Deep Learning-Aided Frequency Offset Estimation Method for 5G System based on Synchronization Signal

We offer an effective frequency offset estimation approach for orthogonal frequency division multiplexing (OFDM) systems in 5G, supported by a deep neural network. The deep fusion architecture was a conventional, fully connected multilayer network divided into two stages: offline training and online inference. Initially, a time-domain cross-correlation frequency offset estimation approach based on PSS and SSS was presented, and the training sample set was created under the initial frequency offset predicted by this method. The deep fusion network is offline trained using the training sample pack to get the ideal network weights. The online inference is employed to execute calculations on the newly selected initial frequency offset group using the trained fusion network to determine the final frequency offset. The simulation results demonstrate that the performance of the suggested approach is significantly better than that of the current methods and that improved performance can be obtained at a relatively low additional cost. The deep learning-assisted frequency offset estimation algorithm also increases the frequency offset estimation range from ±1.752 kHz to ±7 kHz, offering excellent accuracy and stability in frequency offset estimation, particularly concerning having a more vital processing capability of border frequency offset.