本文選題：時(shí)空二維MUSIC算法　切入點(diǎn)：FPGA+DSP實(shí)現(xiàn)　出處：《西安電子科技大學(xué)》2014年碩士論文

【摘要】：本文采用14個(gè)陣元組成的均勻線陣作為天線陣列模型,以時(shí)空二維MUSIC算法以及測距原理為理論,對測頻測向設(shè)備的實(shí)現(xiàn)進(jìn)行了研究。首先,使用Matlab進(jìn)行了仿真。從仿真結(jié)果中得到了時(shí)空二維MUSIC算法在均勻線陣模型下的角度分辨率,確定了影響測距和測向精度的參數(shù)。驗(yàn)證了此方案在給定模型、參數(shù)下理論上的可行性。其次,為了進(jìn)行硬件實(shí)現(xiàn),對時(shí)空二維MUSIC算法進(jìn)行了降維處理。首先對得到的差頻信號(hào)做1024點(diǎn)FFT變換,得到1024個(gè)頻點(diǎn),接著對需要的頻點(diǎn)用一維MUSIC算法進(jìn)行處理。在對一維MUSIC算法的實(shí)現(xiàn)過程中,將它分成了協(xié)方差矩陣的估計(jì)、特征值分解、目標(biāo)信號(hào)個(gè)數(shù)估計(jì)和譜峰搜索四個(gè)模塊。降維處理后的時(shí)空二維MUSIC算法雖然避免了在頻率、方向上同時(shí)進(jìn)行搜索,但是計(jì)算量仍然很大,算法實(shí)現(xiàn)復(fù)雜。針對此問題,本文最終選擇FPGA和DSP協(xié)作作為硬件實(shí)現(xiàn)方案。FFT變換、協(xié)方差矩陣的估計(jì)數(shù)據(jù)量大,且很規(guī)則,因此由FPGA完成。而協(xié)方差矩陣的特征值分解、目標(biāo)信號(hào)個(gè)數(shù)估計(jì)、譜峰搜索的實(shí)現(xiàn)過程復(fù)雜,因此在DSP中完成。為了滿足實(shí)時(shí)性的要求,同時(shí)也基于芯片資源和處理速度的考慮,本設(shè)計(jì)的FPGA芯片選用了Altera的Arria V,DSP芯片采用了TI的TMS320C6455芯片。DSP與FPGA之間主要通過EMIFA接口進(jìn)行通信。FPGA部分采用了乒乓操作、流水線、狀態(tài)機(jī)等方法來進(jìn)行控制和提高算法的處理速度,并且考慮到該設(shè)計(jì)對實(shí)時(shí)性要求很高,多數(shù)情況下犧牲面積來換取速度。由于特征值的分解、信號(hào)源個(gè)數(shù)的估計(jì)方法很多,因此,對各處理算法進(jìn)行了分析,在保證能達(dá)到要求的精度和分辨率的前提下,DSP部分選出了最適合的方案。比如,信號(hào)源個(gè)數(shù)這一參數(shù)的獲得采用MDL準(zhǔn)則完成,協(xié)方差矩陣的分解采用QR算法完成。各部分算法、硬件之間相互配合,FPGA部分給出了FFT變換和得到的協(xié)方差矩陣部分的SigtapII的測試結(jié)果,DSP部分給出了實(shí)現(xiàn)算法的流程圖。所有算法的實(shí)現(xiàn)都在器件上做了測試,能夠成功運(yùn)行�；旧贤瓿闪嘶跁r(shí)空二維MUSIC算法的障礙物的測距測向的分辨率、精度和實(shí)時(shí)性的要求。
[Abstract]:In this paper, a uniform linear array composed of 14 array elements is used as the antenna array model, and the realization of frequency and direction finding equipment is studied based on the space-time two-dimensional MUSIC algorithm and the principle of ranging.First, Matlab is used to simulate.From the simulation results, the angular resolution of the spatio-temporal two-dimensional MUSIC algorithm under the uniform linear array model is obtained, and the parameters affecting the ranging and direction finding accuracy are determined.The theoretical feasibility of this scheme under given model and parameters is verified.Secondly, in order to implement the hardware, the dimension reduction of the two-dimensional MUSIC algorithm is presented.First, the difference frequency signal is transformed by 1024 points FFT, and 1024 frequency points are obtained. Then, the needed frequency points are processed by one-dimensional MUSIC algorithm.In the implementation of one-dimensional MUSIC algorithm, it is divided into four modules: covariance matrix estimation, eigenvalue decomposition, target signal number estimation and spectral peak search.Although the dimensionally reduced two-dimensional MUSIC algorithm avoids simultaneous search in frequency and direction, the computation is still very large and the algorithm is complex.In order to solve this problem, this paper chooses FPGA and DSP cooperation as the hardware implementation scheme. The covariance matrix estimation data is large and regular, so it is completed by FPGA.The eigenvalue decomposition of the covariance matrix, the estimation of the number of target signals, and the implementation of the spectral peak search are complex, so they are completed in DSP.In order to meet the requirements of real-time, but also based on chip resources and processing speed considerations,The FPGA chip of this design uses Altera's Arria VN DSP chip to use TI's TMS320C6455 chip .DSP and FPGA to communicate mainly through EMIFA interface. The part of the design adopts ping-pong operation, pipeline, state machine and other methods to control and improve the processing speed of the algorithm.And considering that the design requires high real-time, in most cases sacrificing the area for speed.Because of the decomposition of the eigenvalue, there are many methods to estimate the number of signal sources. Therefore, the processing algorithms are analyzed, and the most suitable scheme is selected under the premise that the required precision and resolution can be achieved.For example, the number of signal sources is obtained by MDL criterion and the decomposition of covariance matrix by QR algorithm.The FFT transform and the test result of the covariance matrix part are given. The flow chart of the algorithm is given.All the algorithms are tested on the device and can run successfully.The resolution, accuracy and real-time requirements of obstacle ranging and direction finding based on spatio-temporal MUSIC algorithm are basically fulfilled.
【學(xué)位授予單位】：西安電子科技大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類號(hào)】：TN820.15;TN911.7

【參考文獻(xiàn)】

相關(guān)碩士學(xué)位論文 前1條

1 韓丙同;空間譜估計(jì)算法性能分析[D];哈爾濱工業(yè)大學(xué);2010年

，

本文編號(hào)：1716128