Abstract
Integrated sensing and communication (ISAC) in the terahertz (THz) band enables obstacle detection, which in turn facilitates efficient beam management to mitigate THz signal blockage. Simultaneously, a THz radio map, which captures signal propagation characteristics through the distribution of received signal strength (RSS), is well-suited for sensing, as it inherently contains obstacle-related information and reflects the unique properties of the THz channel. This means that communication-assisted sensing in ISAC can be effectively achieved using a THz radio map. However, constructing a radio map presents significant challenges due to the sparse deployment of THz sensors and their limited ability to accurately measure the RSS distribution, which directly affects obstacle sensing. In this paper, we formulate an integrated problem for the first time, leveraging the mutual enhancement between sensed obstacles and the constructed THz radio maps. To address this challenge while improving generalization, we propose an integration framework based on a conditional generative adversarial network (CGAN), which uncovers the manifold structure of THz radio maps embedded with obstacle information. Furthermore, recognizing the shared environmental semantics across THz radio maps from different beam directions, we introduce a novel voting-based sensing scheme, where obstacles are detected by aggregating votes from THz radio maps generated by the CGAN. Simulation results demonstrate that the proposed framework outperforms non-integrated baselines in both radio map construction and obstacle sensing, achieving up to 44.3% and 90.6% reductions in mean squared error (MSE), respectively, in a real-world scenario. These results validate the effectiveness of the proposed voting-based scheme.
1. Introduction
As the telecommunications industry pivots toward 6G, the Terahertz (THz) band has emerged as a critical frontier for achieving ultra-high-speed data transmission. However, the high-frequency nature of THz signals introduces a significant vulnerability: extreme sensitivity to blockages. Unlike lower frequency bands, THz waves exhibit minimal diffraction; an obstacle in the line-of-sight (LoS) path often results in a complete link failure.
To maintain reliable connectivity, researchers utilize Radio Maps—spatial grids representing Received Signal Strength (RSS)—to predict coverage and proactively detect obstacles. Traditional approaches typically treat radio map construction and obstacle sensing as decoupled, sequential tasks. This paper proposes an integrated generative framework that treats these tasks as mutually reinforcing, significantly improving the system's ability to generalize to unseen environments.
2. The Integrated Generative Framework
The core of the proposed solution is a Conditional Generative Adversarial Network (CGAN) designed to exploit the physical manifold of THz propagation.
2.1 CGAN Architecture and U-Net Generator
The framework employs a U-Net architecture as its generator. U-Net is particularly effective here because its encoder-decoder structure with skip connections allows the model to capture both local signal variations and global spatial context. The generator takes sparse, noisy RSS measurements from a limited number of sensors and "inpaints" the high-resolution radio map.
2.2 Voting-Based Sensing Scheme
A key innovation is the multi-beam voting scheme. Since THz base stations often utilize directional beamforming, the system generates multiple radio maps corresponding to different beam directions.
- Hard Voting: During inference, a grid cell is classified as an obstacle only if there is a consensus across all generated beam maps. This reduces false positives caused by noise or localized shadowing.
- Soft Voting: During training, a differentiable soft voting mechanism is used, allowing the sensing error to be backpropagated through the network to refine the generator's understanding of obstacle geometry.
3. Experimental Evaluation
The framework was evaluated using ray-tracing simulations at 300 GHz across three distinct scenarios, ranging from random geometric obstacles to high-fidelity urban models.
The ultimate test of generalization was conducted on a real-world scenario based on Oklahoma City. Despite being trained on synthetic quadrilateral obstacles, the model demonstrated remarkable performance on the complex geometry of a real city.
3.1 Key Results
The integrated approach yielded substantial improvements over state-of-the-art baselines such as RadioUNet and DCRGAN:
- Map Construction Accuracy: The model achieved a 44.3% reduction in Mean Squared Error (MSE) in the Oklahoma City scenario.
- Obstacle Sensing Precision: Sensing error was reduced by 90.6%, demonstrating that the "mutual enhancement" between the map and the sensed obstacles forces the model to adhere to the underlying physics of signal propagation.
As shown in Figure 1, the proposed framework (b) recovers the intricate "shadowing" effects of obstacles much more clearly than baseline methods (a), which often produce blurred or physically inconsistent results in the presence of sparse sensor data.
4. Conclusion and Future Directions
This research validates that integrating radio map construction with obstacle sensing within a generative framework provides a robust path toward 6G ISAC. By leveraging the shared semantics of multiple beam directions through a voting scheme, the system achieves unprecedented accuracy and generalization. Future work will focus on extending this framework to 3D urban environments with varying building heights and addressing the challenges of dynamic, time-varying obstacles.
Summary updated