Signal Synchronization for 5G NR Under Large CFOs Based on Convolutional Neural Network Combined with Long Short-Term Memory

Signal synchronization is one of the core aspects of communication, ensuring that the receiver accurately decodes the signals transmitted by the sender. However, in the diverse application scenarios and broad spectrum range of 5G new radio (NR), the performance of traditional estimation algorithms often deteriorates as frequency offset increases and noise interference intensifies. This work focuses on the estimation of time offset, cell sector identifier (ID), and frequency offset in 5G mobile communication systems. We leverage the advanced learning capabilities and adaptability of a convolutional neural network (CNN) to optimize the estimation process. Additionally, we incorporate a long short-term memory (LSTM) network to capture the dynamic variations in time-varying channels. The results demonstrate that the proposed neural network exhibits significant advantages in estimation performance.