| 摘要: |
| 针对多雷达辐射源脉冲交错背景下，线性调频（Linear Frequency Modulation, LFM）信号低信噪比导致的脉冲分裂带来原始信号参数难以估计的问题，本文提出了基于深度神经网络和直方图统计的LFM信号两阶段提取与参数估计方法。首先利用双向长短时记忆网络挖掘原始脉冲流中LFM信号与非LFM信号的调制模式差异并进行分类；其次通过序列调频斜率直方图寻找LFM信号分裂脉冲序列间隐含的原始信号调频斜率信息，提取不同调频斜率的LFM信号脉冲子序列；最后在每个子序列中分别估计原始信号的参数。仿真实验结果表明，相较于传统的序列差值直方图算法和循环神经网络分选方法，本文所提方法能够更准确地提取出LFM脉冲信号，并得到较为精确的参数估计结果。 |
| 关键词: 脉冲分裂 信号提取 双向长短时记忆网络 序列调频斜率直方图 参数估计 |
| DOI： |
| 分类号:TN95 |
| 基金项目:国家自然科学基金（青年项目） |
|
| An Extraction and Parameter Estimation Method of LFM Signals Under Pulse Splitting Conditions |
|
|
|
|
| Abstract: |
| Aiming at the issue of pulse splitting due to low signal-to-noise ratio of linear frequency modulation (LFM) signals in a multi-radar radiation source pulse interleaving background, which complicates the estimation of original parameters, this paper proposes a two-stage extraction and parameter estimation method for LFM signals based on deep neural networks and histogram statistics. Initially, bidirectional long short-term memory is utilized to mine the modulation pattern differences between LFM and non-LFM signals within the original pulse stream for classification. Subsequently, the sequential frequency modulation slope histogram is used to uncover the intrinsic original signal frequency modulation slope information between split LFM pulse sequences, extracting LFM signal pulse subsequences of different frequency modulation slopes. Finally, the parameters of the original signal in each subsequence are estimated separately. Simulation experiment results indicate that, compared to the traditional sequential difference histogram algorithm and recurrent neural network sorting method, the method proposed in this study can extract LFM pulse signals more accurately and achieve more precise parameter estimation results. |
| Key words: pulse splitting signal extraction BLSTM SFMS parameter estimation |