Research On Adapative Beamforming Technologies For Multi-User Scenarios

The large amount of spectral resources in the millimeter wave(mm Wave)bands have drawn increased attention in the enhancement of 5G communication networks for the rapid growth of user traffic and equipment access requests.To ensure the reliability of the wireless link,mm Wave communication requires the high directivity gain provided by large-scale antenna arrays through beamforming(BF)to compensate the path-loss in the high-frequency transmission.Generally,mm Wave systems employ the hybrid beamforming(HBF)composed of the analog and digital domain for the reduction of high cost and power consumption,where the signal processing can only obtain the finite dimensional observation of the array.Therefore,hardware constraints and different mm Wave propagation characteristics compared to the lower frequencies need to be fully considered in the HBF design.This thesis develops a variety of HBF design methods based on iterative training algorithms and with the assistance of compressed sensing(CS)for the multi-user single-input multiple-output(MUSIMO)uplink mm Wave systems in the multipath environments.Firstly,the MU-SIMO parameterized uplink channel is constructed according to the mm Wave propagation characteristics,and the mathematical model of the required channel parameters is derived.Considering the hardware constraints,the HBF design methods based on the full-subarray training are proposed,where a limited number of radio frequency(RF)chains are allocated to each subarray by time-division connection during the training period,and the signals extracted from the additional reference antenna(RA)are regarded as the phase reference to form the training objective function of the analog beamforming(ABF)which is aimed for the simple gradient update under the phase-only constraint.Then,a subABF iterative training algorithm based on Nesterov Accelerated Gradient(NAG)is proposed for the optimization of phase adjustment.After all subarrays are trained,zero-forcing digital beamforming(ZF-DBF)is performed on the combining output of the trained ABF to finally achieve MU spacedivision multiple access.Simulation results verify that the method based on the full-subarray training outperforms the direct path steering scheme in the multipath environments,and the performance degradation of the proposed method is hardly observed with increasing level of scattering.To reduce the complexity and overhead of the full-subarray training,the sub-connected HBF design methods based on the stage training are further proposed.Except the ZF-DBF,the ABF training is divided into two stages.The first-stage activates a single subarray for the iterative training.Then,the obtained sub-ABF is used for the second-stage training of the virtual equivalent array on the sub-connected output.Finally,all sub-ABFs are derived from the correction vector obtained in the second-stage combined with the trained weight vector in the first-stage.Simulation results in the multipath environments indicate that the stage training has an extent loss of performances and adaptability compared with the full-subarray training,but the time consumption of the training is greatly reduced.According the sparsity of mm Wave channels,the low-dimensional observation of the array is used to directly detect the angle-of-arrival(AOA)of the direct path for each user,so as to obtain the sub-ABF of the first-stage training with fewer observations than iteration training under the stage training framework.The compressed sensing(CS)measurement matrices are constructed in two ways:stochastic and directional.To compensate the inability of the stochastic measurement to provide directivity gain,the Gaussian stochastic measurement(GM)combined with the outlier detection through median absolute deviation(MAD)is proposed to obtain the estimated AOA of the direct path.For the directional measurement based on the hierarchical codebook,continuous basis pursuit(CBP)dictionary and raised cosine with window overlapping are applied to construct the desired angledomain response to generate an ideal beam with a smoother passband.Then,the beams at each level in the codebook are designed by the sparse approximation based on the variable feasible set aimed to consider all sweeping sectors with different widths.Simulation results imply that the proposed hierarchical codebook employed for the AOA detection in the first-stage training can achieve better performances than that of the codebook with the rectangular window and fixed feasible set,and it can finally approach the two-stage iterative training as the signal-to-noise ratio(SNR)increases.