The invention discloses an LLL (Lenstra-Lenstra-LovaszLattice) ambiguity decorrelation algorithm. According to the algorithm, upper triangular decomposition (U<T>U) is performed on a covariance matrix Qa through utilizing Cholesky decomposition method, so that an upper triangular matrix U<T> can be obtained, and with the above decomposition method adopted, the computational efficiency of the LLL algorithm can be improved; before algorithm decomposition every time, descending sorting is performed on column vectors of the matrix Qa according to the magnitude of inner products, and a coefficient matrix can obtain a minimum integer value, and the decorrelation performance of the decorrelation algorithm is better; and a rounding step in an orthogonal transformation process is shifted to a Z-solving matrix, and therefore, calculation quantity increase and error accumulation caused by repeated rounding in an iteration process can be avoided, and the computational efficiency and success rate of the novel algorithm can be improved.