Adoption of hybrid time series neural network in the underwater acoustic signal modulation identification

Abstract

The deep learning methods powerfully enhance the identification performance by retrieving the deep data features in many fields, which can be used in automatic modulation classification (AMC) work for the better results in the acoustic underwater communication. A novel hybrid time series network structure is scheduled for AMC in this paper. It can accommodate the variable-length signal datas to match the fixed-length input request in the common neural network, and there is the ability to suitably deal with the zero data in the signal sequence to alleviate the effect losses. The proposed network has the mix of two time series network styles to enrich the extracted signal modulation classification features, and dramatically improves the recognition capability and owns the low computation complexity. In the meanwhile, the internal network structure is optimized by the well-designed cascading order, which acquires more hidden signal data representations to increase the accuracy. The simulation experiments shows that the proposed network is more effective and robust than the conventional deep learning methods to identify ten modulation modes in the serious interference communication environment.