betaFeedback

WorksComputer Science

Can Reciprocal RIS Perform as Non-Reciprocal RIS?

Abstract

Yes, if and only if some optimal non-reciprocal RIS (i.e., maximizing a given performance metric) makes the round-trip channel transmitter-RIS-transmitter symmetric, which holds if the channel transmitter-RIS has rank one or if single-stream transmission is optimal.

Introduction

Reconfigurable intelligent surface (RIS) has been largely explored as a technology that can enhance wireless communications by shaping the wireless channel [1, 2]. Consider a generic wireless system between NTN_T transmitting antennas and NRN_R receiving antennas, as shown in Figure 1(a). The wireless channel H∈CNR×NT\mathbf{H} \in\mathbb{C}^{N_R \times N_T} with an MM-element RIS is given by

Figure 1. RIS-aided system with NTN_T transmitting antennas and NRN_R receiving antennas where we highlight: (a) the NR×NTN_R \times N_T channel from the NTN_T transmitting to the NRN_R receiving antennas HRT+HRΘHT\mathbf{H}_{RT}+\mathbf{H}_R\boldsymbol{\Theta}\mathbf{H}_T; (b) the NT×NTN_T \times N_T round-trip channel from the NTN_T transmitting antennas to themselves HTTΘHT\mathbf{H}_T^T\boldsymbol{\Theta}\mathbf{H}_T; (c) the K×KK \times K round-trip channel from the KK transmitted streams to themselves FTHTTΘHTF\mathbf{F}^T\mathbf{H}_T^T\boldsymbol{\Theta}\mathbf{H}_T\mathbf{F}.

H=HRT+HRΘHT,(1)\mathbf{H} = \mathbf{H}_{RT} + \mathbf{H}_{R}\mathbf{\Theta}\mathbf{H}_{T}, \tag*{(1)}

where HRT∈CNR×NT\mathbf{H}_{RT} \in\mathbb{C}^{N_R \times N_T} is the direct channel between the transmitting and receiving antennas, HR∈CNR×M\mathbf{H}_{R} \in\mathbb{C}^{N_R \times M} is the channel between the RIS and the receiving antennas, Θ∈CM×M\mathbf{\Theta} \in\mathbb{C}^{M \times M} is the RIS reflection matrix, and HT∈CM×NT\mathbf{H}_{T} \in\mathbb{C}^{M \times N_T} is the channel between the transmitting antennas and the RIS [1]. In this system, the NTN_T transmitting antennas can belong to a single or multiple transmitters, and, similarly, the receiving antennas can belong to a single or multiple receivers. We can control the wireless channel H\mathbf{H} with a RIS by optimizing its reflection matrix Θ\mathbf{\Theta}, which formally is the scattering matrix of the microwave network implementing the RIS. The reader is referred to [3 Chapter 4] for more details on the physical meaning of a scattering matrix.

Conventionally, RIS has been designed by controlling each element independently through a tunable load [1, 2]. Assuming lossless loads, this design results in a scattering matrix

Θ=diag⁡(ejθ1,ejθ2,…,ejθM),(2)\mathbf{\Theta} = \operatorname{diag}(e^{j\theta_1}, e^{j\theta_2}, \ldots, e^{j\theta_M}), \tag*{(2)}

where θm∈[0,2π)\theta_m \in[0,2\pi), for m=1,…,Mm = 1,\ldots,M. Given the structure of this Θ\mathbf{\Theta}, such RIS design has been denoted as diagonal RIS (D-RIS) [4]. Nevertheless, more general RIS designs are possible, which have a scattering matrix Θ\mathbf{\Theta} not limited to being diagonal and go under the umbrella term of beyond-diagonal RIS (BD-RIS) [4, 5]. By implementing a RIS with an arbitrary lossless and reciprocal microwave network, it is possible to optimize

This work has been supported in part by UKRI under Grant EP/Y004086/1, EP/X040569/1, EP/Y037197/1, EP/X04047X/1, EP/Y037243/1.

Matteo Nerini and Bruno Clerckx are with the Department of Electrical and Electronic Engineering, Imperial College London, SW7 2AZ London, U.K. (e-mail: [email protected]; [email protected]).

Juan Vidal Alegría is with the Department of Electrical and Information Technology, Lund University, 221 00 Lund, Sweden. (e-mail: [email protected]).

Θ\mathbf{\Theta} subject to unitary (due to losslessness) and symmetric (due to reciprocity) constraints, i.e.,

ΘHΘ=IM,Θ=ΘT,(3)\mathbf{\Theta}^{H}\mathbf{\Theta} = \mathbf{I}_{M}, \quad\mathbf{\Theta} = \mathbf{\Theta}^{T}, \tag*{(3)}

which are the most general constraints of a lossless reciprocal RIS [4, 5, 6]. Note that the constraints of a reciprocal RIS in (3) are more general than the constraints of a D-RIS in (2), and therefore a reciprocal RIS performs at least as well as a D-RIS. Even more general RIS designs have been proposed, including the use of non-reciprocal microwave components such as circulators, isolators, and gyrators [6, 7, 8]. In particular, by implementing a RIS with an arbitrary lossless microwave network, Θ\mathbf{\Theta} is only constrained to be unitary, i.e.,

ΘHΘ=IM,(4)\mathbf{\Theta}^{H}\mathbf{\Theta} = \mathbf{I}_{M}, \tag*{(4)}

giving the most general constraints of a lossless non-reciprocal RIS. The terminology non-reciprocal RIS does not mean that the RIS is not reciprocal. It means that the RIS is not required to be reciprocal, i.e., it is not constrained to have a symmetric Θ\mathbf{\Theta}, consistent with previous literature [6, 7, 8]. Since the constraints of a non-reciprocal RIS in (4) are more general than those of a reciprocal RIS in (3), a non-reciprocal RIS performs at least as well as a reciprocal RIS.

Related literature explored non-reciprocal RIS mainly for uplink-downlink channel reciprocity attacks [8, 9, 10, 11] and for full-duplex systems [12, 13, 14, 15]. In addition, [6] showed that a reciprocal RIS requires more elements than a non-reciprocal one to arbitrarily reconfigure the channel, e.g., for channel orthogonalization. However, non-reciprocal RIS requires non-reciprocal components (e.g., circulators, isolators, and gyrators), which increase hardware complexity. It is therefore essential to understand when this additional complexity brings a performance gain. In this paper, we therefore answer the fundamental question: Can reciprocal RIS perform as non-reciprocal RIS? Our contributions are as follows.

First, we derive a necessary and sufficient condition for reciprocal RIS to perform as non-reciprocal RIS with the same number of elements. This condition states that there must exist an optimal non-reciprocal RIS (i.e., maximizing the considered performance metric) that makes the round-trip channel from the transmitting antennas, via the RIS, back to the transmitting antennas symmetric. We show that, as a consequence of this condition, reciprocal RIS performs as non-reciprocal RIS when the channel from the transmitting antennas to the RIS, or from the RIS to the receiving antennas, has rank one. This includes all single-user single-input single-output (SISO), single-input multiple-output (SIMO), and multiple-input single-output (MISO) systems.

Second, we refine this necessary and sufficient condition under linear precoding, which is not assumed in our first contribution. In this case, symmetry is only required for the round-trip channel seen by the transmitted streams. Since this round-trip channel is a scalar for single-stream transmission, reciprocal RIS performs as non-reciprocal RIS when single-stream transmission is optimal.

The statements of all lemmas, theorems, and corollaries have been formally verified in Lean 4 with the assistance of Claude (Anthropic), and the code is available at: https://github.com/matteonerini/reciprocal-ris.

Reciprocal Versus Non-Reciprocal RIS

In this section, we formally define when reciprocal RIS performs as non-reciprocal RIS, and we derive two lemmas characterizing the operations that each of them can realize.

Let U={Θ∈CM×M:ΘHΘ=IM}\mathcal{U}=\{\boldsymbol{\Theta}\in\mathbb{C}^{M\times M}:\boldsymbol{\Theta}^{H}\boldsymbol{\Theta}=\mathbf{I}_M\} and S={Θ∈U:Θ=ΘT}\mathcal{S}=\{\boldsymbol{\Theta}\in\mathcal{U}:\boldsymbol{\Theta}=\boldsymbol{\Theta}^{T}\} denote the sets of feasible scattering matrices of a non-reciprocal and a reciprocal RIS, respectively.

