No abstract

The quasi-newton methods are one of the most effective methods for solving unconstrained optimization problems. This letter provides a hardware-efficient massive multiple-input multiple-output (MIMO) detector using an improved quasi-newton (IQN) method. Due to the similarity in the stepsize calculation of Barzilai-Borwein and limitedmemory BFGS, two quasi-newton methods are deeply fused in the proposed IQN algorithm for higher convergence speed. The corresponding efficient detector architecture is also given, in which a dual-track systolic array architecture is employed to diminish the number of required Processing Elements (PE) by nearly half with less computational delay. Furthermore, an approximate divider based on the Goldschmidt method is designed to further reduce hardware overhead. Simulation results show that the proposed IQN algorithm achieves better Bit-Error-Ratio (BER) performance under different antenna configurations, and FPGA implementation results also validate the superiority of the proposed detector in terms of hardware efficiency over the state-of-the-art (SOTA) detectors.

This letter provides an efficient massive multiple-input multiple-output (MIMO) detector based on quasi-newton methods to speed up the convergence performance under realistic scenarios, such as high user load and spatially correlated channels. The proposed method leverages the information of the Hessian matrix by merging Barzilai-Borwein method and Limited Memory-BFGS method. In addition, an efficient initial solution based on constellation mapping is proposed. The simulation results demonstrate that the proposed method diminishes performance loss to 0.7dB at the bit-error-rate of 10 −2 at 128×32 antenna configuration with low complexity, which surpasses the state-of-the-art (SOTA) algorithms.