Sub-THz ISAR Imaging: Techniques & Applications

• Sub-THz ISAR imaging is a radar technique that uses high-frequency signals and motion-induced aperture synthesis to produce high-resolution reflectivity maps of moving targets.
• It employs FMCW chirp signals, advanced processing algorithms, and GPU acceleration to achieve centimeter-level range resolution and detailed structural reconstructions.
• Applications span automotive sensing, space object tracking, and security imaging, with system innovations enabling quasi-real-time 2D and 3D imagery.

Sub-terahertz (sub-THz) Inverse Synthetic Aperture Radar (ISAR) imaging encompasses techniques, algorithms, and hardware for reconstructing high-resolution two-dimensional (2D) and three-dimensional (3D) reflectivity maps of moving targets using electromagnetic signals in the frequency range from roughly 20 GHz up to 300 GHz. Operating at these frequencies provides fine range and cross-range resolution, enabling detailed structure reconstruction of complex objects such as vehicles, spaceborne objects, and humans. The ISAR modality exploits relative motion (target or radar) to synthesize a large virtual aperture, with sub-THz implementations benefiting from wide instantaneous bandwidth and short wavelengths.

1. Fundamental Principles of Sub-THz ISAR Imaging

Sub-THz ISAR leverages the same signal-processing framework as lower-frequency counterparts, but with modified physics and hardware-specific considerations. The transmitted waveform is typically a linear-frequency-modulated (LFM) Frequency-Modulated Continuous Wave (FMCW) chirp, $$s^{(\mathrm{tx})}(\tau) = \mathrm{rect}(\tau/T_{\mathrm{PRI}}) \cdot \exp[j2\pi f_c \tau + j\pi K \tau^2]$$, with center frequency $f_c$ in the sub-THz regime (24–300 GHz) and bandwidth $B$ up to several GHz. The resulting range resolution is $\Delta R = c/(2B)$, often 3–4 cm for $B=4$–5 GHz systems. Doppler or synthetic-aperture processing in the cross-range (azimuth) domain exploits either target motion or controlled platform motion, producing two-dimensional or three-dimensional spatial reconstructions (Pandey et al., 2021, Zhu et al., 2019, Coe et al., 17 Dec 2025, Smith et al., 2023).

At sub-THz, the high carrier frequency yields short wavelengths ($\lambda$ down to ~1 mm at 300 GHz), significantly increasing cross-range sensitivity to rotational or translational motion, enabling detailed micro-structure imaging. The small wavelength also amplifies target scattering effects, including surface roughness and resonance phenomena, and accentuates the sensitivity to system and platform instabilities.

2. System Architectures and Operating Regimes

Several canonical sub-THz ISAR platforms are documented for contemporary research and applications:

• Monostatic Automotive Systems: 77 GHz FMCW radar (e.g., Texas Instruments AWR-1843) with $B=2$ GHz, PRF of 2.5–12 kHz, synthetic-aperture formation via vehicle maneuver, and target RCS models based on facet assemblies (Pandey et al., 2021).
• Spaceborne Monitoring Platforms: 30 cm physical antenna at center frequency $f_o\in\{75, 300\}$ GHz, with $B=5$ GHz, monitoring targets at $R\leq 100$ km, yielding $\Delta x = \Delta y = 3$ cm resolution (Coe et al., 17 Dec 2025).
• Security Imaging Systems: Sparse MIMO arrays operating at 24 GHz, with $N_\mathrm{tx}=8$, $N_\mathrm{rx}=16$, providing real-aperture imaging in elevation and synthesized cross-range via target linear motion (Zhu et al., 2019).
• Laboratory Near-Field R-ISAR: 77 GHz MIMO radar, mechanical target rotation and linear scanning for full 3D imaging in polar format, with Stolt mapping and multistatic-to-monostatic phase compensation for computational efficiency (Smith et al., 2023).

