Wavelength-Multiplexed Raman Spectroscopy
- Wavelength-multiplexed Raman spectroscopy is a method that acquires a range of Raman shifts using multiple excitation and detection channels for enhanced spectral coverage and throughput.
- It employs modalities like time-encoded SRS, multi-window SRS, and multiplex CARS to optimize signal extraction, background suppression, and measurement speed.
- This approach improves chemical contrast in diverse applications—from materials characterization to biomedical imaging—by balancing resolution with rapid, parallelized data acquisition.
Searching arXiv for recent and foundational papers on wavelength-multiplexed Raman spectroscopy and closely related architectures. Wavelength-multiplexed Raman spectroscopy denotes a class of Raman measurement strategies in which spectral information is acquired through multiple wavelength channels rather than through a single fixed-excitation, single-channel readout. In the literature represented here, multiplexing is implemented in several distinct ways: by sweeping probe wavelength and mapping Raman shift into time, by rapidly switching among chemically distinct Raman windows, by using broadband Stokes fields against narrowband pumps in multiplex CARS, by collecting multiple spatial or interferometric channels in parallel, and by acquiring paired spectra at shifted excitation wavelengths for background suppression (Karpf et al., 2014). Across spontaneous Raman, stimulated Raman scattering (SRS), coherent anti-Stokes Raman scattering (CARS), resonant Raman spectroscopy (RRS), and integrated spectrometry, the unifying objective is to combine high throughput, broad or selective spectral coverage, and practical acquisition speed while preserving chemically specific contrast (Ni et al., 2021).
1. Conceptual basis and scope
In Raman spectroscopy, wavelength multiplexing refers to acquisition schemes in which many Raman shifts are accessed within one measurement sequence by exploiting wavelength diversity in the excitation field, detection architecture, or both. The surveyed works show that multiplexing is not confined to one instrumental paradigm. It includes continuous wavelength sweeps with time-domain decoding, multi-window excitation across widely separated Raman bands, broadband multiplex excitation in coherent Raman, multi-aperture or multi-core collection geometries, and dual-wavelength difference methods for suppressing non-Raman background (Karpf et al., 2014).
A central distinction is between multiplexing on the excitation side and multiplexing on the detection side. Excitation-side multiplexing appears in multi-window SRS, where a rapidly tunable fiber optical parametric oscillator accesses the fingerprint, C-D, and C-H regions; in multiplex CARS, where a broadband supercontinuum is combined with a narrowband pump; and in wavelength-dependent resonant Raman spectroscopy, where laser excitation is scanned to select distinct electronic resonances (Ni et al., 2021). Detection-side multiplexing appears in photonic-lantern spectrographs, wavelength-to-time mapping systems using multicore fibers and SPAD arrays, wide-field integral-field Raman imaging, and multi-aperture on-chip Fourier-transform spectrometers (Betters et al., 2020).
The term also spans two related but not identical practices. In nonlinear Raman microscopy, multiplexing often means simultaneous or rapidly switchable access to many vibrational frequencies. In materials characterization, “multi-wavelength Raman” often means using several excitation wavelengths to interrogate resonance behavior, depth sensitivity, clustering, or vibronic structure. The carbyne, graphene nanoribbon, and hydrogenated amorphous carbon studies are examples of this second usage (Martinati et al., 2021).
2. Multiplexing modalities
The recent literature supports at least five technically distinct multiplexing modalities.
| Modality | Operating principle | Representative paper |
|---|---|---|
| Time-encoded SRS | Continuous probe sweep maps Raman shift to acquisition time | (Karpf et al., 2014) |
| Multi-window SRS | Rapid tuning among fingerprint, C-D, and C-H windows | (Ni et al., 2021) |
| Multiplex CARS | Broadband Stokes spectrum interrogates many vibrational lines simultaneously | (Mansuryan et al., 2023) |
| Shifted-excitation Raman difference spectroscopy | Two nearby excitation wavelengths produce background-suppressing difference spectra | (Korinth et al., 2020) |
| Parallel spectrometers and detectors | Multi-core, multi-aperture, or interferometric channels increase throughput and simultaneous acquisition | (Betters et al., 2020) |
In time-encoded Raman, a fixed-wavelength pump interacts with a continuously wavelength-swept probe. As the probe wavelength changes, the pump–probe energy difference scans the Raman axis, and stimulated Raman gain is recorded as a time trace that is calibrated into Raman shift. The demonstrated TICO-Raman system used an all-fiber Fourier Domain Mode Locked probe laser and reported broadband coverage from to , a broadband survey resolution of , a zoomed-in resolution of , and $1565$ spectral points in the broadband survey (Karpf et al., 2014).
