Channel estimation for wideband millimeter wave systems with hybrid architecture

Abstract

Millimeter wave communications have been considered a promising candidate technology of 5th generation communications. To compensate for the high pathloss over millimeter wave bands, millimeter wave systems exploit beamforming gain via large antenna arrays. A popular approach to antenna array processing for millimeter wave systems is to use a hybrid multiple-input multiple-output (MIMO) system consisting of an analog RF beamformer and a digital baseband MIMO processor. Designing an efficient hybrid MIMO system requires accurate channel state information (CSI). In this dissertation, we propose efficient channel estimation techniques for millimeter wave hybrid MIMO systems.

In the first part of this dissertation, we consider the problem of time domain channel estimation for wideband millimeter wave systems with a hybrid architecture. Time domain channel estimators can exploit both angular and delay domain sparsity, and can perform better than frequency domain estimators exploiting only angular domain sparsity

however, the former usually require a heavier computational load than the latter. To overcome this difficulty, we propose an efficient, two-step time-domain channel estimator: in the first step, an effective channel incorporated with beamforming, array responses, and pulse shaping is estimated by the least squares (LS) method

in the second, the desired channel is estimated via an orthogonal matching pursuit (OMP) algorithm. It is analytically shown that the proposed two-step method can yield the same outputs as the original single-step time domain estimator when an identity pulse shaping matrix and a unitary pilot matrix are employed. The proposed algorithm is simpler to implement than the other estimators based on OMP because its sensing matrix is block-diagonal. Simulation results demonstrate that complexity reduction of the proposed method is achieved without (or with minor) performance degradation.

In the second part of this dissertation, we extend the proposed two-step channel estimators to multi-user wideband MIMO systems. Since the compressive sensing (CS) problem for multi-user uplink channel estimation is the same as the one for single-user channel estimation, the T-LS/OMP can also be applied to multi-user systems. We can modify the T-LS/OMP by replacing OMP with the other CS technique. Among the various CS techniques, we propose a two-step approach of weighted-OMP (WOMP) algorithms, called T-LS/WOMP, and study its characteristics. It is analytically shown that, when an identity pulse shaping matrix is employed, the proposed T-LS/WOMP can yield the same outputs as the original single-step WOMP. The computational complexity of the proposed T-LS/WOMP can be significantly reduced when the use of a unitary pilot matrix is assumed. Simulation results show the advantages of the proposed two-step estimators.

In the third part of this dissertation, we consider an autoencoder (AE) with a linear encoder and a nonlinear decoder, which correspond to a training beamformer and a channel estimator, respectively. In the proposed AE, the linear encoder acts as analog beamforming matrices with phased arrays and the nonlinear decoder, which operates at various values of signal-to-noise ratio (SNR), consists of two separate decoders for high/low SNRs. Simulation results show that the proposed methods can outperform conventional OMP algorithms with random beamformers.