Channel Gain Map Construction based on Subregional Learning and Prediction

TL;DR

The paper addresses CGM construction under limited measurements in complex propagation environments by introducing a subregional learning framework. It partitions the area into $K$ subregions via data-driven K-means clustering on augmented sample features, and trains a dedicated MCNN-1D per subregion to predict channel gains, enabling an ensemble that approximates a continuous CGM $oldsymbol{ m ilde S}$. To boost accuracy, it introduces uneven subregion sampling and a training-data-reuse strategy that leverages boundary information, with RMSE-based selection of the optimal number of subregions. Simulation on a ray-traced 470×630 m$^2$ area demonstrates that the proposed method outperforms EM-based partitioning and approaches the accuracy of ray-tracing, while offering lower computational complexity. The approach provides a scalable, environment-aware CGM construction framework suitable for dynamic 6G deployments, with potential extensions via transfer learning for adaptation to changing environments.

Abstract

The construction of channel gain map (CGM) is essential for realizing environment-aware wireless communications expected in 6G, for which a fundamental problem is how to predict the channel gains at unknown locations effectively by a finite number of measurements. As using a single prediction model is not effective in complex propagation environments, we propose a subregional learning-based CGM construction scheme, with which the entire map is divided into subregions via data-driven clustering, then individual models are constructed and trained for every subregion. In this way, specific propagation feature in each subregion can be better extracted with finite training data. Moreover, we propose to further improve prediction accuracy by uneven subregion sampling, as well as training data reuse around the subregion boundaries. Simulation results validate the effectiveness of the proposed scheme in CGM construction.

Figures (4)

• Figure 1: The overall architecture of the subregional learning-based CGM construction.
• Figure 2: Accuracy of the proposed optimization strategies.
• Figure 3: CGM construction with different methods.
• Figure 4: Predicted CGM constructed by the proposed scheme, with $M_{\mathrm{scgm}}=3200$, $N=800$ and $K^{*}=7$.