This thesis deals with the joint maximum likelihood (ML) estimation approach of frequency offset and symbol timing without the known symbols. Under time-varying frequency selective channels, since data-aided synchronization is done through the joint estimation with channel information, its performance is better than that not using a training sequence but its computational complexity is overloaded and computed by much time. Recently in OFDM systems, since non data-aided synchronization is studied mainly over environments of AWGN or time-varying (or time-invariant) flat fading channels, its application to real communication systems is so hard. The proposed algorithm is simply applied to time-varying frequency selective channels by using alpha-function assumed in this paper. The proposed estimator presents the robust characteristics in real environments because of the modified structure of previous methods. And it requires only a few computational loads due to its simple scheme. It is shown by computer simulations that the proposed scheme improves mean square error (MSE) performance under given environments.