---
title: Fourier Plane Tomographic Spectroscopy
url: https://www.emergentmind.com/topics/fourier-plane-tomographic-spectroscopy
type: topic
---

# Fourier Plane Tomographic Spectroscopy

Fourier Plane Tomographic Spectroscopy denotes a class of measurement and reconstruction strategies that combine spectroscopic information with tomography in a Fourier-domain representation. In back-focal-plane microscopy and spectroscopy, it is an angle-resolved, wavelength-resolved measurement carried out at the back focal plane (BFP, or Fourier plane) of a high-NA microscope objective, where the primary observable is the momentum-resolved light distribution \(I(k_x,k_y,\lambda)\) [1806.08280]. In recent single-particle scattering work, the same concept has been extended to simultaneous four-dimensional characterization of scattering as a function of wavelength, incident direction, scattering direction, and polarization, yielding a scattering signature \(S(\lambda,k_i,k_s,p)\) [2507.15760]. Closely related diffraction-tomographic formulations use Fourier-domain mappings on Ewald-sphere manifolds and explicit filtered backpropagation formulas to reconstruct scattering potentials from wave measurements [2407.01793]. A distinct, but terminologically adjacent, usage appears in standoff FPA-FTIR plume sensing, where the Fourier operation is the interferogram-to-spectrum transform and the tomographic output is assimilated as a concentration contour in a PDE-constrained inverse problem [2606.11840]. Rotational slitless spectroscopy extends the same tomography–spectroscopy logic to wide-field datacube recovery through repeated angular projections and matrix inversion [2605.23814].

## 1. Terminology and conceptual scope

In the BFP-based literature, the Fourier plane is the objective pupil plane that encodes angular or wavevector information. Each point in the pupil corresponds to a unique emission or scattering direction, so a BFP image is a k-space image rather than a real-space image. On that basis, Fourier Plane Tomographic Spectroscopy measures the angular spectrum of light together with wavelength, and uses forward modeling or inversion to infer structural, orientational, or modal properties of the source or scatterer [1806.08280].

The Janus-particle implementation makes this scope explicit by defining FPTS as a BFP-based method that resolves scattering simultaneously as a function of wavelength, incident direction, scattering direction, and polarization. In that setting, tomography is performed over illumination angles: the object is probed from many \(k_i\), and the scattered field is recorded over the full \(k_s\) manifold allowed by the objective NA [2507.15760].

The term is not uniform across all subfields. In dual FPA-FTIR standoff sensing, “FPA-FTIR” refers to focal plane array Fourier transform infrared spectroscopy, and the “Fourier transform” is in the spectral domain rather than the optical Fourier plane in spatial frequency. The associated inverse method consumes a tomographically reconstructed threshold contour rather than raw pupil-plane data [2606.11840]. This distinction matters because it separates two non-identical uses of “Fourier” in tomography: Fourier-plane angular spectroscopy in microscopy and Fourier-transform spectral reconstruction in standoff hyperspectral sensing.

A related extension appears in rotational slitless spectroscopy. There, multiple slitless projections acquired at different rotation angles are interpreted tomographically, and the reconstruction recovers a three-dimensional integral-field datacube. The governing language is that of the Radon transform and Fourier slice theorem rather than BFP microscopy, but the operational goal is similar: use spectrally encoded projections to reconstruct a higher-dimensional object [2605.23814].

## 2. Fourier-plane encoding and k-space representations

The core optical principle in BFP methods is the one-to-one mapping between angle and pupil position. For a medium of refractive index \(n\) and free-space wavelength \(\lambda_0\), the wavevector magnitude is \(k = 2\pi n/\lambda_0\), with components
\[
k_x = k\sin\theta\cos\phi,\qquad
k_y = k\sin\theta\sin\phi,\qquad
k_z = k\cos\theta.
\]
For an aplanatic high-NA objective, the pupil radius obeys
\[
\rho_{\mathrm{BFP}} = f_{\mathrm{obj}}\, n \sin\theta,
\]
and, after pupil magnification \(M_p\), the camera-plane radius is \(r=M_p\rho_{\mathrm{BFP}}\). Using the pupil edge calibration \(r_{\max}=M_p f_{\mathrm{obj}}\mathrm{NA}\), one obtains
\[
n\sin\theta = \left(\frac{r}{r_{\max}}\right)\mathrm{NA}.
\]
This gives a direct pixel-to-\((k_x,k_y)\) conversion and defines the collection cone imposed by the objective NA [1806.08280].

