An I2I Inpainting Approach for Efficient Channel Knowledge Map Construction

TL;DR

This work tackles the heavy overhead of constructing environment-aware channel knowledge maps (CKMs) by recasting CKM construction as an image-to-image inpainting task. It introduces a Laplacian pyramid-based CGM reconstruction network (LPCGMN) that splits high-resolution maps into low- and high-frequency components and employs dedicated subnetworks for each to boost efficiency, noting that CGM–geometry differences concentrate in low-frequency bands. The approach combines lightweight LRDC blocks with multi-head self-attention (MHSA) and cross-covariance attention (MHCCA) to fuse local and global environment information, achieving higher reconstruction accuracy at lower computational cost than Unet/WNet baselines and generalizing well to unseen scenarios. This yields a practical, scalable path to ultra-high-resolution CKMs for real-time ISAC and 6G deployments, enabling environment-aware transceiver design with reduced latency and resource use.

Abstract

Channel knowledge map (CKM) has received widespread attention as an emerging enabling technology for environment-aware wireless communications. It involves the construction of databases containing location-specific channel knowledge, which are then leveraged to facilitate channel state information (CSI) acquisition and transceiver design. In this context, a fundamental challenge lies in efficiently constructing the CKM based on a given wireless propagation environment. Most existing methods are based on stochastic modeling and sequence prediction, which do not fully exploit the inherent physical characteristics of the propagation environment, resulting in low accuracy and high computational complexity. To address these limitations, we propose a Laplacian pyramid (LP)-based CKM construction scheme to predict the channel knowledge at arbitrary locations in a targeted area. Specifically, we first view the channel knowledge as a 2-D image and transform the CKM construction problem into an image-to-image (I2I) inpainting task, which predicts the channel knowledge at a specific location by recovering the corresponding pixel value in the image matrix. Then, inspired by the reversible and closed-form structure of the LP, we show its natural suitability for our task in designing a fast I2I mapping network. For different frequency components of LP decomposition, we design tailored networks accordingly. Besides, to encode the global structural information of the propagation environment, we introduce self-attention and cross-covariance attention mechanisms in different layers, respectively. Finally, experimental results show that the proposed scheme outperforms the benchmark, achieving higher reconstruction accuracy while with lower computational complexity. Moreover, the proposed approach has a strong generalization ability and can be implemented in different wireless communication scenarios.

Figures (11)

• Figure 1: The CGM estimation problem is transformed into an I2I inpainting task. We aim to construct the CGM (d) from the wireless environment (a), including buildings, vehicles, etc. The gray dotted arrow represents conventional methods, and the black solid arrow is our method. (b) is the morphological form of the city map. The red pentagram is the location of the BS, these large polygons are buildings, these small yellow rectangles are vehicles, and the yellow shadow is the predicted channel gain at each location.
• Figure 2: The different degrees of spatial discretization corresponding to the geometric location map. (a), (b), and (c) represent the spatial discretization of the target region $A$ into $80 \times 80$, $40 \times 40$, and $10 \times 10$ spatial grids, respectively. Note that ${\Delta _x}$ and ${\Delta _y}$ are taken to be 1, respectively, and the actual geometric location map is gridded into $256 \times 256$ spatial grids in the experimental part of this paper.
• Figure 3: Visual comparison of the gradient descent process for normalized and unnormalized data. The blue and red arrows represent the optimal and actual direction of gradient descent, respectively.
• Figure 4: The LP decomposition obtains the difference between the geometric location map (left) and the CGM (right) for different frequency components. As shown by the MSE, SSIM, PSNR, and the histogram, the differences between the geometric location map and the CGM are more pronounced in the low-frequency components.
• Figure 5: Diagram of the framework of the proposed LPCGMN. When provided with a high-resolution morphological map ${\mathbf{I}_0} \in {\mathbb{R}^{h \times w \times c}}$, we initially decompose it using an $L$-level LP. Green arrow: For the low-frequency component ${\mathbf{I}_L} \in {\mathbb{R}^{{{h \over {{2^L}}}} \times {{w \over {{2^L}}}} \times {{c}}}}$, we reconstruct it into ${{\hat{\mathbf{I}}}_{L}}\in {{\mathbb{R}}^{\frac{h}{{{2}^{L}}}\times \frac{w}{{{2}^{L}}}\times c}}$ through the proposed lightweight network. Red arrow: To adaptively refine the high-frequency component ${\mathbf{r}_{L - 1}} \in {\mathbb{R}^{\frac{h}{{{2^{L - 1}}}} \times \frac{w}{{{2^{L - 1}}}} \times c}}$, we learn an attention map $\tilde{\boldsymbol{\psi}}_{Multihead}^{L-1}\in {{\mathbb{R}}^{\frac{h}{{{2}^{L-1}}}\times \frac{w}{{{2}^{L-1}}}\times 1}}$ based on both high- and low-frequency components. Purple arrows: For the remaining components with higher resolutions, we progressively upsample the acquired attention map and refine it with the proposed LRDC blocks.