
Citation: | WANG Dan, LIANG Jiamin, MEI Zhiqiang, LIU Jinzhi. Millimeter-wave Channel Estimation with Intelligent Reflecting Surface Assisted Based on Vector Approximate Message Passing[J]. Journal of Electronics & Information Technology, 2022, 44(7): 2400-2406. doi: 10.11999/JEIT211271 |
顺应5G无线通信的新时代,对通信系统提出了更高速率、更大容量、更低时延的新要求。随着Sub-6 GHz频段的商用,毫米波也逐渐引起关注并投入研究,为在广域级和局域级提供无缝连接,实现毫米波频谱[1]的利用至关重要。针对5G毫米波传播损耗高、绕射和衍射能力弱、覆盖相对受限等挑战,智能反射面(Intelligent Reflecting Surface, IRS)技术作为一种新技术,通过低功耗甚至无功耗反射电磁波,降低信号损耗,大大地提高了毫米波通信的信道质量,以此更加适用于非视距传播(non-Line-of-Sight, nLoS)环境中。由于IRS低功耗的特点,使得其在很大程度上优于中继设备。IRS不受接收噪声的影响[2]、具有全频带响应、部署灵活性和兼容性[3]等众多优点。将IRS与无人机相结合辅助通信将大大地提高通信质量和速率。针对更多基站的部署将增大电磁辐射水平的问题,IRS的智能性使得部署IRS设备将降低电磁辐射水平。同时IRS也可通过调制来实现低复杂度的大规模多输入多输出(Multiple Input Multiple Output, MIMO)技术。结合IRS的诸多优点,使得该技术在5G毫米波以及未来的6G时代获得越来越多的关注和研究。
对于IRS辅助通信的信道估计问题,文献[4]采用行列块稀疏度来联合估计级联信道,首先利用公共列块稀疏性来估计由出射角阵列控制向量跨越的公共子空间,再利用块稀疏性来制定基于多个测量向量的多用户联合稀疏矩阵恢复问题。但通常优化变量在非凸且难以求解的模拟多用户联合稀疏矩阵恢复问题中是耦合的,此时可通过一种基于替代优化[4]和迭代加权算法[4]的方法来有效地解决问题。文献[5]提出了一种基于双线性广义近似消息传递(Bilinear Generalized Approximate Message Passing, BiG-AMP)的稀疏矩阵分解算法来解决信道的稀疏问题,并且利用黎曼梯度算法实现稀疏矩阵补全,此算法的性能会随着在稀疏矩阵分解阶段估计的随机变量的增加而变差。对比于BiG-AMP算法,有学者提出双线性自适应矢量近似消息传递算法[6](Bilinear Adaptive Vector Approximate Message Passing, BAdVAMP)来解决IRS信道估计,此算法的性能优于前者。在对稀疏信道进行信道估计时,一般可运用压缩感知算法[7,8]。文献[9]通过压缩感知和深度学习对连接到控制器的IRS单元进行控制,从而估计所有经过IRS单元上的信道。文献[10]中利用压缩感知算法的同时,结合AMP算法对IRS辅助的级联信道进行信道估计。在IRS辅助通信系统中,优化线性预编码算法[11]和迭代算法[12]同样也是解决信道估计的方法。在毫米波通信系统中,有研究者提出在IRS辅助毫米波网络中使用波束搜索[13]方法找到所需用户的最佳波束,以增强毫米波的信道质量。但对于以上所有算法都没有针对IRS辅助毫米波通信系统的信道估计问题进行探索,本文将对此系统环境的信道估计问题进行研究。
针对IRS系统中级联信道的复杂性以及IRS辅助毫米波通信的信道稀疏性,本文结合Khatri-Rao积[14]、克罗内克积和矢量近似消息传递算法(Khatri-Rao and Vector Approximate Message Passing, KR-VAMP)提出了KR-VAMP算法来解决系统的信道估计问题。由于加入IRS使得系统信道为级联信道,增加了信道的复杂性,本文通过Khatri-Rao积和克罗内克积[15]将级联信道
全文结构安排如下:第2节介绍IRS辅助毫米波系统的信道模型,以及AMP算法模型。第3节对KR-VAMP算法进行分析和推导。第4节通过KR-VAMP算法和其他算法进行对比,通过仿真验证KR-VAMP算法的有效性。第5节对全文进行总结并给出下一步工作方向。
符号说明:符号
为增强毫米波通信的信道质量,引入IRS辅助毫米波通信。如图1所示的IRS辅助通信的信道示意图,当基站与用户之间的视距(Line-of-Sight, LoS)传播受阻时,通过IRS辅助毫米波信号的nLoS信道传播。基站通过微控制器[17]对IRS系统的每个单元进行控制调节,本文主要研究IRS辅助通信的下行链路系统。在此通信系统中,有
