Abstract: To improve the converging performance of blind equalizer, a novel blind equalization algorithm incorporating the modified support vector machine (SVM) and constant modulus algorithm (CMA) is proposed. The proposed algorithm firstly adopts the support vector machine (SVM) which possesses excellent generalization ability under small training samples to initialize the coefficients of blind equalizer with a short training sequence. After the SVM ini-tialization, it switches to constant modulus algorithm (CMA) to alleviate the computational burden. However, under time-varying underwater acoustic (UWA) channels, the initial coefficients obtained by SVM may still contain mismatch with the channel after algorithm switching, as the classic SVM algorithm is inherently nonadaptive. To deal with this problem, a modified SVM is formulated to initialize the coefficients of blind equalizer at the presence of non-stationary channels to facilitate smooth algorithm switch. In addition, fractional spaced structure (FSE) as well as embedded digital phase lock loop (PLL) is also adopted to further improve the performance of blind equalization. Experimental results performed with numerical simulation and lake-trial data are provided, demonstrating the improvement of the proposed algorithm in convergence rate under time-varying channels.