If the adaptive weights update in the whole range domain, the target signal may be contained in the training data, resulting in signals self-cancelling effect. To cope with such a reality problem, a two stages detection based method is discussed for updating the adaptive weights in the whole range domain when the training data is contaminated by the target signals. The main objective of the first stage is to select proper samples as training data, in which the range cells corrupted by target signal will be deleted from the training data. Then adaptive beamforming and detection processing will be performed in the second stage. The simulations show that the proposed method is capable of updating the adaptive weights in the whole range domain, and is suitable for early warning radar to suppress sidelobe jamming.