为简化模型,假设系统为单用户系统。此系统以毫米波信号为例,基站与IRS之间的信道
V=[V1,V2,⋯,VN]T∈CN×1 | (1) |
Vn=βnejθn | (2) |
其中,
hIB=√NML1L1∑l=1∂lar(ϑl,ωl)aHt(ϕl) | (3) |
hUI=√NL2L2∑l=1∂lat(ϑl,ωl) | (4) |
基站与IRS之间的位置关系图如图2(a)所示,假设
ar(ϑl,ωl)=1√N[1,⋯,ej2πλp(n1v+n2w),⋯,ej2πλp((N1−1)v+(N2−1)w)]T | (5) |
IRS与用户之间的位置关系图如图2(b)所示,即在式(4)中,
由于毫米波信道的稀疏散射特性,路径数量远小于
hIB=AIΨAHB | (6) |
hHUI=AIΓ | (7) |
其中,
AI≜AIn2⊗AIn2=[fpN1(v1),fpN1(v2),⋯,fpN1(vNG,1)]⊗[fpN2(w1),fpN2(w2),⋯,fpN2(wNG,2)] | (8) |
AB≜[fdM(ϕ1),fdM(ϕ2),⋯,fdM(ϕMG)] | (9) |
式(6)和式(7)所示的稀疏信道的信道系数
在如图1所示的IRS辅助毫米波通信的级联信道
H=hUIdiag(V)hIB=VTdiag(hUI)hIB | (10) |
其中,
G=diag(hUI)hIB=(hUI⊙hTIB)T=[(AIΓ)H⊙(AIΨAHB)T]T=[(ΓH⊗A∗BΨT)⋅(AHI⊙ATI)]T=(AHI⊙ATI)T⋅(Γ∗⊗Ψ)⋅(1⊗AHB)=D⋅(Γ∗⊗Ψ)⋅AHB | (11) |
其中,
Y=VT⋅D⋅(Γ∗⊗Ψ)⋅AHBX+Σ=(XT⊗VT)⋅vec[D(Γ∗⊗Ψ)AHB]+Σ=(XT⊗VT)⋅(A∗B⊗D)⋅vec(Γ∗⊗Ψ)+Σ | (12) |
在非凸实例中,AMP不仅可以使用可测试的贝叶斯最优条件精确地描述方法的性能,而且承接了迭代阈值算法的优势,也融入了近似消息传递算法的精髓。本文将结合AMP算法解决信道
Rk=Y−ˆHkX+1δRk−1⟨g′(rk−1,γk−1)⟩ | (13) |
其中,
⟨g′(rk−1,γk−1)⟩=1M′∑i∂[g(ri,k−1,γk−1)]∂ri,k−1 | (14) |
其中,
g\left( {{{\boldsymbol{r}}_{km}},{\gamma _k}} \right) = \mathop {\arg \min }\limits_{{H_m} \in \mathbb{R}} \left[ {\frac{{{\gamma _k}}}{2}{{\left| {{H_m} - {{\boldsymbol{r}}_{km}}} \right|}^2} - \ln p\left( {{H_m}} \right)} \right] | (15) |
由于信道估计值由真实信道冲激响应与信道噪声组成,可假设
\begin{split} & \varepsilon \left( {{\tau _k},{\gamma _k}} \right) \\ & \quad= \frac{1}{{{M_{\text{G}}}}}{\rm{E}}\left[ {{{\left\| {g\left( {{{\boldsymbol{H}}^0} + N\left( {0,{\tau _{k - 1}}{\boldsymbol{I}}} \right),{\gamma _{k - 1}}} \right) - {{\boldsymbol{H}}^0}} \right\|}^2}} \right] \end{split} | (16) |
设定初始条件
\qquad {\tau _k} = \xi _\varSigma ^{ - 1} + \frac{{{M_{\text{G}}}}}{M}\varepsilon \left( {{\tau _{k - 1}},{\gamma _{k - 1}}} \right) | (17) |
\qquad {\gamma _k} = {\left( {{\tau _k}} \right)^{ - 1}} | (18) |
结合传统AMP算法分析和描述,AMP算法的仿真流程如下。
步骤1 设置迭代总次数
步骤2
步骤3
步骤4
步骤5
步骤6 确定
步骤7 将迭代次数
本节将对KR-VAMP 算法的因子转换方式进行分析。首先在式(12)中,定义
p\left( {{\boldsymbol{Y}}|{\boldsymbol{\varPhi}} ,{\boldsymbol{\varGamma}} ,{\boldsymbol{\varPsi}} } \right) = N\left( {{\boldsymbol{Y}};{\boldsymbol{\varOmega}}{\text{vec}}\left( {\boldsymbol{\varPhi}} \right),\xi _{\boldsymbol{\varSigma}} ^{ - 1}{\boldsymbol{I}}} \right) | (19) |
