Sparse Channel Estimation in OFDM System

OFDM is widely used in wireless communications due to its capacity of high data rate transmission over frequency selective channel. For coherent detection of OFDM symbols, channel frequency responses must be estimated, which is usually done with the help of pilot tones. Frequency domain estimation methods are often employed with interpolation methods. Usually, interpolation methods introduce an error floor. The easy solution of increasing the number of pilots decreases the spectral efficiency of the system. Another solution is to work in the time domain. In general, the channel impulse response contains a limited number of significant values having more energy than the noise. In the case of sparse channel, this number is much smaller than that of pilot subcarriers. To improve the sparse channel estimation, some kind of threshold is needed. In this thesis, in the case where the number of pilots is larger than the length of cyclic prefix, a time domain sparse channel estimation method based on an original threshold is proposed. This method with high spectral efficiency, good channel estimation performance, low computational complexity, requires no prior knowledge of both the channel statistics and noise standard deviation. In the case where the channel is sparse with large delay spread, we propose an original channel estimation technique based on compressed sensing theory. The proposed method requires smaller number of pilots than that of classical frequency domain techniques. This work ends with the study of non-sample spaced sparse channel; the idea of smart measurement matrix is proposed to improve the efficiency of the classical CS based estimation methods.