基于神经网络与时域校验的信号分选方法

针对现有雷达信号预分选方法对参数捷变雷达信号分选准确率不高的技术难题,提出了一种基于深度学习算法的全连接神经网络与时域校验的雷达信号预分选方法。该方法首先提取雷达数据库中已知雷达信号的载频、脉宽和脉内调制信息作为单脉冲分选特征,使用全连接神经网络完成单脉冲的识别。为了避免神经网络将未在雷达数据库中的信号(未知雷达信号)识别为已知雷达信号,在神经网络的输出层中加入置信度神经元生成置信指数,将置信指数低于阈值的判定为未知雷达信号进行剔除。最后根据分选结果调用雷达数据库中对应的时域信息(脉冲重复间隔),进行时域校验,完成雷达信号预分选。仿真结果表明,该方法在不同信噪比环境下对参数捷变雷达信号有较高的分选准确率,并且能有效剔除未知雷达信号。

A Method of Radar Signal Sorting Based on Neural Network and Time Domain Verification

Aiming at the technical difficulty that the existing radar signal pre-sorting method has low accuracy of parameter agile radar signal sorting, a fully connected neural network based on deep learning algorithm and time-domain verification radar signal pre-sorting method is proposed. Firstly, this method extracts the carrier frequency, pulse width and intra-pulse modulation information of the known radar signal in the radar database as the single pulse sorting features, and uses the fully connected neural network to complete the identification of the single pulse. Then, in order to avoid that the neural network recognizes the signals (unknown radar signals) that are not in the radar database as the known radar signals, a confidence neuron is added to the output layer of the neural network to generate a confidence index, rejecting the decision that the confidence index is below the threshold as an unknown radar signal. Finally, according to the sorted result, the corresponding time domain information (pulse repetition interval) in the radar database is called to perform time domain verification to complete the radar signal pre-sorting. The simulation results show that the method has higher sorting accuracy for parameter-agile radar signals under different signal-to-noise ratio environments and can eliminate the unknown radar signals effectively.