Abstract
Intra-pulse modulation recognition of radar signals is an important part of modern electronic intelligence reconnaissance and electronic support systems. With the increasing density of radar signals, the analysis and processing of multi-component radar signals have become an urgent problem in the current radar reconnaissance system. In this paper, an intra-pulse modulation recognition approach for single-component and dual-component radar signals is proposed. First, in order to adapt to the time-frequency energy distribution characteristics of various radar signals, we propose to extract the time-frequency images (TFIs) of received signals by Cohen class time-frequency distribution (CTFD) with multiple kernel functions. Besides, the image processing methods are used to suppress noise and adjust the size and amplitude of the TFIs. Second, we design and pre-train a TFI feature extraction network for radar signals based on a convolutional neural network (CNN). Finally, to improve the probability of successful recognition (PSR) of the recognition system in the pulse overlapping environment, a multi-label classification network based on a deep Q-learning network (DQN) is explored. Besides, two sub-networks take TFIs based on special kernel functions as input and re-judge the recognition results of some specific signals to further enhance the recognition effect of the recognition system. The proposed approach can identify 8 kinds of random overlapping radar signals. The simulation results show that the overall PSR of dual-component radar signals and single-component radar signals can reach 94.83% and 94.43%, respectively, when the signal-to-noise ratio (SNR) is −6 dB.
Highlights
Intra-pulse modulation recognition of radar signals is an important part of modern electronic intelligence reconnaissance and electronic support systems [1], [2]
The simulation results show that the overall probability of successful recognition (PSR) of dual-component radar signals can reach 94.83% when the signal-to-noise ratio (SNR) is −6 dB
If the recognition result of the sub-network is that the current pulse does not contain the sinusoidal frequency modulation (SFM) radar signal component, the SFM radar signal will be deleted from the action set, and the recognition result will be reselected according to the Q value output by the deep Q-learning network (DQN)
Summary
Intra-pulse modulation recognition of radar signals is an important part of modern electronic intelligence reconnaissance and electronic support systems [1], [2]. Based on the feature vectors extracted by CNN, the RNN outputs the recognition results of intra-pulse modulation types of single-component and dualcomponent radar signals through multiple cyclic iteration classification processes. Based on the above reward and punishment rules in reinforcement learning, the classification network can choose the order of output of recognition results autonomously to avoid the hard restriction of traditional classification network labels and improve the adaptability of classification network to the recognition of complex dual-component radar signals formed by pulse overlapping. For the recognition of intra-pulse modulation of radar signals in this paper, the RNN obtained in the previous step can be regarded as an agent of reinforcement learning In this part, we train the agent with the deep Q-learning algorithm and get the agent with multilabel classification ability, which is DQN. The training process can be described by Algorithm 1, where the input is the feature vectors of TFIs, the output is Algorithm 1 Deep Q-Learning Network Training Process
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
