Title: Distributed Task-Driven Compression for Locating and Identifying Multi-Emitter Sources Amid Spectral Interference

Abstract: Achieving comprehensive radio frequency spectrum awareness necessitates the capability to detect, pinpoint, and characterize signal sources within crowded and contested wireless landscapes. This study introduces a distributed compression framework designed for task-oriented joint localization and characterization of multiple emitters, leveraging spatially dispersed receiver networks. In this architecture, each receiver captures a brief interval of complex IQ samples, transforms these observations into a time-frequency representation, and compresses them into a concise latent vector. A central fusion decoder then aggregates these latent vectors to reconstruct an unordered set of active emitters, estimating key parameters such as spatial locations, center-frequency deviations, bandwidth occupancy, and waveform classifications. To address the challenge of arbitrary emitter ordering, the model employs a permutation-invariant training objective.

Synthetic experiments involving multi-emitter scenarios with significant spectral overlap reveal that highly compact receiver-side representations are sufficient for maintaining critical information regarding emitter counts and waveform family identification. Conversely, precise localization and the regression of spectral parameters demand higher-dimensional latent spaces. Our findings indicate that expanding the receiver latent dimension from $d_{\mathrm{rx}}=1$ to $d_{\mathrm{rx}}=16$ yields the most substantial performance gains, whereas further increases to $d_{\mathrm{rx}}=64$ result in diminishing returns. These outcomes highlight the efficacy of learned, task-oriented compression in enabling communication-efficient distributed spectrum awareness.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC