脉冲压缩技术是为了解决雷达探测威力与距离分辨力之间的矛盾而提出的。传统匹配滤波方法采用加汉明窗等窗函数的手段抑制脉冲压缩后的副瓣电平,但只能将压缩后的副瓣电平进行一定程度上的抑制,并且带来了不必要的主瓣展宽。近年来,基于卷积神经网络的信号处理技术被广泛应用于各个领域。文中提出了一种新的基于卷积神经网络的线 性调频信号脉冲压缩方法,能够在获取近似理想脉冲压缩主瓣的同时,极大地抑制副瓣电平。该方法共由三个子网络组成,分为两个阶段进行。在第一个阶段,使用两个一维卷积神经网络分别从雷达回波信号的实数部分和虚数部分提取特征,并实现初步脉冲压缩;然后基于第一阶段子网络的输出,在第二阶段中利用含有一层隐藏层的子网络来完成脉冲压缩。通过仿真验证,与传统匹配滤波方法相比,该方法可以在获得极窄的脉冲压缩主瓣情况下将脉冲压缩副瓣电平抑制在-85dB以下,方法有效性得到了充分验证。

Pulse compression technology is proposed to solve the contradiction between radar detection power and range resolution.The traditional matched filtering methods with weighting windows, such as Hamming window, only suppress the sidelobe level in a certain extent, however widen the mainlobe. In recent years, the signal processing methods based on convolutional neural network (CNN) are widely used in various fields. Based on CNN, a new pulse compression method of LFM signal is proposed, which can not only obtain almost ideal mainlobe width, but also obtain ultra-low sidelobe level. The proposed method is consisted of two stages and three subnetworks. In the stage-1, two 1-D CNNs are used to extract features from the real part and imaginary part of the LFM signal sequence and complete preliminary pulse compression. Then, based on outputs of the stage-1 subnetworks, the subnetwork with one hidden layer in stage-2 is used to achieve final pulse compression. Compared with traditional matched filtering methods,the superiority of the proposed method is validated by numerical simulations, that the method can obtain extremely narrow mainlobe and suppress the peak sidelobe level below -85 dB.

TN957. 51