Fast Beam Training In Mmwave Multiuser MIMO Systems With Finite-bit Phase Shifters

As a result of rapidly increasing data rate demands,the sub-6 GHz spectrum is becoming very crowded.Millimeter wave(mmWave)communications operating in the super high frequency(SHF)and extremely high frequency(EHF)bands has been deemed to be a promising technology for future wireless communications due to the enormous available bandwidths.For example,mmWave can be used for multi-Gbits/s data transmission in both wireless backhaul and access networks as part of 5G systems.Nevertheless,mmWave communications has the characteristics of high path loss.Therefore,based on multi-input multi-output(MIMO)systems,beamforming technology has been widely utilized in mmWave communications.The antenna arrays employed in the BS and the MS would form some beam pairs which would increase the receive Signal-to-Noise-Ratio(SNR).Hybrid analog/digital precoding has been widely utilized in mmWave MIMO systems to relieve the severe power consumption of gigasample/s mixed-signal devices.With conventional exhaustive beam search,however,the beams formed by large antenna arrays can lead to a large time overhead at beam training stage.In this paper,we propose a novel multiuser downlink beam training strategy to estimate the angle of arrival(AoA)and the angle of departure(AoD)between the BS and the MSs.The proposed algorithm realizes much reduction of beam training overhead and releases more time resources for users' data transmission.Firstly,We propose a fast beam training algorithm for mmWave downlink multiuser MIMO systems with an extended orthogonal analog codebook inspired by previous works,which is constituted by finite-bit analog phase shifters(APS)with constant amplitude.We focus on reducing the total training overhead when the BS communicates simultaneously with M MSs.The algorithm proves to be much faster in finding the optimum beam pairs for all users compared with the exhaustive search method.To reduce the spectral efficiency(SE)loss caused by quantization error,we further improve the accuracy of AoA/AoD estimation with a very low additional overhead.The improved algorithm is able to realize approximately the same spectral efficiency compared with the previous works,where the analog precoders are constructed by accurate AoA/AoD values.Secondly,We also present a method to train all the paths in mmWave channel rather than just for a dominant path.Considering the sparse multi-path characteristic of mmWave channel,we propose an algorithm to train for several paths for all users rather than just a dominant path.The existence of multi-path interference would reduce the success rate compared with the single-path beam training.Our paper reduces the multi-path interference by rotating basis angles.The numerical results show that the proposed beam search algorithm can realize a better performance compared with the existing alternatives yet with lower time overhead,while the multi-path beam training algorithm can reach a more satisfactory success rate.