Phase-Locked LLM Inference and Swarm Coherence for Autonomous UAVs under Electronic Warfare via Rössler Dynamics and Dynamic F-LoRA

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Autonomous UAV operations under Electronic Warfare (EW) demand continuous on-boardintelligence without reliance on cloud connectivity. We introduce Phase-Locked DynamicSampling (PLDS) and Dynamic F-LoRA to maintain LLM stability and execution efficiencyon resource-constrained embedded systems. PLDS couples hidden state trajectories with aRössler dynamic manifold, modulating decoding temperature according to phase divergenceand the maximal Lyapunov exponent (λmax). To reduce VRAM and bandwidth overhead,Dynamic F-LoRA parameterizes adapter weights in the 2D-DCT frequency domain usingGumbel-Sigmoid STE gating, achieving a parameter sparsity of 78.4%±1.2%. On an NvidiaJetson Orin Nano, the hybrid architecture reduces VRAM footprint to 5.7 ± 0.1 GB andinference latency from 185 ms to 42 ± 1.8 ms per token (∼ 24 tokens/s). Under simulatedEW jamming, PLDS lowers hallucination rates to 2.1%±0.3% on tactical navigation benchmarks,while Kuramoto phase synchronization detects anomalous nodes within 115±8.2 ms,sustaining a 96.4% ± 1.2% mission success rate across 5–10 node swarms.