Iterative estimation of sparse and doubly-selective multi-input multi-output (MIMO) channel

The estimation of doubly-selective channels is challenging since long channel impulse response should be estimated with a fast tracking speed. Provided that a structure of the channel response is sparse, i.e., only a few of channel gains are nonzero, tracking performance of the channel estimator can be improved significantly by avoiding estimation of zero taps. In this paper, we study estimation of fast time-varying channels that have a sparse structure in multi-input multi-output (MIMO) systems. In order to exploit the sparse structure, we parameterize locations of nonzero channel taps using a deterministic binary vector and incorporate it into the state-space form built upon auto-regressive (AR) time-varying channel model. Then, we derive a joint estimate of the binary vector and channel gains based on maximum likelihood (ML) criterion. Expectation maximization (EM) algorithm is derived to find a sparse structure and channel gains iteratively. According to the simulation study performed over MIMO Rician fading channels, the proposed sparse channel estimator outperforms the previous channel estimation schemes, especially when Doppler rate is high.