To compare these two architectures in any wireless system under any performance metric, we first observe how a RIS affects performance. The transmitted signals reach the RIS only through the channel HT\mathbf{H}_T, and the signals reflected by the RIS reach the receivers only through the channel HR\mathbf{H}_R. Therefore, most, if not all, performance metrics depend on Θ\boldsymbol{\Theta} through ΘHT\boldsymbol{\Theta}\mathbf{H}_T, through HRΘ\mathbf{H}_R\boldsymbol{\Theta}, or through both. In this work, we consider metrics of the form f(ΘHT)f(\boldsymbol{\Theta}\mathbf{H}_T), where ff can be any real-valued metric, such as capacity, sum-rate, max-min fairness, or energy efficiency, in which all other design variables, such as precoders and combiners, have already been optimized.

For metrics of the form f(HRΘ)f(\mathbf{H}_R\boldsymbol{\Theta}), all results of this work hold with HT\mathbf{H}_T replaced by HRT\mathbf{H}_R^T, since HRΘ=(ΘTHRT)T\mathbf{H}_R\boldsymbol{\Theta}=(\boldsymbol{\Theta}^T\mathbf{H}_R^T)^T and both U\mathcal{U} and S\mathcal{S} are closed under transposition. Metrics depending on HRΘHT\mathbf{H}_R\boldsymbol{\Theta}\mathbf{H}_T, such as any function of the channel H\mathbf{H}, are of both forms, so all results of this work apply to them both as they are and with HT\mathbf{H}_T replaced by HRT\mathbf{H}_R^T. Note that metrics that are functions of ΘHT\boldsymbol{\Theta}\mathbf{H}_T but not of the whole HRΘHT\mathbf{H}_R\boldsymbol{\Theta}\mathbf{H}_T are rare but exist. For example, in a two-operator system [16], a RIS controlled by one operator must not alter the channels of the other operator, imposing a fixed ΘHT,2\boldsymbol{\Theta}\mathbf{H}_{T,2} in [16 Equation (2)]. These constraints involve ΘHT\boldsymbol{\Theta}\mathbf{H}_T but not the whole HRΘHT\mathbf{H}_R\boldsymbol{\Theta}\mathbf{H}_T.

We assume that the maximum of the considered metric over U\mathcal{U} is attained, which holds, e.g., when the metric is continuous, since U\mathcal{U} is compact. Since S⊂U\mathcal{S}\subset\mathcal{U}, a non-reciprocal RIS always performs at least as well as a reciprocal RIS. We are interested in when a reciprocal RIS is enough to achieve optimal performance, as formalized by the following definition.

Definition 1. We say that reciprocal RIS performs as non-reciprocal RIS when there exists Θ⋆∈S\boldsymbol{\Theta}^{\star}\in\mathcal{S} such that

f(Θ⋆HT)=max⁡Θ∈Uf(ΘHT).(5)f(\boldsymbol{\Theta}^{\star}\mathbf{H}_T)=\max_{\boldsymbol{\Theta}\in\mathcal{U}}f(\boldsymbol{\Theta}\mathbf{H}_T). \tag*{(5)}

Definition 1 suggests a simple strategy to verify whether reciprocal RIS can perform as non-reciprocal RIS. Let ΘU⋆∈U\boldsymbol{\Theta}^{\star}_{\mathcal{U}}\in\mathcal{U} be an optimal non-reciprocal RIS, i.e., a solution to the right-hand side of Eq. (5). Since ff depends on Θ\boldsymbol{\Theta} only through the matrix ΘHT\boldsymbol{\Theta}\mathbf{H}_T, if a reciprocal RIS ΘS⋆∈S\boldsymbol{\Theta}^{\star}_{\mathcal{S}}\in\mathcal{S} can produce the same matrix, i.e., ΘS⋆HT=ΘU⋆HT\boldsymbol{\Theta}^{\star}_{\mathcal{S}}\mathbf{H}_T=\boldsymbol{\Theta}^{\star}_{\mathcal{U}}\mathbf{H}_T, then it achieves optimal performance, and (5) holds. Therefore, comparing reciprocal and non-reciprocal RIS reduces to answering the following question: which matrices ΘHT\boldsymbol{\Theta}\mathbf{H}_T can be produced with Θ∈U\boldsymbol{\Theta}\in\mathcal{U}, and which with Θ∈S\boldsymbol{\Theta}\in\mathcal{S}? We answer this question for non-reciprocal RIS in Lemma 1 and for reciprocal RIS in Lemma 2.

