Efficient Distributed Randomized Iterative Detection for Decentralized XL-MIMO Systems

With the rapid scaling up of system dimensions, distributed signal detection based on decentralized baseband architectures has become an important problem in extremely large-scale MIMO (XL-MIMO). In this paper, to reduce the complexity cost of distributed detections for XL-MIMO, the efficient distributed randomized iterative detection (EDRID) algorithm is proposed. First of all, by removing the computationally expensive calculation of matrix pseudoinverse, the randomized iteration in EDRID is still able to exponentially converge into a close range around the target least squares (LS) solution but with significant complexity reduction. Then, to overcome the convergence bias of EDRID, we adopt a dynamic step-size strategy into EDRID to ensure its exact convergence to the LS solution. Theoretical analysis demonstrates that the proposed EDRID with dynamic step-size not only returns the LS solution with arbitrary accuracy for sufficient iterations, but also features the global convergence for different practical scenarios. Finally, to obtain further improvements in both convergence and efficiency, the technique of conditional sampling is also introduced to EDRID, making it a low-complexity, flexible, scalable detection choice for XL-MIMO systems.