A Decoupled Channel Estimation Method for Beyond Diagonal RIS
Bruno Sokal, Fazal-E-Asim, André L. F. de Almeida, Hongyu Li, and Bruno Clerckx · 2024
Read A Decoupled Channel Estimation Method for Beyond Diagonal RIS by Bruno Sokal, Fazal-E-Asim, André L. F. de Almeida, Hongyu Li, and Bruno Clerckx (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-06683.txt.
Introduction
Bruno Sokal and four co-authors proposed a channel estimation method in twenty twenty-four for a new kind of smart surface that shapes radio waves. These surfaces, called reconfigurable intelligent surfaces, bounce signals to boost coverage and speed. The newer beyond-diagonal design links its elements together, gaining flexibility but making the radio channel harder to measure.
Their method first estimates the combined channel with a standard least squares fit. It then reshapes that result and splits it into its parts with a factorization trick, using simple rank-one fits step by step. Separate part estimates combine into a more accurate whole. Computer tests show the split method beats the plain least squares approach. Better channel knowledge is a base need for these surfaces to work well in real networks.
Catalog record
| CREATOR | Bruno Sokal, Fazal-E-Asim, André L. F. de Almeida, Hongyu Li, and Bruno Clerckx |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-06683
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Decoupled Channel Estimation Method for Beyond Diagonal RIS Bruno Sokal, Fazal-E-Asim, André L. F. de Almeida, Hongyu Li, and Bruno Clerckx Abstract Beyond diagonal reconfigurable intelligent surface (BD-RIS) is a new architecture for RIS where elements are interconnected to provide more wave manipulation flexibility than traditional single-connected RIS, enhancing data rate and coverage. However, channel estimation for BD-RIS is challenging due to the more complex multiple-connection structure involving their scattering elements. To address this issue, this paper proposes a decoupled channel estimation method for BD-RIS that yields separate estimates of the involved channels to enhance the accuracy of the overall combined channel by capitalizing on its Kronecker structure. Starting from a least squares estimate of the combined channel and by properly reshaping the resulting filtered signal, the proposed algorithm resorts to a Khatri-Rao Factorization (KRF) method that teases out the individual channels based on simple rank-one matrix approximation steps. Numerical results show that the proposed decoupled channel estimation yields more accurate channel estimates than the classical least squares scheme. I Introduction A recent advancement over the conventional reconfigurable intelligent surface (RIS) technology, which uses diagonal phase shift matrices [ 1 , 2 ] , is the beyond diagonal reconfigurable intelligent surface (BD-RIS) which proved to achieve enhanced channel gain and enlarged coverage [ 3 , 4 , 5 ] . Interconnecting elements via additional tunable components enables BD-RIS to mathematically generate scattering matrices with nonzero off-diagonal entries, increasing the flexibility to manipulate waves. It is worth noting that the enhanced performance of BD-RIS architectures and modes depends highly on the accuracy of the channel state information (CSI). However, it is difficult to effectively and efficiently acquire the CSI for BD-RIS-aided wireless systems due to the lack of signal processing capabilities, from which, as the conventional RIS, the combined TX-RIS-RX channel is often estimated [ 6 , 7 , 8 ] . Also, the BD-RIS has additional physical constraints due to the physical connections, imposing a challenge on the design of the training BD-RIS matrix. For conventional RIS-assisted systems, a traditional least squares (LS) based method has been proposed for combined TX-RIS-RX channel estimation. Although LS methods demonstrate good performance in terms of normalized mean square error (NMSE) , LS methods often suffer from long pilot overhead and do not exploit the intrinsic structure of the combined TX-RIS-RX channel. In [ 9 , 8 ] , the authors proposed a decoupled channel estimation method that exploits the structure of the combined TX-RIS-RX channel. Their proposed method relies on tensor-based signal processing due to the ability of tensor tools to fully exploit different signal diversity dimensions (e.g., space, time, frequency, polarization, etc.) [ 10 , 11 , 12 , 13 ] . In this sense, several works have adopted tensor-based methods for channel estimation [ 14 , 15 , 16 ] . In [ 14 ] , the authors proposed a tensor-based approach for RIS-assisted multi-user MISO systems. The work [ 15 ] proposed a tensor-based framework for parametric estimation of low-rank channels. In [ 16 ] , the authors proposed a tensor-based method for channel estimation and data in time-varying channels. Also, tensor-based methods can be employed in different applications in RIS-assisted systems, e.g., in [ 17 ] , the authors introduced a tensor-based method that decouples the optimum phase shift vector of the RIS into multiple smaller factors, reducing the total required control signaling overhead between the network control and the RIS. For BD-RIS, few works in the literature addressed the channel estimation problem [ 18 , 19 ] . The work [ 18 ] has proposed a baseline LS