Deep Learning Aided Blind Synchronization Word Estimation

In this paper, we address a blind frame synchronization problem where the receiver acts as an eavesdropper in wiretap channel. A challenging condition is considered, has completely no prior information except that unknown word (SW) repeated nonperiodic fashion. Although autocorrelation method was proposed for fixed length scenario, scheme not applicable to variable-length scenario. To solve problem, propose deep learning-aided SW estimation recurrent neural networks (RNNs) are used symbol predictor predicts from observation of preceding symbols. The prediction confidence RNN-based localization symbols received signal. Two RNNs fed with signal forward and backward accurate localization. It been verified by simulation schemes estimate well when amount sufficiently large. straightforward get correctly estimated SW.