The weakly damped low frequency oscillation is one of the main factors that affect the stabile operation of interconnected power grids. Mode shape is a key characteristic of low frequency oscillation. Based on the measured data， mode shape can be identified after the oscillation happened， but the analysis method applicable in steady state is lacking. It has been observed from the wide area measurement data that small fluctuations exist continuously in power grids， caused by random changes of loads， which can also be called ambient signals. A method named auto regressive moving average-Prony（ARMA-P） is proposed to process such ambient signals. Based on the low frequency oscillation mode parameters identified by the ARMA model， the Prony model of ambient signal is further built， then the mode shape can be estimated. This method is used to analyze the simulation data from the New England system， and the result is compared with the small signal stability calculation result. It is further employed to process the measured ambient signals in China Southern Power Grid to validate the feasibility of this method.