Abstract: The minimum mean squared error ordered successive interference cancellation (MMSE-OSIC) detection algorithm is a bank of linear receivers, with the detected signal components successively canceled from the received signal at each stage. More specifically, the detected signal in each stage is subtracted from the received signal so that the remaining signal with the reduced interference can be used in the subsequent stage. By analyzing the MMSE-OSIC detection algorithm with the error propagation, we find out the estimation error can be reduced with the order, and the estimation of reliability can be increased, while the performance is related to detection order. In order to mitigate the impact of the error propagation, the paper proposes a detection algorithm of sorting group user interference elimination between groups. At the same time, the Oblique projection is a type of parallel projection, it projects an image by intersecting parallel rays from the three-dimensional source object with the drawing surface. The algorithm adopts the Oblique Projection Principle to the user in reverse order, chooses the user data with maximum correlation to detect finally. Meanwhile, one of the shortcomings in the MMSE-OSIC is the high computational complexity, in order to alleviate the drawback, the paper groups the user data and then to achieve user interference offset between groups. And then it tests the users in the group by the maximum likelihood (ML) detection again after the minimum mean squared error (MMSE) detection to reduce the error propagation and the algorithm complexity. The theoretical analysis and experimental results show that the improved algorithm improves the user equivalent signal-to-noise ratio, reduce the detection error rate