit retains one spatial coordinate along itself; multielement and integral-field instruments organize additional spatial channels.

From a slit to separate focal points

From a slit to separate focal pointsOur normal-incidence transmission grating: d=10 µm,m=1; wavelengths 450/550/650 nm. Slit x=0, collimator x=100 mm, grating x=200 mm, camera x=300 mm, detector x=400 mm; both ideal paraxial lenses have f=100 mm and pupil ray heights −5/0/+5 mm. Diffraction angles are 2.579°/3.153°/3.727°, all <4°; every color focuses independently of pupil height. The upper geometry has equal x/y scales: 1.8 SVG units/mm. Below is a separate enlarged one-dimensional focal y scale (80 SVG units/mm), rather than continued rays, at 4.505/5.508/6.514 mm. Lens and grating symbols mark ideal planes, not real surface shapes. One order is shown; aberrations, thickness, efficiency and actual hardware are omitted. Rays are selected geometric directions, not three photons. This is not the reflective STIS/ESO layout.SlitCollimatorGratingCameraDetectorParaxial geometry · equal x and y scalesEnlarged focal scale · separate window4567450 nm · 4.505 mm550 nm · 5.508 mm650 nm · 6.514 mmy_f · mm
Our normal-incidence transmission grating: d=10 µm,m=1; wavelengths 450/550/650 nm. Slit x=0, collimator x=100 mm, grating x=200 mm, camera x=300 mm, detector x=400 mm; both ideal paraxial lenses have f=100 mm and pupil ray heights −5/0/+5 mm. Diffraction angles are 2.579°/3.153°/3.727°, all <4°; every color focuses independently of pupil height. The upper geometry has equal x/y scales: 1.8 SVG units/mm. Below is a separate enlarged one-dimensional focal y scale (80 SVG units/mm), rather than continued rays, at 4.505/5.508/6.514 mm. Lens and grating symbols mark ideal planes, not real surface shapes. One order is shown; aberrations, thickness, efficiency and actual hardware are omitted. Rays are selected geometric directions, not three photons. This is not the reflective STIS/ESO layout. Source and limits.

A filter can restrict unwanted wavelengths and help avoid overlapping grating orders. Slit width involves a trade-off: a narrower entrance may narrow the instrumental line, while rejecting light and increasing dependence on pointing and the atmospheric image. After assembly, slit position, collimation, camera focus, throughput and wavelength calibration against known lines are checked. Secure mechanical seating cannot replace a spectrum check. ESO: slit, seeing and detector, pp. 19–20.

Branch · With equationsOptional detail; the main explanation continues below.

Spectral spot positions follow geometry

In our diagram, light arrives normally at a transmission grating with period d_(g) = 10 μm. For order m and wavelength λ, the path difference between neighboring grooves gives:

d_(g)sin⁡β = mλ.

We show m=1 and 450/550/650 nm. Angles β are approximately 2.579°/3.153°/3.727°. With paraxial focusing camera f=100 mm, the hypothetical foci lie 4.505/5.508/6.514 mm from the axis. Model lenses act in the paraxial approximation, with angles below 4°; thickness, aberrations and order efficiency are not calculated. The diffraction angle is checked through inter-groove phase difference, and ray convergence through an independent transfer matrix.

This is not an exact copy of STIS or ESO’s reflective layout. A reflection grating uses another sign geometry; its equation cannot be transferred without defining incidence and the normal. The zeroth order gives no such separation, while other orders can overlap in wavelength. ESO: diffraction and grating equation, pp. 10–12.

Spectral grid spacing is not resolving power

Even an extremely narrow input line emerges with finite width. This response’s shape is the instrumental profile, determined by the slit, diffraction, optics and detector. A convenient width measure, FWHM, is the distance between the edges at half the peak height. At the same wavelength, a narrower response corresponds to higher spectral resolving power R. Two nearby lines can merge into one broad spot. Increasing image scale or adding software samples cannot recover a distinction already lost.

Dispersion relates position to wavelength, while sampling gives nanometers per pixel or spectral sample. Describing an instrumental profile requires enough samples, but fine spacing alone does not narrow it. The figure passes the same two hypothetical lines through two instrumental profiles, with the same total flux and shared axes. These profile parameters are specified separately from the preceding grating geometry; they are not two claimed modes of the same instrument.

The same flux, different instrumental widths

The same flux, different instrumental widthsOur input lines at 499.85/500.15 nm each carry 100 photons/s. Two normalized Gaussian instrumental responses have FWHM=0.5/0.1 nm, or R=1000/5000 at 500 nm. Axes and integrated flux of 200 photons/s are identical; density is photons/(s·nm), and peaks are not individually normalized. Fine numerical curve samples do not imply improved resolution. These are separately specified LSF models, not two modes of the preceding grating; continuum, noise, actual detector sampling and intrinsic line width are omitted.ph s⁻¹ nm⁻¹05001000499499.5500500.5501499.85500.15λ · nm

Blue: R=1000, FWHM=0.5 nm. Rust: R=5000, FWHM=0.1 nm. Vertical dashed lines mark input wavelengths; both cases retain the same total flux.