In multi-window SRS, the goal is not dense continuous coverage of one narrow spectral span but rapid access to chemically complementary windows. The fiber-laser platform based on a fiber optical parametric oscillator reported a spectral tuning range of $1050$ to , spectral resolution of about , and tuning speed between arbitrary wavenumbers of . This enabled cross-window comparison of the C-H region, which reports general biomass composition, the C-D region, which reports deuterium-labeled metabolic activity, and the fingerprint region, which contains chemically specific bands such as lipid unsaturation and cholesterol/sterol signals (Ni et al., 2021).
In multiplex CARS, multiplexing arises from a broadband Stokes field interrogating many vibrational frequencies against a narrowband pump. A few-mode microstructured-fiber source demonstrated a self-referenced architecture in which one spatial mode generated a supercontinuum spanning to 0 while another mode preserved a nearly monochromatic 1064 nm component to serve as the pump. This configuration eliminated the external delay line otherwise required when residual pump light and supercontinuum are generated in separate optical paths (Mansuryan et al., 2023).
In shifted excitation Raman difference spectroscopy, multiplexing is implemented through a paired-wavelength measurement. Two closely spaced laser lines are used so that Raman bands shift on the detector while fluorescence, room light, and most background features do not. The difference spectrum
1
suppresses the non-shifting background. Demonstrated excitation pairs were 2 and 3, corresponding to about 4 and 5 shifts, respectively (Korinth et al., 2020).
A further modality appears in burst-driven stimulated Raman spectroscopy, where the Raman shift is scanned not by wavelength tuning but by digitally stepping the offset phase 6 of amplified femtosecond pulse bursts. The direct mapping between burst offset phase and Raman-shift frequency was proposed to provide hyper spectral resolution of 7 with a pixel dwell time of 8, giving a resolution–speed product of about 9 (Hu et al., 2023). This suggests a spectrally multiplexed SRS regime in which phase-programmable comb structure replaces conventional wavelength detuning.
3. Optical architectures and source technologies
A major branch of wavelength-multiplexed Raman instrumentation aims to reconcile multimode collection efficiency with diffraction-limited spectral analysis. A compact spectrograph built around a 19-core multi-core fiber photonic lantern exemplifies this direction. In that device, a tapered lantern converts a 0 multimode input into discrete outputs that form a pseudo-slit with a near-diffraction-limited 1 aperture at the spectrograph entrance. The hexagonal geometry permits rotation into a Photonic TIGER configuration, and the spectrograph uses the MCF end directly as the slit, without a complex image reformatter, slit-fiber bundle, or coded mask. The demonstrated Raman interval was 2, corresponding to 3, with a reported resolution of 4 or about 5, in a volume of about 6 (Betters et al., 2020).
A related but detection-oriented architecture is wavelength-to-time mapping of multimode light through a photonic lantern and a 121-channel SPAD/TDC array. Here a 121-core multicore fiber performs both multimode-to-single-mode conversion and dispersive propagation, so wavelength is mapped into photon arrival time according to
7
The system demonstrated multiplexed single-mode wavelength-to-time conversion with a 8 CMOS SPAD array, per-pixel timing resolution of 9, and an effective fill factor increase of about 0, compared with a physical SPAD fill factor of about 1 (Chandrasekharan et al., 2016). The authors explicitly identified Raman spectroscopy as a target application.
