Aiming at blind estimation of the combination code and the information sequences for binary offset carrier signal under low signal-to-noise ratio, the unitary fuzzy phenomenon might exist in the largest singular vector and the secondary largest vector when using singular value decomposition (SVD) algorithm to estimate the combination code and the information sequences under unknown delay. An improved SVD method is proposed to solve this problem. Firstly, the largest singular vector and the secondary largest vector are linearly combined. Then, the coefficient matrix can be selected in an optimal way. Finally, combined with the coefficient matrix, it can estimate the combination code sequence and the information sequence. The simulations show that the improved SVD algorithm can be applied to both the combination code and the information sequences estimation under unknown delay, and it has the characteristics of good unitary fuzzy elimination performance, less data required and strong anti-noise performance.