Lemma 1. Let X,Y∈CM×K\mathbf{X},\mathbf{Y}\in\mathbb{C}^{M\times K}. There exists Θ∈U\boldsymbol{\Theta}\in\mathcal{U} such that ΘX=Y\boldsymbol{\Theta}\mathbf{X}=\mathbf{Y} if and only if

XHX=YHY.(6)\mathbf{X}^{H}\mathbf{X}=\mathbf{Y}^{H}\mathbf{Y}. \tag*{(6)}

Proof. Although Lemma 1 is a classical result [17 Theorem 7.3.11], we report its proof for completeness, as its structure is mirrored in the proof of Lemma 2.

Necessity. If ΘX=Y\boldsymbol{\Theta}\mathbf{X}=\mathbf{Y} with ΘHΘ=IM\boldsymbol{\Theta}^{H}\boldsymbol{\Theta}=\mathbf{I}_M, then YHY=XHΘHΘX=XHX\mathbf{Y}^{H}\mathbf{Y}=\mathbf{X}^{H}\boldsymbol{\Theta}^{H}\boldsymbol{\Theta}\mathbf{X}=\mathbf{X}^{H}\mathbf{X}.

Sufficiency. We show that if (6) holds, then it is always possible to construct a unitary Θ\boldsymbol{\Theta} such that ΘX=Y\boldsymbol{\Theta}\mathbf{X}=\mathbf{Y}. Let rr be the rank of X\mathbf{X} and write its singular value decomposition (SVD) as X=U1ΣV1H\mathbf{X}=\mathbf{U}_1\boldsymbol{\Sigma}\mathbf{V}_1^H, where U1∈CM×r\mathbf{U}_1\in\mathbb{C}^{M\times r} and V1∈CK×r\mathbf{V}_1\in\mathbb{C}^{K\times r} have orthonormal columns and Σ∈Rr×r\boldsymbol{\Sigma}\in\mathbb{R}^{r\times r} is diagonal with positive entries. Complete U1\mathbf{U}_1 to an M×MM\times M unitary matrix U=[U1,U2]\mathbf{U}=[\mathbf{U}_1,\mathbf{U}_2].

Step 1 (introduce a unitary W). Define W1=YV1Σ−1∈CM×r\mathbf{W}_1=\mathbf{Y}\mathbf{V}_1\boldsymbol{\Sigma}^{-1}\in\mathbb{C}^{M\times r}. We have

W1HW1=Σ−1V1HYHYV1Σ−1(7)\mathbf{W}_1^H\mathbf{W}_1=\boldsymbol{\Sigma}^{-1}\mathbf{V}_1^H\mathbf{Y}^H\mathbf{Y}\mathbf{V}_1\boldsymbol{\Sigma}^{-1} \tag*{(7)}
W1HW1=Σ−1V1HXHXV1Σ−1(8)\phantom{\mathbf{W}_1^H\mathbf{W}_1}=\boldsymbol{\Sigma}^{-1}\mathbf{V}_1^H\mathbf{X}^H\mathbf{X}\mathbf{V}_1\boldsymbol{\Sigma}^{-1} \tag*{(8)}
W1HW1=Σ−1V1HV1Σ2V1HV1Σ−1(9)\phantom{\mathbf{W}_1^H\mathbf{W}_1}=\boldsymbol{\Sigma}^{-1}\mathbf{V}_1^H\mathbf{V}_1\boldsymbol{\Sigma}^{2}\mathbf{V}_1^H\mathbf{V}_1\boldsymbol{\Sigma}^{-1} \tag*{(9)}
W1HW1=Ir.(10)\phantom{\mathbf{W}_1^H\mathbf{W}_1}=\mathbf{I}_r. \tag*{(10)}

where (a) follows from YHY=XHXY^{H}Y=X^{H}X, (b) follows from XHX=V1Σ2V1HX^{H}X=V_{1}\Sigma^{2}V_{1}^{H}, and (c) follows from V1HV1=IrV_{1}^{H}V_{1}=I_{r}. Therefore, W1W_{1} has orthonormal columns, and we can complete it to an M×MM \times M unitary matrix W=[W1,W2]W=[W_{1},W_{2}].

