AEGIS

AEGIS computes absorbed power density on human body surfaces in wireless environments. It replaces volumetric EM simulation (10¹² voxels in FDTD) with O(MN) surface operations by exploiting the fact that at mmWave frequencies, the skin depth is so shallow (< 0.5 mm) that absorption is entirely a surface phenomenon.

The core equation:

S_ab(r) = S_inc * T_0 * ReLU[n_hat(r) . (-k_hat)]

Nine fidelity levels (0-8) provide a controlled accuracy-cost tradeoff, from O(1) worst-case bounds to exposure-constrained MIMO beamforming. The geometric computation matches the full Fresnel solution to within 0.35%.

Built for: Researchers in dosimetry and wireless exposure, compliance engineers evaluating ICNIRP 2020 limits, and antenna designers optimizing MIMO precoders under safety constraints.

pip install -e ".[dev]"

from aegis import DosimetryEngine, BodyMesh, TissueModel, PropagationPaths

skin = TissueModel.from_database("Skin")
body = BodyMesh.load("thelonious.stl")
paths = PropagationPaths.from_powers(k_hat=[[0, 0, -1]], power=[1.0])

engine = DosimetryEngine(skin, frequency=28e9)
result = engine.compute(body, paths, level=2)

print(f"P_abs = {result.p_abs:.4f} W")
print(f"Peak S_ab = {result.peak_sab:.2f} W/m2")

For coherent MIMO with exposure-constrained beamforming:

from aegis import Precoder

precoder = Precoder.mrt(channel_matrix)
result = engine.compute(body, paths, level=8, precoder=precoder)

print(f"ICNIRP compliant: {result.is_compliant}")

Launch the 3D viewer:

python -m aegis.viewer --location "Ghent, Belgium"
python -m aegis.viewer --config configs/outdoor_urban.json
python -m aegis.viewer --scenario open_ground

Levels 0-6 use scalar power per path (incoherent). Levels 7-8 use complex amplitudes and MIMO antenna structure (coherent).

graph LR
    A[Ray tracer] --> B[PropagationPaths]
    B --> C[DosimetryEngine.compute]
    C --> D[DosimetryResult]

    style A fill:#2196F3
    style B fill:#4CAF50
    style C fill:#FF9800
    style D fill:#9C27B0

1. Ray tracer produces propagation paths (directions k_hat, powers, or complex amplitudes psi)
2. PropagationPaths wraps the ray data with .from_powers() for incoherent or full complex fields for coherent
3. DosimetryEngine dispatches to the appropriate fidelity kernel (levels 0-8)
4. DosimetryResult contains absorbed power density S_ab, total absorbed power P_abs, whole-body SAR, and ICNIRP compliance

The engine accepts any triangle mesh as a body model and any tissue model from the IT'IS v5.0 database or custom Cole-Cole parameters.

src/aegis/
    engine.py          Main entry point, dispatches to kernel by level
    paths.py           PropagationPaths (k_hat, psi, element_index)
    result.py          DosimetryResult (sab, p_abs, sar_wb, Q, rho)
    precoder.py        Precoder for MIMO beamforming vector x
    tissue/            EM properties, Cole-Cole model, Fresnel coefficients
    geometry/          Body mesh, ambient occlusion, directivity, spatial averaging
    kernels/           Fidelity levels 0-8, one file per level
    coherent/          Field channel, exposure operator Q, ECBF solver
    optim.py           Differentiable loss functions for jax.grad
    compliance/        ICNIRP 2020 limits
    integration/       DiffeRT ray tracer bridge
    viewer/            Flask REST backend (compute, data, config APIs)
    viz/               Matplotlib/Plotly dashboards
aegis-web/             React + Three.js frontend (Vite, R3F, Zustand)

git clone https://github.com/rwydaegh/aegis.git
cd aegis
git lfs install
git lfs pull
pip install -e ".[dev]"

If git-lfs is not installed yet, install it before git lfs pull. The deploy and full-test workflows already check out with lfs: true, and the Docker image copies data/ into the backend image, so local clones need the same data present if you want parity with deploys.

pip install -e ".[viz]"      # matplotlib, plotly, pyvista dashboards
pip install -e ".[rt]"       # DiffeRT ray tracer integration
pip install -e ".[gpu]"      # JAX with CUDA support
pip install -e ".[docs]"     # MkDocs documentation site
pip install -e ".[all]"      # everything

• Python 3.11+
• NumPy >= 1.24, SciPy >= 1.10
• Data assets under data/ (set AEGIS_DATA_DIR if you keep them elsewhere)

The animated phantom viewer assets live in data/phantoms/. The committed runtime payload is the .glb outputs plus preview renders. Large FBX build inputs are intentionally kept local.

pytest tests/ -m "not slow" -x
pytest tests/
pytest tests/ --collect-only -q
pytest tests/ --cov=aegis

The suite includes golden tests against monograph tables, Mie regression as the CI canary, Hypothesis property tests, engine and coherent pipeline tests, Flask viewer routes, and visualization smoke tests. See docs/developer_guide/testing.md for defaults (-n 2 workers), CI, and coverage omissions.

Build the docs locally:

pip install -e ".[docs]"
mkdocs serve

1. Fork the repo and create a feature branch
2. Follow code style (Ruff, type annotations on public API)
3. Add tests for new features. Never weaken assertions to make tests pass.
4. Submit a PR with a clear description

ruff check src/ tests/
ruff format src/ tests/
pytest tests/ -m "not slow"

@software{Wydaeghe_AEGIS,
  title   = {{AEGIS: Adaptive Electromagnetic Geometric Illumination \& Safety}},
  author  = {Wydaeghe, Robin},
  url     = {https://github.com/rwydaegh/aegis},
  license = {Proprietary},
  version = {0.41.0}
}

Proprietary. All rights reserved. See LICENSE for details.