PaperlayerView exact revision on arXiv

Source-linked AI summary

Efficient and Accurate Surrogate-Assisted Electromagnetic Parameter Calibration for 6G Digital Twin Channels

Xiaofan Zou, Pan Tang, Peijie Liu, Changyou Tai

arXiv:2608.12872v1eess.SP

TL;DR

Calibrating electromagnetic material parameters for high-fidelity ray-tracing digital-twin channels is computationally expensive and vulnerable to premature convergence. SGWO-IM combines surrogate pre-screening with adaptive convergence and individual memory, reducing real RT simulations from 600 to 225 and RMSE from 3.65 dB to 2.97 dB versus GWO.

  • Problem

    Electromagnetic parameter calibration requires costly forward RT simulations and can prematurely converge because coupled channel parameters reduce population diversity.

  • Method

    SGWO-IM combines an online RBF surrogate for elite-candidate pre-screening with adaptive convergence and individual memory for parameter search.

  • Results

    225 RT simulations versus 600 for GWO reduced RMSE from 3.65 dB to 2.97 dB in measured urban-channel calibration.

  • Takeaways & Limitations

    SGWO-IM balances computational efficiency and calibration accuracy for electromagnetic environment reconstruction in a high-density urban scenario.

Abstract

from arXiv · show

High-fidelity ray-tracing (RT) channel reconstruction is a fundamental step toward building digital twins for the era of 6G wireless communications. However, precise calibration of complex electromagnetic material parameters remains a dual challenge characterized by massive computational overhead and strict accuracy requirements. To overcome this bottleneck, we propose a Surrogate-assisted Grey Wolf Optimizer with Individual Memory (SGWO-IM) algorithm that simultaneously improves computational efficiency and calibration accuracy. In terms of computational efficiency, an online surrogate model is seamlessly embedded into the evaluation workflow for candidate pre-screening, substantially reducing the reliance on highly time-consuming real RT simulations. Regarding calibration accuracy, adaptive convergence and individual memory strategies are incorporated to optimize the global parameter search path, effectively enhancing the consistency between the reconstructed channel and measured data. Validated against measured channel data from a high-density urban scenario, the proposed algorithm requires only 225 real RT simulation calls compared to the 600 calls needed by the standard Grey Wolf Optimizer (GWO), cutting computational overhead by 62.5%. Concurrently, the final Root Mean Square Error (RMSE) is substantially reduced from the 3.65 dB of GWO to 2.97 dB. The results demonstrate that the SGWO-IM algorithm achieves significant advancements in both efficiency and precision, providing a solution that effectively balances efficiency and accuracy for electromagnetic environment reconstruction.

I. INTRODUCTION

Accurate 6G digital-twin channel reconstruction requires calibrating difficult electromagnetic parameters through a computationally expensive, non-convex optimization process. The proposed SGWO-IM addresses this bottleneck with surrogate-assisted RT evaluation and dynamic search mechanisms, reducing RMSE from 3.65 dB to 2.97 dB relative to standard GWO.

  • Motivation: 6G channel foundation models require massive, highly accurate digital-twin channel data generated through ray-tracing reconstruction.Electromagnetic parameters vary significantly with frequency bands and are difficult to measure directly, motivating reverse calibration using measured channel data.
  • Challenge: Reverse calibration is a highly challenging multi-variable, non-convex, and computationally expensive non-linear optimization task.Measured channel data are used because the relevant parameters are difficult to measure directly.
  • Computational bottleneck: Standard heuristic and swarm-intelligence methods incur massive computational overhead by running time-consuming forward RT simulations for every population member at every iteration.Their performance is further challenged by highly coupled natural parameters, limiting practical deployment.
  • Surrogate-assisted calibration: SGWO-IM introduces an online Radial Basis Function (RBF) neural-network surrogate that pre-screens candidates and restricts expensive forward RT simulations to predicted elite individuals.Initial samples support low-cost pre-screening within the surrogate-assisted calibration framework.
  • Dynamic search and accuracy: 3.65 dB to 2.97 dB: SGWO-IM reduces final Root Mean Square Error (RMSE) relative to the standard Grey Wolf Optimizer (GWO).A non-linear cosine convergence factor and individual cognitive component improve exploration beyond blindly following the current best solution, validated with V2V measurements from a typical urban street scenario.

II. PROPOSED METHOD · A. Overall Process

The proposed calibration approach operates as a continuous feedback loop linking scenario inputs, RT-based path-loss simulation, RMSE evaluation, and iterative SGWO-IM parameter updates. RBF surrogate pre-screening improves efficiency, while dynamic search targets more accurate electromagnetic parameter estimation.