Step 2 (construct a unitary Θ\Theta). Define

Θ=WUH=W1U1H+W2U2H,(11)\Theta=WU^{H}=W_{1}U_{1}^{H}+W_{2}U_{2}^{H}, \tag*{(11)}

which is unitary as a product of unitary matrices.

Step 3 (verify that ΘX=Y\Theta X=Y). Since U1HU1=IrU_{1}^{H}U_{1}=I_{r} and U2HU1=0U_{2}^{H}U_{1}=0, we have

ΘX=W1ΣV1H=YV1V1H.(12)\Theta X=W_{1}\Sigma V_{1}^{H}=YV_{1}V_{1}^{H}. \tag*{(12)}

It remains to show YV1V1H=YYV_{1}V_{1}^{H}=Y. Let P=IK−V1V1HP=I_{K}-V_{1}V_{1}^{H}, so that XP=0XP=0. Using YHY=XHXY^{H}Y=X^{H}X, we obtain

(YP)H(YP)=PYHYP(13)(YP)^{H}(YP)=PY^{H}YP \tag*{(13)}
=PXHXP(14)=PX^{H}XP \tag*{(14)}
=(XP)H(XP)=0.(15)=(XP)^{H}(XP)=0. \tag*{(15)}

Hence YP=0YP=0, which gives YV1V1H=YYV_{1}V_{1}^{H}=Y. □

Lemma 1 states that a unitary Θ\Theta can map the columns of X=[x1,…,xK]X=[x_{1},\ldots,x_{K}] to the columns of Y=[y1,…,yK]Y=[y_{1},\ldots,y_{K}] if and only if all their norms and inner products are preserved, i.e., xk1Hxk2=yk1Hyk2x_{k_{1}}^{H}x_{k_{2}}=y_{k_{1}}^{H}y_{k_{2}}, for k1,k2=1,…,Kk_{1},k_{2}=1,\ldots,K. For K=1K=1, condition (6) reduces to ∥x∥=∥y∥\lVert x\rVert=\lVert y\rVert. Hence, a unitary Θ\Theta can map any vector xx to any vector yy with the same norm. For K≥2K\geq2, it is instructive to consider the case where the columns of XX are orthogonal, so that XHXX^{H}X is diagonal. In this case, (6) holds if and only if the columns of YY are also orthogonal and ∥xk∥=∥yk∥\lVert x_{k}\rVert=\lVert y_{k}\rVert for k=1,…,Kk=1,\ldots,K. Hence, a unitary Θ\Theta can map an arbitrary set of orthogonal vectors to an arbitrary set of orthogonal vectors with the same norms. We now show how this flexibility is reduced when Θ\Theta is also required to be symmetric.

Lemma 2. Let X,Y∈CM×KX,Y\in\mathbb{C}^{M\times K}. There exists Θ∈S\Theta\in\mathcal{S} such that ΘX=Y\Theta X=Y if and only if

XHX=YHY,(16)X^{H}X=Y^{H}Y, \tag*{(16)}
XTY=(XTY)T.(17)X^{T}Y=(X^{T}Y)^{T}. \tag*{(17)}

Proof. This lemma implicitly appears in the proof of [18 Theorem 2] (see [18 Section IV-B]). Here, we report a self-contained proof that constructs a Θ∈S\Theta\in\mathcal{S} solving ΘX=Y\Theta X=Y, similar to the proof of Lemma 1.

Necessity. If ΘX=Y\Theta X=Y with ΘHΘ=IM\Theta^{H}\Theta=I_{M}, then YHY=XHΘHΘX=XHXY^{H}Y=X^{H}\Theta^{H}\Theta X=X^{H}X. If, in addition, Θ=ΘT\Theta=\Theta^{T}, then XTY=XTΘXX^{T}Y=X^{T}\Theta X is symmetric.

