Constructing characteristic basis functions (CBFs) is a key step of characteristic basis function method (CBFM). But it is required to set adequate plane wave excitations in each sub-block, which leads to the increased number of characteristic basis functions and the longer time consumed in singular value decomposition of traditional method. In order to accelerate the construction of CBFs, an improved CBFM is presented, which fully considers the mutual coupling effects among sub-blocks and then the secondary level characteristic basis function (SCBF) is obtained, therefore the number of plane wave excitations is reduced greatly, and so is the number of characteristic basis functions. The adaptive cross approximation algorithm is also used to accelerate the matrix-vector multiplication procedure of generating SCBF and constructing the reduced matrix. Numerical results demonstrate that the proposed method is accurate and efficient.