Integrated spectrometry provides another architectural route. A multi-aperture SiN waveguide-based Fourier-transform spectrometer fabricated on a 200 mm wafer used 160 edge-coupled apertures/interferometers in a 2 footprint. It was designed for 785 nm-excited Raman signals in the 3 collection band, achieved an experimentally characterized spectral range of 4, and delivered about 5 measured resolution, close to the 6 design target (Kerman et al., 2024). In this architecture, multiplexing is inherent to Fourier-transform detection and is amplified by the 160-aperture spatial fan-in.
On the source side, multiplexed Raman systems benefit from laser platforms that either generate multiple excitation lines or support rapid wavelength agility. A gas-filled Raman laser based on cascaded rotational SRS in hydrogen-filled nested anti-resonant fiber produced distinct lines at 1683 nm, 1868 nm, 2100 nm, and 2400 nm, with pulse energies of 7, 8, 9, and $1565$0, respectively. The line energies could be controlled by tuning hydrogen pressure from 1 bar to 20 bar (Adamu et al., 2020). A different source-oriented development demonstrated Raman-based nonlinear power combining of two Yb fiber lasers into a single output near the $1565$1 band, reaching $1565$2 total output with $1565$3 in-band, corresponding to about $1565$4 absolute conversion efficiency and about $1565$5 of quantum-limited efficiency (Aparanji et al., 2017). These source papers are not spectrometers, but they supply wavelength-agile or power-scalable excitation infrastructure relevant to multiplexed Raman implementations.
4. Signal extraction, calibration, and computational reconstruction
Because multiplexing redistributes spectral information across time bins, interferometric channels, detector pixels, or multiple core outputs, calibration and reconstruction are integral parts of the measurement itself rather than post hoc conveniences.
In time-encoded Raman, accurate recovery depends on the deterministic time-to-wavenumber characteristic of the swept probe laser. The TICO-Raman measurement chain synchronized pump pulses to known points in the probe sweep, detected probe transmission with balanced photodetection, removed residual artifacts digitally, and mapped each measurement time point to Raman shift using the sweep calibration (Karpf et al., 2014). The same system used balanced photodetection and digital balancing to suppress probe intensity offset, acoustic waves, thermal lensing, and interference artifacts.
In multi-window SRS with fiber lasers, relative intensity noise is the dominant obstacle. The auto-balanced detection scheme split the noisy beam into signal and reference arms and subtracted the electrical outputs, with a PID controller automatically balancing amplitudes. Reported performance included pump noise suppression of about $1565$6 at $1565$7, stimulated Raman loss SNR improvement of about $1565$8, and detection sensitivity better than $1565$9 DMSO (Ni et al., 2021). By contrast, stimulated Raman gain improved only about $1050$0, because the oscillator output was already much cleaner than the FOPO output.
In photonic-lantern spectrographs, the multi-core output itself becomes the multiplexing basis. The 19 simultaneous core channels of the compact Raman spectrograph were processed using an optimal extraction algorithm, calibrated with an Hg-Ar lamp, and combined onto a common linear wavelength scale (Betters et al., 2020). The measured resolving power was $1050$1 at $1050$2 and $1050$3 at $1050$4, lower than the theoretical sampling-based estimate because the telecom-designed MCF operated in a few-mode regime at the Raman wavelengths.
For wavelength-to-time mapping through a multicore photonic lantern, calibration is core-specific. In the 121-core system, arrival-time differences across the $1050$5 core array reached about $1050$6 at 532 nm, indicating nonuniform group velocity, likely from fabrication variation or spooling-induced strain. Each core therefore required an independent wavelength calibration, implemented with 4th-order polynomial fits of arrival time versus wavelength (Chandrasekharan et al., 2016). About $1050$7 of SPAD pixels were excluded because of high dark-count rates, and SPAD IRF FWHM varied from 137 ps to 174 ps.
In integrated Fourier-transform Raman spectrometry, reconstruction is a linear inverse problem,
$1050$8
where $1050$9 is the unknown spectrum, 0 is the measured interferometer output vector, and 1 is the experimentally characterized transform matrix. Because the measured matrix had dimensions 2, recovery was underconstrained, and regularization became decisive. The SiN FTS study compared pseudoinverse, Ridge, Elastic-Net, and LASSO, concluding that LASSO gave the best Raman reconstructions for sparse spectral features. The explicit LASSO formulation was
3
followed by Savitzky–Golay smoothing to suppress residual artifacts (Kerman et al., 2024).
