ScholarWorks@중앙대학교

초록

Integrated sensing and communication has emerged as a key paradigm in next-generation wireless systems. In orthogonal frequency-division multiplexing (OFDM)-based radar systems with time-division duplexing (TDD), the separation of downlink (DL) and uplink (UL) slots leads to spectral replicas along the Doppler axis, degrading sensing performance. To address this issue, we propose a convolutional long short-term memory autoencoder that predicts the channel state information (CSI) associated with unobserved UL slots between DL transmissions in a TDD frame. By estimating the CSI in unobserved UL slots, the proposed method effectively suppresses spectral replicas in the range-Doppler (RD) map. Simulation results under 3rd Generation Partnership Project New Radio numerologies (μ=1,2,3) demonstrate that the proposed model accurately predicts CSI values across diverse channel conditions. The model achieves a magnitude mean squared error (MSE) of 0.014 and a phase MSE of 0.038, consistently outperforming other methods. In addition, under equal parameter budgets, it achieves superior accuracy–efficiency trade-offs over alternative deep learning models, requiring only 7.2 ms per frame on a modern Internet-of-Things edge device. RD map analysis further confirms replica suppression, improving velocity coverage from 13.05 m/s to 613.5 m/s and signal-to-noise ratio from 45.16 dB to 60.91 dB compared with conventional methods, while maintaining target detection rates above 0.99. These results confirm that accurate CSI reconstruction by the proposed model significantly improves sensing fidelity in TDD-based OFDM radar systems.

키워드