针对相移键控(MPSK)相干光通信系统中的频率偏移，最常用的两种算法是基于差分相位或者FFT最大值的算法，但是当数据长度较短时，两种算法均很难实现MHz的估计误差，这将使得后续载波相位恢复估计很难恢复原始数据。为满足MPSK系统中频率偏移估计算法高精度和实时性的要求，首次将多重信号分类算法引入该问题，提出一种基于快速傅里叶变换和多重信号分类的频率估计算法，该算法在数据较短时精度很高。利用基本原理和流程图对算法加以说明，并进行了20-GBaud/\sQPSK相干光系统仿真实验验证算法切实可行。

For optical coherent M-ary phase-shift-keying (M-PSK) system, the frequency offset algorithm based on differential phase or FFT maximization which was widely used is difficult to achieve MHz estimation error when the data length is short, which is difficult for the following carrier phase estimation to recover the data. To meet the needs of high accuracy and real-time performance for frequency offset estimation in the M-PSK system, a frequency estimator based on fast Fourier transform and multiple signal classification (MUSIC) was proposed and investigated. For the first time, MUSIC algorithm was used in this area. The proposed algorithm is accurate especially when the data length is short. The principle and flowchart were proposed to illustrate the algorithm. Numerical simulations of 20-Gbaud QPSK coherent systems were carried out to demonstrate this algorithm.