Introduction
Spectrometers achieve spectral-to-spatial mapping which allows spectral decomposition of
the input light. Conventional spectrometers use dispersive media such as gratings or prisms,
and they can achieve resolving powers around R < 104, and can reach up to R < 105only
with the installation of complex triple-grating systems (the spectral resolution is δλ =λ0/R,
where λ0is the operating wavelength). However, such spectrometers generally require moving
parts (grating, mirrors) and a line array detector for scanning whole wavelengths of interest.
Moreover, inverse proportionality between spectral resolution and optical path length leads to
bulky systems when high-resolution is demanded. The fundamental need for high resolution
spectral analysis in various lines of research and applications triggers new concepts that are
built on the basis of holography [1], scattering of light by a photonic crystal [2, 3], a random
scattering medium [4], or a multimode fiber (MMF) [5, 6, 7, 8] to form a complex spatial
intensity distribution (a speckle pattern) on a multipixel detector such as a charged couple
device (CCD) or a focal plane array (FPA). In such systems, wavelengths experience different
propagation constants inside the scattering medium, thus forming distinct spatial intensity
profiles on the detector, which provides the required one-to-one spectral-to-spatial mapping.
Before the use of spectrometer, a calibration matrix is measured by scanning all wavelengths
in the operational range and it stores the corresponding speckle patterns. The calibration
matrix is then utilized to reconstruct an arbitrary input spectrum based on the measured
intensity distribution. However, the increased cost of CCD and FPA sensors especially in
infrared regime limits the deployment of high-resolution spectral analysis tools.
Among all scattering-based systems, the multimode fiber spectrometers have been particu-
larly attractive by offering high-resolutions with reduced scattering losses (keeping the light
collimated inside the fiber and preventing scattering to higher angles). Since fibers can be
wrapped, higher spectral resolution can be achieved without enlarging the system. It was
shown in [7] that high resolving powers R > 106in near-infrared regime is possible with
fibers of 100 m long. Yet the signal-to-noise ratio (SNR) is the main limiting factor for the
resolution at low signal levels and increased fiber length [9].
Single pixel detection together with compressed sensing [10, 11, 12] have been revolutionizing
imaging methods. Surprisingly, the penetration of this methods in spectroscopy have been
very limited due to mechanical resolution limits. The single-pixel imaging (SPI) systems are
based on the use of a spatial light modulator (SLM) and a single-pixel detector. Employing
the SPI method is particularly useful when working in the infrared regime since FPAs get
extremely expensive at longer wavelengths [13].
In this paper, we develop a high-resolution single-pixel multimode fiber spectrometer and
demonstrate its ability to reconstruct arbitrary spectra. The single-pixel detection is achieved
by focusing light on a selected target region of a focal plane array which is employed as
a bucket detector. The input wavefronts are structured using a spatial light modulator
which provides distinct output intensities at the detector [14, 15, 16, 17, 18]. The intensity
variations at the target position as a function of input wavelength are used to reconstruct the
spectra at a resolution of 20 pm. The increased intensity at the focused point also increases
SNR which removes low signal barrier in reaching high resolutions at low signal levels. This,
2