采用STFT幅度值的长信号相位恢复算法

摘要

针对以往基于短时傅里叶变换( STFT)幅度值的相位恢复( STFTMPR)算法仅能处理一组较短的信号,并且信号的STFT幅度测量值的长度只能是质数的情况,提出了一种基于STFT幅度值的长信号相位恢复( LS-STFTMPR)算法.把一组长信号变化成几组较短的数组信号,通过改进的最小二乘( LS)法获取梯度下降( GD)法的迭代初始值,然后最小化各数组信号的非凸损失函数,并最终收敛到全局最小值.实验结果表明,在恢复一组较长信号的时候,所提算法的性能明显优于STFTM-PR算法,并且具有较强的抗噪声能力.%The phase recovery algorithm based on the magnitude of Short-Time Fourier Transform( STFT) called STFTMPR algorithm,can only process a relatively short set of signals,and the length of the magni-tude of the signal can only be prime number. An algorithm of Phase Retrieval of Long Signal Based on STFT Magnitude Measurements( LS-STFTMPR) is presented. A group of long signals are divided into sev-eral sets of short array signals,and the iterative value of the Gradient Descent( GD) method is initialized by an improved Least Squares( LS) method,then the non-convex loss function is minimized to make it con-verge to the global minimum. The experimental results show that when recovering a group of relatively long signals,the performance of the LS-STFTMPR algorithm with a high noise-resistance capability is obviously better than that of STFTMPR algorithm.