In this paper, fast independent vector analysis (FastIVA) based on convolutional aliasing model is proposed to separate the aliased signals collected by distributed acoustic sensing (DAS) system, and the time-frequency entropy is used to judge the separation performance. The results show that the time-frequency entropy interval of the separated signals and the source signals correspond to each other, which means that the aliased signals are separated. This method is used to improve the accuracy of DAS system signal detection and recognition in complex real environment, and reduce the number of false alarm events.