基于差分演化-MP的快速信号稀疏分解

Fast Signal Sparse Decomposition Based on Differential Evolution-MP

利用差分演化算法具有鲁棒性强和全局收敛性好的优点,提出了一种基于差分演化的匹配追踪算法(DE-MP)。算法使用差分演化(DE)算法替换传统匹配追踪(MP)算法中的遍历搜索策略,优化了寻找稀疏分解最优原子的过程,从而大大降低了算法复杂度。此外,DE算法特殊的搜索策略很好地提高MP的全局收敛性,进一步提高了稀疏分解的准确性。通过对雷达仿真信号和语音信号仿真实验结果表明:与传统MP算法相比,差分演化匹配追踪算法(DE-MP)在计算速度上提高了两个数量级,在收敛精度上也有明显提高,且收敛精度优于其他改进MP算法。

To resolve the problem of traditional sparse decomposition algorithm which is complex in computation and time-consuming, a matching pursuit （MP） algorithm based on differential evolution （DE） is proposed on the advantage of strong robustness and good global convergence. In the algorithm, DE algorithm replaces the traversing search strategy of the traditional MP algorithm, which great- ly reduces the algorithm complexity by optimizing the process of finding the best sparse decomposition atomic. Also the special search strategy of DE algorithm is good to improve the global convergence of the MP and the accuracy of the sparse decomposition. The simula- tion results of the radar simulation signal and the speech signal test show that the DE-MP is increased two orders of magnitude in com- puting speed and improved obviously in the convergence accuracy in comparison with traditional algorithms of MP, and the convergence accuracy is even superior to the other improved algorithms.