In this paper, a new underdetermined mixing matrix estimation method for Blind Source Separation (BSS) from linear time delay mixture is developed based on Single-Source Point (SSP) detection. Firstly, prior knowledge of linear time delay mixtures are extracted with the help of receive antennas, and then the rule is derived to determine the SSPs. After selecting the SSPs, clustering method is applied for clustering, and estimates all the elements of mixing matrix. The proposed method can be used to estimate the Direct of Arrival (DOA) and signal extraction, which can be used in radar array signal processing areas. Simulation results reveal that the proposed method can estimate the mixing model with higher accuracy and better performance when SNR is low and number of sources is large, which is more reliable in the real non-cooperative environments.