A Hybrid Attention and SE-Net Architecture for CFO Estimation in MIMO-OFDM Ambient Backscatter Communication Systems

Abstract

Ambient backscatter communication (ABC) is a promising technology for future wireless systems due to its ultra-low-power operation and simplicity. To further enhance its performance, multiple input multiple output-orthogonal frequency division multiplexing (MIMO-OFDM) has been adopted in this work. However, carrier frequency offset (CFO) can disrupt the orthogonality of the MIMO-OFDM signal, which significantly degrades the performance of the reader’s receiver. To address this issue, we consider an asynchronous ABC system and propose a deep learning (DL) algorithm to estimate and compensate the CFO at the reader, ensuring reliable communication in MIMO-OFDM-based ABC systems. The proposed architecture integrates a hybrid attention mechanism, which captures inter-dependencies between the two types of extracted features, and a squeeze-and-excitation network that adaptively refines channel-wise feature responses. The proposed architecture can also work for conventional OFDM and MIMO-OFDM systems. Furthermore, to validate the model’s robustness and generalization capability, additional test datasets are generated for various modulation schemes and different 3GPP TR38.901 tapped delay line channel models. The simulation results indicate its outperformance over existing approaches and has lower computational complexity than the existing methods.