The phase array radar has been widely used in military due to the dramatic advance in computer technique and signal processing technology. However, the phase array always has thousands of antenna elements. It is a huge amount cost of hardware resource. In order to reduce the cost, we need to perform subarray level signal processing for phase array radar. In this paper, we propose the linear minimum variance adaptive beamforming approach to calculate the adaptive weighs at the subarray level for digital adaptive beamforming. The proposed method helps to decrease the hardware cost and make the phase array radar available to suppress the interference adaptively. During the calculation of the adaptive weights, the inverse of the matrix needs to be derived. The float round-off error in the calculation is accumulated. It is very important to analyse the impact of the float round-off error on the result accuracy. In this paper, the float round-off error in the proposed method is analysed.