Deep Learning-Based Joint Channel Estimation and Frequency Offset Compensation under Large Frequency Offset

• 1、Key Laboratory of Universal Wireless Communications, Ministry of Education, Beijing University of Posts and Telecommunications

Abstract：To address the frequency offset sensitivity and limited pilot overhead in 5G n79 IIoT systems, this paper proposes a deep learning-based joint scheme for frequency offset and blind channel estimation. Focusing on the large frequency offsets from 4000 to 8000Hz caused by low-cost crystal oscillators (approximately 10ppm), a cascaded residual CNN architecture is designed to decouple deterministic phase rotation from random multipath fading by performing compensation before estimation. Simulation results in TDL-A industrial channels demonstrate that the proposed scheme achieves high robustness against offsets exceeding 10ppm without pilots. Compared to the conventional Least Squares (LS) method, this approach significantly enhances BLER performance and spectral efficiency, providing a robust physical-layer solution for low-cost, high-reliability industrial terminals.

Keywords： wireless communication systems blind channel estimation deep learning OFDM IIoT frequency offset compensation