Feature Analysis and Service Classification for Encrypted Sparklink Basic Traffic in Emergency Communications

Abstract

Next-generation emergency short-range wireless communication systems are expected to provide adaptive and service-aware connectivity under severely constrained radio and computing resources. Encrypted service traffic classification addresses this challenge by identifying heterogeneous service demands without accessing packet payloads, thereby supporting dynamic resource allocation. However, existing classification methods based on flow statistical features generally depend on high-dimensional and redundant feature representations, which introduce substantial computational overhead and limit their deployment on resource-constrained emergency communication terminals. To overcome this limitation, this paper shows that only three interpretable link-layer features are sufficient to achieve near-full-feature classification performance for encrypted SparkLink Basic (SLB) traffic. Finally, the proposed method is evaluated on an experimentally collected SLB testbed dataset covering four service categories and multiple communication distances. With three top-ranked core features, the proposed framework achieves an overall accuracy of 98.48%, with only a 0.12-percentage-point accuracy difference compared to the full feature set obtained with the full-dimensional feature set, indicating that a compact feature subset can retain near-baseline classification performance while maintaining high classification accuracy.