Companion code for the Digital Signal Processing (Elsevier, 2025) article on single-LED VLP positioning through an intelligent reflecting surface (IRS).

Authors: Efe Tarhan · Furkan Kökdoğan · Sinan Gezici

Paper: 📄 IRS aided visible light positioning with a single LED transmitter — Digital Signal Processing, Vol. 156, 104799 (2025). DOI: 10.1016/j.dsp.2024.104799

This repository contains MATLAB code that reproduces the results of the paper. The study investigates visible-light positioning (VLP) in indoor environments where the line-of-sight (LoS) between a single LED and the receiver is blocked. A wall of individually-tilted micro-mirrors — an Intelligent Reflecting Surface (IRS) — is used to redirect the LED signal into the room, enabling non-line-of-sight (NLoS) localization. The IRS profile (one orientation per mirror group) is optimized to minimize the Cramér–Rao lower bound (CRLB) of the position estimator.

Key elements:

• Time-division multiplexing scheme that reuses a single LED across multiple IRS configurations to produce a sequence of independent measurements.
• CRLB analysis of the resulting NLoS channel as a function of receiver position, mirror count, and noise variance.
• IRS-profile optimization with both gradient-based (fmincon) and metaheuristic (PSO) solvers, in bounded, unbounded and ensemble variants.
• Maximum-likelihood position estimator evaluated against the CRLB via Monte-Carlo RMSE simulations.

ml-vlp-irs/
├── README.md
├── LICENSE
├── setup_paths.m         <- run once before any script
├── functions/            <- core simulation + optimization functions
├── scripts/              <- entry-point scripts that reproduce the paper figures
│   └── legacy/           <- scratch / unused scripts kept for reference
├── data/                 <- precomputed .mat results (CRLB, optimized profiles, RMSE)
│   ├── bounded_single_optim/
│   ├── bounded_ensemble_optim/
│   ├── unbounded_single_optim/
│   ├── data_rmse_crlb_all/
│   ├── grid_results/
│   └── figures/          <- raw figure exports tied to the data
├── figures/              <- paper figures (png/eps/fig + overleaf source)
└── fingerprinting/       <- side project (ML-based VLP fingerprinting)

Tested with MATLAB R2022b or newer. The Optimization Toolbox and Global Optimization Toolbox are required for fmincon / particleswarm.

% from the repository root
>> setup_paths              % adds functions/ and fingerprinting/functions/ to the path
>> cd scripts
>> demo_room_and_irs        % minimal IRS scene + power-map demonstration
>> rmse_focused_scenario    % CRLB / RMSE curve for the optimized focused-IRS scenario
>> plot_paper_figures       % reproduces the RMSE / CRLB comparison figures

Most heavy results (optimized profiles, ensemble runs, Monte-Carlo RMSE) are already cached as .mat files under data/, so the plotting scripts run in seconds without re-running the optimization.

All scripts are in scripts/ and assume setup_paths has been run.

The scripts/legacy/ folder holds files that were kept for historical reference but are not used to produce the paper's figures:

• sketch_rectangles.m — early 2-D layout sketch
• seed_sensitivity_check.m — sanity check on RNG seed influence
• scratch_optim_sweeps.m — author's scratch script with redundant / speculative sweeps

N = 4	N = 9	N = 16	N = 25

N = 4	N = 9	N = 16	N = 25

The fingerprinting/ folder contains an independent, exploratory study on machine-learning-based VLP (KNN and fully-connected neural-network fingerprinting on IRS power measurements). It is not part of the published paper — see fingerprinting/README.md for details.

If you use this code or its results, please cite:

@article{tarhan2025irs,
  title   = {IRS aided visible light positioning with a single LED transmitter},
  author  = {Tarhan, Efe and Kokdogan, Furkan and Gezici, Sinan},
  journal = {Digital Signal Processing},
  volume  = {156},
  pages   = {104799},
  year    = {2025},
  publisher = {Elsevier},
  doi     = {10.1016/j.dsp.2024.104799},
  url     = {https://doi.org/10.1016/j.dsp.2024.104799}
}

See LICENSE.