  • A. Overall Process: The process begins by feeding XML-based 3D environment data, material categories, Tx/Rx locations, and simulation settings into the RT engine.
  • A. Overall Process: The RT engine executes electromagnetic wave propagation simulations to produce simulated path loss for comparison with empirical V2V data.
  • A. Overall Process: RMSE evaluation quantifies the discrepancy between simulated results and measured data through the stated objective function.
  • A. Overall Process: The objective function uses M as the total number of measurement points and distinguishes RT-simulated path loss from actual measured path loss.
  • A. Overall Process: Error feedback drives SGWO-IM to iteratively update candidate electromagnetic parameters and return them to the RT engine for the next evaluation.
  • A. Overall Process: An RBF surrogate model pre-screens candidates to enhance efficiency, while a dynamic search mechanism improves parameter estimation accuracy.
  • A. Overall Process: The closed-loop calibration process repeats until the convergence criteria are satisfied.

B. Proposed SGWO-IM Algorithm for Parameter Calibration

The SGWO-IM algorithm addresses the computational bottleneck and premature-convergence risks in electromagnetic parameter calibration by improving computational efficiency and calibration accuracy within the standard framework.

  • Challenges: Traditional swarm-intelligence calibration faces a severe computational bottleneck from massive forward RT simulations.
  • Challenges: Highly coupled channel parameters create a risk of premature convergence during calibration.
  • Proposed approach: SGWO-IM optimizes the standard framework along two dimensions: computational efficiency and calibration accuracy.The detailed procedure is summarized in Algorithm 1.

1) Computational Efficiency Enhancement via Online Surrogate:

An online RBF surrogate reduces optimization overhead by predicting candidate RMSE values before real RT evaluation. After initial real-RT sampling and online training, only the top-K predicted candidates proceed for further evaluation.

  • Computational Efficiency Enhancement via Online Surrogate: An RBF neural network is integrated as an online surrogate to predict evaluation errors for candidate parameter combinations.This replaces the need to evaluate every population candidate with the RT engine during each iteration.
  • Computational Efficiency Enhancement via Online Surrogate: After an initial phase of real RT evaluations, historical samples train the RBF model online once an iteration threshold is reached.The early evaluations accumulate reliable initial data samples before surrogate-based screening begins.
  • Computational Efficiency Enhancement via Online Surrogate: The trained RBF model predicts RMSE for all N candidates at negligible computational cost, then ranks them for screening.Only the top-K elite individuals continue in the subsequent evaluation workflow, with K = 5.

2) Calibration Accuracy Enhancement via Dynamic Search:

SGWO-IM enhances calibration accuracy through dynamic search combining a non-linear cosine convergence factor with individual historical memory. The design supports broad early exploration, rapid late-stage fine-tuning, and continued candidate diversity around precise electromagnetic parameters.

  • Dynamic Search Mechanism: A dynamic search mechanism combines a non-linear cosine convergence factor with individual historical best memory to prevent stagnation at suboptimal parameter combinations.The mechanism is designed specifically to enhance electromagnetic parameter calibration accuracy.
  • Dynamic Search Mechanism: The non-linear factor maintains a large early-stage search step for thorough parameter-space exploration before rapidly decaying for late-stage fine-tuning.This supports exploration across wide permittivity bounds, including those of concrete and glass.
  • Individual Historical Memory: Each candidate solution retains its historically best parameter combination, complementing the standard algorithm’s reliance on the current best solution.The individual historical best is denoted Pbest and is incorporated into the parameter update process.
  • Individual Historical Memory: The dynamic memory weight uses wmin = 0.1 and wmax = 0.3 to fuse global guidance with each candidate’s independent historical optimum.This fusion directs candidates toward the global optimal region while preserving reference to their individual best solutions.
  • Individual Historical Memory: Dynamic fusion preserves candidate-solution spatial diversity, directly improving the search for precise electromagnetic parameters.Candidates simultaneously move toward the global optimal region and consult their independent historical optimums.

III. MEASUREMENT AND SIMULATION SETTINGS

The study combines high-density urban V2V channel measurements with a MATLAB-based forward ray-tracing platform for electromagnetic propagation simulation. The SGWO-IM implementation uses an online RBF surrogate, candidate pre-screening, individual memory, adaptive convergence, and RT-based RMSE evaluation.

