Regularized Diffusion-based Contract Model for Covert Semantic Entropy Control in LAENets

TL;DR

This work proposes an incentive-aware semantic entropy control framework for covert communications in LAENets, which regulates semantic uncertainty at the receiver by adjusting the semantic abstraction level at the UAV side, thereby enabling reliable task information delivery under extreme covert constraints.

Abstract

Low-Altitude Economy Networks (LAENets) have emerged as a critical communication paradigm for operation-critical and regulation-aware applications, where Unmanned Aerial Vehicles (UAVs) transmit task-related information under stringent low-probability-of-detection constraints. These constraints severely limit the available transmission power and bandwidth, rendering conventional bit-level communication inefficient when task performance depends on high-level semantic understanding rather than raw data fidelity. Fortunately, Semantic Communication (SemCom) can be a promising solution by prioritizing task-relevant information over bit-level accuracy. However, different levels of semantic abstraction inherently introduce different degrees of information loss and redundancy, which may either compromise task reliability or incur excessive transmission overhead if not properly controlled. To this end, we propose an incentive-aware semantic entropy control framework for covert communications in LAENets. Specifically, we regulate semantic uncertainty at the receiver by adjusting the semantic abstraction level at the UAV side, thereby enabling reliable task information delivery under extreme covert constraints. Since the Base Station (BS) cannot directly observe the semantic processing capabilities and abstraction-dependent transmission costs of UAVs, information asymmetry naturally arises in SemCom service provision. Accordingly, we propose a contract theoretic model, where we adopt Prospect Theory (PT) to capture the subjective utility of the BS toward personalized semantic services. Furthermore, we design a Regularized Diffusion-based Soft Actor-Critic (RDSAC) algorithm for optimal contract design under PT. This algorithm enhances contract design by introducing diffusion entropy regularization together with action entropy regularization.

Figures (8)

• Figure 1: Incentive-aware semantic entropy control framework in LAENets. The BS generates regularized diffusion-based contracts according to subjective utility preferences and covert semantic communication requirements. UAVs select contracts and adjust their transmission strategies accordingly. By adopting hierarchical semantic abstraction levels, UAVs exhibit heterogeneous subjective utilities.
• Figure 2: The architecture of the proposed RDSAC algorithm for optimal contract design under PT, where action entropy regularization and diffusion entropy regularization are incorporated into the diffusion-based policy to improve policy expressiveness and training stability.
• Figure 3: Evolution of CLIP feature distributions for hierarchical semantic representations under varying SNR levels.
• Figure 4: Average reward comparison of our scheme with other schemes under PT, where reference point $U_\mathrm{ref}=160$.
• Figure 5: Difference in average rewards between the complete-information contract and the asymmetric-information contract during training under PT, where the reference point $U_\mathrm{ref}=160$.

Theorems & Definitions (5)

• Definition 1
• Definition 2
• Lemma 1
• proof
• Remark 1