These systems may operate in ground, airborne, or space-based regimes, with near-field or far-field geometries, and utilize MIMO, SISO, or rotating-platform approaches depending on application requirements.

3. Signal, Scattering, and Imaging Models

ISAR imaging relies on modeling both the transmitted waveform and the electromagnetic scattering phenomenology of the target:

• Radar Signal Model: The received baseband signal for a point scatterer incorporates round-trip delay, Doppler shift, and matched filter (or stretch) processing. For ISAR, the 2D FFT over fast (range) and slow (Doppler) time produces a range–Doppler ($\chi(f_D, r)$) surface, which is remapped to cross-range using the relationship $cr = f_D\,\,(\lambda_c/2\omega)$, with cross-range resolution $\Delta cr = \lambda_c/(2\omega T_{\mathrm{CPI}})$.
• Scattering Phenomenology: Detailed CAD-based facet models (triangular surfaces) are used to compute aspect-dependent RCS for automotive or satellite targets. For satellites and RSOs, the backscattered field is expressed as $$E^s(k,\phi) = \sum_{i=1}^K A_i\,\exp[-j2k \cdot r_i]$$ over a set of $K$ dominant scattering centers (Coe et al., 17 Dec 2025).
• 3D Imaging and MIMO: For MIMO or near-field arrays, mechanics involve comprehensive phase models for transmit and receive paths, multistatic–monostatic phase compensation, and polar-format resampling (Stolt mapping). Back-projection algorithms are implemented for 3D imaging, taking into account both near-field spherical-wave propagation and platform motion (Zhu et al., 2019, Smith et al., 2023).

Clutter and noise modeling is essential, with explicit inclusion of surface clutter cross-sections, Doppler clutter spectra, and additive receiver noise, leading to robust simulation and algorithm validation (Pandey et al., 2021).

4. Sequence Processing and Feature Extraction

High-fidelity ISAR image sequences enable persistent feature recognition, tracking, and structural analysis of complex targets. Several key algorithmic tools are used:

• Frame Alignment: Each ISAR frame is coarsely segmented and registered to a "body-fixed" coordinate system using affine transforms: $$A = R(\theta)\cdot \mathrm{Sh}(d_x,d_y)\cdot \mathrm{Sc}(c_x,c_y)$$ plus translation $S$ (Coe et al., 17 Dec 2025).
• Edge Detection: The Ratio of Exponentially Weighted Averages (ROEWA) is employed for edge detection in the presence of speckle, producing magnitude and orientation maps $G(x,y),\,\Theta(x,y)$ for mask formation.
• Double-weighted Hough Transform: Edge pixels identified by ROEWA vote in Hough space, with weights proportional to both gradient magnitude $W_M(x,y) = G(x,y)$ and orientation consistency $$W_D(x,y;\theta) = \exp[-(\theta - \Theta(x,y))^2/(2\sigma^2)]$$, robustly extracting line features and preserving edge polarities (Coe et al., 17 Dec 2025).
• Feature Tracking: Detected features are clustered in parameter space, e.g., $(\rho,\theta)$ across frames using DBSCAN. Persistent clusters correspond to physical structure edges; tracking their temporal evolution allows distinction between rigid features and moving/shadow-induced edges, as in cast-shadow detection for satellites (Coe et al., 17 Dec 2025).

Statistical analysis over aligned sequences, including PCA and temporal span, refines feature classification, yielding high-confidence composite images and enabling downstream analysis such as space domain awareness (SDA).