In wide-field SERDS imaging, reconstruction has a different purpose: removal of non-Raman background rather than inversion of a multiplexing matrix. The integral-field system extracted raw data with p3d, calibrated wavelength with Hg/Ne and Ar lamps, calibrated intensity with a white-light source, and formed SERDS images by subtracting paired spectra for each spaxel. Reconstructed conventional Raman spectra were then obtained by summation or integration and baseline-corrected with the SNIP algorithm (Korinth et al., 2020).
5. Analytical and imaging applications
The demonstrated applications span compact chemical sensing, hyperspectral microscopy, tissue imaging, and resonance-sensitive materials characterization.
Portable and compact spontaneous Raman systems were illustrated by the photonic-lantern spectrograph and the integrated SiN Fourier-transform spectrometer. The lantern-based instrument recorded Raman spectra of ibuprofen, paracetamol, heat-shrink polyolefin, Blu-Tack, Norland 61 UV adhesive, spinach leaf, and an Hg-Ar calibration lamp, and the paper reported “clear signatures of multiple pharmaceuticals and other compounds detected with short integrations” (Betters et al., 2020). The on-chip FTS demonstrated Raman measurements of isopropyl alcohol, glucose, Paracetamol, and Ibuprofen with a 785 nm excitation laser; Pearson correlation coefficients up to 4 were reported for some reconstructed spectra relative to a conventional spectrometer (Kerman et al., 2024).
Hyperspectral and multi-window coherent Raman imaging broadened the application space to biology. TICO-Raman measured a solvent mixture of cyclohexane, benzene, and toluene over 5, showing good agreement with spontaneous Raman spectra and linearity in concentration, and also acquired 64-point spectra from 6 to 7 in geranium phaeum stem immersed in olive oil, separating lignin and olive oil signals about 8 apart (Karpf et al., 2014). Multi-window SRS used the C-H, C-D, and fingerprint regions to study fungi, ovarian cancer cells, and Caenorhabditis elegans. In fungi and OVCAR5 cells, LASSO unmixing of C-H hyperspectral stacks separated protein-rich and lipid-rich regions, while the metabolic-labeling ratio
9
revealed deuterium incorporation and metabolic heterogeneity (Ni et al., 2021). In C. elegans, the lipid unsaturation ratio
0
was used together with channels at 1650, 1670, and 1 to map unsaturated fatty acids, sterol-related signal, and overall lipid storage.
Wide-field SERDS imaging addressed Raman imaging under strong fluorescence and ambient background. A 2 fiber array yielded 400 simultaneous spectra, and synchronized nod-and-shuffle acquisition reduced the impact of CCD readout time and photobleaching. Demonstrations included paracetamol and aspirin tablets, mixed polystyrene and PMMA beads, and pork tissue. The paper showed that room-light bands introduced deliberately during tablet imaging were almost completely removed after SERDS subtraction, and that rapid alternating short exposures in pork tissue produced a much flatter baseline than a single long exposure of the same total dose (Korinth et al., 2020).
In multiplex CARS, the few-mode-microstructured-fiber source was applied to paraffin peaks at 1296, 1432, 2848, 2874, 2940, and 3, and multimodal imaging was demonstrated on a curcumin superparticle with M-CARS, second harmonic generation, and two-photon fluorescence (Mansuryan et al., 2023).
A second application domain is excitation-dependent materials spectroscopy. Wavelength-dependent resonant Raman spectroscopy of ultralong linear carbon chains confined in double-walled carbon nanotubes used excitation wavelengths from 400 to 800 nm in 5 nm steps. Raman maps revealed four resonances, R1–R4, in the C-mode resonance profile, and joint experiment–DFT analysis assigned them to one optical-gap transition, a second optically allowed transition, and a vibronic progression built on the higher state (Martinati et al., 2021). The reported separation of the two optically allowed states in the infinite-chain limit was 4.
