Research And Implementation Of Approximate Message Passing Algorithm Based On FPGA

Compressed sensing is a new type of signal processing method,which contains two core components of "compression" and "perception".Compression refers to making full use of the prior information of the signal in the process of information collection,and achieving signal compression while collecting information;perception is to achieve distortion-free perception of signals through a small amount of observation data,combined with efficient sparse signal restoration algorithms.In compressed sensing,reconstruction algorithms are the focus of research.Approximate message passing algorithm is a reconstruction algorithm with low complexity and good reconstruction performance.Metamaterial aperture imaging technology is a new real-aperture microwave imaging technology developed in recent years.By optimizing the design and layout of the metamaterial aperture resonant unit,it can generate randomly varying spatial radiation under the excitation of electromagnetic signals of different frequencies.Field,the target scene is detected in order to obtain the information of the scene.In metamaterial aperture imaging,due to various factors,the number of measurement modes that can be generated is very limited.Therefore,metamaterial aperture radar imaging technology needs to recover the original signal from a small amount of observations.The main research content of this paper includes the research of imaging principle based on metamaterial aperture radar,the research of metamaterial aperture radar imaging algorithm based on approximate message passing algorithm,and the realization and optimization of VLSI of approximate message passing algorithm.The article does not do too much research on the specific design and simulation of the metamaterial aperture antenna.The specific research content includes the following aspects:(1)The imaging principle of metamaterial aperture radar: firstly,the generation mechanism of the random radiation field of metamaterial aperture antenna is studied,the imaging model of metamaterial aperture radar is built,and its imaging measurement equation is analyzed;then the basic theory of compressed sensing and mainstream compression reconstruction algorithm are introduced,On this basis,the Approximate Message Passing(AMP)algorithm is introduced.(2)Application of Approximate Message Passing Algorithm in Metamaterial Aperture Radar Imaging: Introduced the principle and derivation process of CAMP algorithm,the complex number domain extended form of AMP algorithm,introduced the principle of median complex approximate message passing(MC-AMP)algorithm and its application in radar imaging;introduced the principle of vector approximation message passing(VAMP)algorithm and combined it with the theory of group sparse signal reconstruction,and deduced the complex VAMP algorithm.Perform performance analysis and verification by building a simulation environment,and draw conclusions after simulation verification:Compared with the MC-AMP algorithm,the CVAMP algorithm has fast iteration speed and strong anti-noise ability,and is suitable for sparse signal reconstruction of metamaterial aperture radar.(3)Optimization strategy of CVAMP algorithm in FPGA: Analyze the computational complexity of CVAMP algorithm,decide to use optimization acceleration methods such as data format optimization,parallel expansion optimization,systolic array acceleration,etc.to accelerate the CVAMP algorithm,And designed a complex matrix multiplication module with three structures: parallel pipeline,systolic array,and preprocessed systolic array.Perform performance evaluation of three complex matrix multipliers through FPGA simulation verification: the parallel pipeline structure consumes the least resources and has the slowest calculation speed;the systolic array structure has the fastest calculation speed.The resource utilization is the highest.Compared with the systolic array structure,the preprocessing systolic array structure saves 25% of DSP resources,but the calculation speed is almost the same.It is concluded that the pre-processing systolic array complex matrix multiplier is a high-efficiency,low-overhead matrix multiplication structure.(4)Design of CVAMP Algorithm Fast Implementation Scheme in FPGA: Combining the characteristics of CVAMP algorithm,design a parallel rapid implementation architecture,and Introduced the specific implementation methods and simulation verification results of important modules such as its internal systolic array data flow control,Cordic algorithm to achieve modulo operation,and soft threshold algorithm.Finally,the overall implementation effect of FPGA is compared and analyzed,and it is concluded that,compared with Matlab operation,the computational efficiency of CVAMP's FPGA rapid implementation architecture has increased by about 30 times.