因子转化图如图3所示,黑色方框表示因子节点,白色圆形表示变量节点,通过图3因子图可知
\begin{split} p\left( {{\boldsymbol{Y}},{\boldsymbol{\varPhi}} ,{\boldsymbol{\varGamma}} ,{\boldsymbol{\varPsi}} } \right) =& p\left( {\boldsymbol{\varGamma}} \right)p\left( {\boldsymbol{\varPsi}} \right)\delta \left( {{\boldsymbol{\varPhi }} - {\boldsymbol{\varGamma}} \otimes {\boldsymbol{\varPsi}} } \right)\\ & \cdot N\left( {{\boldsymbol{Y}};{\boldsymbol{\varOmega}}{\text{vec}}\left({\boldsymbol{ \varPhi}} \right),\xi _{\boldsymbol{\varSigma}} ^{ - 1}{\boldsymbol{I}}} \right) \end{split} | (20) |
在因子转换图中,设定变量节点
\begin{split} {u_{\varGamma \to \delta }}\left( \varGamma \right) & = \frac{{N\left( {{\boldsymbol{\varGamma}} ;{{\hat {\boldsymbol{\varGamma}} }_k},\eta _{{{\boldsymbol{\varGamma}} }k}^{ - 1}{\boldsymbol{I}}} \right)}}{{N\left( {{{\boldsymbol{\varGamma}} };{{\boldsymbol{r}}_k},\xi _{{{\boldsymbol{\varGamma}} }k}^{ - 1}{\boldsymbol{I}}} \right)}} \\ & \propto N \left( {{{\boldsymbol{\varGamma}} };\frac{{\left( {{{{\hat {\boldsymbol{\varGamma}} }}_k}{\eta _{{\varGamma }k}} - {{\boldsymbol{r}}_k}{\xi _{{{\boldsymbol{\varGamma}} }k}}} \right)}}{{\left( {{\eta _{{{\boldsymbol{\varGamma}} }k}} - {\xi _{{{\boldsymbol{\varGamma}} }k}}} \right)}},{{\left( {{\eta _{{{\boldsymbol{\varGamma}} }k}} - {\xi _{{{\boldsymbol{\varGamma}} }k}}} \right)}^{ - 1}}{\boldsymbol{I}}} \right)\\ \end{split} | (21) |
同理设定变量节点
\begin{split} {u_{{\varPsi } \to \delta }}\left( {{\boldsymbol{\varPsi}} } \right) & = \frac{{N\left( {{{\boldsymbol{\varPsi}} };{{{\hat {\boldsymbol{\varPsi}} }}_k},\eta _{{{\boldsymbol{\varPsi}} }k}^{ - 1}{\boldsymbol{I}}} \right)}}{{N\left( {{{\boldsymbol{\varPsi}} };{{\boldsymbol{r}}_k},\xi _{{{\boldsymbol{\varPsi}} }k}^{ - 1}{\boldsymbol{I}}} \right)}} \\ & \propto N \left( {{{\boldsymbol{\varPsi}} };\frac{{\left( {{{{\hat {\boldsymbol{\varPsi}} }}_k}{\eta _{{{\boldsymbol{\varPsi}} }k}} - {{\boldsymbol{r}}_k}{\xi _{{{\boldsymbol{\varPsi}} }k}}} \right)}}{{\left( {{\eta _{{{\boldsymbol{\varPsi}} }k}} - {\xi _{{\boldsymbol{{\varPsi}} }k}}} \right)}},{{\left( {{\eta _{{{\boldsymbol{\varPsi}} }k}} - {\xi _{{{\boldsymbol{\varPsi}} }k}}} \right)}^{ - 1}}{\boldsymbol{I}}} \right)\\ \end{split} | (22) |
