Wigner distribution is a joint time-frequency representation for nonstationary signals. Many techniques are proposed to synthesize signals from the desired Wigner distribution. In this paper, we first use the clustering technique to achieve signal separation in the fractional Fourier domain. Then the result is used as the initial guess for the weighted cross Wigner distribution synthesis (WCWDS) algorithm proposed by Boudreaux-Bartels to obtain better signal separation result. Based on the simulation results, the proposed technique provides the recovered signal that is 5.3% closer to the desired signal compared with WCWDS algorithm.