Probabilistic quality-of-service constrained robust transceiver designin multiple antenna systems

Abstract

In downlink multi-user multiple-input multiple-output (MU-MIMO)

systems, different users, even multiple data streams serving one user,

might require different quality-of-services (QoS). The transceiver should

allocate resources to different users aiming at satisfying their QoS

requirements. In order to design the optimal transceiver, channel

state information is necessary. In practice, channel state information

has to to be estimated, and estimation error is unavoidable. Therefore,

robust transceiver design, which takes the channel estimation

uncertainty into consideration, is important. For the previous robust

transceiver designs, bounded estimation errors or Gaussian estimation

errors were assumed. However, if there exists unknown distributed interference,

the distribution of the channel estimation error cannot be

modeled accurately a priori. Therefore, in this thesis, we investigate

the robust transceiver design problem in downlink MU-MIMO system

under probabilistic QoS constraints with arbitrary distributed channel

estimation error.

To tackle the probabilistic QoS constraints under arbitrary distributed

channel estimation error, the transceiver design problem is expressed

in terms of worst-case probabilistic constraints. Two methods are

then proposed to solve the worst-case problem. Firstly, the Chebyshev

inequality based method is proposed. After the worst-case probabilistic

constraint is approximated by the Chebyshev inequality, an

iteration between two convex subproblems is proposed to solve the

approximated problem. The convergence of the iterative method is

proved, the implementation issues and the computational complexity

are discussed.

Secondly, in order to solve the worst-case probabilistic constraint more

accurately, a novel duality method is proposed. After a series of reformulations

based on duality and S-Lemma, the worst-case statistically

constrained problem is transformed into a deterministic finite

constrained problem, with strong duality guaranteed. The resulting

problem is then solved by a convergence-guaranteed iteration between

two subproblems. Although one of the subproblems is still nonconvex,

it can be solved by a tight semidefinite relaxation (SDR).

Simulation results show that, compared to the non-robust method, the

QoS requirement is satisfied by both proposed algorithms. Furthermore,

among the two proposed methods, the duality method shows a

superior performance in transmit power, while the Chebyshev method

demonstrates a lower computational complexity.