The Janus-particle formulation uses the same mapping in scattering notation. In that work, each BFP point \((u,v)\) corresponds to a unique scattered direction \((\theta_s,\phi_s)\), with
\[
k_x = k\sin\theta_s\cos\phi_s,\qquad
k_y = k\sin\theta_s\sin\phi_s,\qquad
k_z = k\cos\theta_s,
\]
and \(\sin\theta_s\le \mathrm{NA}/n\). The BFP image is spectrally dispersed along one camera axis, so that one detector coordinate carries wavelength and the orthogonal coordinate carries a fixed in-plane BFP coordinate. Scanning a slit across the BFP reconstructs the full angular distribution versus \(\lambda\) [2507.15760].

Diffraction tomography expresses the same k-space logic through the generalized Fourier diffraction theorem. For plane-wave incidence \(\psi_{\mathrm{inc}}(x)=e^{ik_0 s\cdot x}\) and detector plane \(x_d=a\) outside the object support, the partial Fourier transform of the scattered field satisfies
\[
\tilde{\mathcal{F}}u(\xi,a)
=
\sqrt{\frac{\pi}{2}}
\frac{e^{\pm i\kappa(\xi)a}k_0^2}{\kappa(\xi)}
\,
\widehat{f}\!\left(h^\pm(\xi)-k_0 s\right),
\qquad |\xi|<k_0,
\]
with \(h^\pm(\xi)=(\xi,\pm\kappa(\xi))^\top\) and \(\kappa(\xi)=\sqrt{k_0^2-|\xi|^2}\) for \(|\xi|\le k_0\). The accessible object-space sample is therefore
\[
q = k_s-k_i = h^\pm(\xi)-k_0 s,
\]
which lies on a hemisphere of the Ewald sphere centered at \(-k_0 s\) [2407.01793].

These formulations differ in physical setting, but they share the same structural feature: the detector does not measure the object directly in real space. It measures either a momentum distribution or a wavefield whose Fourier-domain geometry determines which parts of object k-space are accessible.

## 3. Tomographic inverse models and reconstruction formalisms

The inverse problem in diffraction tomography is organized around Fourier coverage. With acquisition parameters collected in
\[
\mathcal{U}=\{(\xi,t)\in\mathbb{R}^{d-1}\times[0,L]: |\xi|<k_0(t)\},
\]
the k-space mapping is
\[
T(\xi,t)=R(t)\big(h^\pm(\xi,t)-k_0(t)s(t)\big),
\]
and the coverage set is
\[
\mathcal{Y}=\{T(\xi,t):(\xi,t)\in\mathcal{U}\}.
\]
The resulting filtered backpropagation reconstructs the \(L^2\)-best approximation whose Fourier support lies in \(\mathcal{Y}\), with explicit Jacobian and multiplicity corrections through \(\det(\nabla T)\) and the Banach indicatrix \(\mathrm{Card}(T^{-1}(\cdot))\) [2407.01793]. In that framework, missing-cone structure, finite aperture, and limited angular diversity are properties of the coverage set itself rather than merely numerical artifacts.

Rotational slitless spectroscopy casts reconstruction as a linear inverse problem. After discretizing the datacube \(x\in\mathbb{R}^{UVL}\) and stacking all detector images into \(b\in\mathbb{R}^{n_rXY}\), the forward model is
\[
b = A x + n.
\]
The least-squares formulation minimizes \(\|Ax-b\|_2^2\), with normal equations \(A^\top A x = A^\top b\). The paper emphasizes on-the-fly forward and adjoint operators \(G\) and \(G^\*\), gradient descent, and Lucy–Richardson updates,
\[
f'_{t+1} = f'_t \cdot \left[ G^\*\!\left(\frac{f}{Gf'_t}\right)\right],
\]
rather than explicit matrix storage. Tomographically, each rotated projection corresponds to a different family of plane integrals through the \((u,v,\lambda)\)-cube, and the Fourier slice theorem links the 2D Fourier transform of each projection to a central plane in the 3D Fourier transform of the datacube [2605.23814].

The standoff FTIR source-localization framework uses a different inverse model again. The contaminant concentration \(c(\mathbf{x},t)\) evolves under the advection–diffusion PDE
\[
\partial_t c(\mathbf{x},t) + \mathbf{v}(\mathbf{x})\cdot\nabla c(\mathbf{x},t) = \kappa \Delta c(\mathbf{x},t),
\]
with initial condition \(c(\mathbf{x},0)=m(\mathbf{x})\). The tomographic measurement is a threshold contour
\[
\Gamma_\tau[c] = \{\mathbf{x}\in\Omega \mid c(\mathbf{x},\tau)=C_{\mathrm{th}}\},
\]
and the inverse problem is written over nonnegative Radon measures,
\[
\min_{\mu\in \mathcal{M}^+(\overline{\Omega})}
\;
\frac12\big\|\widehat F(\mu)-C_{\mathrm{th}}\big\|^2_{L^2(\Gamma_\tau)}
+
\alpha\,\mu(\overline{\Omega}).
\]
The optimization is solved by the Primal-Dual Active Point (PDAP) algorithm, with forward PDE solves, adjoint solves driven by a line delta source on \(\Gamma_\tau\), greedy activation by a dual certificate, and convex amplitude updates on the active set [2606.11840].

