Physics-Enabled Frequency-Domain Non-Stationary Channel Estimation Based on Multimodal Attention Mechanism

Abstract

In large-scale multiple-input multiple-output (U-MIMO) orthogonal frequency division multiplexing (OFDM) systems, the frequency domain channel exhibits non-stationarity, exacerbated by limited pilot positions which complicate channel estimation. This letter extends single-band channel estimation to multi-band scenarios, focusing on frequency domain non-stationary sub-6 GHz - millimeter-wave(mmWave) systems. A novel framework leveraging a multi-modal attention mechanism is proposed: it employs a multi-modal input network to extract sub-6 GHz channel details and fine-tune with limited mmWave data. An adaptive attention mechanism dynamically adjusts information weights, facilitating effective feature fusion for mmWave channel estimation. The algorithm’s reliability and effectiveness in non-stationary channel estimation are validated through theoretical analysis and numerical simulations in adaptive frequency domains.