In this thesis, we first compare the least squares (LS) channel estimation performance in terms of mean squared error (MSE) with the perfect knowledge of channel length L versus the estimation without the knowledge of L. This comparison motivates us to estimate the channel in two steps by exploiting the instantaneous delay spread. We show that the channel estimation performance can be improved by 5 dB even in the presence of channel mismatch. Then we propose the optimal training sequence which achieves not only the minimum MSE of LS channel estimation, but also the minimum peak-to-average power ratio (PAPR) at transmitter with a low computational complexity. Finally we describe all the necessary synchronization functions and FPGA implementation in the SISO OFDM receiver. And this receiver serves as the foundation to build up a wireless testbed to verify the new algorithms.