A plausible implication is that “tomographic spectroscopy” in current usage is less a single inversion algorithm than a family of inverse problems distinguished by the geometry of the measurement operator: Ewald-sphere sampling in diffraction tomography, Radon-type projections in rotational slitless spectroscopy, and level-set constraints in standoff plume reconstruction.

## 4. Acquisition architectures and optical implementations

Classical BFP microscopy requires relay optics because the objective’s BFP lies inside the objective body. The review literature describes three common configurations for projecting a pupil conjugate to the detector: a Bertrand lens before the image plane, a Bertrand lens after the image plane, and a modified \(4f\) relay. Telecentric pupil relays are emphasized because they stabilize the mapping against sample axial motion. For spectroscopy, a slit can be placed at a pupil conjugate and coupled to an imaging spectrograph, or the BFP can be projected directly onto the spectrograph entrance slit [1806.08280].

The Janus-particle implementation realizes this architecture in a dark-field scattering microscope. Illumination uses a dark-field condenser with Fourier-plane apertures \(B1\) and \(B2\) to restrict incident wavevectors to a narrow cone of approximately \(0.096\,\mathrm{sr}\); detection uses an oil-immersion objective with adjustable NA up to \(1.3\), a relay system that images either the sample plane or the BFP, a \(0.9\,\mathrm{mm}\) spatial pinhole \(B4\) to isolate a single particle, a blazed transmission grating in front of an sCMOS camera, and an adjustable slit \(B5\) that is translated to reconstruct the full BFP at each wavelength. Spectral response \(R(\lambda)\) is obtained from the spectrally dispersed illumination pattern, and datasets with and without a long-pass filter are merged to cover approximately \(400\text{–}1100\,\mathrm{nm}\) [2507.15760].

Rotational slitless spectroscopy uses a different acquisition geometry. Independent projections are generated either by rotating the dispersion direction relative to the detector or by rotating the entire telescope or field. The dispersion law is modeled as
\[
d(\lambda)=k(\lambda-\lambda_c),
\]
and the rotation-dependent coordinate mappings determine how each datacube voxel contributes to detector pixels. The number of required independent rotations follows the counting relation
\[
n_r \cdot X \cdot Y \ge U \cdot V \cdot L,
\qquad
n_{\min}=\left\lceil \frac{UVL}{XY}\right\rceil,
\]
which directly links instrument geometry, chosen pixelization, and observing strategy [2605.23814].

The standoff FTIR configuration uses dual FPA-FTIR hyperspectral imaging systems operated with a suitable opening angle so that their fields of view overlap the suspected release area. Each system acquires interferograms that are Fourier transformed into calibrated radiance spectra; tomographic processing then yields spatial maps and, for the inverse model, a threshold contour \(\Gamma_\tau\). The paper emphasizes that representing the measurement as a contour rather than a volumetric field allows a mesh-independent formulation via a curve discretization and a trace operator, avoiding remeshing [2606.11840].

## 5. Applications and demonstrated regimes

BFP imaging and spectroscopy have been used to study angular emission patterns of fluorescence and Raman signals from molecules, elastic scattering from nanostructures, energy–momentum dispersion of guided or leaky modes, metasurface emission, and polarization-resolved momentum-space observables. The review specifically highlights single-molecule dipole orientation retrieval, Raman radiation patterns in graphene, plasmonic nanorod and nanowire scattering, Yagi–Uda antenna emission, and E–k mapping in photonic crystals and microcavities [1806.08280].

The Janus-particle study gives a detailed contemporary example of FPTS as four-dimensional scattering tomography. The particles are polystyrene spheres of diameter \(1\,\mu\mathrm{m}\) with hemispherical Au caps, immersed in oil of refractive index approximately \(1.51\). The measurements and finite-element simulations identify three distinct multipolar families up to fifth order: axial-propagating transverse-electric, transverse-propagating transverse-electric, and transverse-propagating axial-electric. A shoulder at \(\lambda \approx 550\,\mathrm{nm}\) is prominent for Au-side illumination, weaker for side-on illumination, and largely absent for PS-side illumination; side-on illumination shows a cluster of weaker peaks between \(700\text{–}800\,\mathrm{nm}\). The measured angular patterns serve as orientation fingerprints, and the reported spectra show progressive red-shifts and linewidth narrowing of higher-order resonances [2507.15760].