Sufficiency. We show that if (16) holds, then it is always possible to construct a symmetric unitary Θ\Theta such that ΘX=Y\Theta X=Y. To this end, we exploit the fact that ΘX=Y\Theta X=Y also implies ΘY∗=X∗\Theta Y^{*}=X^{*} when Θ\Theta is symmetric unitary, since Θ−1=Θ∗\Theta^{-1}=\Theta^{*}. We therefore construct a symmetric unitary Θ\Theta satisfying both equations at once, i.e., ΘX~=Y~\Theta\widetilde{X}=\widetilde{Y}, where X~=[X,Y∗]\widetilde{X}=[X,Y^{*}], Y~=[Y,X∗]∈CM×2K\widetilde{Y}=[Y,X^{*}]\in\mathbb{C}^{M\times2K}. Let rr be the rank of X~\widetilde{X} and write its SVD as X~=U1ΣV1H\widetilde{X}=U_{1}\Sigma V_{1}^{H}, where U1∈CM×rU_{1}\in\mathbb{C}^{M\times r} and V1∈C2K×rV_{1}\in\mathbb{C}^{2K\times r} have orthonormal columns and Σ∈Rr×r\Sigma\in\mathbb{R}^{r\times r} is diagonal with positive entries. Complete U1U_{1} to an M×MM\times M unitary matrix U=[U1,U2]U=[U_{1},U_{2}]. Introducing the symmetric permutation matrix

Π=[0IKIK0],(18)\Pi=\begin{bmatrix} 0 & I_{K}\\ I_{K} & 0 \end{bmatrix}, \tag*{(18)}

we have

Y~=X~∗Π=U1∗ΣV1TΠ.(19)\widetilde{Y}=\widetilde{X}^{*}\Pi=U_{1}^{*}\Sigma V_{1}^{T}\Pi. \tag*{(19)}

From the definitions of X~\widetilde{X} and Y~\widetilde{Y}, if (16) holds, we have

X~HX~=Y~HY~.(20)\widetilde{X}^{H}\widetilde{X}=\widetilde{Y}^{H}\widetilde{Y}. \tag*{(20)}

Substituting the SVD expressions of X~\widetilde{X} and Y~\widetilde{Y} into (19) gives

V1Σ2V1H=ΠV1∗Σ2V1TΠ.(21)V_{1}\Sigma^{2}V_{1}^{H}=\Pi V_{1}^{*}\Sigma^{2}V_{1}^{T}\Pi. \tag*{(21)}

Step 1 (introduce a symmetric unitary CC). Define the r×rr\times r matrices

N=V1TΠV1,C=ΣNΣ−1.(22)N=V_{1}^{T}\Pi V_{1},\qquad C=\Sigma N\Sigma^{-1}. \tag*{(22)}

NN is symmetric because Π\Pi is symmetric. In the following, we show that CC is symmetric unitary. Multiplying (20) by V1TΠV_{1}^{T}\Pi on the left and by V1V_{1} on the right gives NΣ2=Σ2NN\Sigma^{2}=\Sigma^{2}N. Multiplying it on the left by Σ−1\Sigma^{-1} and on the right by Σ−1\Sigma^{-1} yields Σ−1NΣ=ΣNΣ−1\Sigma^{-1}N\Sigma=\Sigma N\Sigma^{-1}, showing that CC is symmetric. Furthermore, multiplying (20) by V1HV_{1}^{H} on the left and by V1V_{1} on the right gives Σ2=NHΣ2N\Sigma^{2}=N^{H}\Sigma^{2}N. Thus, CHC=Σ−1NHΣ2NΣ−1=IrC^{H}C=\Sigma^{-1}N^{H}\Sigma^{2}N\Sigma^{-1}=I_{r}, proving that CC is unitary.

