Research map / Architectures and efficiency / RNNs and normalization
Understanding Multi-layered Transmission Matrices
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Analyzes how many planar aberration layers (multi-slice model) are needed to approximate tissue transmission matrices, showing the missing cone limits required layers so sparse layer counts still expand the wavefront-shaping field of view.
Problem
Multi-conjugate wavefront shaping through scattering tissue requires many SLM layers to model 3D aberrations, but such systems are hard to build; it is unclear how many layers are needed and what field of view a few layers can correct.
Approach
Shows that the missing cone of microscope frequency coverage turns into an advantage for transmission matrices: the butterfly-shaped sampled spectrum means far fewer layers than the Nyquist prediction suffice, and bounded-support matrices need even fewer layers.
Key results
- Nyquist theory predicts about 100 layers for a 200µm tissue at λ=0.5µm and NA=0.5, but transmission matrices are well fit with far fewer layers.
- On a 170µm chicken breast transmission matrix with 13×13µm support, about 3 layers focus across the full 13×13µm window, versus about 3×3µm with a single layer (≈18× larger area).
- For a 40×40µm support, about 6 layers give good focusing according to the simulated fitting curves.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Increase the correction from 1 to 3 layers | Multi-slice wavefront-shaping correction · Chicken breast, 170µm thick | correctable support: 3 × 3µm to 13 × 13µm; ×18 larger | Helped |
| Use fewer layers to fit a smaller transmission-matrix support | Multi-slice transmission-matrix fit · Synthetic transmission matrices | number of layers required for a good fit: 40 × 40µm: M = 6; 1 × 1µm: M = 1; 2 × 2µm: M = 2; 5 × 5µm: M = 3 | Helped |
| Use a sparse multi-slice approximation rather than dense layer sampling | Synthetic transmission-matrix fitting · Toy volume transmission matrix | transmission-matrix fitting error | Mixed |
| Increase tissue thickness | Multi-slice transmission-matrix fit · Synthetic transmission matrices | number of layers required for a good fit | Hurt |
| Increase optical density at the same tissue thickness | Multi-slice transmission-matrix fit · Synthetic transmission matrices | number of layers required for a good fit | No clear effect |
| Limit the fitted transmission matrix to a smaller support | Multi-slice transmission-matrix fit · Lab-captured chicken breast transmission matrix | number of layers required for a good fit | Helped |
Surprising findings
- The missing cone limitation, usually a drawback in 3D microscopy, reduces the number of layers needed for transmission matrix approximation well below the Nyquist limit.
Limitations
- Analysis mainly relies on weak-scattering first-Born approximation; multiple scattering is handled only empirically.
- Synthetic volumes limited in thickness by computational constraints; full few-hundred-micron fields of view likely need many more layers.
Experimental setup
Datasets: synthetic toy volume of spheres with varying refractive indices lab-captured transmission matrices (chicken breast, mouse brain tissue, parafilm)
Benchmarks: synthetic transmission matrices lab-captured transmission matrices
Metrics: fitting error E_M (L2 norm) focusing spot area / support size
Baselines: Nyquist sampling prediction single-layer correction
Builds on: diffraction tomography multi-conjugate wavefront shaping first-Born approximation
Closest papers by meaning
- WaveMo: Learning Wavefront Modulations to See Through Scattering (CVPR 2024 · similarity 0.85)
- TransmissiveGS: Residual-Guided Disentangled Gaussian Splatting for Transmissive Scene Reconstruction and Rendering (NeurIPS 2026 · similarity 0.83)
- LENS: Low-Frequency Eigen Noise Shaping for Efficient Diffusion Sampling (NeurIPS 2026 · similarity 0.82)
- Subsurface Scattering for Gaussian Splatting (NeurIPS 2024 · similarity 0.81)
- CD-RCM: Generalizable Continuous-Depth Novel View Synthesis for Reflectance Confocal Microscopy (NeurIPS 2026 · similarity 0.81)
- Single Exposure Quantitative Phase Imaging with a Conventional Microscope Using Diffusion Models (AAAI 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.