In standoff plume reconstruction, the synthetic test case uses kinematic viscosity \(\nu=\mathrm{0.25\,m^2/s}\), inflow velocity \(\mathrm{1\,m/s}\) at the southern boundary, diffusion coefficient \(\kappa=\mathrm{2.0\,m^2/s}\), measurement time \(\tau=\mathrm{5\,s}\), and threshold \(C_{\mathrm{th}}=10^{-3}\). PDAP identifies a sparse set of 6 active Dirac components; the post-processed source achieves a localization error of approximately \(0.26\,\mathrm{m}\), total reconstructed intensity \(\lambda_{\mathrm{post}}\approx 1.02\) for ground truth \(1.0\), reconstruction error below \(1\%\) across the domain at measurement and forecast times, and convergence in 16 forward and 16 adjoint solves [2606.11840].

In rotational slitless spectroscopy, numerical experiments generate toy datacubes containing 100 stars with uniform spectra, sampled by \(10^8\) photons per star. With datacube size \((160,160,160)\), detector size \((160,160)\), and \(n_r=160\) rotations, both rotation modes reconstruct the datacube accurately. Flux profiles along \(u\), \(v\), and \(\lambda\) agree with ground truth to within approximately \(2\%\) across the central region, with overall deviations \(2\text{–}5\%\) at worst; 500 iterations run in less than \(0.5\,\mathrm{h}\) on a 10-core M2 laptop with \(16\,\mathrm{GB}\) RAM [2605.23814].

## 6. Limitations, ambiguities, and future development

A persistent ambiguity is terminological. In optical microscopy, the Fourier plane is the BFP of an objective and directly encodes wavevector content. In FPA-FTIR plume sensing, the Fourier transform is performed on interferograms to obtain spectra, and the tomographic inversion is driven by a geometric contour rather than by optical Fourier-plane imaging [2606.11840]. The two usages are compatible only at a high level of abstraction: both combine spectroscopy with tomography, but they do so through different physical measurement operators.

Coverage limitations are central in every branch of the subject. In diffraction tomography, the generalized Fourier diffraction theorem is derived in the scalar Helmholtz model under Born or Rytov approximations, and finite aperture or limited angle diversity produces missing-cone structure and reduced low-frequency coverage. The coverage remains bounded by radius \(2k_{\max}\), and non-uniform sampling requires Jacobian compensation and multiplicity correction [2407.01793].

In BFP-based FPTS on single particles, the angular range is limited by the objective NA and the incident-angle range by the condenser NA. The reported Janus-particle measurements are intensity-only; no phase is recorded, and polarization channels are not separated in the reported experiments, even though the platform supports polarization-resolved operation. Signal-to-noise degrades beyond approximately \(900\,\mathrm{nm}\) because of detector quantum efficiency, and sample-to-sample variability, roughness, and thickness variation broaden or weaken higher-order resonances [2507.15760].

In rotational slitless spectroscopy, ill-conditioning arises from correlated projection angles, finite detector sampling, and edge truncation. The paper mitigates boundary artifacts by using only the central \(60\%\times 60\%\) of the detector as an active region and leaving a \(20\%\) border as a buffer. The authors identify regularization, adaptive voxelization, Bayesian inference, angle-set optimization, filtered backprojection adapted to realistic sampling, and hybrid rotation modes as natural extensions [2605.23814].

In standoff contour-based spectroscopy, the inverse problem is severely ill-posed and underdetermined without sparsity. Accuracy depends on the quality of the extracted contour, the assumed advection–diffusion model, known wind and diffusion parameters, and sufficient multi-view coverage. The paper notes that strong winds, rapidly changing conditions, sensor noise, and occlusions can make the contour incomplete or ambiguous, and it proposes future work on simultaneous calibration of turbulence, reduced-order surrogates, and noise-aware formulations based on
\[
\Gamma^{\mathrm{obs}} = \{\mathbf{x}\mid c(\mathbf{x},\tau)+\eta(\mathbf{x}) = C_{\mathrm{th}}\},
\qquad
\eta(\mathbf{x})\sim \mathcal{N}(0,\sigma^2)
\]
[2606.11840].

Taken together, these developments show that Fourier Plane Tomographic Spectroscopy is best understood as a technically heterogeneous field organized around a common principle: spectroscopic measurements are acquired in a representation tied to angular, Fourier, or projection geometry, and tomography reconstructs latent spatial, modal, or dynamical structure from incomplete but physically structured data.

Source: https://www.emergentmind.com/topics/fourier-plane-tomographic-spectroscopy