Computing Radar Cross Section (RCS) via High-Frequency Physical Optics in StrikeCEM

Aug 15, 2026 · 6 min read

Simulation
Electromagnetics
C++20
CUDA
StrikeCEM

Radar Cross Section (RCS) is the primary metric defining the stealth characteristics of aerospace vehicles. It quantifies how much electromagnetic energy an aircraft, drone, or missile scatters back toward a radar receiver, measured in square meters ( $m^2$ ) or decibels relative to a square meter ( $\text{dBsm}$ ).

In air combat simulations like StrikeEngine, a missile’s active radar seeker needs to determine target signal returns at every millisecond timestep. However, computing electromagnetic wave scattering in real time is computationally impossible.

To solve this, I designed StrikeCEM—a high-frequency computational electromagnetics (CEM) engine written in C++20 and CUDA that pre-computes high-density radar signature databases from 3D CAD meshes for offline use during flight simulation.

1. The High-Frequency Computational Bottleneck

The exact physics of electromagnetic wave propagation is governed by Maxwell’s equations. Traditional full-wave numerical solvers—such as Finite-Difference Time-Domain (FDTD), Finite Element Method (FEM), or the Method of Moments (MoM)—discretize space or target boundaries into microscopic elements proportional to the illumination wavelength:

\Delta x \le \frac{\lambda}{10}

At typical radar frequencies (such as X-band, $f = 10\text{ GHz}$ , $\lambda = 3\text{ cm}$ ), illuminating a 15-meter fighter jet requires dividing space into tens of billions of 3D grid cells. Solving the resulting dense linear systems requires terabytes of RAM and days of supercomputer time for a single angular direction.

Because tactical radar simulations require full 360° sweeps across both azimuth and elevation at fine angular increments (e.g. 0.5° steps), full-wave methods are unusable. We must turn to high-frequency asymptotic methods, where target dimensions are much larger than the radar wavelength ( $L \gg \lambda$ ).

Physical Optics simplifies wave scattering by treating the target’s surface as locally planar. When a radar plane wave strikes a Perfect Electric Conductor (PEC), it induces an electrical surface current density $\mathbf{J}_s$ :

\mathbf{J}_s = \begin{cases} 2 (\mathbf{n} \times \mathbf{H}_{\text{inc}}) & \text{on illuminated surfaces } (\mathbf{n} \cdot \mathbf{k}_{\text{inc}} < 0) \\ \mathbf{0} & \text{on shadowed surfaces} \end{cases}

Where:

$\mathbf{n}$ is the outward facet unit normal vector.
$\mathbf{H}_{\text{inc}}$ is the incident magnetic field vector.
$\mathbf{k}_{\text{inc}}$ is the wave propagation vector.

// Surface illumination testing in StrikeCEM
inline bool isFacetIlluminated(const Vector3& normal, const Vector3& k_inc) noexcept {
    return normal.dot(-k_inc) > 0.0f;
}

// Compute induced surface current under Physical Optics approximation
inline Vector3 computeSurfaceCurrent(const Vector3& normal, const Vector3& H_inc) noexcept {
    return 2.0f * normal.cross(H_inc);
}

3. The Transverse Projector & Far-Field Scattering

A subtle but critical mathematical trap in Physical Optics implementation concerns far-field projection. Many simplified textbooks state the far-field scattered electric field as proportional to a single cross product $\hat{\mathbf{r}} \times \mathbf{J}_s$ .

However, as documented in StrikeCEM’s architectural decision record (ADR-0001), a single cross product yields the far magnetic field direction rather than the electric field amplitude. The scattered far electric field requires the transverse projector:

\mathbf{E}_s(\mathbf{r}) = \frac{j k \eta}{4\pi r} e^{-j k r} \iint_S \left[ \hat{\mathbf{r}} \times (\hat{\mathbf{r}} \times \mathbf{J}_s) \right] e^{j k \hat{\mathbf{r}} \cdot \mathbf{r}'} dS

Why This Matters

Consider a flat conducting plate lying in the $z=0$ plane illuminated at normal incidence from above:

Using the naive single-cross form $\hat{\mathbf{r}} \times \mathbf{J}$ yields zero co-polarized return ( $S_{HH} = 0$ ) and dumps all energy into the cross-polarization channel ( $S_{VH}$ ).
Using the double-cross transverse projector $\hat{\mathbf{r}} \times (\hat{\mathbf{r}} \times \mathbf{J})$ correctly yields zero cross-polarization by physical symmetry, producing the exact analytical broadside peak:

\sigma_{\text{peak}} = \frac{4\pi A^2}{\lambda^2}

Where $A$ is the facet area and $\lambda$ is the radar wavelength. This distinction is vital for accurately modeling stealth shaping, where co-polarization and cross-polarization channel separation dictate radar detectability.

