Novel Approach to Design Time-Domain Training Sequence for Accurate Sparse Channel Estimation

Nowadays, orthogonal frequency division multiplexing system plays a more and more important role in telecommunication systems where the training sequence (TS) is usually adopted for synchronization and channel estimation, such as in the digital television/terrestrial multimedia broadcasting system. However, to achieve a channel estimation scheme with both high accuracy and spectrum efficiency is still challenging due to noise interference and delay spread of the propagation channel. In this paper, by applying the compressive sensing (CS) theory into the sparse channel estimation process for time-domain TS, a thorough investigation on the TS design criteria is carried out. Three criteria to optimize the TS design, which are to minimize the hyper-factors for coherence, the cumulative coherence, and the coherence variance, respectively, are proposed to improve the recovery performance. To minimize the corresponding merit factors of the proposed criteria, we first investigate a CS-based inverse discrete Fourier transform pattern of TS with cyclic structure, and then a genetic algorithm is proposed to further lower the merit factors. The simulation results show that by using the proposed optimized TSs, the channel estimation performance outperforms those obtained by either conventional pseudo-random noise sequence or brute force searching sequence in correct recovery probability, mean square error, and bit error rate. Moreover, the proposed criteria II and III have better performance than criterion I, while criterion III has the lowest computational complexity and is the most suitable for application.