work modes of airborne phased array radar using deep learning

As the precursor and foundation of electronic warfare, electronic reconnaissance is an important factor in determining the outcome of the war. How to use the intercepted intelligence in the complex electromagnetic environment to provide support for the electronic warfare, so as to improve the ability of electronic warfare, has become a key issue to be solved in modern electronic warfare. Taking the confrontation among space movement platforms as our research background, we mainly study the simulation and identification of airborne phased array radar under different work mode. Based on the basic principle of radar, four typical work modes of airborne phased array radar are simulated, and the deep neural network is used to study and identify the characteristics of signal, hoping this work can provide support for the development of electronic countermeasure technology. The main work of this paper are as follows:

1. To meet the simulation requirements, the model of airborne phased array radar is established to simulate typical work modes of radar, according to different intercept scene. The simulation contains four kinds of work mode such as track while scan, track and scan, multi-target tracking and single target tracking. Simulation results are the pulse description words intercepted by the reconnaissance aircraft, which provide input data for work mode identification of airborne phased array radar.

2. A signal amplitude fitting method based on multi-level modeling is proposed to solve the problem that the characteristics of the radar scaning are easily interrupted by noise. Multi-level modeling method is used to model the pulse description words at pulse level, pulse group level and the work mode level, effectively overcome the influence of dropped or false pulse on amplitude fitting.

3. The method based on denoising autoencoder is proposed to recognize work modes of airborne phased array radar. Also we make a thorough analysis of the parameters belong to deep neural network, such as the number of hidden layers, the number of hidden nodes and the ratio of noise. On this basis, the identification rate of different deep neural network are compared. Experimental results show that denoising autoencoder can adaptively extract the essential characteristics of different modes, and the recognition rate reaches 96.68%, which verified the effective of the proposed method.

4. In view of the disadvantage of denoising autoencoder, costing lots of training time, a new method based on marginalized stacked denoising autoencoder is proposed to recognize work modes of radar. Also an information fusion method combining SVM and DS evidence theory is proposed to further improve the classification accuracy. Qualitative experiments show that the proposed method combining marginalized stacked denoising autoencoder and DS evidence theory, can not only get a high recognition rate, but also achieve good result on small training set, which verified its the effectiveness and superiority.

The research is supported by the Key Project of National Science Foundation of China with monitoring data of high-speed train service security state assessment of the key questions (No.61134002).

Key Words: airborne phased array radar; work mode; functional level simulation; multi-level modeling; autoencoder; information fusion