Method for functional state recognition of multifunction radars based on recurrent neural networks

Abstract

Radar signal recognition plays a vital role in electronic warfare. For the multifunction radars (MFRs) with complex dynamical modes, the signal recognition needs to identify not only the emitter but also its current functional state. Existing research on MFR recognition mainly focuses on hierarchical modelling approaches. Inspired by recent progress of deep neural networks, the authors propose to further develop radar signal modelling with recurrent neural networks. Here, the authors propose a more efficient method for functional state recognition of MFRs based on a gated recurrent unit (GRU). The method makes full use of the ability of GRUs to automatically learn the characteristics of radar signal sequences, and the nonlinear modelling of GRUs can deal with the corrupted data well. Simulation results on the Mercury radar show that the proposed method achieves a great recognition accuracy with little prior information, and the designed GRU model outperforms the predictive state representation model under the same condition.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

An End-to-End Deep Learning Approach for State Recognition of Multifunction Radars

“…We have applied a GRU to the state recognition of radar word sequences in our paper [21], achieving pretty good results. This time, we also apply the GRU to the radar word recognition and offer a complete end-to-end recognition network.…”

“…This allows it to gradually supersede the feature engineering [17] and shallow neural network. Inspired by its success, we introduced deep learning into the actual radar signal processing and used RNNs in our MFR state recognition [21]. RNNs are neural networks with the hidden states that can capture the historical information of previous time steps.…”

“…This time, we also apply the GRU to the radar word recognition and offer a complete end-to-end recognition network. Compared with the previous state recognition methods, the advantages of GRU-based state recognition methods are the following [21]: (1) GRUs can process raw data directly without the need of other designs, and its ability to learn the features from the input data reduces the dependence on prior information; (2) GRUs can capture long dependencies in sequences and are more suitable for modeling long sequences such as radar pulses; (3) GRUs are nonlinear models that can better tackle the corrupted radar signal as well as the sequence alignment problems caused by missing data.…”

“…For MFR recognition, more and more papers use recurrent neural networks (RNNs) [17] to process the sequence data of MFRs [18][19][20]. Inspired by them, we introduced RNN to identify MFR states from extracted radar word sequences in our paper [21]. Compared with OOM or PSR, the advantage of RNNs is that they can process raw data directly and learn the effective features from it without other designs.…”

“…Based on the work of our paper [21], we add the radar word recognition to the network of MFR state recognition in this paper and propose a novel end-to-end recognition model for MFR state recognition. The whole network is the concatenation of two RNNs: the first RNN extracts the representation vectors of radar words from the input pulse sequences, which can be regarded as a feature extractor; the second RNN acts as a state recognizer to recognize the MFR states from the extracted representation vectors.…”