Improved Iterative Reweighted Multiuser Detectors for Massive Machine-Type Communications

Grant-free non-orthogonal multiple access (GF-NOMA) is a promising technique for 6G massive machine-type communications (mMTC) to support massive connectivity and reduce access latency. However, the multi-user detection (MUD) method for single measurement vector (SMV) models in GF-NOMA still poses challenges due to its low accuracy and high computational complexity. This paper studies the improvement space of iterative reweighted (IR) detectors and proposes two aspects of improvement, error performance optimization and runtime optimization. In terms of error performance, the convergence of regularization parameters is introduced to solve the problem of not being able to approach optimal performance when the regularization parameters are fixed. In terms of running time, firstly, based on the conjugate gradients (CG) iterative method, IR-CG improves the accuracy of IR-based MUD by replacing the fixed regularization parameters with variable ones to avoid overfitting of noise, and also accelerates the iteration convergence with conjugate gradient. Subsequently, based on the preconditioned conjugate gradient (PCG) method, IR-PCG improves the computational efficiency of IR-CG with the help of the incomplete Choleski decomposition to make the coefficient matrix with higher condition number and sparsity property. Simulation results show that the improved IR detector proposed in this paper, by combining two improvement measures, not only slightly improves the error performance, but also significantly shortens the time required for the detector to operate.