Multi-wavelength Raman spectroscopy also probed structural heterogeneity in graphene nanoribbons and hydrogenated amorphous carbon. For ultra-narrow graphene nanoribbons, excitation energies from 1.57 to 2.71 eV showed that D-peak dispersion depends strongly on edge functionalization, with typical values of 5 for 4-cGNRs and 6 for 9/15-aGNRs, except for the fully brominated case, which showed about zero D dispersion (Rizzo et al., 2019). For a:C-H films, five excitation wavelengths—633, 514, 407, 325, and 266 nm—were used to track annealing-induced changes in 7, 8, 9, and 0, while comparison with EELS showed that empirical Raman models were qualitatively useful but not sufficiently reliable quantitatively under strong wavelength-dependent resonance, oxidation, and depth inhomogeneity (Lajaunie et al., 21 Jan 2025).
6. Performance limits, trade-offs, and interpretation
The surveyed literature makes clear that wavelength multiplexing does not remove the standard Raman trade-offs; rather, it redistributes them among source complexity, calibration burden, detector noise, and reconstruction assumptions.
A first recurring limitation is the balance between spectral resolution and practical throughput. In the photonic-lantern Raman spectrograph, the theoretical estimate of 1 and 2 was not achieved because the MCF, designed for telecom wavelengths, operated in a few-mode regime at 785 nm excitation, broadening the point-spread function (Betters et al., 2020). In the wavelength-to-time mapping system, spectral resolution in the visible was only about 3, limited by fiber dispersion and detector timing response (Chandrasekharan et al., 2016). In self-referenced multiplex CARS, the paraffin peak at 4 broadened from 5 FWHM in a standard setup to 6 in the self-referenced configuration (Mansuryan et al., 2023).
A second limitation is noise. Fiber-based SRS sources are compact and rapidly tunable, but their relative intensity noise can dominate the measurement unless actively balanced. The 7 SRL SNR gain obtained with auto-balanced detection in the multi-window SRS system indicates that multiplexing performance can depend as much on noise engineering as on wavelength agility (Ni et al., 2021). Photon-starved operation is likewise evident in wavelength-to-time mapping, where acquisitions of 480 s were required and count rate per SPAD was about 0.0002 photons per pulse (Chandrasekharan et al., 2016).
A third limitation is reconstruction bias. The on-chip FTS study found that pseudoinverse recovery produced mirrored artifacts, while LASSO plus Savitzky–Golay smoothing improved Raman reconstruction but could slightly reduce effective resolution (Kerman et al., 2024). In SERDS, the choice of excitation shift is intrinsically sample dependent: if the wavelength shift is too large for narrow Raman bands, reconstruction broadens features and introduces artifacts; if too small, the difference signal becomes less informative (Korinth et al., 2020). In multi-wavelength Raman of a:C-H, 8 was linear with hydrogen content only for 633, 514, and 407 nm excitation; for 325 and 266 nm, the relation broke down because of strong UV enhancement of the G-band Raman cross section (Lajaunie et al., 21 Jan 2025).
A fourth issue concerns how “multiplexed” should be interpreted physically. In coherent Raman, multiplexing often means simultaneous access to many vibrational frequencies in one optical event. In resonant Raman materials studies, it usually means excitation-energy selectivity across a set of measurements. The latter does not deliver simultaneous spectral channels, but it can reveal otherwise hidden electronic structure, as shown by the carbyne resonance profiles and the excitation-dependent D and G dispersions of graphene nanoribbons (Martinati et al., 2021). This suggests that wavelength multiplexing is best treated as a family resemblance concept rather than a single instrumental recipe.
Across the surveyed works, the dominant trajectory is toward compactness, parallelization, and algorithmically mediated spectroscopy. Fiber-based swept sources, photonic-lantern pseudo-slits, SPAD-based wavelength-to-time mapping, wide-field SERDS, and on-chip Fourier-transform spectrometers all pursue the same practical goal: to increase the chemical information obtained per unit acquisition time and per unit instrument volume (Karpf et al., 2014). A plausible implication is that future Raman systems will increasingly combine multiple multiplexing layers—source agility, spatial parallelism, and computational inversion—rather than relying on any one mechanism alone.