Channel Estimation in TDD and FDD-Based Massive MIMO Systems

There are three parts to this thesis. In the first part, we study the channel estimation problem in frequency-selective multi-user (MU) multi-cell massive multiple-input multiple-output (MIMO) systems, where, a time-domain semi-blind channel estimation technique is proposed. Compared to frequency-domain, the time-domain channel estimation requires fewer parameters be estimated. Importantly, the time-domain estimation has enough samples for an accurate channel estimate. Given this many samples in the time-domain, the proposed channel estimation technique obtains a better estimate of the channel. Here, there is no assumption on orthogonality of users' channels, knowledge of large-scale fading coefficients, and the orthogonality between the training symbols of the users in all cells. The second part of the thesis studies the channel estimation problem in correlated massive MIMO systems with a reduced number of radio-frequency (RF) chains. Leveraging the knowledge of channel correlation matrices, we propose to estimate the channel entries in its eigen-domain. Due to the limited number of RF chains, channel estimation is typically performed in multiple time slots. Using the minimum mean squared error (MMSE) criterion, the optimal precoder and combiner in each time slot are aligned to transmitter and receiver eigen-directions, respectively. Meanwhile, the optimal power allocation for each training time slots is obtained via a waterfilling-type expression. In the final part, we study the downlink channel estimation for frequency-division-duplex (FDD) massive MIMO systems. Acquiring downlink channel state information in these systems is challenging due to the large training and feedback overhead. Motivated by the partial reciprocity of uplink and downlink channels, we first estimate the frequency-independent channel parameters, i.e., the path gains, delays, angles-of-arrivals (AoAs) and angles-of-departures (AoDs), via uplink training, since these parameters are common in both uplink and downlink. Then, the frequency-specific channel parameters are estimated via downlink training using a very short training signal. To efficiently estimate the channel parameters in the uplink, the underlying distribution of the channel parameters is incorporated as a prior into our estimation algorithm. This distribution is captured using deep generative models (DGMs). The proposed channel estimation technique significantly outperforms the conventional channel estimation techniques in practical ranges of signal-to-noise ratio (SNR).Ph.D