Radar Emitter Classification With Attention-Based Multi-RNNs

Abstract

Analyzing and recognizing radar signals are important tasks for effective Electronic Support Measurement (ESM) system operation. The electromagnetic environment is highly complex nowadays, however, resulting in non-uniformed distributed pulse streams. The high-dimensional features of the radar emitters are also overly complicated. Isolating useful information of the pulse streams and removing noise can assist in the emitter classification process. This letter proposes an attention-based approach for radar emitter classification using recurrent neural networks (RNNs). Several RNNs assigned to individual features exploit the intrinsic patterns of the radar pulse streams via supervised learning; the learned patterns are then used to identify patterns of interest in the test pulse streams and place them into different categories. The attention mechanism demonstrates effective treatment of high missing and spurious pulse ratios, especially in cases of multiple consecutive missing pulses and multifunctional radar pulses. Simulation results also show that the proposed model outperforms other state-of-the-art neural network structures.