基于广义投影梯度下降算法的深度学习大规模MIMO信号检测  

Generalizing projected gradient descent algorithm for massive MIMO detection based on deep-learning

作　　者：黄永明 王正[1,2] Huang Yongming;Wang Zheng(National Mobile Communications Research Laboratory,Southeast University,Nanjing 211189,China;School of Information Science and Engineering,Southeast University,Nanjing 211189,China)

机构地区：[1]东南大学移动通信国家重点实验室,南京211189 [2]东南大学信息科学与工程学院,南京211189

出　　处：《东南大学学报（自然科学版）》2024年第4期961-971,共11页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(62225107,61720106003)。

摘　　要：为了提升大规模MIMO系统的信号检测性能,对由投影和梯度下降(gradient descent,GD)这2个基础操作构成的投影梯度下降(projected gradient descent,PGD)算法进行研究.在基于PGD算法的大规模MIMO检测器中,由于投影和GD操作的损失函数不同,迭代时需要使两者达到平衡,因此通过广义投影梯度下降(generalized projected gradient descent,GPGD)方法实现了投影和GD操作的灵活选取.GPGD方法中在多次的GD步骤后执行1次投影,与传统方式中交替进行投影和GD操作相比,具有显著优势;同时为了保证算法的收敛效率,也对GD操作的步长进行了探究.另外,通过对GPGD算法进行基于深度神经网络的迭代展开,进一步构建了自纠错自动检测器的检测框架,有效提升检测性能和效率.仿真结果表明,GPGD方法带来了明显的系统增益,具有显著的优越性.The projected gradient descent(PGD)-based detector,which consists of two basic operations,projection and gradient descent(GD),was studied to achieve the performance improvement for massive multiple input multiple output(MIMO)detection.In a PGD-based detector for massive MIMO system,since the projection and GD step have different loss functions,necessary compromise has to be made to balance them during iterations.For this reason,the generalized PGD(GPGD)method was proposed with flexible choices of projection and GD.Different from traditional way of performing projection and GD alternatively,GPGD implements projection after every multiple GD steps offers significant advantages.Meanwhile,the step-size of GD was also investigated for convergence efficiency.After that,by unfolding the proposed GPGD method with deep neural networks,the self-corrected auto-detector was established to achieve better decoding performance and efficiency.The simulation results show that the GPGD method achieves an apparent system gain and has a significant superiority.

关 键 词：大规模MIMO检测 投影梯度下降 去噪自动编码器 深度学习 

分 类 号：TN929.5[电子电信—通信与信息系统]