A Hybrid Frequency Offset Estimation Combining Data-Driven Method and Model-Driven Method for 6G OFDMA Systems

The 6G radio access networks may include an OFDMA based physical layer. The OFDMA system is vulnerable to frequency offsets due to its own physical characteristics. They interfere the synchronization of OFDM packets as well as cause the phase noises and I/Q imbalance. Thus, it is essential to estimate frequency offset accurately in the 6G OFDMA system. The conventional carrier frequency offset (CFO) estimation is based on a maximum likelihood estimation (MLE). The MLE is used to estimate the parameters of a statistical model based on an observed dataset. As the name said, it finds the parameter values maximizing the likelihood function while observing the probability of the dataset under the statistical model. In the OFDMA system, The MLE uses a repetitive preamble and perform correlation with two received preamble symbols. However, the weakness of MLE is based on one strong assumption: datasets must be independently and identically distributed. In real world, the wireless channel can be correlated. If the assumption is not satisfied, the MLE is not consistent. On the other hands, the recurrent neural network (RNN) uses temporal correlations between the historical data and the current data. It is well matched with the CFO estimation when the CFO values of sequential inputs affects to the current CFO estimation. In addition, it doesn’t rely on the preamble signals. Thus, they two will make a good combination. In this paper, we develop a combined method using a data driven approach and a model driven approach to estimate the CFO and attempt to have the optimal values of the CFO estimation with a wider SNR range.