4. Edge Diffraction (PTD) & Multi-Bounce Cavities

While Physical Optics accurately captures specular reflections from flat and gently curved surfaces, it suffers from two major limitations:

Edge Currents: PO assumes currents terminate abruptly at facet boundaries, underestimating edge scattering along wing leading edges and control surface gaps.
Cavity Reflections: Radar waves entering engine intake ducts or corner reflectors bounce multiple times, creating intense retro-reflections.

StrikeCEM enhances pure Physical Optics with two advanced mechanisms:

Physical Theory of Diffraction (PTD)

StrikeCEM integrates PTD fringe wave currents along wedge boundaries. By subtracting the non-uniform Physical Optics edge contribution from exact canonical wedge diffraction solutions (such as Ufimtsev fringe currents), StrikeCEM computes edge diffraction contributions that prevent radar signatures from vanishing at edge-on viewing angles.

Multi-Bounce Geometric Optics (GO / SBR)

To capture cavity reflections, StrikeCEM implements geometric optics ray shooting with up to 3 bounces. Reflected rays propagate along specular directions and illuminate secondary facets, correctly accounting for self-shadowing and corner-reflector returns.

5. GPU Acceleration & Sweep Execution

Evaluating an aircraft model composed of 200,000 triangular facets across full 360° azimuth and elevation sweeps (e.g. 720 azimuth steps $\times$ 360 elevation steps $\times$ 6 polarization channels) requires over 1.5 billion facet integrations.

StrikeCEM parallelizes this on modern multi-core CPUs via OpenMP and on NVIDIA GPUs using a dedicated CUDA compute backend:

// CUDA kernel for parallel facet scattering evaluation
__global__ void evaluateFacetScatteringKernel(
    const FacetData* facets,
    const SweepAngle* angles,
    Complex64* out_scattered_field,
    int facet_count,
    int angle_count
) {
    int angle_idx = blockIdx.x;
    int facet_idx = threadIdx.x + blockIdx.y * blockDim.x;

    if (facet_idx >= facet_count || angle_idx >= angle_count) return;

    // Evaluate facet visibility, phase term, and transverse projection
    Complex64 contrib = evaluateFacetPO(facets[facet_idx], angles[angle_idx]);

    // Atomic accumulation into observation bin
    atomicAddComplex(&out_scattered_field[angle_idx], contrib);
}

The CLI executable (strikecem) provides preflight validation and time estimation before launching compute-heavy jobs:

# Validate CAD mesh topology and JSON configuration schema
strikecem validate configs/su30_flanker_xband.json

# Estimate solve time and memory usage before starting
strikecem estimate configs/su30_flanker_xband.json

# Run Physical Optics sweep on CUDA device and export HDF5 database
strikecem run configs/su30_flanker_xband.json --output su30_rcs.h5

6. HDF5 Storage & StrikeEngine Ingestion

The resulting radar signature is written to a flat-row HDF5 database featuring:

Full Polarimetric Channels: Stores returns across linear (HH, VV, HV, VH) and circular (RHCP, LHCP) bases.
Cryptographic Chunk Checksums: Protects multi-gigabyte databases against disk corruption during long batch sweeps.
Output-Stage Resume: Interrupted sweeps can resume from the last completed angular chunk without re-evaluating the entire mesh from scratch.

During flight simulation in StrikeEngine, the missile’s active radar seeker queries this HDF5 table via zero-copy bilinear interpolation:

\sigma(\theta, \phi, f) \to \text{Seeker Receiver Power } P_r = \frac{P_t G_t G_r \lambda^2 \sigma}{(4\pi)^3 R^4}

This allows StrikeEngine to simulate realistic radar detection ranges, stealth aspect sensitivity, and evasive target maneuvers with microsecond lookup speeds.

Summary

By combining the Physical Optics surface approximation with transverse projector corrections, PTD fringe currents, and CUDA GPU acceleration, StrikeCEM bridges the gap between electromagnetic theory and real-time flight simulation.

The project is available on GitHub: yashwk/StrikeCEM.