Step 2 (construct a symmetric unitary Θ\Theta). Define

Θ=U∗[C00IM−r]UH,(23)\Theta=U^{*} \begin{bmatrix} C & 0\\ 0 & I_{M-r} \end{bmatrix} U^{H}, \tag*{(23)}

which is unitary as a product of unitary matrices, and symmetric because the middle matrix is symmetric and UH=(U∗)TU^{H}=(U^{*})^{T}.

Step 3 (verify that ΘX~=Y~\Theta\widetilde{X}=\widetilde{Y}). Since UHU1=[Ir,0]TU^{H}U_{1}=[I_{r},0]^{T}, we have

ΘX~=U∗CΣV1H=U1∗ΣV1TΠV1V1H=Y~V1V1H.(24)\Theta\widetilde{X}=U^{*}C\Sigma V_{1}^{H}=U_{1}^{*}\Sigma V_{1}^{T}\Pi V_{1}V_{1}^{H}=\widetilde{Y}V_{1}V_{1}^{H}. \tag*{(24)}

It remains to show Y~V1V1H=Y~\widetilde{Y}V_{1}V_{1}^{H}=\widetilde{Y}. Let P=I2K−V1V1HP=I_{2K}-V_{1}V_{1}^{H}, so that X~P=0\widetilde{X}P=0. Using (19), we get (Y~P)H(Y~P)=0(\widetilde{Y}P)^{H}(\widetilde{Y}P)=0. Hence, Y~P=0\widetilde{Y}P=0, which gives Y~V1V1H=Y~\widetilde{Y}V_{1}V_{1}^{H}=\widetilde{Y}. We have finally obtained ΘX~=Y~\Theta\widetilde{X}=\widetilde{Y}, whose first KK columns are ΘX=Y\Theta X=Y. □

Lemma 2 states that ΘX=Y\Theta X=Y has a symmetric unitary solution if and only if it satisfies, in addition to (16a), also the symmetry condition (16b), i.e., xk1Tyk2=xk2Tyk1x_{k_{1}}^{T}y_{k_{2}}=x_{k_{2}}^{T}y_{k_{1}}, for k1,k2=1,…,Kk_{1},k_{2}=1,\ldots,K. For K=1K=1, condition (16a) reduces to ∥x∥=∥y∥\lVert x\rVert=\lVert y\rVert, and (16b) is a scalar identity that always holds. Hence, a symmetric unitary Θ\Theta can map any vector xx to any vector yy with the same norm, exactly as a unitary Θ\Theta. For K≥2K\geq2, (16b) imposes K(K−1)/2K(K-1)/2 additional constraints. Therefore, unlike a unitary Θ\Theta, a symmetric unitary Θ\Theta cannot map an arbitrary set of orthogonal vectors to an arbitrary set of orthogonal vectors with the same norms.

A Necessary and Sufficient Condition

In this section, we derive a necessary and sufficient condition under which reciprocal RIS performs as non-reciprocal RIS.

Notice that reciprocal RIS performs as non-reciprocal RIS when it can reproduce the matrix Θ⋆HT\boldsymbol{\Theta}^{\star}\mathbf{H}_{T} produced by an optimal non-reciprocal RIS Θ⋆\boldsymbol{\Theta}^{\star}. Lemma 2 states when this is possible. Introducing X=HT\mathbf{X}=\mathbf{H}_{T} and Y=Θ⋆HT\mathbf{Y}=\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}, condition (16a) is always satisfied since Θ⋆\boldsymbol{\Theta}^{\star} is unitary. Therefore, the only obstacle for a reciprocal RIS is condition (16b), which requires XTY=HTTΘ⋆HT\mathbf{X}^{T}\mathbf{Y}=\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T} to be symmetric. This observation is formalized in the following theorem.

Theorem 1. Reciprocal RIS performs as non-reciprocal RIS if and only if there exists an optimal non-reciprocal RIS Θ⋆∈U\boldsymbol{\Theta}^{\star}\in\mathcal{U} for which HTTΘ⋆HT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T} is symmetric.

Proof. Necessity. If (5) holds, Θ⋆∈S\boldsymbol{\Theta}^{\star}\in\mathcal{S} is an optimal non-reciprocal RIS (since S⊂U\mathcal{S}\subset\mathcal{U}), and HTTΘ⋆HT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T} is symmetric because Θ⋆\boldsymbol{\Theta}^{\star} is.

