A Material-Aware Channel Model for Efficient CKM Generation via Environment Reconstruction

I. INTRODUCTION

• Environment-reconstruction CKM generation first recovers 3D geometry and materials, then generates channel knowledge, while material acquisition remains challenging despite LiDAR geometry being readily available.The approach is described as offering explainability, generalizability, and improved data efficiency compared with direct generation methods that require abundant measurements.Source: §I · p. 1
• Material inference from sparse channel measurements requires an explicit link to channel response, which prior statistical models omit and other cited models represent only implicitly through radar cross section.The implicit radar cross section depends on material, directions, and frequency.Source: §I · p. 1
• The proposed electromagnetic-theory channel model explicitly represents scatterer materials through relative permittivities and conductivities, alongside an iterative gradient descent reconstruction algorithm.Full-wave simulation is used to validate the model and algorithm.Source: §I · p. 1

II. SYSTEM MODEL

• With precise scatterer geometries as prior knowledge, the system model uses sparse channel measurements to infer the material vector for S uniform scatterers.The measurements are channel impulse responses at N spatially sparse receiver positions; the material vector has dimension 2S.Source: §II · p. 1

A. Channel response relative to materials

• The far-field radar equation relates each scatterer’s channel component to its bistatic RCS; its delay is the total transmitter-to-scatterer-to-receiver path length divided by c.The component’s phase and received-to-emitted power ratio are also specified in the model.Source: A. Channel response relative to… · pp. 1–2
• Scatterer s is partitioned into Ms rectangular facets, whose parameter sets include facet position, orientation, dimensions, and the scatterer’s permittivity and conductivity.Each facet’s bistatic RCS is evaluated from its parameter set and incident and observation directions.Source: Figure 1 · p. 2, A. Channel response relative to… · p. 2
• The model defines the scatterer’s bistatic RCS using the facet-level RCS contributions, with each facet represented by its own parameter set.The passage identifies the facet RCS as a function of the facet parameters and incident and observation directions.Source: A. Channel response relative to… · p. 2

B. The closed-form expression of facet’s RCS

• The material properties enter the reflection coefficients through relative permittivity and conductivity, alongside the incidence angle.The section defines the vertical and parallel-polarization reflection coefficients and the material-dependent quantity χ(ε, σ).Source: B. The closed-form expression of… · p. 2
• The facet RCS is derived from the incident and scattered electric-field phasors, with the scattered wave partitioned into two orthogonal components under far-field conditions.The simple case uses a single rectangular facet at the origin; its dimensions and material properties are included in its parameter set.Source: B. The closed-form expression of… · pp. 2–3
• Once the environment geometry is known, the RCS depends explicitly on material properties, enabling inference of those properties from channel measurements.The section extends the facet RCS to the scatterers in the environment and uses it to obtain the channel-related quantity Ξ(qT, qR).Source: B. The closed-form expression of… · p. 3
• For an arbitrarily positioned facet, the incident and observation angles are obtained from its position and orientation, subject to far-field distance constraints.The facet RCS is then evaluated using these angles and the expressions for the material and angular terms.Source: B. The closed-form expression of… · p. 3

IV. MATERIAL RECONSTRUCTION

• The proposed algorithm addresses the material reconstruction problem.The section identifies material reconstruction as the algorithm’s target.Source: §IV · p. 3
• The algorithm is iterative.The section describes the proposed approach as iterative gradient descent-based.Source: §IV · p. 3
• The proposed approach is based on gradient descent.The section names gradient descent as the basis of the material reconstruction algorithm.Source: §IV · p. 3

A. Evaluation metric and loss function

• The channel difference is evaluated using path gain and delay, with RMS delay spread defined as a metric.The modeled channel is compared with measurements across samples.Source: A. Evaluation metric and loss function · p. 3
• MASE is introduced as a second metric for comparing the modeled channel with measurements.The passage places MASE alongside the RMS delay spread evaluation.Source: A. Evaluation metric and loss function · p. 3
• The metrics are combined into a weighted loss, with weights intended to emphasize positions with large deviation.The loss uses the modeled and measured channels; the weight is specified as positive.Source: A. Evaluation metric and loss function · p. 3

B. Optimization problem and the proposed algorithm

• The reconstruction objective minimizes MASE across N samples by adjusting the material vector m.The section introduces this as the optimization problem for material reconstruction.Source: B. Optimization problem and the… · p. 4
• Closed-form solutions are difficult because the model has complex expressions and includes square roots in the complex domain.The paper therefore uses an iterative algorithm to approximately solve the optimization problem.Source: B. Optimization problem and the… · p. 4
• The algorithm obtains the loss gradient with respect to material parameters using automatic differentiation, then updates the materials by gradient descent.The update uses a learning rate and utility functions that restrict the learning rate and material values.Source: B. Optimization problem and the… · p. 4
• Each iteration computes modeled channels and loss for the current materials, updates the learning rate and materials, and continues until the iteration limit.The algorithm outputs the optimized materials and final loss.Source: B. Optimization problem and the… · p. 4

• CKM accuracy is evaluated using achievable rate R(H), comparing ground-truth H with channels estimated using default ITU or reconstructed materials.Source: C. Quantitative metric for CKM… · p. 4
• The imperfect-channel rate uses water-filling power allocation derived from the channel’s singular values.Source: C. Quantitative metric for CKM… · p. 4
• The normalized error err(H, ˆH) lies in [0, 1), and errors are averaged over Nsam samples.Source: C. Quantitative metric for CKM… · p. 4

• The proposed RCS model agrees well with full-wave simulations, achieving a MASE of −13.88 dB relative to PO.Source: §V · p. 4
• 11.66% to 4.39%: average percentage error decreases when reconstructed materials replace default ITU materials for six scatterers.The median position-wise MASE of HR and H is −16.34 dB.Source: §V · p. 4
• The material-reconstruction evaluation uses a 3D environment with six scatterers whose ground-truth materials follow ITU specifications.The evaluation uses 3.5 GHz, SNR = 90 dB, and 1,000 samples.Source: §V · p. 4

• The paper presents a material-aware channel model that explicitly links material properties to the channel response.Source: §VI · p. 4
• An iterative gradient descent algorithm is proposed to reconstruct materials for CKM generation.Source: §VI · p. 4
• Fig. 5 compares err(H, ˆHITU) with err(H, ˆHR).Source: Figure 5 · p. 5

APPENDIX A PROOF OF LEMMA 1

• Lemma 1’s proof relates the incidence angle ρ to the material-dependent expression and uses Snell’s law to obtain equations (7) and (8).Source: Appendix A · p. 5

APPENDIX B PROOF OF THEOREM 1

• Theorem 1’s proof derives the reflected electric wave using Snell’s law, then substitutes the reflected-wave relations into the induced surface-current expression.Source: Appendix B · p. 5

APPENDIX C PROOF OF THEOREM 2

• Theorem 2’s proof uses the far-field approximated Stratton-Chu equation to derive the scattering wave phasors in equation (16).Source: Appendix C · p. 5