Performance Analysis of Maximum-Likelihood Semiblind Estimation of MIMO Channels

Iterative channel estimation and data detection is a useful method to improve the channel estimation quality without sacrificing the bandwidth efficiency. Since both the known training symbols (non-blind) and the unknown data symbols (blind) are used for channel estimation, corresponding techniques are referred to as semiblind. If the channel estimator and data detector are both optimal in the sense of maximum-likelihood criterion, we may call the algorithm as maximum-likelihood (ML) semiblind channel estimation (SBCE). This paper deals with ML-SBCE for frequency-flat multi-input multi-output systems with focus on the channel estimation mean squared error (MSE) analysis. Through semi-analytical efforts, we showed that ML-SBCE is biased at low SNR and tends to be unbiased at high SNR. The reasons of biasing are the erroneous data detection and the correlation between the noise and the detection errors. Besides, we showed that the MSE performance of ML-SBCE is also influenced by the noise-error correlation. Based on these analyses, possibilities to compensate the biasing as well as improve the MSE performance is pointed out