Stochastic channel model¶

AEGIS includes a 3GPP TR 38.901 cluster-based channel generator that produces multipath propagation paths without a ray tracer. It draws large-scale fading parameters (delay spread, K-factor, angular spreads, shadow fading) from log-normal distributions parameterized by the chosen scenario preset, then expands them into per-cluster and per-subpath arrival directions and powers. The output is a PropagationPaths object identical to one produced by DiffeRT or Sionna RT, so any incoherent fidelity level (0-6) can consume it directly.

The model is the default path source in the interactive viewer when ray tracing is disabled.

When to use it¶

Use the stochastic channel when you want fast dosimetry estimates across many antenna placements without building or loading a geometric scene. It is appropriate for sensitivity studies, fidelity-level comparisons, and compliance screening where the exact multipath geometry is unknown. Use ray tracing when building geometry is available and you need spatially consistent paths tied to specific reflectors.

The two modes are mutually exclusive in the viewer. Enabling stochastic channel disables ray tracing and vice versa.

Presets¶

Presets are QuaDRiGa-format .conf files in data/channel_presets/. The library contains over 80 entries organized by standard:

• 3GPP 38.901 (14 presets): UMi, UMa, Indoor, InF, RMa with LOS/NLOS/O2I variants. 0.5-100 GHz.
• 3GPP 37.885 (5 presets): V2X highway and urban scenarios.
• 3GPP 3D (8 presets): 3D channel models for UMa/UMi.
• QuaDRiGa (12 presets): Industrial, NTN (satellite), and urban device-to-device.
• WINNER (12 presets): Indoor A1, SMa C1, UMa C2, UMi B1, and indoor-to-outdoor transitions.
• mmMAGIC (6 presets): Millimeter-wave indoor and UMi, including O2I.
• 5G-ALLSTAR (8 presets): Rural, suburban, urban, and dense urban.
• MIMOSA (8 presets): Frequency-banded (10-45 GHz range).
• BERLIN/DRESDEN (8 presets): Measurement-based UMa and UMi campus/square.
• Canonical (4 presets): Freespace, LOSonly, TwoRayGR, Null.

The viewer groups presets by standard in two dropdowns: pick the standard first, then the scenario. The default preset is 3GPP_38.901_UMi_LOS. To list all installed presets programmatically:

from aegis.channel import list_presets
from pathlib import Path

names = list_presets(Path("data/channel_presets"))
print(names)  # sorted list of all .conf stems

Key parameters¶

Each preset defines a set of large-scale parameters sampled once per realization:

• K-factor (KF_mu, KF_sigma): Ricean K-factor in dB. Controls the power ratio of the dominant (LOS) cluster to the diffuse multipath. The UMi LOS preset uses \(K = 9 \pm 5\) dB. NLOS presets set KF_mu = -100 dB (effectively zero LOS component).
• Delay spread (DS_mu, DS_gamma): RMS delay spread in log\(_{10}\)(s), frequency-scaled as \(\mu + \gamma \log_{10}(f/f_0)\).
• Azimuth spread of arrival (AS_A_mu, AS_A_gamma): Log\(_{10}\)(degrees) at the body. For UMi LOS at 28 GHz, the mean ASA is about \(10^{1.56} \approx 36°\).
• Elevation spread of arrival (ES_A_mu): Elevation equivalent of ASA. Typically smaller than ASA (UMi LOS mean ESA \(\approx 7°\) at 28 GHz).
• NumClusters: Number of scattering clusters. The UMi preset uses 12 clusters.
• NumSubPaths: Sub-paths per NLOS cluster (default 20). Each cluster is expanded using the fixed QuaDRiGa offset table with per-cluster angular spreads PerClusterAS_A and PerClusterES_A.

Shadow fading (SF_sigma) is drawn independently and applied as a multiplicative offset on the total received power.

Generation procedure¶

generate_channel runs the following steps:

1. Draw large-scale parameters (DS, KF, SF, ASA, ESA) from their log-normal distributions, applying frequency scaling.
2. Generate cluster powers from an exponential power delay profile scaled by the delay spread ratio r_DS. Add per-cluster log-normal shadowing (LNS_ksi). Boost cluster 0 power by the K-factor.
3. Draw cluster arrival angles (azimuth and elevation), then scale them to match the drawn ASA/ESA.
4. Rotate all cluster angles so cluster 0 points toward the body center.
5. Expand each NLOS cluster into NumSubPaths sub-paths using the fixed offset table. The LOS cluster remains a single ray.
6. Compute path loss from the antenna-to-body distance using the preset's path loss model (logdist, dual_slope, or nlos). Apply the shadow fading offset.
7. Return PropagationPaths.from_powers(k_hat, power).

The LOS cluster points toward the body by construction. The remaining clusters form a diffuse halo whose angular spread matches the drawn ASA/ESA.

Path loss models¶

The generator supports three path loss models selected by PL_model in the preset:

• logdist: \(\text{PL} = A \log_{10}(d) + B + C \log_{10}(f)\)
• dual_slope: two-segment LOS model with a breakpoint distance \(d_{\mathrm{BP}} = E(h_{\mathrm{BS}} - h_E)(h_{\mathrm{MS}} - h_E) f\)
• nlos: \(\max(\text{dual\_slope}, A_n \log_{10}(d) + B_n + C_n \log_{10}(f))\)

Unknown models fall back to free-space path loss \(20 \log_{10}(d) + 20 \log_{10}(f) + 32.45\) dB.

Python API¶

Basic usage¶

from aegis.channel import generate_channel, load_preset
from aegis import DosimetryEngine
from aegis.tissue.dielectric import SKIN_28GHZ
import numpy as np

preset = load_preset("3GPP_38.901_UMi_LOS", "data/channel_presets")

paths = generate_channel(
    params=preset["params"],
    freq_ghz=28.0,
    antenna_pos=np.array([10.0, 0.0, 1.5]),
    body_center=np.array([0.0, 0.0, 1.0]),
    power_dbm=23.0,
    seed=42,
)

engine = DosimetryEngine(SKIN_28GHZ)
result = engine.compute(body, paths, level=2)

Overriding preset parameters¶

Pass an overrides dict to generate_channel to fix specific parameters without editing the .conf file:

paths = generate_channel(
    params=preset["params"],
    freq_ghz=28.0,
    antenna_pos=antenna_pos,
    body_center=body_center,
    power_dbm=23.0,
    seed=42,
    overrides={
        "KF_mu": 15,          # stronger LOS component (15 dB)
        "AS_A_mu": 1.0,       # narrower azimuth spread (~10 deg)
    },
)

When a parameter key appears in overrides, its variance (sigma) is set to zero, so the override becomes the deterministic value.

Loading a custom preset¶

Any QuaDRiGa-format .conf file can be used. Pass a Path to the directory containing your file:

from aegis.channel import parse_conf

params = parse_conf("my_scenarios/custom_scenario.conf")
paths = generate_channel(params, freq_ghz=60.0, ...)

Viewer configuration¶

The stochastic channel is controlled by the dosimetry.stochastic block in the viewer config:

{
  "dosimetry": {
    "stochastic": {
      "preset_dir": "data/channel_presets",
      "default_preset": "3GPP_38.901_UMi_LOS",
      "default_seed": 42
    }
  }
}

All presets in the preset directory are available in the viewer, grouped by standard. The viewer UI also exposes K-factor, azimuth/elevation spread, and number of clusters as overridable fields per realization.

Comparison with ray tracing¶

Ray tracing supports coherent levels 7-8 because it produces complex path amplitudes \(\psi\). The stochastic generator only produces scalar powers and therefore feeds incoherent kernels only.