  • Measurement scenario: 2.1 km × 1.7 km: Measurements covered a high-density urban street scenario near Golden Beach, Qingdao, with buildings generally 60–70 m high.The campaign followed an established urban V2V channel-characterization methodology.
  • SGWO-IM implementation: Top-K elites: When available, the RBF surrogate predicted RMSE for N candidates and selected an evaluation subset before real RT calls.Without an available surrogate, the algorithm evaluated the entire population and added actual (X, RMSE) results to the RBF training set.
  • SGWO-IM implementation: Individual memory: SGWO-IM updated candidate positions by fusing standard GWO guidance with each candidate’s personal best, using nonlinear convergence and dynamic memory factors.The RBF model was retrained online, and the algorithm returned the optimal electromagnetic parameters and minimum RMSE.
  • Measurement hardware: -10 dBm: The transmitter sent a periodic 511-sample PN9 sequence through a vector signal generator, power amplifier, and omnidirectional antenna.The receiver used an omnidirectional antenna, low-noise amplifier, spectrum analyzer, and laptop for data processing.
  • RT simulation platform: MATLAB: The forward RT platform imported .xml geometric data and systematically searched geometric-optics propagation paths between transmitters and receivers.The platform then used the identified path geometries for electromagnetic calculations and field-strength determination.
  • Measurement procedure: 40 valid spatial sample points comprised 27 receiver points for Tx 1 and 13 for Tx 2 during fixed-point measurements.GPS synchronously recorded transceiver coordinates and timestamps, while the receiver continuously recorded 100 sets of samples for measurement consistency.

IV. RESULTS AND ANALYSIS · A. Computational Efficiency Analysis

The computational-efficiency analysis evaluates cumulative execution time over 20 iterations using Fig. 4 and Table III. It shows that SGWO-IM reduces reliance on costly RT simulations, with its cost curve flattening after surrogate intervention.

  • A. Computational Efficiency Analysis: RT-based channel calibration faces substantial computational overhead, motivating an evaluation of algorithmic acceleration.The analysis uses cumulative execution time over 20 iterations and quantitative cost comparisons.
  • A. Computational Efficiency Analysis: 600 RT simulations are required by GWO to complete the calibration process.These simulations are described as time-consuming.
  • A. Computational Efficiency Analysis: 831.68 s is the total execution time reported for GWO.The value accompanies GWO’s 600 required RT simulations.
  • A. Computational Efficiency Analysis: SGWO-IM’s cumulative computational-cost curve flattens significantly after the 5th iteration.This behavior contrasts with GWO’s steep linear growth in computational cost.
  • A. Computational Efficiency Analysis: The vertical dashed line in Fig. 4 marks the transition labeled “Surrogate Intervention”.It identifies the critical transition point in the cumulative-cost comparison.
  • A. Computational Efficiency Analysis: The RBF surrogate model is sufficiently trained at the surrogate-intervention transition point.The passage identifies this training condition as the critical transition in the workflow.

B. Calibration Accuracy Analysis

SGWO-IM improves calibration accuracy by avoiding GWO’s premature local convergence and achieving lower RMSE while maintaining stronger agreement with measured path-loss data across the receiver indices. Its global optimization behavior can slightly sacrifice fit at an isolated point to prevent overfitting and preserve spatial fidelity.

  • Convergence accuracy: 3.65 dB was GWO’s stagnation point at the 15th iteration, whereas SGWO-IM reached 3.02 dB by iteration 15 and ultimately converged to 2.97 dB.SGWO-IM benefits from a non-linear convergence factor and individual history mechanism to escape local traps and continue exploring the electromagnetic parameter space.
  • Measured-data consistency: 40 Rx indices showed SGWO-IM consistency with the measured curve, particularly in the high-loss regions at Rx 10–16 and Rx 25–27.At Rx 15, GWO produced a local deviation of 4.5 dB by failing to capture the measured signal fluctuation.
  • Measured-data consistency: 4.5 dB was GWO’s local deviation at Rx 15, while SGWO-IM adhered more closely to the measured fluctuation in that high-loss region.The comparison highlights SGWO-IM’s stronger robustness where GWO fails to reproduce local signal behavior.
  • Robustness and limitations: At Rx 9, GWO aligned slightly better with measurement than SGWO-IM because premature convergence coincidentally matched that point’s multipath superposition.SGWO-IM instead prioritizes global spatial fidelity and avoids overfitting isolated noise points.

V. CONCLUSION

The paper proposes SGWO-IM for electromagnetic-parameter calibration in complex urban 6G digital-twin channel reconstruction, combining online surrogate pre-screening with individual memory. The approach addresses the traditional tradeoff between computational efficiency and global calibration accuracy.

  • SGWO-IM is proposed for electromagnetic-parameter calibration supporting 6G digital-twin channel reconstruction in complex urban environments.
  • The method integrates online surrogate pre-screening and an individual memory strategy to improve computational efficiency and global calibration accuracy.
Loading 2608.12872v1…