短时高斯包络线性调频基自适应信号分解算法

Short Time-Adaptive Gaussian Chirplet Signal Decomposition

作为一种参数化时频分析的方法,基于高斯包络线性调频基自适应信号分解的快速算法具有分辨力高、零交叉项和计算量小的优点,在信号时频分析中具有独特的优势和广阔的应用前景。然而该快速算法却存在由于采样点初值选择不当而造成分解失效的缺点,虽然后来的基于优化初值选择的自适应高斯包络线性调频基信号分解对初值选择算法进行了改进,提高了分解性能的稳定性,但仍存在较多的问题没有解决。本文将对这些问题进行研究和改进,并提出短时自适应高斯包络线性调频基信号分解算法。算法通过加短时窗来增强时频中心定位的准确性,通过控制采样基时宽来获取有效的初始方差取值范围,从而提高了分解的自适应性和稳定性。对仿真信号和语音信号的分解结果表明了该算法的有效性。

As a parametric time-frequency analysis method, the fast algorithm of adaptive Gaussian chirplet decomposition has high resolution, zero cross term, fast speed and wide prospect in time-frequency analysis. Yet the result of the algorithm will become invalid when the test points and the variance chosen inappropriately. Although the optimized initial adaptive Gaussian chirplet decomposition improves algorithm in some degree, there are still some problems not solved. The paper studies such problems and proposes the short time adaptive Gaussian chirplet decomposition algorithm. It improves the accuracy of positioning of time-frequency center by adding short time window to the signal, obtains the valid variance range by controlling the time width of test basis, and improves the adaptability and stability of decomposition. The effectiveness of the algorithm introduced is demonstrated by numerical simulations and the decomposition results of the real voice signal.