Paper Submitted to IEEE Wireless Communications Letters

We submitted a new paper to IEEE Wireless Communications Letters:

Geometry-Aware Regularization for Deep-Unfolded MIMO-OTFS Detection

Nikolaos Ntavanelos and Evangelos Vlachos

MIMO-OTFS systems suffer from non-uniform scaling across symbol components due to spatial and fractional Doppler–Delay coupling. This work introduces a geometry-aware coordinate-wise regularizer within a deep-unfolded Conjugate Gradient detector to exploit this structure.

Key contributions:

• Gram matrix analysis: We prove that all diagonal variation in the MIMO-OTFS Gram matrix arises exclusively from inter-path coupling, analytically characterizing its non-uniform structure.
• Deep-unfolded CG detector: A geometry-aware regularizer is embedded directly into the CG linear system, with per-component weights learned in a data-driven manner.
• Robust performance: The proposed detector outperforms low-complexity baselines and approaches heavier neural architectures at medium-to-high SNR under both low- and high-mobility scenarios.