在图3所示的因子图中,可将
\;\; {{\boldsymbol{r}}_{{\varPhi }k}} = \frac{{\left( {{{{\hat {\boldsymbol{\varGamma}} }}_k}{\eta _{{{\boldsymbol{\varGamma}} }k}} - {{\boldsymbol{r}}_k}{\xi _{{{\boldsymbol{\varGamma}} }k}}} \right)}}{{\left( {{\eta _{{{\boldsymbol{\varGamma}} }k}} - {\xi _{{{\boldsymbol{\varGamma}} }k}}} \right)}} + \frac{{\left( {{{{\hat {\boldsymbol{\varPsi}} }}_k}{\eta _{{{\boldsymbol{\varPsi}} }k}} - {{\boldsymbol{r}}_k}{\xi _{{{\boldsymbol{\varPsi}} }k}}} \right)}}{{\left( {{\eta _{{{\boldsymbol{\varPsi}} }k}} - {\xi _{{{\boldsymbol{\varPsi}} }k}}} \right)}} | (23) |
\;\; \xi _{{{\boldsymbol{\varPhi}} }k}^{ - 1} = {\left( {{\eta _{{{\boldsymbol{\varGamma}} }k}} - {\xi _{{{\boldsymbol{\varGamma}} }k}}} \right)^{ - 2}} + {\left( {{\eta _{{{\boldsymbol{\varPsi}} }k}} - {\xi _{{{\boldsymbol{\varPsi}} }k}}} \right)^{ - 2}} | (24) |
变量节点
\begin{split} {b_{{\text{sp}}}}\left( {{\boldsymbol{\varPhi}} } \right) \propto & N\left( {{\text{vec}}\left( {{\boldsymbol{\varPhi}} } \right);\left( {{{{\boldsymbol{\varOmega}} }^{\text{ + }}}{\boldsymbol{Y}}} \right),{{\left( {{{{\boldsymbol{\varOmega}} }^{\text{T}}}{\xi _{{\boldsymbol{\varSigma}} }}{{\boldsymbol{\varOmega}} }} \right)}^{ - 1}}} \right)\\ & \cdot N\left( {{{\boldsymbol{\varPhi}} };{{\boldsymbol{r}}_{{{\boldsymbol{\varPhi}} }k}},\xi _{{{\boldsymbol{\varPhi}} }k}^{ - 1}{\boldsymbol{I}}} \right) \\ \propto &N\left( {\text{vec}}\left( {{\boldsymbol{\varPhi}} } \right);{{\left( {{{{\boldsymbol{\varOmega}} }^{\text{T}}}{\xi _{{\boldsymbol{\varSigma}} }}{{\boldsymbol{\varOmega}} } + {\xi _{{{\boldsymbol{\varPhi}} }k}}} \right)}^{ - 1}}\right.\\ & \cdot\left( {{{{\boldsymbol{\varOmega }}}^{\text{T}}}{\xi _{{\boldsymbol{\varSigma}} }}{{\boldsymbol{\varOmega}} }{{{\boldsymbol{\varOmega}} }^{\text{ + }}}{\boldsymbol{Y}} + {\xi _{{{\boldsymbol{\varPhi}} }k}}{{\boldsymbol{r}}_{{{\boldsymbol{\varPhi}} }k}}} \right),\\ \\ & \left.{{\left( {{{{\boldsymbol{\varOmega}} }^{\text{T}}}{\xi _{{\boldsymbol{\varSigma}} }}{{\boldsymbol{\varOmega}} } + {\xi _{{{\boldsymbol{\varPhi}} }k}}} \right)}^{ - 1}} \right) \end{split} | (25) |