Our input lines at 499.85/500.15 nm each carry 100 photons/s. Two normalized Gaussian instrumental responses have FWHM=0.5/0.1 nm, or R=1000/5000 at 500 nm. Axes and integrated flux of 200 photons/s are identical; density is photons/(s·nm), and peaks are not individually normalized. Fine numerical curve samples do not imply improved resolution. These are separately specified LSF models, not two modes of the preceding grating; continuum, noise, actual detector sampling and intrinsic line width are omitted. Source and limits.

Branch · With equationsOptional detail; the main explanation continues below.

What exactly is in the denominator of λ/Δλ?

Spectral resolving power is often written as a dimensionless quantity:

R = (λ_(0))/(Δλ).

Δλ must be defined. Here it is FWHM, the full width at half maximum of the Gaussian response to one infinitely narrow line. This is a practical instrumental-profile convention, rather than a claim that all two-line resolution criteria are identical. ESO also describes estimating R from calibration-line FWHM. ESO Data Products Standard v6, p. 37, SPEC_RES.

At λ₀=500 nm, widths of 0.5 nm and 0.1 nm give R=1000 and R=5000 respectively. Our input lines at 499.85 and 500.15 nm are separated by 0.30 nm; they merge under the broader response, while the narrower response shows two peaks. Numerical integration of both spectra retains the same flux. A spacing of 0.025 nm would provide 20 samples across a 0.5 nm width without changing R=1000. These are spectral quantities; arcseconds per pixel describe a different coordinate.

A polarimeter compares orientations rather than coloring a frame

An ordinary intensity receiver does not fully determine light’s polarization state. Optical elements placed before it convert the difference into intensity differences. A linear polarizer passes the corresponding field projection. Rotating the analyzer provides comparable measurements; a Wollaston prism can simultaneously split two orthogonal states into different images.

A real polarimeter often has a wave plate, or retarder, before the splitter. It changes the relative phase of field components, allowing the state read by the analyzer to be modulated. EFOSC2 uses a half-wave plate for linear-polarization measurements and a quarter-wave plate for circular polarization. Simply rotating a linear polarizer does not determine the circular component. ESO: retarders and Wollaston prisms.

An angular reference, matched exposures and channel throughput are also required. Tilted mirrors and detectors may introduce instrumental polarization or phase changes. Observing known polarized and unpolarized standards helps distinguish object from instrument. That is why the entire chain needs calibration, rather than just one purchased polarizer. STScI: ACS instrumental effects.

Branch · With equationsOptional detail; the main explanation continues below.

Four analyzer angles recover I, Q and U

I describes total intensity, Q and U two linear components relative to specified directions, and V the circular component. An ideal analyzer at angle α has the expected intensity in one output:

I_(α) = (I + Qcos⁡2α + Usin⁡2α)/(2).

Adding its orthogonal companion output always gives I. Our synthetic source has I=200,Q=40,U=20 photons/s and V=0, with ideal efficiency 1. Analyzers at 0°/45°/90°/135° give 120/110/80/90 photons/s. From readings normalized to the same time and throughput:

I = I_(0) + I_(90),; Q = I_(0) − I_(90),; U = I_(45) − I_(135).

The linear-polarization fraction and angle are:

p_(L) = (√((Q)^(2) + (U)^(2)))/(I),; χ = (1)/(2)atan2⁡(U,Q).

We obtain pₗ≈0.224, or 22.36%, and χ≈13.283° from the specified zero, with a 180° ambiguity. The angle is undefined when the linear component is zero. For a fully linearly polarized state, the equation reduces to Malus’s law; unpolarized or purely circular light gives a constant half of I through a simple linear analyzer. That is why V cannot be recovered from these four numbers.

These are our ideal angles and noiseless reconstruction. ACS, for example, has nominal 0°/60°/120° polarizers and requires its own efficiency, leakage and orientation coefficients. Statistical uncertainty also matters, particularly at low polarization fractions; actual analysis is not simply substituting raw codes into these equations. STScI: IQU and calibration corrections.

Choose the question, then check the whole measurement chain

Finding a faint object needs a suitable band, low background and enough signal. Rapid variations demand appropriate timing, readout and overhead. Nearby spectral lines need a narrow, known instrumental profile and adequate sampling. Polarimetry requires comparable channels and controlled instrumental effects. No single label, “more megapixels” or “lower noise,” solves every task.

Useful acceptance separates the checks. Unilluminated frames reveal electronics background and dark signal under specified conditions. Uniform illumination checks pixel differences and linearity. Known spectral lines test scale and response width; polarization standards test the whole analyzer with the telescope. Values must be tied to temperature, mode and calibration date. Turning these measurements into a final image and color needs its own detailed discussion.

Our numbers and diagrams are teaching models, while fabrication descriptions analyze documentation. We have not physically manufactured a camera or measured a particular camera’s performance. The main distinction is clear: the telescope delivers light, while the selected measurement chain determines which properties become testable data.