Multiple-input multiple-output wireless system designs with imperfect channel knowledge
Abstract
Empowered by linear precoding and decoding, a spatially multiplexed
multiple-input multiple-output (MIMO) system becomes a convenient
framework to offer high data rate, diversity and interference
management. While most of the current precoding/decoding designs
have assumed perfect channel state information (CSI) at the
receiver, and sometimes even at the transmitter, in this thesis we
design the precoder and decoder with imperfect CSI at both the
transmit and the receive sides, and investigate the joint impact of
channel estimation errors and channel correlation on system
structure and performance. The mean-square error (MSE) related
performance metrics are used as the design criteria.
We begin with the minimum total MSE precoding/decoding design for a
single-user MIMO system assuming imperfect CSI at both ends. Here
the CSI includes the channel estimate and channel correlation
information. The structures of the optimum precoder and decoder are
determined. Compared to the perfect CSI case, linear filters are
added to the transceiver structure to improve system robustness
against imperfect CSI. The effects of channel estimation error and
channel correlation are quantified by simulations.
With imperfect CSI at both ends, the exact capacity expression for a
single-user MIMO channel is difficult to obtain. Instead, a tight
capacity lower-bound is used for system design. The optimum
structure of the transmit covariance matrix for the lower-bound has
not been found in the existing literature. By transforming the
transmitter design into a joint precoding/decoding design problem,
we derive the expression of the optimum transmit covariance matrix.
The close relationship between the maximum mutual information design
and the minimum total MSE design is also discovered assuming
imperfect CSI.
For robust multiuser MIMO communications, minimum average sum MSE
transceiver (precoder-decoder pairs) design problems are formulated
for both the uplink and the downlink, assuming imperfect channel
estimation and channel correlation at the base station (BS). We
propose improved iterative algorithms based on the associated
Karush-Kuhn-Tucker (KKT) conditions. Under the assumption of
imperfect CSI, an uplink--downlink duality in average sum MSE is
proved. As an alternative for the uplink optimization, a sequential
semidefinite programming (SDP) method is proposed. Simulation
results are provided to corroborate the analysis.