结合
g\left( {{{\boldsymbol{r}}_{{\varPhi }k}},{\xi _{{\varPhi }k}}} \right) = {{\boldsymbol{V}}^{\text{T}}}{\boldsymbol{D}} \cdot {\text{reshape}}\left( {{\text{vec}}\left( {{\hat {\boldsymbol{\varPhi}} }} \right),N_{\text{G}}^{\text{2}},{M_{\text{G}}}} \right) \cdot {\boldsymbol{A}}_{\text{B}}^{\text{H}} | (26) |
结合式(16)和式(26),可将
步骤1 利用Khatri-Rao积以及克罗内克积的特性将信道
步骤2 设置初始
步骤3
步骤4
步骤5
步骤6
步骤7
步骤8 将迭代次数
由上述AMP算法的仿真流程和KR-VAMP算法的仿真流程可知,阈值函数
本文主要验证IRS辅助毫米波通信系统的有效性,以及KR-VAMP算法与其他AMP算法在IRS辅助毫米波的系统环境中信道误差数值。结果主要对比了AMP算法、BiG-AMP算法[5]和KR-VAMP算法的信道误差。仿真数据设置IRS的单元数为16,各IRS单元之间的距离为15 mm。基站处的天线数为2,天线之间的距离设置为15 mm。完备矩阵维度
在图4所示的3维坐标中,基站、IRS和用户的坐标分别为
在上述的前提条件下,当噪声平均功率
当迭代初始值
综合上述两个变量,当噪声平均功率
本文利用IRS辅助毫米波通信,以此降低毫米波在大气中的损耗,提高毫米波nLoS传播路径的性能。在此系统环境中,本文基于VAMP算法提出了KR-VAMP算法,通过Khatri-Rao积和克罗内克儿积将级联信道转换为稀疏信号恢复问题,以及VAMP算法良好的迭代阈值优势,减少了训练迭代次数,提高了系统的性能,降低了整个系统的信道误差。最后仿真验证,与传统的AMP相比,使用KR-VAMP算法可以更好地降低系统的信道误差,对毫米波信道具有一定的性能提升。在本文信道估计过程所使用的VAMP算法还存在一定的复杂度。于是,后续工作将重点围绕降低算法复杂度问题展开研究。
[1] |
SAAD W, BENNIS M, and CHEN Mingzhe. A vision of 6G wireless systems: Applications, trends, technologies, and open research problems[J]. IEEE Network, 2020, 34(3): 134–142. doi: 10.1109/MNET.001.1900287
|
[2] |
BASAR E, DI RENZO M, DE ROSNY J, et al. Wireless communications through reconfigurable intelligent surfaces[J]. IEEE Access, 2019, 7: 116753–116773. doi: 10.1109/ACCESS.2019.2935192
|
[3] |
YOU Changsheng, ZHENG Beixiong, and ZHANG Rui. Intelligent reflecting surface with discrete phase shifts: Channel estimation and passive beamforming[C]. The ICC 2020–2020 IEEE International Conference on Communications (ICC), Dublin, Ireland, 2020: 1–6.
|
[4] |
CHEN Jie, LIANG Yingchang, CHENG H V, et al. Channel estimation for reconfigurable intelligent surface aided multi-user MIMO systems[EB/OL]. https://arxiv.org/abs/1912.03619, 2019.