based receiver to estimate the combined channel based on an orthogonal BD-RIS training matrix design that fulfills the BD-RIS physical constraints. In [ 19 ] , the authors proposed a tensor-based approach for channel estimation and design of the BD-RIS training matrix, where an iterative and a closed-form-based receiver is proposed. In this paper, we formulate a decoupled channel estimation method for MIMO BD-RIS systems. We consider the same system aspect and BD-RIS training design as the one proposed in the baseline LS solution [ 18 ] . Starting from the LS estimate of the combined channel and by exploiting its Krenecker decomposition structure, we resort to a Khatri-Rao Factorization (KRF) method that reveals the individual channels based on simple rank-one matrix approximation steps. The proposed method is conceptually simple and can be executed at low complexity since it relies on permute and reshaping operations on the filtered signal followed by multiple independent rank-one approximations. Interestingly, it is built up on the same principle as the Khatri-Rao Factorization (KRF) algorithm proposed earlier in [ 8 ] for traditional single-connected RIS. Otherwise stated, the proposed decoupled BD-RIS channel estimation method can be viewed as a general approach that encompasses the single-connected RIS to the group and fully-connected RIS (BD-RIS). Numerical results corroborate the effectiveness of the proposed method to provide accurate and individual channel estimates for BD-RIS. II Notation and properties Scalars are represented as non-bold lower-case letters a , column vectors as lower-case boldface letters \boldsymbol{a} , matrices as upper-case boldface letters \boldsymbol{A} . The superscripts \{\cdot\}^{\text{T}} , \{\cdot\}^{\text{*}} , \{\cdot\}^{\text{H}} stand for transpose, conjugate, conjugate transpose. An identity matrix of dimension K is denoted as \boldsymbol{I}_{K} . The operator \Arrowvert\cdot\Arrowvert_{\text{F}} denotes the Frobenius norm of a matrix or tensor, and \mathbb{E}\{\cdot\} is the expectation operator. The operator \text{diag}\left(\boldsymbol{a}\right) converts \boldsymbol{a} into a diagonal matrix, while \text{diag}(\boldsymbol{A}) returns a vector whose elements are the main the diagonal of \boldsymbol{A} . Given a matrix \boldsymbol{A}\in\mathbb{C}^{I\times R} , the operator \text{D}_{i}(\boldsymbol{A}) defines a diagonal matrix of size R\times R constructed from the i -th row of \boldsymbol{A} , for i\in\{1,\ldots,I\} . From a set of Q matrices \boldsymbol{X}^{(q)}\in\mathbb{C}^{M\times N} , q=\{1,\ldots,Q\} , we can construct a block diagonal matrix as \boldsymbol{X}=\text{bdiag}\big(\boldsymbol{X}^{(1)},\ldots,\boldsymbol{X}^{(Q)}\big)\in\mathbb{C}^{MQ\times NQ} . Moreover, \text{vec}\left(\boldsymbol{A}\right) converts \boldsymbol{A}\in\mathbb{C}^{I\times R} to a column vector \boldsymbol{a}\in\mathbb{C}^{IR\times 1} by stacking its columns on top of each other, while the operator \text{unvec}\big(\boldsymbol{a}\big)_{I\times R} returns to the matrix \boldsymbol{A}\in\mathbb{C}^{I\times R} . The symbols \circ , \otimes , and \diamond denote the outer product, the Kronecker product, and the Khatri-Rao product (also known as the column-wise Kronecker product), respectively. For example, the Khatri-Rao product of matrices \boldsymbol{X}\in\mathbb{C}^{I\times R} and \boldsymbol{Y}\in\mathbb{C}^{J\times R} , is defined as \boldsymbol{Z}=\boldsymbol{X}\diamond\boldsymbol{Y}=[\boldsymbol{x}_{1}\otimes\boldsymbol{y}_{1},\ldots,\boldsymbol{x}_{R}\otimes\boldsymbol{y}_{R}]\in\mathbb{C}^{JI\times R}, (1) where \boldsymbol{x}_{r} and \boldsymbol{y}_{r} are the r -th column of \boldsymbol{X} and \boldsymbol{Y} , respectively, for r=\{1,\ldots,R\} . Useful properties that will be exploited in this paper are \displaystyle\text{vec}\left(\boldsymbol{A}\boldsymbol{B}\boldsymbol{C}\right) \displaystyle=\left(\boldsymbol{C}^{\text{T}}\otimes\boldsymbol{A}\right)\text{vec}\left(\boldsymbol{B}\right), (2) \displaystyle\text{vec}\left(\boldsymbol{A}\text{diag}\left(\boldsymbol{b}\right)\boldsymbol{C}\right) \displaystyle=\left(\boldsymbol{C}^{\text{T}}\diamond\boldsymbol{A}\right)\boldsymbol{b}, (3) \displaystyle\text{vec}(\boldsymbol{A}\otimes\boldsymbol{B}) \displaystyle=\boldsymbol{P}\big(\text{vec}(\boldsymbol{A})\otimes\text{vec}(\boldsymbol{B})\big) (4) \displaystyle\text{vec}\left(\boldsymbol{a}\boldsymbol{b}^{\text{T}}\right) \displaystyle=\boldsymbol{b}\otimes\boldsymbol{a} (5) where the involved vectors and matrices have compatible dimensions in each case, and \boldsymbol{P} is a permutation matrix. Fig. 1 : BD-RIS assisted communication scenario. III System Model We consider a multiple-input multiple-output (MIMO) system assisted by a BD-RIS , where the transmitter and the receiver are equipped with M_{T} and M_{R} antennas, respectively, and a BD-RIS with N elements, illustrated in Figure 1 . We assume the direct link between the transmitter and the receiver is blocked. Assuming a length- T pilot sequence, the received signal at the t time slot is given by \boldsymbol{y}_{t}=\boldsymbol{G}\boldsymbol{S}_{t}\boldsymbol{H}^{\text{T}}\boldsymbol{x}_{t}+\boldsymbol{b}_{t}\in\mathbb{C}^{M_{R}\times 1} , where \boldsymbol{G}\in\mathbb{C}^{M_{R}\times N} and \boldsymbol{H}\in\mathbb{C}^{M_{T}\times N} are the IRS-RX and the TX-IRS