Abstract:The channel estimation accuracy could be improved significantly by sparse channel estimation methods with a priori sparsity,that is,a small number of significant paths and large time delays in wireless channels.In view of non-optimal selection of atoms when minimizing residual norm error in greedy algorithms and the performance degradation of generalized orthogonal matching pursuit (gOMP) caused by selection of excessive atoms,an algorithm named optimized generalized orthogonal matching pursuit (OgOMP) was proposed.In the procedure of selecting atoms in OgOMP,the criterion of selecting an atom which had the maximal absolute inner product with residual was substituted by a new criterion which selected an atom with minimum residual norm error.In order to delete wrongly chosen atoms,an atom refine procedure which selected the atoms with maximal channel fading factors was added to make a second choice for selected atoms at each iteration.The number of selected atoms was the same as the channel sparsity and then wrongly chosen atoms were deleted for guaranteeing the sparsity consistency between reconstructed signals and original signals.Finally,the performance of gOMP and OgOMP were compared in terms of the channel estimation mean square error,the bit error rate, the convergence of residual with different pilot numbers and selected atom numbers.It was shown that OgOMP could save 4 dB signal to noise ratio at same bit error rate and gain maximum 5 dB at 20 dB signal to noise ratio compared with gOMP;The convergences of residual of gOMP and OgOMP were superior to MP;More pilots could improve channel estimation mean square error in both two algorithms and OgOMP needed fewer pilot overhead with the same channel estimation performance;When different atom numbers were used at each iteration the performance of OgOMP was almost the same and it was more stable compared with gOMP.Experiments verified the effectiveness of OgOMP.