基于L2O的MU-MIMO容量算法优化

Capacity Optimization of MU-MIMO Algorithm Based on L20

在求解MU-MIMO容量优化问题时,传统数值优化算法的收敛速度较慢,而利用机器学习与传统数值优化相结合的方法可以有效提升求解效率,这种基于机器学习技术开发优化算法的方法被称为L2O。首先对MU-MIMO场景下容量优化相关的传统算法进行了讨论分析。然后,基于L2O思想,构建了嵌入DNN的有限步梯度投影算法。最后,通过大量仿真实验表明,相较于传统梯度投影算法,改进后的有限步梯度投影算法在具有较高求解质量的同时极大地提升了求解效率,能够在MU场景下更加有效地求解容量优化问题。

In the context of solving the MU-MIMO capacity optimization problem,traditional numerical optimization algorithms often exhibit a slow convergence speed.However,the combination of machine learning and traditional numerical optimization effectively improves the solution efficiency.This method of developing an optimization algorithms based on machine learning technology is termed as learning to optimize(L2O).This paper analyzes and discusses the traditional algorithms pertaining to the capacity optimization in the MU-MIMO scenario.Subsequently,based on the L2O concept,the paper constructs a limited-step gradient projection algorithm.Finally,through a plethora of simulation experiments,the improved limited-step gradient projection algorithm has a higher solution quality and significantly enhance the solution efficiency in comparison with the traditional gradient projection algorithm,and thus it can effectively resolve the capacity optimization problem in the MU scenario.