Joint Channel Estimation and Detection for Multi-Carrier MIMO Communications

In MIMO OFDM systems, channel estimation and detection are very important. Pilot-based channel estimation using BEMs is widely used for approximating time-frequency variations
of doubly-selective channels. BEMs can provide high estimation performance with low computational load. Data-aided channel estimation outperforms the pilot-based estimation. The data-aided estimation iteratively improves estimates using tentative data symbols and corresponding
adaptive weights (reweighted channel estimation). These weights are computed assuming Gaussian data errors, which is inapplicable to OFDM. In this thesis, this assumption
is however shown to improve the channel estimation performance. The reweighted channel estimation is shown to significantly outperform the unweighted estimation. Most often used mismatched receivers assume perfect channel estimates when detecting data symbols. However, due to limited pilot symbols and data errors, the channel estimates are imperfect, resulting in a degraded detection performance. The optimal receiver without explicit channel
estimation significantly outperform mismatched receivers. However, its complexity is high. To reduce the complexity, a receiver that combines mismatched and optimal detection is proposed. The optimal detection is only applied to data symbols unreliably detected by the mismatched detector, identified using weights computed in the reweighted estimator. The channel estimator and the optimal receiver require the knowledge of channel statistics, which
are unavailable and difficult to acquire. To overcome this, an adaptive regularization using the cross-validation criterion is introduced, which finds a regularization matrix providing best channel estimates. The proposed receiver has a reduced complexity than the optimal receiver and provides close-to-optimal detection performance without the knowledge of channel PDP. The adaptive regularization is extended to joint estimation of the Doppler-delay spread and channel. The Doppler and delay spread corresponding to the optimal regularization are selected as their estimates. This approach outperforms other known techniques and provides channel estimation performance close to that obtained with perfect channel statistics.