Research On Massive MIMO Signal Detection Algorithm

In recent years,the number of terminals explosively grows with the development of mobile internet.The higher demand for wireless data service has been a continuous process.The concept of Massive MIMO has been put forward as one of the 5G's key technologies.Massive MIMO has not only the great gain of diversity and multiplexing,but also the significant improvement of channel's capacity and spectral efficiency.Among many important technologies of Massive MIMO,signal detection at the receiving terminal of the system is crucial.However,traditional signal detection algorithms are difficult in finding the reasonable tradeoff between complexity and performance.In this thesis,efficient algorithms are tried to be found through studying and analyzing different Massive MIMO signal detectection algortihms.The main research contents can be summarized as follows:For the aspect of traditional signal detection algorithms applied in Massive MIMO system,the algorithms are introduced in detail by separating them into tree-search algorithms and non-tree-search algorithms.The former one is mainly studied.The improved depth-first-search sphere decoder(IDFS)is put forward to resolve redundant search problem of traditional tree-search algorithm.The performance of signal detection algorithms is simulated in the system of the fixed and different number of antennas.Simulation results show that IDFS can achieve almost the same performance as maximum likelihood detection with lower complex than fixed complexity sphere decoding algorithm(FSD).FSD and K-best have the characteristic of Massive MIMO which means that the performance improves with the growing number of antennas.For the situation that likelihood ascent search(LAS)algorithms are easily trapped into local optimum,the improved multiple search candidates LAS(IMSCS_LAS)is proposed by combining multiple initial vectors LAS(MIV_LAS)and MSCS LAS to solve the problem.The proposed IMSCS_LAS algorithm is inspired by the ideology of FSD.In each iteration,it only reserves the fixed number of survival vectors.Then the proposed algorithm is compared with the class algorithms of tabu search(TS).Simulation results show that LAS and TS have good characteristic of Massive MIMO.SISO has only 0.6dB signal-to-noise ratio(SNR)gain more than LAS and 0.4dB SNR gain more than TS to achieve the error bit rate(BER)of 10-4 when the number of antenna reaches 128.The performance of proposed IMSCS_LAS algorithm is close to the performance of FSD.The complexity and needed SNR of IMSCS_LAS reaching BER of 10-2 are less than other search algorithms when 16QAM modulation is adopted.Considering the stalling problem which exits in the most Markov Chain Monte Carlo(MCMC)detection algorithms based on Gibbs sampling,mixed Gibbs sampling MCMC(MGS-MCMC)and MGS-MCMC with restart(MGS-MCMCR)algorithms are introduced in detail to solve the problem efficiently.In order to solve the low complexity of MGS-MCMC in high QAM modulation,the random parallel MGS-MCMC(RMGS-MCMC)based on multiple Markov Chain is proposed to solve this chanllenge.The performance of MCMC algorithms has been simulated.Simulation results show that MGS-MCMC has good characteristic of Massive MIMO.SISO has only 0.6dB SNR gain than MGS-MCMC to achieve the BER of 10-4 when the number of antenna reaches 64.Proposed RMGS-MCMC can achieve the performance of FSD.It can be easily implemented by hardware.In conclusion,this thesis studies the Massive MIMO signal detection algorithm in the receiving end from the above three aspects.Heuristic and MCMC methods are mainly studied.The improved algorithms based on existing ones are proposed.The performance and complexity of the proposed algorithms are also simulated.