Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beamforming-Codebook-Aware Channel Knowledge Map Construction for Multi-Antenna Systems

Haohan Wang, Xu Shi, Hengyu Zhang, Yashuai Cao, Jintao Wang

TL;DR

This work tackles the problem of constructing beamforming-aware CKMs for multi-antenna systems by introducing a TransUNet-based framework that integrates DFT precoding vectors and environment-aware attention. The proposed encoder–decoder architecture combines CNN, Transformer, and UNet components, paired with a composite loss (L2, Laplacian pyramid, and edge-aware terms) to capture both fine-scale details and beamforming-induced patterns. A beamforming-codebook-specific formulation enables CKMs to be conditioned on finite codebooks, improving practicality and robustness. Experiments on a 10,000-scenario ray-tracing dataset show that TransUNet outperforms state-of-the-art DL approaches with a 17% RMSE reduction and achieves real-time inference (~0.017 s), highlighting its potential to reduce pilot overhead and enhance beam management in mmWave/THz systems.

Abstract

Channel knowledge map (CKM) has emerged as a crucial technology for next-generation communication, enabling the construction of high-fidelity mappings between spatial environments and channel parameters via electromagnetic information analysis. Traditional CKM construction methods like ray tracing are computationally intensive. Recent studies utilizing neural networks (NNs) have achieved efficient CKM generation with reduced computational complexity and real-time processing capabilities. Nevertheless, existing research predominantly focuses on single-antenna systems, failing to address the beamforming requirements inherent to MIMO configurations. Given that appropriate precoding vector selection in MIMO systems can substantially enhance user communication rates, this paper presents a TransUNet-based framework for constructing CKM, which effectively incorporates discrete Fourier transform (DFT) precoding vectors. The proposed architecture combines a UNet backbone for multiscale feature extraction with a Transformer module to capture global dependencies among encoded linear vectors. Experimental results demonstrate that the proposed method outperforms state-of-the-art (SOTA) deep learning (DL) approaches, yielding a 17\% improvement in RMSE compared to RadioWNet. The code is publicly accessible at https://github.com/github-whh/TransUNet.

Beamforming-Codebook-Aware Channel Knowledge Map Construction for Multi-Antenna Systems

TL;DR

This work tackles the problem of constructing beamforming-aware CKMs for multi-antenna systems by introducing a TransUNet-based framework that integrates DFT precoding vectors and environment-aware attention. The proposed encoder–decoder architecture combines CNN, Transformer, and UNet components, paired with a composite loss (L2, Laplacian pyramid, and edge-aware terms) to capture both fine-scale details and beamforming-induced patterns. A beamforming-codebook-specific formulation enables CKMs to be conditioned on finite codebooks, improving practicality and robustness. Experiments on a 10,000-scenario ray-tracing dataset show that TransUNet outperforms state-of-the-art DL approaches with a 17% RMSE reduction and achieves real-time inference (~0.017 s), highlighting its potential to reduce pilot overhead and enhance beam management in mmWave/THz systems.

Abstract

Channel knowledge map (CKM) has emerged as a crucial technology for next-generation communication, enabling the construction of high-fidelity mappings between spatial environments and channel parameters via electromagnetic information analysis. Traditional CKM construction methods like ray tracing are computationally intensive. Recent studies utilizing neural networks (NNs) have achieved efficient CKM generation with reduced computational complexity and real-time processing capabilities. Nevertheless, existing research predominantly focuses on single-antenna systems, failing to address the beamforming requirements inherent to MIMO configurations. Given that appropriate precoding vector selection in MIMO systems can substantially enhance user communication rates, this paper presents a TransUNet-based framework for constructing CKM, which effectively incorporates discrete Fourier transform (DFT) precoding vectors. The proposed architecture combines a UNet backbone for multiscale feature extraction with a Transformer module to capture global dependencies among encoded linear vectors. Experimental results demonstrate that the proposed method outperforms state-of-the-art (SOTA) deep learning (DL) approaches, yielding a 17\% improvement in RMSE compared to RadioWNet. The code is publicly accessible at https://github.com/github-whh/TransUNet.

Paper Structure

This paper contains 14 sections, 14 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model
  3. Beamforming Codebook-Specific CKM
  4. TransUNet-based CKM construction
  5. CNN-Transformer Hybrid as Encoder
  6. Cascaded UNet-Structured Upsampler
  7. Loss Function Design
  8. L2 (Mean Squared Error) Loss
  9. Laplacian Pyramid Loss
  10. Edge-Aware Loss
  11. Simulations and discussion
  12. Dataset and Simulation Setup
  13. Discussion
  14. Conclusion

Figures (5)

  • Figure 1: Comparison of beam-aware and beam-unaware pathloss.
  • Figure 2: Schematic diagram of real-world modeling and CKM construction based on the TransUNet.
  • Figure 3: The detailed structural schematic of TransUNet for CKM construction: CNN extracts hidden features, the Transformer layer learns contextual relationships, and the upsampling module reconstructs the CKM.
  • Figure 4: Visualization of predicted CKM and ground truth with specific beamforming.
  • Figure 5: Overall comparison between different loss functions.