Bandwidth efficiency is a critical concern in wireless communications. To fully utilize the available bandwidth, the superimposed training (ST) scheme is adopted in this thesis for orthogonal frequency division multiplexing (OFDM) systems without using the cyclic prefix (CP) and the guard interval (GI). It is shown that the performance of the channel estimation using the ST scheme is the same for both the proposed architecture, denoted as OFDM-ST, and the conventional OFDM system with CP, denoted as CP-OFDM-ST. In addition, since the CP is not added in the proposed system, leading to substantial increase in both the inter-symbol interference (ISI) and the inter-carrier interference (ICI), an interference cancellation scheme is derived. To further improve the performance of channel estimation using ST scheme, the joint ML data detection and channel estimation method is investigated. The simulation results illustrate that the proposed algorithm enhances the systems performance significantly. Finally, it is demonstrated that the proposed structure has a much better effective data rate than the CP-OFDM-ST system.