Distributed Semantic Sensing via Axial-Attention Encoding for the 6G Edge–Cloud Continuum

Abstract

Realizing the IoT-edge-cloud continuum in 6G requires distributed intelligence services that dynamically orchestrate communication, computation, and inference across heterogeneous resource tiers, yet existing approaches treat sensing, transmission, and fusion as isolated design problems. We present a distributed semantic sensing service that instantiates the continuum paradigm for low-altitude UAV surveillance, jointly designing edge-side encoding, bandwidth-constrained semantic transmission, and cloud-assisted cooperative inference as a unified, resource-aware pipeline. Unlike conventional approaches that separately optimize feature extraction, communication efficiency, and multi-node fusion, the proposed framework derives all three components from a unified design principle that couples radar measurement structure, information bottleneck theory, and distributed network geometry. The framework jointly addresses the computational, bandwidth, and reliability constraints of this continuum service through three co-designed components. At the edge, a radar-structure-aware axial-attention encoder exploits the approximate axis-wise separability of the radar point spread function to extract compact 3D features while preserving micro-Doppler signatures essential for UAV discrimination. Across the backhaul, an information-bottleneck-motivated vector quantization scheme discards background information that does not contribute to the downstream sensing task, achieving a $16 \times$ link-rate reduction over scalar quantization baselines. At the fusion center, a geometry-aware decoder converts packet erasures into structural uncertainty via pure position tokens, enabling graceful degradation under unreliable links. Evaluated on the SYN-ISAC-UAV benchmark, the proposed framework retains 99.1% of the unquantized fusion benchmark at only 0.2 Mbps per node, demonstrating that task-oriented semantic networking can realize high-precision distributed sensing as a scalable and resource-efficient continuum service for future 6G systems.