A Learning Approach to Design Binary Sequences with Good Correlation Properties

Employing binary sequences with good autocorrelation properties leads to performance improvement for active sensing systems. Designing such sequences has been studied widely in the literature from optimization point of view. In this paper we consider the problem of designing binary sequences with good aperiodic/periodic auto-correlation functions in terms of Integrated Sidelobe Level (ISL) using machine learning method. Specifically, we propose a novel neural network structure which can learn an algorithm for designing binary sequences with small ISL. We also extend the resulting network, which we refer to as BiSCorN, by designing Low-Correlation Zone (LCZ) binary sequences. Numerical experiments show that our proposed method outperforms state-of-the-art algorithms in terms of ISL, and interestingly Peak Sidelobe Level (PSL).