Dsp-based Radar Target Recognition Algorithm For High-speed Implementation

The radar target recognition is an important developing field of moder radar technology. The range profile of target formed by high-range resolution radar is a host issue in the field of radar target recognition. With the application and development of pattern recognition and signal processing, DSP has played an important role in the signal and data processing of radar target recognition. Algorithm framework and high-speed implementation of range profile radar target recognition based on floating-point DSP is intensively studied in this paper.First, the model and characteristic of range profile are discussed. Starting from the feature extraction and transform of radar target recognition based on rule-function, usual two subspace target recognition methods are reviewed. The characteristic and frame of feature subspace based on singular value decomplsition (SVD) are intensively studied. Accordingly, SVD is the key of whole target recognition algorithm.Secondly, some algorithms of singular value decomplsition about floating-point matrix are analysed and reviewed seriously. Then, a modified orthonormal decopmplsition is employed to decrease complexity and operation of the algorithm based on the classical algorithm. Furthermore, the speeds of classical and modified SVD algorithms are estimated theoretically in terms of the framework and characteristic of TMS320C6713B DSP. Besides, the theoretical simulation and compare of two algorithms are accomplished.Finally, the design of target recognition system based on TMS320C6713B DSP is discussed. With this hardware platform, the software-design and implementation of two singular value decomposition algorithms are discussed, during which process various optimal methods of software are used to improve the speed and efficiency of program. Then, the performance of two SVD algorithms and feature subspace radar target recognition algorithm based on SVD are evaluated according to real data of planes. The experimental results show the feasibility of modified singular value decomposition algorithm in allowable range of precision error, which can improve the speed of feature subspace algorithm efficiently.