This repository implements a multi-camera coherent imaging system for characterizing multimode optical fibers through holographic measurements. The system synchronizes three Basler GigE cameras using an Arduino Due microcontroller to capture spatially and angularly resolved interference patterns, performs real-time Fourier transform processing, and conducts offline statistical analysis on multi-dimensional datasets.
The system supports three Basler GigE cameras capturing synchronized images for holographic fiber characterization experiments. Multiple acquisition paths coexist to serve different purposes: the primary real-time system using the proprietary camera SDK with X11 visualization, alternative scanning patterns, an open-source Aravis-based alternative, and C++ implementations for performance testing.
The image processing subsystem provides array manipulation, format conversion, and mathematical operations for processing camera data in real-time and preparing it for FFT analysis and storage. This page covers the Common Lisp image processing utilities found in the image-processing package.
The Julia-based offline analysis pipeline that processes multi-dimensional ICS files generated by the camera acquisition systems. The scripts perform compute-intensive post-processing operations including endface reconstruction, angular throughput analysis, Pearson correlation mapping, and SVD decomposition.
This page documents the hardware control subsystem that manages the Arduino Due for camera synchronization and Fast Steering Mirror positioning. The Arduino Due runs a FemtoLisp interpreter and communicates with the Common Lisp acquisition system via serial port to coordinate precise timing of camera triggers and mirror angle control through DAC outputs.
The system implements two distinct approaches: true REPL-driven interactive development for Common Lisp components, and signal-triggered dynamic library reloading for C/C++ components. Both mechanisms preserve critical runtime state (camera connections, buffers, hardware control) across code updates, eliminating the need for process restarts during algorithm development.
The system models light propagation through a perturbed multimode fiber by decomposing the problem into a sequence of linear operations on the mode amplitude vector E = (A₀, A₁, ..., Aₙ). Each physical component transforms this vector through matrix multiplication.
Propagation Matrix: A straight fiber segment of length L applies phase shifts exp(iβᵢL) to each mode i, represented by diagonal matrix P_L: P_L(i,i) exp(iβᵢ L) P_L(i,j), i≠j 0
Coupling Matrix: A bend with angle θ couples modes through the symmetric matrix T_θ. For weakly-guiding circular fibers, only modes with the same azimuthal quantum number l couple, making T_θ sparse.
Real fibers have continuous curvature, which the model approximates by discretizing into N equidistant bends, each characterized by angles (θᵢ, φᵢ)
The total transfer matrix becomes a product: M = P_Λ T_θN ... P_Λ T_θ2 P_Λ T_θ1
This formulation enables reconstruction of fiber shape from measured transmission properties if sufficient modes are measured. For a fiber with 10,000 modes, the number of measurable parameters scales with the number of bends that can be resolved.
The experimental system measures both transmitted and reflected fields to characterize the fiber. The reflection measurement captures:
E_out = M' R M E_in
where:
M = forward transfer matrix (fiber input → exit)
R = reflection matrix at fiber end face (diagonal with exp(iπ))
M' = return transfer matrix = M^T (by reciprocity)
For a holographically measured field U(x) = A(x) exp(iΦ(x)), the phase derivative is computed via:
Φ'(x) = Im[ (1/U) × (dU/dx - d|U|/dx × U/|U|) ]
This expression avoids 2π discontinuities inherent in computing d(arg(U))/dx directly, making it numerically stable for Wigner distribution calculation.
The Fast Steering Mirror (FSM) scans the input field u^i across a 2D grid of angles (θₙ, φₘ) ∈ Ω, where Ω is the fiber's numerical aperture. Each scan position effectively launches a different modal distribution.
The theoretical quantities map to 4D/5D array structures in the codebase:
| Theoretical Object | Array Dimensions | ICS File Organization |
|---|---|---|
| u^t_∥(x',y',θ,φ) | [64, 64, N_θ, N_φ] | o0.ics (Camera 1) |
| u^t_⊥(x',y',θ,φ) | [64, 64, N_θ, N_φ] | o2.ics (Camera 3) |
| u^o_⊥(x,y,θ,φ) | [128, 128, N_θ, N_φ] | o1.ics (Camera 2) |
| FT domain | [90, 90, 3, N_θ, N_φ] | FFT ROI extraction |
The scan generates ~167×127 angle points.
This is code I use to control my holographic setup to image through multimode fibers.
A significant part is a wrapper for the Aravis library to read out multiple GigE Vision cameras. The file dependencies are described in arv.asd. I chose to use Clozure Common Lisp for this library because it comes with a foreign function generator that can parse C header files. To simplify development I added syntax completion for foreign functions to slime.
Realtime control is done with an Arduino Due. To ease development (in particular I don't like the long waiting time when uploading new code to the Arduino) I ported femtolisp to this Arduino. This is sufficient to generate the two DAC channels for my XY-scanning mirror and a bunch of digital trigger signals for the cameras and a shutter. See arduino-femtolisp/. From Clozure Common Lisp I talk to the Arduino using the code in arduino-serial-ccl/.
My main experiment is controlled from test.lisp. I also use the binding for FFTW https://github.com/plops/cl-cffi-fftw3 to do Fourier transforms.
Copyright (c) 2014 Martin Kielhorn kielhorn.martin@gmail.com Licensing: GPL v2