Abstract: UAV swarms face the dual constraints of communication limitations and imperfect information, making efficient collaborative countermeasures a critical challenge to be addressed. To tackle the collaborative decision-making problem in electromagnetic countermeasures for heterogeneous UAV swarms under star topology constraints and partial observation conditions, this paper proposes an intelligent decision-making method based on spatio-temporal spectrum awareness to enhance the collaborative effectiveness of heterogeneous swarms in electromagnetic confrontation. First, a Dec-POMDP model is constructed that integrates role differentiation of heterogeneous nodes, a closed-loop "communication-reconnaissance-jamming" interaction mechanism, and intelligent adversarial gaming, where multi-agent outputs contain continuous decision vectors across communication, motion, reconnaissance, and other dimensions. Subsequently, a Spatio-Temporal Spectrum Awareness Multi-Agent Proximal Policy Optimization (STSA-MAPPO) algorithm is proposed, which decouples frequency-domain features and spatial topological relationships from high-dimensional observations through parallel spectrum awareness and spatial coordination modules, aggregates friend-enemy interaction information in stages via serial cross-attention mechanisms, and incorporates GRU memory units to model temporal dependencies for inferring global situations and predicting adversarial strategies. Simulation results demonstrate that the proposed algorithm significantly outperforms baseline algorithms in key metrics including electromagnetic advantage degree and spectrum conflict rate.

Key words: heterogeneous UAV swarm, electronic countermeasures, multi-agent reinforcement learning, star topology, spatio-temporal spectrum awareness