5. Experimental Demonstrations and Performance Metrics

Sub-THz ISAR has been validated in several application domains with quantitative performance results:

Application	Carrier (GHz)	Bandwidth (GHz)	Range/CR Resolution	Acquisition Mode	Key Validation Results
Automotive Target Recognition	77	2	3.7 cm	Vehicle motion, ground-based	>90% classification; fidelity to physical dimensions (Pandey et al., 2021)
RSO Feature Mapping (Space)	300	5	3 cm	Space-monitoring, synthetic aspect	Persistent line recovery; shadow/structure separation (Coe et al., 17 Dec 2025)
3D Security Human Imaging	24	4	3.75/1.2/1.8 cm	Sparse MIMO, linear motion	Concealed object detection, 96% recognition (Zhu et al., 2019)
Near-Field MIMO R-ISAR	77	4	3.75 mm-4 cm	Mechanical scan, rotary + vertical	<2 s 3D recon on CPU; 20 dB CNR; shape detail (Smith et al., 2023)

Resolution is bandwidth-limited: $\Delta R = c/(2B)$ for range, $\Delta_{CR} = \lambda/(2\Delta\theta)$ for cross-range (azimuthal span $\Delta\theta$). Real-world measurements confirm the simulation models, with features such as car dimensions, wheel micro-Doppler signatures, and detailed satellite structure all evident at sub-THz frequencies.

Classification metrics (where reported) show ISAR-based recognition accuracy exceeding 90%. For security imaging, CNN-based recognition achieves 96% accuracy with 8% false alarm and 0.3% missed detection rates (Zhu et al., 2019). Feature-tracked ISAR for SDA enables mapping of satellite panels, bus, and functional appendages with resilience to artefacts and occlusions (Coe et al., 17 Dec 2025).

6. Computational Architectures and Algorithmic Efficiency

Efficient reconstruction of sub-THz ISAR data, especially in 3D or with large MIMO arrays, demands optimized computational pipelines:

• GPU-based Acceleration: Range FFTs with CUFFT, bulk complex processing with CUBLAS, and custom CUDA kernels for voxel summations and phase compensation result in 400× speedup over single-CPU implementations, enabling $200 \times 100 \times 100$ volumes in ≈1 s (Zhu et al., 2019).
• Spectral Resampling and Inverse FFTs: Polar-format reconstructions, Stolt mapping, and 3D IFFT on large grids ($256 \times 64 \times 256$) allow 3D imagery within $<2$ s on modern CPUs (Smith et al., 2023).
• Algorithmic Innovations: Multistatic-to-monostatic phase compensation enables SISO-grade speed for MIMO arrangements. Efficient clustering (e.g., DBSCAN in edge-feature space) allows real-time structure extraction even in high-clutter or low-SNR settings (Coe et al., 17 Dec 2025).

These advancements facilitate "quasi-real-time" imaging, making sub-THz ISAR suitable for operational scenarios including moving vehicle classification, security screening, and in-orbit satellite surveillance.

7. Application Domains and Future Directions

Sub-THz ISAR imaging demonstrates utility across multiple domains:

• Automotive Sensing: Classification and identification of vehicles and pedestrians with resilience to noise, clutter, and varying trajectories (Pandey et al., 2021).
• Space Domain Awareness: Persistent feature reconstruction and shadow analysis on RSOs, yielding robust structural mapping without extensive training datasets (Coe et al., 17 Dec 2025).
• Security Screening: 3D identification and recognition of concealed objects on moving subjects with high spatial resolution and rapid throughput (Zhu et al., 2019).
• Scientific and Industrial Inspection: Laboratory-based high-fidelity imaging of objects with complex geometry using efficient MIMO R-ISAR platforms (Smith et al., 2023).

A notable trend is the integration of advanced signal-processing (feature tracking, robust edge detection, spectral compensation) with hardware innovations (MIMO, wideband FMCW, mechanical scanning, GPU/CPU acceleration). Persistent monitoring, feature-level temporal analysis, and multimodal fusion are emerging research directions, expanding ISAR's utility for autonomous vehicles, orbital object management, and intelligent security systems.

A plausible implication is that as bandwidths and carrier frequencies continue to increase, detailed microstructure imaging and automatic feature extraction will become routine, contingent on ongoing development of computationally efficient, robust reconstruction algorithms and affordable high-frequency hardware.