Deep Learning Aided Blind Synchronization Word Estimation

doi:10.1109/ACCESS.2021.3058351

Deep Learning Aided Blind Synchronization Word Estimation

2021-01-01

4

OA

AI

DOI：10.1109/ACCESS.2021.3058351

Original

Original request for help

Save

Abstract

En 中文

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

Keywords:

Synchronization

Estimation

Receivers

Signal to noise ratio

Standards

Recurrent neural networks

Logic gates

Blind frame synchronization

deep learning

frame recognition

non-cooperative communication

synchronization word

Deep Learning Aided Blind Synchronization Word Estimation

Abstract

Journal

Organization