Multimodal Deep Learning Model for Specific Emitter Identification

Abstract

Specific Emitter Identification (SEI) is an approach to distinguish similar emitters, especially those of the same type and transmitting parameters, based on revealing the underlying physical association between the emitter hardware and the received signals of it. The process of conducting SEI consists of feature extraction and classifier designing, which directly determines the performance of SEI. Traditional SEI is realized by expert knowledge and machine learning techniques. Recently, general and self-adapted feature extraction methods and deep learning techniques have been widely studied to improve the SEI performance under certain scenarios. To provide a universal SEI method, the authors provide a novel multimodal deep learning model based on various signal features and deep network branches. Received signal features are extracted with multiple techniques and then are treated as independent inputs to corresponded deep model branches. The outputs of each model branch are further merged in fully-connected layer in the multimodal network. The proposed the final method is tested on a signal source dataset; the experimental results indicated that this novel scheme outperforms the typical single modal deep learning-based SEI methods.