Optimization of Collaborative Semantic Communication Network Performance with Channel and Content Preference Feedback

A value-decomposition actor-critic scheme with dynamic neighborhood construction for preference-aware semantic image transmission

Defeng Zhou; Dongyu Wei; Siyao Li; Mingzhe Chen· 2026· DOI 10.48550/arXiv.2607.25011

The core problem

Existing semantic communication frameworks treat and transmit all image regions with equal importance, an assumption that is impractical for real-world applications where different content within an image may be prioritized differently. To address this limitation, the authors propose a novel semantic communication framework that enables a transmitter to use limited channel and content feedback to prioritize the transmission of important image regions. In the proposed framework, a base station (BS) divides each image into sub-images, extracts their semantic information, and transmits them to users according to their preferences. Users reconstruct the image based on the received sub-images and cooperatively decide when to send channel state information (CSI) or content-preference feedback under dynamic channels and limited resources. The central problem is formulated as an optimization that minimizes the semantic-weighted mean square error between the original image and the regenerated image by jointly optimizing sub-channel allocation, users' power allocation, and feedback selection.

Simulation results show that the proposed VDAC-DNC scheme can improve performance by up to 5.04% compared to the standard multi-agent QAC method. Furthermore, it achieves up to 18.55% improvement compared to the proposed method without feedback transmission. These gains demonstrate the effectiveness of combining value decomposition actor-critic learning with dynamic neighborhood construction for preference-aware semantic communication under dynamic channels and limited resources.

Existing semantic communication frameworks treat and transmit all image regions with equal importance, an assumption that is impractical for real-world applications where different content within an image may be prioritized differently. To address this limitation, the authors propose a novel semantic communication framework that enables a transmitter to use limited channel and content feedback to prioritize the transmission of important image regions.

In the proposed framework, a base station (BS) divides each image into sub-images, extracts their semantic information, and transmits them to users according to their preferences. Users reconstruct the image based on the received sub-images and cooperatively decide when to send channel state information (CSI) or content-preference feedback under dynamic channels and limited resources. The central problem is formulated as an optimization that minimizes the semantic-weighted mean square error between the original image and the regenerated image by jointly optimizing sub-channel allocation, users' power allocation, and feedback selection.

The results indicate that limited channel and content feedback can be exploited effectively to prioritize important image regions, yielding substantial performance improvements over both a standard multi-agent QAC baseline and a no-feedback variant. The VDAC-DNC scheme's continuous action approximation reduces output dimensionality, while dynamic neighborhood construction avoids exhaustive search over large discrete action spaces, jointly improving training efficiency. The framework's reliance on cooperative user decisions for feedback selection highlights a practical trade-off between feedback overhead and reconstruction quality. The reported improvements of 5.04% and 18.55% suggest that feedback-aware semantic prioritization is a promising direction for collaborative semantic communication networks, though the evaluation is limited to simulation settings and specific channel dynamics.

CS practitioners and researchers

Opening member content…