Decentralized Semantic Federated Learning for Real-Time Public Safety Tasks: Challenges, Methods, and Directions

TL;DR

The paper tackles the energy and heterogeneity challenges of real-time public safety tasks by proposing the Decentralized Semantic Federated Learning (DSFL) framework, which combines a hierarchical semantic communication scheme with a hybrid computation structure to enable efficient intra-BS semantic learning and inter-BS semantic exchange. It uses a SwinJSCC-based local semantic model and an $SNR$-aware, top-$k$ compression strategy to achieve energy-efficient transmission while maintaining semantic accuracy across a multi-BS network. A BoWFire-based case study demonstrates DSFL's ability to reduce communication energy and improve fire-detection performance under varying wireless conditions, suggesting robustness in real-time edge scenarios. The authors also outline open directions in robust and multimodal semantic communication and privacy-preserving real-time edge AI, highlighting DSFL’s potential to enhance scalable, safe, and responsive public safety operations.

Abstract

Public safety tasks rely on the collaborative functioning of multiple edge devices (MEDs) and base stations (BSs) in different regions, consuming significant communication energy and computational resources to execute critical operations like fire monitoring and rescue missions. Traditional federated edge computing (EC) methods require frequent central communication, consuming substantial energy and struggling with resource heterogeneity across devices, networks, and data. To this end, this paper introduces a decentralized semantic federated learning (DSFL) framework tailored for large-scale wireless communication systems and heterogeneous MEDs. The framework incorporates a hierarchical semantic communication (SC) scheme to extend EC coverage and reduce communication overhead. Specifically, the lower layer optimizes intra-BS communication through task-specific encoding and selective transmission under constrained networks, while the upper layer ensures robust inter-BS communication via semantic aggregation and distributed consensus across different regions. To further balance communication costs and semantic accuracy, an energy-efficient aggregation scheme is developed for both intra-BS and inter-BS communication. The effectiveness of the DSFL framework is demonstrated through a case study using the BoWFire dataset, showcasing its potential in real-time fire detection scenarios. Finally, we outlines open issues for edge intelligence and SC in public safety tasks.

Figures (6)

• Figure 1: Integration of FL with SC in public safety tasks. The MEDs are used for semantic encoding, while the BS, functioning as a central server, aggregates the semantic data and facilitates data recovery.
• Figure 2: The framework of DSFL. It comprises MEDs and multiple BSs, organized into two layers: the lower layer enables intra-BS SC (BS-to-MED interactions within a coverage area), while the upper layer supports inter-BS SC (BS-to-BS coordination for global semantic fusion).
• Figure 3: Local SC model in DSFL framework.
• Figure 4: Energy-efficient compression schemes include intra-BS and inter-BS communication compression.
• Figure 5: Performance comparison of the DSFL framework for transmitting fire images under varying SNR conditions. The left panel illustrates reconstructed images after transmission through channels with SNRs of 1 dB and 13 dB, while the right panel presents the corresponding quantitative metrics (MS-SSIM and PSNR) to evaluate transmission quality.