Research On Decentralized Architecture For Massive MIMO System

Massive multiple input multiple output(MIMO)is one of the key technologies to achieve high frequency spectral efficiency and high reliability in 5G communication system.It improves the spatial resolution and spectral efficiency by configuring hundreds of antenna elements at the base station(BS),so that the BS can serve more users at the same time.However,linear data-detection/ precoding algorithms in traditional massive MIMO systems typically rely on centralized architecture,which results in excessive raw data interaction.Thus,the central processing units suffer from high inter-connection data rates and computational complexity.Moreover,to support a variety of antenna scale scenarios,the BS based on centralized baseband processing architecture is required to design different connection schemes according to different connection forms,which introduces additional cost and lead to poor scalability.Therefore,centralized baseband processing may become the bottleneck of restricting the application of larger scale antennas.To cope with the challenges in centralized architecture,we aim to design algorithms for both uplink detection and downlink precoding based on decentralized daisy-chain architecture in the massive MIMO system.Firstly,we investigate the uplink/downlink massive MIMO system based on traditional centralized architecture.The thesis first introduces traditional uplink/downlink system model for Massive MIMO,and then derives the expressions of three classical linear equalization/precoding algorithms.Further,we conduct MATLAB experiments to compare their bit error rate.Finally,taking 5G NR system as an example,It's confirmed that the sampling rate required by base station with centralized architecture can not afford as antenna scales.In order to solve the problem of high traffic bandwidth,distributed architectures are proposed in this thesis.Secondly,the decentralized unidirectional daisy-chain equalization algorithms in the uplink massive MIMO system is studied.According to way that information exchanges between modules,we propose unidirectional-chain equalization algorithm based on Gauss-Seidel recursive algorithm.Further,the amount of information exchange,computational complexity and transmission latency are compared between the centralized equalization algorithm and unidirectional-chain equalization algorithm in both formulation phase and filtering phase.Simulation results show that the BER curves of unidirectional chain equalization algorithm approximates to the centralized one.Thirdly,the thesis studies decentralized bidirectional daisy-chain equalization algorithms in the uplink massive MIMO system.Since bidirectional-chain equalization algorithm under unconstrained condition in uplink decentralized daisy-chain architecture is non-convergent,proximal modification and time-average modification are adopted to eliminate the curve fluctuation.The mathematical expressions of the modification algorithms are derived in a much detail.In addition,computational complexity,traffic bandwidth and latency are compared between these two modification algorithms.Simulation results have show that the two modified equalization algorithms can effectively suppress the oscillation caused by parallel iteration under unconstrained conditions,and obtain BER performance equivalent to the unidirectional-chain equalization algorithm with a lower delay.Finally,this thesis also gives the solution of precoding algorithm for decentralized daisy-chain architecture in downlink massive MIMO systems.The analytical formulas of centralized precoding algorithm,unidirectional-chain precoding algorithm and bidirectional-chain precoding algorithm are derived.As for bidirectional-chain precoding algorithm based on Jacobi recursive method,the non-convergence still occurs when the optimize functions are not strongly convex,therefore,we employ the same two modifications as uplink does to ensure their convergence.Simulation results show that all the proposed precoding algorithms approximately approach to the centralized one with lower I/O bandwidth requirements.It's verified that our proposed daisy-chain architecture and corresponding algorithms provide solid support for the practical deployment of massive MIMO system.