Spectrometer

Reports centre wavelength, detected spectral range, bandwidth, and a qualitative spectrum.

Click the spectrometer to see its live specs and try its parameters — this mini canvas can't be moved, deleted, or added to.

In the real world

A spectrometer answers one question: how is this light's power distributed across wavelength? That is enough to characterise a laser or an LED, to check the channels of a wavelength-division-multiplexed link and their signal-to-noise ratios, to measure a component's transmission by comparing spectra taken with and without it, and to read the wavelength-dependent gain and noise figure of a fibre amplifier^[1].

What is worth knowing is that instruments answering that one question are built on several quite different principles, and the principle decides what the instrument is good at^[1].

Spectrograph

A grating disperses the light and a detector array — a photodiode array, or a linear CCD — catches all the wavelengths at once. Nothing moves, so acquisition is fast, and the resolution is set by the detector rather than by the optics. The costs are that spatially resolving detectors exist only for limited spectral regions, poorly into the infrared, and that stray light inside the instrument caps the dynamic range^[1]. This is the compact instrument most people mean by "a spectrometer", and the pattern behind the small, inexpensive designs that put a grating and a scanning reflector in a package a few centimetres across^[3].

Scanning monochromator

Rather than catching every wavelength at once, send the light through a tunable bandpass filter and measure the transmitted power with a single detector, sweeping the filter across the range of interest^[1]. The filter is a grating monochromator — Czerny–Turner, typically — turned by a precise motor, and the resolution is set by its slit width and grating.

This is how high-performance instruments are built, and the reason is dynamic range. One monochromator manages perhaps 30 dB, because strong light at one wavelength scatters inside it and lifts the reading everywhere else. Two in series, held on the same wavelength, reach beyond 70 dB^[1]. The price is time: a sweep takes longer for finer resolution, and longer again for more sensitivity.

Fourier transform

A Michelson interferometer measures something else entirely — the output power against arm-length difference — and Fourier transforms it^[1]. Monochromatic light gives a sinusoid whose period is the wavelength, which is how a wavemeter works. Here the resolution is set by how far the arm was scanned, not by any slit: the wavenumber resolution is simply the inverse of the path-difference range, so a 15 mm scan gives about 10 GHz, roughly 0.03 nm at 1 µm^[1].

Its weakness is instructive. A strong line does not produce a perfectly clean sinusoid, and the noise on it transforms into a background spread across the whole spectrum — so sensitivity to a weak line gets worse when a strong one is present, and no amount of extra scan range fixes it. Dynamic range lands around 30–40 dB^[1].

Acousto-optic

A diffraction grating is not the only way to disperse light. A Bragg cell driven by a surface acoustic wave diffracts each optical frequency to its own angle, and integrated-optic spectrum analysers were built on exactly that — a guided wave interacting with a surface acoustic wave on a single chip^[5]. The same interaction, run as a filter rather than as a disperser, is the AOTF.

\text{PSD}(\lambda) = \frac{P_{\text{measured}}}{\text{RBW}}

Power spectral density is the measured power divided by the resolution bandwidth the instrument was set to. It is the honest vertical axis, and the one that lets a narrow line and a broad band be compared.

That division is subtler than it looks, and it is the single most common way of misreading a spectrum. An analyser's vertical axis often shows measured power, not power spectral density^[1]. Converting between them means dividing by the resolution bandwidth — but the calibration is usually done for quasi-monochromatic light, and when the bandwidth is quoted as a full width at half maximum, how well power-divided-by-bandwidth matches the true density depends on the shape of the instrument's filter^[1]. Log scales in dBm are common precisely because the interesting range spans orders of magnitude.

One warning from the same source is worth repeating: a spectrum analyser is not the instrument to measure optical power with. Coupling efficiencies in the delivery path are rarely known well enough. Use a power meter^[1].

In OpticalSetup

The Spectrometer reports the centre wavelength, the detected range, the bandwidth, and a plotted spectrum of everything reaching its face. Wire it to a Detector screen to see the spectrum drawn.

Its vertical axis offers exactly the choice above. Spectral density is the honest one — power per nanometre, so a band's height does not depend on how finely it happened to be sampled. It carries the consequence that makes real instruments awkward too: a laser line has no width of its own, so it is spread over a nominal 0.1 nm to give it a height at all, and it then towers over any continuum beside it. That is what a real spectrometer shows, and it is useless when the point is to see a weak Raman line next to its own pump — so relative mode scales each source to its own peak instead.

Two behaviours are worth knowing because they were built deliberately. Bands that do not touch stay apart. One source can arrive carrying several disjoint bands — an AOTF selecting three lines out of a supercontinuum is the standard case — and each is measured and drawn on its own, rather than being summarised across the gaps between them into a single smear. Overlapping passbands are one band, correctly, and the grid inside it is fine enough to keep whatever structure it has: several narrow lines cutting a pulsed laser's envelope come back as separate peaks whose heights still trace that envelope.

And the axis is sized from the measurement, spanning whatever clears a thousandth of each feature's own peak. Per feature, not against one global maximum — otherwise a line's towering density would push a perfectly real broadband source off the plot for the crime of sharing a detector with a laser. A manual range is available when a fixed window is wanted.

Simplified vs. reality

This is not an instrument, it is a readout. There is no monochromator, no slit, and so no resolution bandwidth: a real spectrometer shows the true spectrum convolved with its own filter function, and reports something broader than reality for anything narrower than that filter. Here the modelled spectrum is reported directly. Nothing sets a sweep time, and there is no distinction between a spectrograph, a scanning instrument and a Fourier-transform one — all of which would answer differently.

There is no dynamic range and no noise floor. Stray light does not exist, so the 30 dB that limits a single monochromator and the 30–40 dB that limits a Fourier-transform instrument have no counterpart, and a weak line beside a strong one is read as easily as if it were alone — which is precisely the measurement real instruments find hardest. There is no logarithmic or dBm scale.

The plotted samples are a display budget, not a physical resolution, and wavelengths are keyed to 0.1 nm, so two lines closer together than that are reported as one. Readings are fractions of a source's emitted power rather than absolute values in watts; for power, use the power meter, which is the advice for real instruments too.