Signal Processing Of LFMCW Radar And Implementation Based On GPU

In the early era,the development of the linear frequency modulated continuous wave(LFMCW)radar was not comparable with that of pulse radar both in the theoretical basis and key technology,which greatly constrained the application fields of LFMCW radar.However,in recent years,the performance of LFMCW radar in short-range target detection is outstanding,which gradually attracts the attention of massive researchers in this field.Moreover,with the rapid improvement in the performance of digital signal processing devices and the incresing development of high-speed digital signal processing technology,the research on key theory and technology of LFMCW radar has also entered a new historical period.This thesis conducts a series of study on the LFMCW radar signal processing algorithm and its parallel implementation techniques.Firstly,this thesis introduces the system structure of LFMCW radar,posterior to the understanding on the development history of the LFMCW radar,the current state of research,and its main advantages and disadvantages.We can find out that the distance and speed information of the target to be detected is contained in the center frequency corresponding to the spectrum peak of the target echo beat signal through detailed derivation and study on a single-period sawtooth LFMCW signal.However,due to the presence of distance-velocity coupling problem,the real distance information of the target cannot be accurately estimated.From the analysis on the sawtooth LFMCW signals of several periods,it is concluded that the spectrum peak of the echo beat signal of the moving target appears as the cosine envelope with the change of the frequency modulation period,which contributes to the distance-velocity decoupling method.Further analysis on the detection performance of the sawtooth LFMCW radar is also implemented.Then,a complete signal processing scheme for the target parameter estimation in LFMCW radar is given based on the signal analysis of the preceding chapter.The MTI is conducted by subtracting the frequency spectrum of the beat signal of the adjacent frequency modulation period to suppress the fixed clutter.On the basis of MTI,an FFT-equivalent doppler filter bank is utilized for MTD processing to attain multi-period signal accumulation and velocity measurement.Then,a series of integrated target detection methods,including CFAR target detection,twice ranking selection,condensation processing and doppler compensation,are used to extract the target parameter information.And the performance of the proposed LFMCW radar signal processing algorithm is verified and analyzed by simulated experiments for multiple sets of targets with different parameters.Finally,the parallel implementation technology of GPU-based LFMCW radar signal processing algorithm is presented.At the beginning,a brief introduction to GPU and CUDA is given.Then we discuss in detail the parallel implementation process of these key algorithms,including FFT with window,MTI,MTD,CFAR,and twice ranking selection in LFACW radar signal processing under CUDA architecture,which are also implemented on different hardware and software platforms with multiple sets of simulated data.Through comparison and analysis of the estimated results of target parameters and the running time of processing algorithm,it can be demonstrated that GPU has stronger ability than that of CPU to implement data processing and parallel computing,which can greatly improve the real-time performance and effectiveness of the system under the condition of large computational workload.