本文采用一种经过特殊处理的时变自回归滑动平均（ARMA）模型对非平稳随机信号进行分析．将这种模型左边的时变参数假设为一组基时间函数的线性组合，右边时变参数简化为常数，并用反馈线性估计法进行参数估计。该方法的主要特点是简单，计算量小，占用存储空间少．并用仿真的方法对算法予以验证，可用于一些常用的非平稳随机信号的分析． In this paper, a specially treated time-varying autoregressive moving average (ARMA) model is used to analyze nonstationary random signals. The time-varying parameters on the left side of the model are assumed to be a linear combination of a group of time functions. The time-varying parameters on the right are simplified to constants and the parameters are estimated using the feedback linear estimation method. The main features of the method are simple, small amount of computation, less storage space. The simulation method is used to verify the algorithm, which can be used to analyze some commonly used nonstationary random signals.