On Parameter Estimation Over MIMO Channels

Multi-input and multi-output (MIMO) technology is an important means to im-prove the performance of high-speed wireless communications, such as multi-mediacommunications and wireless Internet access. It is widely understood that over multi-path fading channels, a MIMO system has much higher spectral efficiency than aconventional single-input and single-output (SISO) system. MIMO technology offers avariety of research fields, including the channel capacity, parameter estimation, signaldetection, and so on. This dissertation deals with the problem of parameter estimationover MIMO channels, the main contributions are as follows.1) The problem of periodic training sequence design for joint channel andfrequency-offset estimation is considered. The design criterion for periodic training se-quences is to minimize jointly the mean square error (MSE) of the maximum-likelihood(ML) channel estimation and the asymptotic Cramér-Rao Lower Bound (CRLB) forfrequency-offset estimation. It is shown that all the training sequences minimizing theMSE of channel estimation have the same asymptotic CRLB for frequency-offset es-timation. Moreover, they also minimize the asymptotic CRLB as long as the channelis independent Rayleigh fading with any power-delay profile. We also consider thedesign of low-complexity frequency-offset estimators based on the proposed periodictraining sequence and show that the correlation-based frequency-offset estimator isa sub-optimal estimator; however, it allows a flexible tradeoff between the estimationrange and performance.2) Concatenated training sequences, which consist of two parts: short blocksand long blocks, are commonly used for frequency-offset estimation to achieve a goodtradeoff between the estimation range, accuracy, and complexity. In this dissertation,we use the CRLB as a metric to analyze the performance of concatenated training se-quences. It is shown that in general, concatenated training sequences for frequency-offset estimation lead to the threshold effect, which can seriously degrade the esti-mation performance. Furthermore, we show that the optimal concatenated trainingstructure, which minimizes the threshold and the CRLB and maximizes the estimationrange, is just dividing the complete sequence into the shortest blocks without longblocks.3) The problem of large carrier frequency-offset estimation is considered. Sincethe well-known ML frequency-offset estimator is rather complex, many complexity- reduced frequency-offset estimators have been proposed. However, most of them arebased on periodic training sequences and have limited estimation range. We proposea general framework called composite frequency-offset estimator (CFE), which con-sists of an existing estimator and a Bayesian/GLRT detector. The CFE can extend theestimation range of any range-limited frequency-offset estimator up to the full trans-mission spectrum. It is shown that the composite frequency-offset estimate is the MLestimate as long as the original range-limited estimator is unbiased and attains theCRLB and that the performance of the CFE is asymptotically equal to that of the orig-inal estimator. The CFE also allows a flexible tradeoff between the estimation rangeand computational complexity.4) In the general MIMO channel, the frequency-offsets between different transmit-receive pairs are different. The computational complexity of the ML estimator is time-consuming; however, existing low-complexity methods estimate the channel param-eters in a time-division mode, which is bandwidth wasting and results in the peak-average-power ratio problem. By considering a MIMO system as equivalent SISOsub-systems with multi-antenna interference, a novel iterative parameter estimationmethod with interference cancellation is proposed. The proposed method has a rela-tively low complexity and achieves good performance.