|
[5] |
HE Zhenqing and YUAN Xiaojun. Cascaded channel estimation for large intelligent metasurface assisted massive MIMO[J]. IEEE Wireless Communications Letters, 2020, 9(2): 210–214. doi: 10.1109/LWC.2019.2948632
|
[6] |
MIRZA J and ALI B. Channel estimation method and phase shift design for reconfigurable intelligent surface assisted MIMO networks[J]. IEEE Transactions on Cognitive Communications and Networking, 2021, 7(2): 441–451. doi: 10.1109/TCCN.2021.3072895
|
[7] |
WANG Peilan, FANG Jun, and LI Hongbin. Joint beamforming for intelligent reflecting surface-assisted millimeter wave communications[EB/OL]. https://arxiv.org/abs/1910.08541v1, 2019.
|
[8] |
TSAI C R, LIU Y H, and WU A Y. Efficient compressive channel estimation for millimeter-wave large-scale antenna systems[J]. IEEE Transactions on Signal Processing, 2018, 66(9): 2414–2428. doi: 10.1109/TSP.2018.2811742
|
[9] |
TAHA A, ALRABEIAH M, and ALKHATEEB A. Enabling large intelligent surfaces with compressive sensing and deep learning[J]. IEEE Access, 2021, 9: 44304–44321. doi: 10.1109/ACCESS.2021.3064073
|
[10] |
LIU Hang, YUAN Xiaojun, and ZHANG Y J A. Message-passing based channel estimation for reconfigurable intelligent surface assisted MIMO[C]. 2020 IEEE International Symposium on Information Theory (ISIT), Los Angeles, USA, 2020: 2983–2988.
|
[11] |
NADEEM Q U A, ALWAZANI H, KAMMOUN A, et al. Intelligent reflecting surface-assisted multi-user MISO communication: Channel estimation and beamforming design[J]. IEEE Open Journal of the Communications Society, 2020, 1: 661–680. doi: 10.1109/OJCOMS.2020.2992791
|
[12] |
ZHANG Jinming, QI Chenhao, LI Ping, et al. Channel estimation for reconfigurable intelligent surface aided massive MIMO system[C]. The 21st International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Atlanta, USA, 2020.
|
[13] |
TAN Xin, SUN Zhi, KOUTSONIKOLAS D, et al. Enabling indoor mobile millimeter-wave networks based on smart reflect-arrays[C]. The IEEE INFOCOM 2018 - IEEE Conference on Computer Communications, Honolulu, USA, 2018: 270–278.
|
[14] |
DE ARAÚJO G T and DE ALMEIDA A L F. PARAFAC-based channel estimation for intelligent reflective surface assisted MIMO system[C]. The 11th Sensor Array and Multichannel Signal Processing Workshop (SAM), Hangzhou, China, 2020: 1–5.
|
[15] |
WANG Peilan, FANG Jun, DUAN Huiping, et al. Compressed channel estimation for intelligent reflecting surface-assisted millimeter wave systems[J]. IEEE Signal Processing Letters, 2020, 27: 905–909. doi: 10.1109/LSP.2020.2998357
|
[16] |
BARON D, RUSH C, and YAPICI Y. mmWave channel estimation via approximate message passing with side information[C]. The 21st International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Atlanta, USA, 2020.
|
[17] |
ABEYWICKRAMA S, ZHANG Rui, WU Qingqing, et al. Intelligent reflecting surface: Practical phase shift model and beamforming optimization[J]. IEEE Transactions on Communications, 2020, 68(9): 5849–5863. doi: 10.1109/TCOMM.2020.3001125
|
[18] |
王丹, 梁家敏, 刘金枝, 等. 6G可重构智能表面的相移模型设计[J]. 计算机应用, 2021, 41(9): 2694–2698. doi: 10.11772/j.issn.1001-9081.2020111836
WANG Dan, LIANG Jiamin, LIU Jinzhi, et al. Phase shift model design for 6G reconfigurable intelligent surface[J]. Journal of Computer Applications, 2021, 41(9): 2694–2698. doi: 10.11772/j.issn.1001-9081.2020111836
|
[19] |
RANGAN S, SCHNITER P, and FLETCHER A K. Vector approximate message passing[C]. 2017 IEEE International Symposium on Information Theory (ISIT), Aachen, Germany, 2017: 1588–1592.
|