Massive MIMO System Low-complexity Detection Algorithm And Implementation

With the development of 5G,massive multiple-input multiple-output(Massive MIMO)system has drawn more and more attention.Massive MIMO system has hundreds of antennas at the base stations serving for multi-users,thus it has high spectral efficiency and high data rates.However,with the dimension of massive MIMO system going large,the complexity in signal processing,especially in signal detection,increases exponentially,which decreases the throughput and makes it hard to implement.Moreover,the existing MIMO detection algorithms need to pass accurate information during the calculation,which further increase the difficulty of the hardware implementation.A low-complexity signal detection approach based on Kaczmarz algorithm is proposed to iteratively realize MMSE detection for uplink massive MIMO systems.By applying an equivalent augmented matrix in MMSE detection,the proposed algorithm avoids the exact matrix inversion and Gram matrix computation.Furthermore,we modify the iterative form of Kaczmarz algorithm to iterate according to the characteristics of massive MIMO channel matrix,and reduce the complexity of iterative computation.Moreover,promising initial estimation and an approximate method to compute soft-output information is utilized to further reduce the complexity.Simulation results demonstrate that the proposed approach outperforms recently proposed Neumann Series,Conjugate Gradient,and Gauss-Seidel algorithms in complexity and performance.The proposed algorithm approaches the optimal solution within three iterations,while its computational complexity has 20% advantage than other methods.Meanwhile,the FPGA implementation results verify that our proposed method can efficiently compute the approximate inverse with 20% lower resource consumption.In order to further reduce the overall computational complexity,we propose to use stochastic computing unit to complete matrix multiplication,aiming at the shortcomings of transmitting accurate information in massive MIMO detection.We use deterministic sequence calculating and dual part calculating in basic stochastic computing elements,basing on which we propose stochastic matrix multiplier that performs all calculating in stochastic aspects.The simulation results demonstrate that the probability matrix multiplier can further reduce DSP resource consumption with only 0.1dB of performance loss.