Cognitive Radar Waveform Selection Technology Based On Machine Learning And Environment Perception

The concept of cognitive radar provides a new way of thinking and direction for the future development of Radar – intellectualized.Radar adjusts the waveform according to the change of the environment,which allows radar to adapt to the environment,and carries out more effective and accurate detection to the target.The main research and core of cognitive radar is the perception of working environment and the optimization design of transmitting waveform.Based on the basic structure of radar,this paper adopts the methods of standard function,off-line learning and online learning to realize the perception and waveform optimization of environment information.The tracking effect of general linear motion in the environment without interference,the effect of target tracking in the presence of interference and the prior knowledge are obtained.The tracking effect under the condition is analyzed.Firstly,based on the basic structure of cognitive radar,the component module and its realization function of cognitive radar are analyzed,neural network and reinforcement learning are introduced,and the process of signal processing is analyzed with Bayesian filter as the core.The model of transmitting waveform is established,and the influence of waveform parameters on measurement is analyzed,including the influence on SNR and observation value.The functional model of cognitive radar tracking task is established,including the state space of the target and the Kalman Filter algorithm under linear motion,and the waveform library is established.Based on the mean square error minimum criterion,the linear motion of the extended target is simulated and analyzed.Through the performance of the cognitive closed-loop radar and the traditional open-loop radar,the influence of the waveform parameter selection on the observation precision and the tracking precision is analyzed.This paper introduces the supervised learning method of BP neural network,uses the neural network to learn the corresponding decision rules,establishes the relationship between the environment information and the selection of waveform parameters,and analyzes the difference between the supervised learning method and the method of waveform selection using the criterion,and the performance of the method in the environment with interference.This paper introduces the reinforcement learning theory,and proposes two algorithms for waveform selection,such as the approximate dynamic programming Q learning algorithm and a state independent cognitive learning algorithm introducing neural network learning.The former is an off-line learning algorithm,while the latter adopts both the off-line learning method and the online learning method.The above method is compared with the method of waveform selection using the criterion function.The performance of the Q learning method depends on the state division,while the state independent cognitive learning algorithm,which is introduced into the neural network,has better adaptability and stability.Based on the machine learning method,this paper realizes the perception of the cognitive radar to the environment,establishes the relationship between the tracking and observation and the waveform parameters,so as to improve the tracking precision.