In this thesis we devise a novel iterative orthogonal frequency-division
multiplexing receiver algorithm that exploits the assumption that the wireless
channel is sparse, i.e. its impulse response consists of a few dominant
components. The task of the sparse channel estimation is posed as a sparse
signal estimation problem. Using the approach of sparse Bayesian learning with
hierarchical prior modeling the channel estimation problem is integrated with
the other receiver tasks through a factor graph representation of the whole
system. The iterative algorithm for joint channel estimation and decoding
is thus analytically derived by applying the combined belief propagation and
mean field inference framework as message-passing on this factor graph.

Our numerical results show that the proposed algorithm outperforms, in terms of
bit-error-rate, an analogous receiver that uses a robust channel assumption, but
does not exploit the sparsity of the channel. As
the channel estimation part is of high computational complexity we propose two
different methods for reducing the complexity:
1) heuristic modification of the message-computation scheduling such that only
part of the soft data information is used in channel estimation 2) grouping
channel variables of the factor graph into vectors of a certain size.