Abstract:

To address challenges such as the difficulty in correcting time-varying amplitude-phase errors caused by polarization multi-channel reception， the difficulty in obtaining the number of radiation sources in non-cooperative monitoring scenarios， and the poor performance of polarization feature extraction and clustering under low signal-to-noise ratio conditions， an efficient blind clustering algorithm for polarization features of communication radiation sources is proposed. A single-channel time-division dual-polarization reception system is designed to mitigate the adverse impact of amplitude-phase errors across multiple reception channels on polarization feature extraction， thereby enhancing the estimation quality of individual communication radiation source polarization features. Leveraging the adaptive mechanism of the Affinity Propagation （AP） algorithm， the number of radiation sources is accurately estimated using minimal data. Building upon this foundation， a deep neural network capable of rapidly extracting latent correlation features between polarization amplitude and phase is designed， incorporating deep embedded learning principles. The Kullback-Leibler （KL） divergence between the initial distribution of polarization features and the target distribution is calculated， enabling adaptive clustering of individual communication radiation sources. Experiment with simulation data and anechoic chamber measurements demonstrates that the proposed algorithm achieves communication radiation source clustering under conditions of unknown source count. Both clustering efficiency and average accuracy surpass existing algorithms， fully validating the novel algorithm’s effectiveness.

Key words: communication radiation sources, individual identification, polarization features, deep embedding, adaptive clustering