Source-linked AI summary

Accelerated S-NFC for Million-Chaff RCS Computation Using Low-Rank Compression of Concatenated Block Rows

Dong-Yeop Na, Somyeong Lee, Chung Hyun Lee

arXiv:2608.27936v1physics.comp-phcs.IT

TL;DR

Large chaff clouds make full-wave RCS computation costly even after S-NFC sparsification leaves substantial retained coupling. The paper jointly low-rank compresses concatenated receiving-side block rows while preserving exact self interactions and reusing template factorizations. For a one-million-chaff plume, it achieves 6.92× end-to-end speedup with 6.60% retained-coupling storage and 0.253% complex-far-field error.

  • Problem

    S-NFC still retains off-diagonal coupling that creates substantial memory and iterative-solver costs at million-element scale.

  • Method

    The method jointly factorizes concatenated retained interactions for each receiving chaff, sharing a receiving-side basis while preserving exact self interactions.

  • Results

    For a one-million-chaff plume, the method achieves 6.92× end-to-end speedup over uncompressed S-NFC with 6.60% retained-coupling storage and 0.253% complex-far-field error.

  • Takeaways & Limitations

    Low-rank-accelerated S-NFC supports large-scale full-wave chaff analysis, with required rank governed by local morphology and coupling density.

Abstract

from arXiv · show

Sparsification via neglecting far-field coupling (S-NFC) enables fast full-wave radar-cross-section analysis of large-scale chaff clouds by retaining only significant local electromagnetic interactions. This letter further accelerates S-NFC by concatenating the retained off-diagonal interaction blocks associated with each receiving chaff element and applying a joint low-rank factorization with a shared receiving-side basis. Exact self interactions are preserved, while repeated chaff templates reuse precomputed lower--upper factorizations of the self-interaction blocks. The compressed formulation reduces retained-coupling storage and matrix--vector multiplication cost and also decreases the number of iterations required by the generalized conjugate residual solver. Numerical tests with 100,000 chaff elements demonstrate sub-$1\%$ complex-far-field error for low-rank approximations in sparse regimes and identify a practical self-only limit at sufficiently large mean spacing. For a one-million-chaff plume, the proposed compressed S-NFC achieves a $6.92\times$ end-to-end speedup over uncompressed S-NFC while storing only $6.60\%$ of the retained coupling, with a complex-far-field error of $0.253\%$.

I. INTRODUCTION

Large chaff-cloud RCS prediction requires resolving full-wave induced currents and interactions, but dense impedance systems become computationally and memory intensive at scale. The paper accelerates sparsified S-NFC by jointly compressing retained interactions associated with each receiving chaff.

  • Full-wave EFIE/MoM formulations directly resolve induced currents and inter-element interactions but incur high computational and memory costs for large chaff clouds.
  • S-NFC exploits progressively weaker interactions between sufficiently separated chaff elements to reduce the retained interaction problem.
  • Existing surrogate approaches accelerate analysis but rely on S-NFC-generated training data, motivating optimization of the underlying physics-based solver.
  • Retained off-diagonal S-NFC blocks can still impose substantial memory and iterative-solver costs at million-element scale.
  • The proposed receiving-chaff-wise factorization concatenates retained interactions and shares a receiving-side basis across neighboring source interactions.
  • Validation covers 100,000-chaff density and morphology studies, template-shape tests, and one-million-chaff uniform and plume clouds.

II. S-NFC WITH EXACT REUSE AND CONCATENATED-ROW COMPRESSION

The method preserves exact self interactions while applying low-rank compression to retained mutual-coupling block rows. Shared receiving-side bases, template-level LU reuse, and adaptive ranks reduce storage and solver work.

  • System formulation: The Galerkin TWA-EFIE system is written as Zx = b and split into exact block-diagonal self interactions plus retained mutual coupling.
  • S-NFC sparsification: S-NFC selects retained source elements geometrically using a cutoff distance, and compression changes only the representation of that retained coupling.
  • Exact self reuse: Repeated chaff templates share self blocks and precomputed LU factors, reducing storage while preserving exact template solves.
  • Concatenated-row compression: Concatenating all retained blocks for one receiving element exposes shared column-space redundancy and motivates a common receiving-side basis.
  • Rank selection: Adaptive row ranks are selected from singular values using a discarded singular-value-energy tolerance τE, with zero rank representing a diagnostic block-diagonal-only limit.
  • Diagnostic limit: At r = 0, mutual coupling is removed while exact self-block solutions remain, quantifying when mutual coupling is negligible for the selected observable.
  • Computational effect: The compressed formulation preserves exact self interactions and block-diagonal preconditioning while reducing retained-coupling storage and matrix–vector multiplication cost.

III. NUMERICAL RESULTS

The numerical evaluation compares compressed and uncompressed S-NFC under matched workloads and measures complex-far-field error and end-to-end runtime. Accuracy errors isolate compression effects relative to the uncompressed S-NFC reference.

  • Evaluation protocol: The uncompressed S-NFC solution for the same realization, excitation, cutoff, and stopping criterion serves as the quantitative reference.
  • Accuracy metric: Complex-far-field accuracy uses 720 complex-field observation samples, with error isolating low-rank compression or block-diagonal-only effects.
  • Runtime metric: End-to-end runtime includes assembly, 360 right-hand-side solves, monostatic and bistatic evaluation, preconditioning, and I/O under identical workloads.

A. Density and Morphology Dependence

Compression accuracy depends on mean spacing and local morphology: sparse clouds can support very low ranks or even self-only modeling, while denser local coupling increases error and difficulty.

  • 0.187% complex-far-field error is achieved by rank one at measured mean nearest-neighbor spacing 2λ.
  • 0.919% complex-far-field error is achieved by adaptive τE = 0.01 at 0.5λ, compared with 2.64%, 1.86%, and 0.951% for fixed ranks 1, 2, and 4.
  • 27.17%, 1.15%, 0.602%, and 0.383% are the block-diagonal-only errors at mean spacings 0.5λ, 2λ, 3λ, and 4λ, respectively.
  • The block-diagonal-only model is a practical self-only approximation in the tested regime d̄NN ≥ 3λ, where complex-far-field error is below 1%, but the threshold is not universal.
  • At matched spacing 2λ, the plume has 5.67 mean neighbors within 3λ versus 2.41 for the uniform sphere, indicating different local coupling graphs.
  • For fixed rank one, uniform/plume errors are 0.187%/0.845%, with end-to-end speedups of 3.17×/4.74×; the plume is more difficult because of denser local coupling.

B. Template-Shape Robustness and Iterative Behavior

The shape-robustness tests show that rank-one compression closely preserves bistatic RCS for twisted-bent chaff, while compressed operators can also reduce iterative-solver work.

  • Template-shape robustness: Rank-one compression produces nearly coincident full and compressed RCS curves for the 100,000-chaff twisted-bent cloud across the 720-point bistatic cut.The test fixes sphere centers and orientations while changing only the straight, bent, or twisted-bent template shape.
  • Cloud morphology: Fig. 2 encodes morphology using color for the acute axial angle from the vertical direction at matched mean spacing 2λ.
  • Template-shape robustness: 0.188% complex-far-field error is reported for the twisted-bent run.
  • Iterative behavior: 10 iterations are required at ranks one and two, compared with 56 for full S-NFC in the one-million-chaff straight-cloud test.Rank four requires 15 iterations; these counts refer to the respective approximate operators.
  • Morphology dependence: Fig. 3 compares field error and end-to-end speedup with black and red curves on identical axis scales across the morphology cases.

C. Million-Chaff Validation

The million-chaff validation compares uniform and plume clouds and evaluates adaptive compression using field error, storage, speedup, and full-sweep costs. The plume has a larger retained graph and achieves substantial compression and acceleration at low error.

  • Cloud setup: M1 and M2 are uniform and plume clouds matched at 2.5λ mean nearest-neighbor spacing, with 1,236,058 and 3,029,936 directed retained blocks.Their mean neighbor counts within 3λ are 1.236 and 3.030, respectively.
  • Solver convergence: Figure 5 reports GCR residual histories for the one-million-chaff straight-cloud baseline at f = 8.427 GHz.The supplied figure materials identify the residual-history comparison and operating frequency.
  • Baseline cost: The plume has a 2.45× larger retained graph and higher full-sweep cost, peak memory, and average GCR iterations than the uniform cloud.Full sweeps take 1.405/5.707 h, peak memory is 19.81/38.57 GiB, and average iterations are 9.44/25.84 for M1/M2.
  • Adaptive tradeoff: For M1, τE = 0.5 yields 0.043% error, 8.28% retained-coupling storage, and 2.05× speedup, while τE = 0.01 yields 0.018% error and 1.73× speedup.The corresponding mean ranks are 1.000 and 1.404.
  • Adaptive tradeoff: For M2, τE = 0.5 achieves 0.253% error, 6.60% retained-coupling storage, and 6.92× speedup.The corresponding mean rank is 1.001.

IV. CONCLUSION

Concatenated-row low-rank compression accelerates S-NFC by sharing a receiving-side basis across retained neighboring interactions while preserving exact self interactions. In a one-million-chaff plume, it combines low storage with high speedup and low complex-far-field error.

  • Conclusion: The method shares a receiving-side basis across retained neighboring interactions while preserving exact self interactions.Stronger compression also reduces matrix–vector multiplication cost and generalized conjugate residual iterations.
  • Conclusion: 6.60% retained-coupling storage, 6.92× end-to-end speedup, and 0.253% complex-far-field error are achieved for a one-million-chaff plume.The comparison is against uncompressed S-NFC.
Loading 2608.27936v1…