DA-ISCC: Dynamic Adaptation for Integrated Sensing, Communication, and Computation in SAGINs

Abstract

Integrated sensing, communication, and computation (ISCC) systems in space-air-ground integrated networks (SAGINs) offer a promising foundation for next-generation intelligent infrastructure. However, their highly dynamic links, heterogeneous computing resources, and stringent real-time requirements pose critical challenges, including task offloading instability, satellite handover disruptions, and cross-tier coordination inefficiency. To address these issues, we propose DA-ISCC, a dynamic adaptive architecture supporting heterogeneous node collaboration and ISCC-aware task offloading optimization. We design a multi-tier network architecture combining low earth orbit satellites, unmanned aerial vehicles, and terrestrial edge nodes, and formulate a unified model for heterogeneous tasks. We develop two optimization methods for binary and partial task offloading, respectively. A dynamic adaptive predictive alternating direction method of multipliers (DAP-ADMM) enhances robustness and convergence efficiency in binary task offloading through dynamic step-size adjustment, long short-term memory-based resource prediction, and asynchronous communication mechanism. For partial task offloading, a method based on weighted minimum mean square error (WMMSE) handles non-convex objectives via auxiliary variables and block coordinate updates, improving computational efficiency. Extensive experiments spanning small to large-scale scenarios demonstrate that DAP-ADMM achieves convergence within 0.05 residual under satellite handover and UAV mobility, outperforming greedy, DQN-based and other baselines while improving convergence speed by over 60% compared to standard ADMM with 1–2 orders of magnitude reduction in residuals. The WMMSE-based method maintains a stable computation rate exceeding 480 Mbit/s within 100 iterations.