| 摘要: |
| 受益于信道硬化现象和高维渐近特性，传统线性信号检测算法在大规模多输入多输出(Multiple-Input Multiple-Output，MIMO)系统中可获得优良的检测性能，但高维矩阵求逆的繁重计算负担将导致实际应用困难。借助于信号检测领域知识和深度学习技术，提出了一种Kaczmarz深度网络(Kaczmarz Network，Kaczmarz-Net)大规模MIMO上行链路信号检测方法。首先，对综合性能最优的默认降序Kaczmarz检测算法实施深度网络结构设计，将算法迭代运算过程映射为深度网络。其次，结合Kaczmarz算法自身特有的循环迭代更新特性，引入可学习参数并改进算法内部更新结构。最后，利用简化对数似然比计算软信息，将深度网络引入软判决提升检测精确度。实验结果表明，在硬判决检测条件下，所提出的Kaczmarz-Net在天线配置为64×64、误码率（Bit Error Rate,BER）为10-2时，相较最小均方误差(Minimum Mean Square Error，MMSE)算法可获得1 dB的性能增益，且仅需发射端天线数平方级的计算开销；在软判决条件下，Kaczmarz-Net深度网络可获得与MMSE软检测算法相当的BER性能表现。 |
| 关键词: 大规模MIMO 信号检测 深度学习 软输出 Kaczmarz算法 |
| DOI：10.20079/j.issn.1001-893x.241018001 |
|
| 基金项目:国家自然科学基金资助项目 (U23A20279) |
|
| Signal Detection Method Based on Kaczmarz-Net Deep Network for Massive MIMO Systems |
| JIN Longkang,WANG Bianzheng,SHEN Bin |
| (1.School of Communications and Information Engineering，Chongqing University of Posts and Telecommunications，Chongqing 400065，China;2.College of Electronics and Information Engineering，Sichuan University，Chengdu 610065，China) |
| Abstract: |
| Benefiting from the channel hardening phenomenon and high-dimensional asymptotic properties，traditional linear signal detection algorithms can obtain excellent detection performance in massive multiple-input-multiple-output (MIMO) systems，but the heavy computational burden of the inverse of high-dimensional matrices leads to difficulties in practical applications.With the benefit of signal detection domain knowledge and deep learning techniques，a Kaczmarz-Net deep network(Kaczmarz-Net) massive MIMO uplink signal detection method is proposed.First，the default descending Kaczmarz detection algorithm with optimal comprehensive performance is implemented to design the deep network structure，and the iterative operation process of the algorithm is mapped into a deep network.Second，by combining the unique cyclic iterative update characteristics of the Kaczmarz algorithm，learnable parameters are introduced and the internal update structure of the algorithm is improved.In general，the simplified log-likelihood ratio is used to calculate the soft information，and the deep network is introduced into the soft judgment to improve detection accuracy.Experimental results demonstrate that under hard-decision detection conditions，the proposed Kaczmarz-Net achieves a performance gain of 1 dB compared to the minimum mean square error(MMSE) algorithm when the antenna configuration is 64×64 and the bit error rate (BER) is 10-2，while requiring only computational overhead quadratic in the number of transmit antennas.Under soft-decision conditions，the Kaczmarz-Net exhibits BER performance comparable to the MMSE soft detection algorithm. |
| Key words: massive MIMO signal detection deep learning soft-output Kaczmarz algorithm |
《电讯技术》编辑部 