Precoder design with statistical channel state information over MIMO Rician channels

Abstract

A precoder is a processing block that exploits channel state information (CSI) at the transmitter (CSIT) by transforming the input signal prior to transmission. By acting as a combination of the input matching matrix, power allocator, and beamformer, the precoder matches the input signal to the channel and distributes transmit power over eigen-beams. When a perfect CSIT is available, the system performance can be enhanced by dynamically adjusting transmission parameters according to CSIT. However, the acquisition of a perfect CSIT is costly and sometimes infeasible due to delayed feedback, channel estimation, and quantization errors. The bandwidth overhead for CSI feedback is also a vital concern. So the design of an optimal precoder that is based on imperfect or partial CSIT, i.e., statistical CSIT, becomes important in practice. The objective of this thesis is to investigate precoder design methods that are based on statistical CSIT for multiple-input multiple-output (MIMO) wireless communication systems. We consider two MIMO systems: point-to-point communication and cooperative relay networks. First, we propose a statistically robust precoding scheme for a point-to-point MIMO system with orthogonal space-time block codes (OSTBC) to minimize the Chernoff bound of pairwise error probability (PEP). The conventional Kronecker channel model with general transmit and receive correlation matrices is adopted. We assume that the receiver obtains instantaneous CSI, whereas only statistical CSI, i.e., the channel mean and correlation matrices, is available on the transmitter side. By defining new power control variables and exploiting the eigenmode structure of the precoder, the optimization problem becomes convex when the power allocation and beamforming matrix are iteratively solved. An iterative procedure for constructing the eigen-structure, which includes an efficient algorithm for determining the number of eigen-beams with positive power, their power allocation, and a simple scheme using Givens rotations for updating the beamforming matrix, is derived. Based on a low signal-to-noise ratio (SNR) asymptotic analysis, a unitary matrix that reduces the precoder design to a power allocation problem, which contains significantly lower computational complexity, is derived for the PEP criterion. For a given Rician K-factor, the asymptotic analysis reveals that the precoder approaches a single-beam mode that is dependent on the channel mean at a low SNR. Conversely, for high SNR, the power allocation approaches an equal power scheme. These asymptotic behaviours are affected by the constellation size. However, for a given SNR, as the value of the Rician K-factor increases, the optimal precoder tends to use fewer eigen-modes and finally becomes a single-mode scheme. This contrary effect of a large SNR and K-factor on the design is dependent on the specific channel mean and correlation. The analysis also indicates that the low-rank channel mean matrix yields better PEP performance than the high-rank channel mean matrix in the low SNR region. The numerical results demonstrate better performance of the proposed method compared with existing methods with significantly lower computational complexity. We propose a novel precoding scheme for space-time coded MIMO relay networks. We assume that the destination terminal perfectly knows the CSI, whereas only statistical CSI is available at the source and relay nodes. The majority of the existing methods use mean square error (MSE) criterion with the Wiener filter as an equalizer in the design of the precoding system. However, the optimum design derived from this approach may not provide satisfactory PEP. To directly relate the optimization to error probability in this thesis, a lower bound of an accurate approximation of the total average SNR at the destination is adopted as the performance criterion. The optimal receiver at the destination terminal is demonstrated to be a maximum ratio combiner (MRC) and the precoding matrix for relay nodes should be block diagonal with its main diagonal matrices as the precoding matrices at the relay nodes. Two system configurations-a configuration with the direct channel and a configuration without the direct channel-are examined. These configurations can be established by controlling the combined weight factors for the two received signals in two different time slots at the destination terminal: one signal is received directly from the source and another signal is received from the relays. By appropriately expanding the SNR expression, the relay precoder can be solved with a generalized Rayleigh quotient. The maximization of the SNR results in an optimal source precoder of a single-beam structure. To reduce the feedback overhead and processing burden on the relays, we also investigated the scenario in which a single antenna is used in the relay nodes. The numerical results demonstrate that the proposed method outperforms the equal power allocation scheme and existing methods. The error performance improves with the number of relays or the total antennas on relays. The bit error rate (BER) performance of a robust design for statistical CSI is similar to the perfect CSI case, in which the Rician K-factors for all channels are sufficiently high. When the Rician Kfactors of the cascade channel are high and the Rician K-factor of the direct channel is low, the cooperative network with a sufficiently large number of relays can provide adequate system performance without a direct channel as if a direct channel were deployed, which implies a reduction in feedback overhead.

Date of Award	15 Jul 2014
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Shu Hung LEUNG (Supervisor)