Robust Streaming Tensor Train Completion for Dynamic Space-based Spectrum Situation Map Construction

Leveraging the broad coverage and collaborative sensing capabilities of low Earth orbit satellite constellations, space-based spectrum monitoring enables the construction of spectrum situation map (SSM) for dynamic interference tracking and precise spectrum management. However, the limitations of predetermined satellite orbits and long satellite-to-ground transmission result in spectrum measurements highly incomplete and contaminated with unexpected anomalies. To better explore spatio-temporal-frequency correlations, tensor train (TT) decomposition is exploited in this paper to capture subspace information within high-order correlations for accurate SSM construction. As space-based spectrum measurements are incrementally collected with a large scale and high rate, conventional batch-based optimization limits real-time analysis and time variation tracking in dynamic changing spectrum environments. To address this issue, we propose a robust streaming tensor train completion framework that decomposes high-order incomplete streaming spectrum into the TT format and performs online tensor factor tracking and outlier removal to construct the SSM over time. Moreover, by exploiting the latent temporal structure extracted through TT format, the proposed scheme can further forecast future SSM evolution. Numerical analyses demonstrate that the proposed algorithm achieves better accuracy and scalability than state-of-the-art schemes in dealing with dynamic space-based SSM construction and prediction.