CKMImageNet: A Comprehensive Dataset to Enable Channel Knowledge Map Construction via Computer Vision

Di Wu, Zijian Wu, Yuelong Qiu, Shen Fu, Yong Zeng

Abstract

Environment-aware communication and sensing is one of the promising paradigm shifts towards 6G, which fully leverages prior information of the local wireless environment to optimize network performance. One of the key enablers for environment-aware communication and sensing is channel knowledge map (CKM), which provides location-specific channel knowledge that is crucial for channel state information (CSI) acquisition. To support the efficient construction of CKM, large-scale location-specific channel data is essential. However, most existing channel datasets do not have the location information nor visual representations of channel data, making them inadequate for exploring the intrinsic relationship between the channel knowledge and the local environment, nor for applying advanced artificial intelligence (AI) algorithms such as computer vision (CV) for CKM construction. To address such issues, in this paper, a large-scale dataset named CKMImageNet is established, which can provide both location-tagged numerical channel data and visual images, providing a holistic view of the channel and environment.

Introduction: The Vision Problem in Wireless

As we transition toward 6G, wireless networks face a daunting challenge: the massive overhead required for Channel State Information (CSI) acquisition in complex environments like mmWave and XL-MIMO. A promising solution is environment-aware communication, which uses a Channel Knowledge Map (CKM)—a location-specific database of channel characteristics—to predict signal behavior and reduce real-time training overhead.

However, existing wireless datasets often lack precise location data or visual representations of the environment. This makes it nearly impossible to use Computer Vision (CV) algorithms to explore the intrinsic relationship between the physical world (buildings, trees, streets) and signal propagation. Without this link, we cannot easily build "channel twins" or high-resolution maps that "see" how a signal will behave before a device even connects.

The CKMImageNet Framework

The authors developed CKMImageNet using a high-fidelity pipeline centered on commercial ray-tracing software (Wireless InSite). The methodology follows a hierarchical structure that organizes data by scenario, location, and base station, bridging the gap between raw electromagnetic simulations and visual data representations.

Hierarchical structure and data components of CKMImageNet

The dataset generation follows a structured pipeline:

  1. Scenario Construction: GIS data is used to digitize real-world urban (e.g., Beijing, Boston), rural, and indoor environments.
  2. Ray Tracing Simulation: Electromagnetic wave propagation is modeled at 28 GHz, accounting for reflections and diffractions.
  3. Visual CKM Generation: Raw numerical data is converted into "Visual CKMs"—64x64 pixel heatmaps representing parameters like signal strength and path loss.
  4. Environment Integration: Channel heatmaps are paired with physical environment maps, allowing AI models to learn the direct correlation between physical obstacles and channel quality.

Dataset Characteristics and Comparison

CKMImageNet is a multi-scenario dataset that significantly expands upon existing benchmarks, offering over 25 million channel data entries and 40,000 visual channel images across 10 different environment types. In the Beijing urban scenario alone, the authors generated 1.68 million data entries from 42 Base Stations and nearly 500,000 UE locations.

Comparison of CKMImageNet with existing wireless datasets

As shown in the comparison above, CKMImageNet is uniquely positioned as the only dataset to simultaneously provide location information, detailed channel parameters, visual CKMs, and integrated environment maps. This multi-modal nature is what enables the application of CV algorithms to wireless problems.

Key Insights: Wireless as a Vision Task

By representing channel data as heatmaps, CKMImageNet enables the use of mature CV architectures (like CNNs or Transformers) for wireless tasks. This treats signal prediction as an image-to-image translation or regression problem. The dataset's "secret sauce" is the explicit link between User Equipment (UE) locations and channel parameters, allowing for sophisticated spatial interpolation—predicting the channel at one point based on data from another and the visual layout of the intervening space.

Conclusion and Future Work

CKMImageNet provides a holistic view of the channel and environment, supporting the verification of various communication and sensing algorithms. While currently based on high-fidelity ray-tracing simulations, future work will focus on expanding the dataset to more diverse frequencies and incorporating real-world measurement data to further refine the models. This dataset marks a significant step toward realizing environment-aware 6G networks through the fusion of wireless communications and computer vision.

Summary updated

Abstract

Environment-aware communication and sensing is one of the promising paradigm shifts towards 6G, which fully leverages prior information of the local wireless environment to optimize network performance. One of the key enablers for environment-aware communication and sensing is channel knowledge map (CKM), which provides location-specific channel knowledge that is crucial for channel state information (CSI) acquisition. To support the efficient construction of CKM, large-scale location-specific channel data is essential. However, most existing channel datasets do not have the location information nor visual representations of channel data, making them inadequate for exploring the intrinsic relationship between the channel knowledge and the local environment, nor for applying advanced artificial intelligence (AI) algorithms such as computer vision (CV) for CKM construction. To address such issues, in this paper, a large-scale dataset named CKMImageNet is established, which can provide both location-tagged numerical channel data and visual images, providing a holistic view of the channel and environment. Built using commercial ray tracing software, CKMImageNet captures electromagnetic wave propagation in different scenarios, revealing the relationships between location, environment and channel knowledge. By integrating detailed channel data and the corresponding image, CKMImageNet not only supports the verification of various communication and sensing algorithms, but also enables CKM construction with CV algorithms.
BibTeX
@misc{wu2024ckmimagenet,
  author = {Di Wu and Zijian Wu and Yuelong Qiu and Shen Fu and Yong Zeng},
  title = {{CKMImageNet: A Comprehensive Dataset to Enable Channel Knowledge Map Construction via Computer Vision}},
  year = {2024},
  eprint = {2410.07219},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2410.07219},
}
Figure