Research On Algorithms And Architectures For Fast Beamforming With Millimeter Wave Massive Antenna Arrays

As a key technology for the 5G and B5 G wireless communication systems,the massive arrays operating on mm Wave frequency bands have been put into commercial use,where the number of antennas has been increased to hundreds even thousands.Apart from broadening the frequency resources,the mm Wave massive array can produce beams with high directivity to combat the high path loss of mm Wave frequency bands,enabling the ultra-reliable,low-latency,broad band communications.With the increase in number of antennas,the power consumption of the beamforming architectures also increase,making the conventional fully digital beamforming architectures hardly practical,hence the hybrid beamforming architectures where digital precoding and analog precoding work jointly,are of more interest.Due to the large dimensions of the beamforming problems,conventional methods which depend on perfect channel state information suffer from a high computational complexity of exponential growth.This is detrimental to system latencies and has inspired the beamforming technologies based on predefined codebooks.Therefore,it is essential to study the fast and low complexity beamforming technologies for mm Wave massive arrays with hundreds to thousands of antennas.This dissertation focuses on the fast beamforming technologies for mm Wave massive arrays,and aims to address the issues of high computational complexity,feedback overhead and power consumption.Considering the adaptive and codebook beamforming schemes,both algorithms and hardware architectures are studied.The core technical requirement is lower computational complexity,feedback overhead,power consumption and high spectrum efficiency during the study,and the impacts brought by the channel state information and hardware components are analyzed and corresponding algorithms and architectures are proposed in order to establish the solutions to issues in terms of ultrareliable,low-latency and broad band communications.The research in this dissertation is summarized in the following aspects:Firtstly,the optimization of computational complexity for the adaptive beamforming schemes is studied.Considering that adaptive beamforming schemes require complete channel state information to perform matrix decomposition and calculate the optimal precoding matrices,which are of high complexity for massive arrays,the partially and dynamically connected hybrid beamforming architectures are studied.The corresponding beamforming problem for the former is then converted to independent subproblems that can be solved in parallel at the same time,while the one for the latter is converted to a greedy search problem,according to the relation between RF chains and connected subarrays.The sum of complexities of all subproblems is much smaller than the original problem.Furthermore,the subproblems can be solved in a parallel manner,saving more computation time.Secondly,the fast codebook beamforming for multiuser scenario is studied.The limited resolution of the phase shifters(PSs)and the potential mis-selection for codewords when the angles of arrival(Ao As)or angles of departure(Ao Ds)fall within the boundaries of two adjacent beams are both taken into consideration for FDD systems,which lead to a primary-auxiliary codebook design.The auxiliary codebook exploits the maximum PS resolution and compensate for the potential mis-selection while the primary codebook supports the hierarchical beam training which is of low complexity.To further reduce the feedback between the base station and mobile stations,a synchronized broadcasting beam training scheme is then proposed,during which the codewords that no mobile station selects are ignored for the next layer.Furthermore,when the number of RF chains is larger than the number of mobile stations,the multipath beamforming scheme for different distributions of propagation paths is proposed,which exploits the other paths on top of the dominant path for more diversity gain.Thirdly,the partial reciprocity aided fast codebook beamforming approach is studied.Considering the large amount of feedback overheads between the base station and mobile stations,the partial reciprocity between uplink and downlink channels is employed to avoid feedback overheads.As the dominant uplink path may not be the same as dominant link for the downlink channel,the discrete Fourier transform of the channel is applied to perform a multipath extraction and reconstruction scheme.The proposed scheme reconstructs the single path channels which can act as the training data set to establish a quick mapping between the channels and correspond optimal codewords using machine learning tools.Fourthly,the fast beamforming under combined PS resolutions is studied.In order to address the non-convex constraints brought by conventional single PS architecture and the high power consumption of the recent double PS architecture,a novel architecture where for one RF chain one high resolution PS is employed in parallel to multiple low resolution PSs,is proposed.The proposed combined architecture performs closely to a double PS architecture and greatly outperforms the conventional single PS architecture.Based on the proposed novel combined PS architecture,an alternate codebook design for flatter main beam gain and sharp cutting edge is proposed with low complexity.Finally,an energy efficient fast beamforming strategy for asymmetric uplinkdownlink traffic is studied.Considering the asymmetric uplink-downlink traffic for TDD systems,a PS deactivation strategy for user-wise adaption to asymmetric traffic is proposed,which meets the performance demands for the link with lighter traffic.The proposed strategy overcomes the disadvantage of the dynamic TDD scheme that changes the ratio of uplink-downlink time slots of all users(i.e.treating all the users identically),and is able to achieve a user-wise adaption.A sequential algorithm of low complexity,which deactivates one PS for each iteration,is then proposed.In order to support the assumptions of the proposed sequential algorithm,an alternate codebook design is also proposed.Different from conventional methods that treat the average main beam gain as a constant,the proposed design updates the average main beam gain,leading to flatter main beam and faster convergence.During the iterations some mediate results are also generated which can compensate for the broadened beams after PS deactivation.