【作者】 成彬彬；

【导师】 邵贝贝； 陈惠连；

【作者基本信息】 清华大学 ， 核科学与技术， 2009， 博士

【摘要】 目标的高分辨一维距离像可以用来进行目标特征提取,对目标识别具有重要意义。传统的成像多采用对线性调频信号进行匹配滤波的方法,这种处理方法受回波波形畸变的影响较大,系统稳健性不高。本研究将仿生学的波形设计和信号处理方法引入到成像雷达的信号检测中,可显著克服传统处理方法存在的上述缺陷。本文以蝙蝠的回声定位系统为参考,研究了基于速度选通的一维多尺度高分辨距离像的目标成像方法。自适应波形发射和仿生智能信号处理是该成像系统的两个重要特性,发射波形采用脉内恒频加线性调频(CF+FM)的信号模式,CF窄带信号主要用于获取目标速度,而FM信号用于获取细节信息,并对目标成像,CF获取的速度信息一方面用来对FM信号处理模块进行补偿,另外一方面可以对一维距离像成像进行速度选通,发射波形的自适应性表现在:在搜索阶段,发射长周期脉冲,脉内以CF为主,一旦发现目标,在接近目标的过程中,脉冲重复频率增加,FM成分增多,以便实时获取目标信息和进行精确成像;对FM回波的智能仿生处理以听觉神经为模型,主要包括谱相关和谱变换两个模块,可以实现两个不同尺度的成像精度。谱相关模块包括用于频率编码的“耳蜗滤波器组”,用于半波整流的“内耳毛细胞”,用于低通滤波和峰值检测的“星形细胞”,用于互相关运算的“相合细胞”以及用于通道间信息融合的“神经中枢”等,谱变换模块在谱相关模块的基础上提高测距分辨率,从而建立目标的精确一维距离像。采用这种基于动力学方程的仿生建模处理方法,可显著提高处理系统的稳健性,抗噪声等性能。本研究建立了上述仿生处理算法模型,并通过仿真验证了其正确性,同时仿真结果也表明与传统的处理方法相比,该方法具有更高的距离分辨力和更好的灵活性及稳健性。论文最后利用System Generator完成了算法的FPGA硬件描述语言设计,并在Xilinx Virtex-5XCV5LX110上对算法进行了验证,表明了系统的硬件可实现性,实验结果表明,算法可对回波信号实现实时处理。更多还原

【Abstract】 High-resolution range profiles are used to extract target features for recognition and classification. The traditional method for imaging is using matched filter for LFM (Linear Frequency Modulated) pulse compression. The performance of compression is affected by distortion and Doppler frequency shift of the echo. The method is not robust. In the thesis, a bionic waveform design and signal processing method is introduces into imaging radar signal detection and the performance is further improved against traditional methods.A velocity selected imaging method for multi-scale high-resolution range profile is proposed based on bats’echolocation system. Adaptive hybrid waveform and intelligent bionic signal processing are two important characteristics of the system. The emissions are adaptive hybrid waveform containing CF (Constant Frequency) and FM (Frequency Modulated) signals in a single pulse. The CF component is used to get speed information of the target while FM is for fine detection and imaging. The Doppler frequency shift getting from CF processing is used for compensation of FM signal processing and also a speed gate for imaging. The waveform design is adaptive because in searching phase, the pulse period is long with most CF component and gets shorter and shorter with more FM component in tracking phase for real-time imaging; The intelligent bionic signal processing model based on auditory nervous contains two major blocks called spectrogram correlation and spectrogram transformation. These two blocks are used to get low and high resolution range profiles respectively for multi-scale detecting. In the spectrogram correlation block, a cochlear filter bank is adopted for encoding the bats’transmission and multiple echoes, in which the input is divided into several parallel pathways. Each frequency channel consists of an inner hair for half-wave rectification, followed by a stellate cell for low-pass filtering and peak-detection, and then a coincidence cell used as cross-correlator. The outputs of all frequency channels are fused in the auditory nervous centre to get the absolute range of the targets. The spectrogram transformation takes place across all frequency channels to improve range resolution and reconstruct fine range structure of the target.Bionic modeling and signal processing method for each block are studied in this thesis. The model is then simulated with MATLAB to check the correctness of the model.and the simulation results imply that the bionic method is more robust and flexible compared with traditional methods with higher resolution. At last the model is realized in FPGA (Field Programmable Gate Arrays). The FPGA hardware description code for this model is developed in Simulink@MATLAB based on System Generator for DSP toolbox and then downloaded to Xilinx Virtex-5XCV5LX110 FPGA. The experiment results improve that it is hardware realizable for the bionic model in a single FPGA chip and real-time processing can be achieved.更多还原