| The combination of massive multiple-input multiple-output(MIMO)and orthogonal frequency division multiplexing(OFDM),namely massive MIMO-OFDM technology,is capable of providing extremely high spectral efficiency and high-rate data transmission,thereby has been adopted as the key physical-layer transmission architecture of the fifth-generation mobile communication technology(5G).Nevertheless,during uplink and downlink transmissions,signal detection and precoding involve large-scale matrix or vector operations,and the acquisition of accurate channel information consumes a large amount of pilot overhead.As the system dimensions increase,the complexity of link transmission becomes unacceptable.Although the sparse features in massive MIMO systems provide new perspectives for designing link transmission techniques,the problem of high computational complexity during signal detection and precoding,as well as the problem of high pilot overhead for the time-varying channel estimation still needs further investigations.To this end,this dissertation takes full advantage of sparse features in the massive MIMO-OFDM system to design low-complexity link transmission techniques,which aim at reducing the computational complexity during link transmissions and the training overhead in the channel estimation stage.The main contributions are summarized as follows.(1)To reduce the computational complexity of linear precoding during the downlink transmission,we propose a low-complexity beamspace precoding algorithm based on the precondition-based conjugate gradient method.By modeling the linear precoding problem in the beam domain,the sparsity of beam-domain channels can be used to decompose the matrix-vector multiplication into multiplications of sparse matrices and vectors during iterations.Thus the proposed algorithm does not require large-scale matrix computations,which leads to extremely low computational complexity and delay.Based on the asymptotic orthogonality of massive MIMO channels,a low-complexity preconditioner is proposed to improve the algorithm convergence.Moreover,a beam selection strategy can be utilized to balance the performance and complexity of the beamspace precoding scheme.Thus,we first analyze the relationship between the beam width and beam selection error given the beam centers of multipath components,and then design a beam selection strategy for minimizing the upper bound of the normalized mean squared error.The simulation results under a single-cell system setup verify the superior performance of the proposed precoding algorithm and beam selection strategy.(2)To reduce the computational complexity of nonlinear signal detection during the uplink transmission,we propose a low-complexity groupwise signal detection scheme based on the Variational Bayes framework.Motivated by the sparsity of channel correlations,the proposed groupwise signal detection scheme assumes a factorized approximation on the posterior distribution of transmitted signals,which significantly reduces the dimension of nonlinear detection.Then the groupwise detection algorithm and user grouping algorithm are jointly designed by minimizing the Kullback-Leibler divergence between the posterior distribution and its factorized approximation.By ignoring the finite-alphabet constraint of the transmitted signal,the intractable optimization problem of user grouping is transformed into a tractable one that depends on the channel correlation.And it can be solved by an increment greedy-based method.The complexity analysis shows that the proposed groupwise detection algorithm does not need large-scale matrix operations during iterations,and user grouping can be performed on a wideband basis.In addition,the probabilities of high-correlated channels are derived,and then the random graph theory is used to analyze the maximum group size in massive MIMO systems.Theoretical results demonstrate that the maximum group size grows logarithmically with the number of users in the worst case of line-of-sight propagation,thereby the complexity of nonlinear groupwise detection is affordable in massive MIMO.The simulation results show that the proposed groupwise detection scheme achieves higher detection efficiency under both Rayleigh fading and standard channel models.(3)To reduce the training overhead of time-varying channel estimation,we propose a slow-down channel estimation scheme based on the joint beam-delay-Dopplerdomain sparsity of massive MIMO-OFDM channels.The proposed channel estimation scheme assumes the beam-delay domain channel follows the single-tone timevarying model,thereby decoupling the time-varying channel estimation problem into subproblems of Doppler parameter estimation and slow-down channel tracking.Since Doppler parameters can be viewed as constant over a long period of time,tracking the slow-down channel can significantly reduce the training overhead.To address the coupling issue between unknown variables and parameters in the Doppler parameter estimation stage,the expectation maximization(EM)algorithm is adopted for the joint estimation of Doppler parameters and slow-down channels.Specifically,in the E-step,based on the pattern-coupled Bayesian prior,a low-complexity variational Bayes inference algorithm is proposed to estimate the slow-down channel coefficients.In the M-step,an optimization approach based on Riemannian manifold optimization and single-tone frequency estimation is developed to solve the non-convex optimization problem for Doppler frequencies.The simulation results based on standard channel models verify that the proposed slowdown channel estimation scheme can achieve the same estimation performance as the existing methods with far less pilot overhead. |
