Robust Cognitive Beamforming Based On Convex Optimization

The increasing of wireless communication demand has made the nonrenewablefrequency more valuable. In this case, cognitive radio technology which can effectivelytake advantage of idle spectrum receive widespread attention. The core idea is to makethe devices have the ability of discover spectrum holes, then in an opportunistic way,rationally use the unoccupied frequency or the frequency only with a small business.This borrowing way achieve the reutilization of radio spectrum, thus effectivelyimprove the spectrum efficiency.Multiple Input Multiple Output which takes multiple antennas at both thetransmitter and receiver for wireless data transmission. Without the need for additionalbandwidth and transmission power, MIMO can significantly improve systemperformance. Multiple antennas can allocate transmitting dimensions in space.Consequently, take MIMO into cognitive radio network, can provide additional spacefreedom to the secondary user except in time and frequency, as well as control theinference between the secondary user and primary user.This dissertation considers MIMO cognitive radio network, studies cognitivebeamforming in the downlink. As the channel state information estimated at thetransmitter could never be perfect, we focus on the robust beamforming. The Originaloptimization problems are almost non-convex, and solve the problem directly is difficultwith a large complexity. So it is necessary to convert the non-convex into convex in thefirst step. At first, we study the SINR balancing problem bounded in an ellipsoid,through S-procedure combined with rank relaxation, we get a quasi-convex problem,then we use bisection to solve the beamforming vectors, at the same time,we makesimulation analysis about robust and non-robust algorithms. Then we focus on the totalpower minimization problem, take S-procedure to limit the constraints, then weintroduce a new method named as convex iteration to solve to get the optimalbeamforming vector. Simulation results show that the proposed method compared to theinequality relaxation randomized algorithm has better performance.