Sufficiency. Assume there exists an optimal non-reciprocal RIS ΘU⋆∈U\boldsymbol{\Theta}^{\star}_{U}\in\mathcal{U} such that HTTΘU⋆HT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T} is symmetric. We show that a reciprocal RIS can produce the same matrix ΘHT\boldsymbol{\Theta}\mathbf{H}_{T}, and therefore achieve optimal performance. To this end, consider X=HT\mathbf{X}=\mathbf{H}_{T} and Y=ΘU⋆HT\mathbf{Y}=\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T}. Since ΘU⋆\boldsymbol{\Theta}^{\star}_{U} is unitary, we have YHY=HTHHT=XHX\mathbf{Y}^{H}\mathbf{Y}=\mathbf{H}_{T}^{H}\mathbf{H}_{T}=\mathbf{X}^{H}\mathbf{X}. In addition, XTY=HTTΘU⋆HT\mathbf{X}^{T}\mathbf{Y}=\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T} is symmetric. Therefore, by Lemma 2, there exists a reciprocal RIS ΘS⋆∈S\boldsymbol{\Theta}^{\star}_{S}\in\mathcal{S} such that ΘS⋆HT=ΘU⋆HT\boldsymbol{\Theta}^{\star}_{S}\mathbf{H}_{T}=\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T}, achieving optimal performance.

Theorem 1 can be physically interpreted as follows. Since HTT\mathbf{H}_{T}^{T} is the channel from the RIS back to the transmitting antennas, HTTΘHT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}\mathbf{H}_{T} is the round-trip channel from the transmitting antennas, via the RIS, back to the transmitting antennas, as shown in Figure 1(b). Its (k1,k2)(k_{1},k_{2})th entry is the channel from the k2k_{2}th to the k1k_{1}th transmitting antenna via the RIS, so its symmetry means that this channel is the same in both directions, i.e., the round-trip channel is reciprocal. A reciprocal RIS can never break this reciprocity, and Theorem 1 shows that this is the only limitation of reciprocal RIS that matters. In other words, reciprocal RIS performs as non-reciprocal RIS if and only if optimal performance does not require breaking the reciprocity of this round-trip channel. Note that for metrics of the form f(HRΘ)f(\mathbf{H}_{R}\boldsymbol{\Theta}), an analogous result holds with HRΘ⋆HRT\mathbf{H}_{R}\boldsymbol{\Theta}^{\star}\mathbf{H}_{R}^{T} in place of HTTΘ⋆HT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}, i.e., the round-trip channel from the receiving antennas, via the RIS, back to the receiving antennas. For metrics that are functions of HRΘHT\mathbf{H}_{R}\boldsymbol{\Theta}\mathbf{H}_{T}, which are of both forms f(ΘHT)f(\boldsymbol{\Theta}\mathbf{H}_{T}) and f(HRΘ)f(\mathbf{H}_{R}\boldsymbol{\Theta}), either of the two conditions can be equivalently used.

A notable case where the round-trip channel HTTΘHT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}\mathbf{H}_{T} is always symmetric is when HT\mathbf{H}_{T} has rank one, as stated in the following corollary.

Corollary 1. If rank⁡(HT)=1\operatorname{rank}(\mathbf{H}_{T})=1, then reciprocal RIS performs as non-reciprocal RIS.

Proof. If rank⁡(HT)=1\operatorname{rank}(\mathbf{H}_{T})=1, we have HT=abT\mathbf{H}_{T}=\mathbf{a}\mathbf{b}^{T} for some a∈CM×1\mathbf{a}\in\mathbb{C}^{M\times1} and b∈CNT×1\mathbf{b}\in\mathbb{C}^{N_{T}\times1}, so that

HTTΘHT=b(aTΘa)bT=(a)(aTΘa)bbT,(25)\mathbf{H}_{T}^{T}\boldsymbol{\Theta}\mathbf{H}_{T}=\mathbf{b}\left(\mathbf{a}^{T}\boldsymbol{\Theta}\mathbf{a}\right)\mathbf{b}^{T}\overset{(a)}{=}\left(\mathbf{a}^{T}\boldsymbol{\Theta}\mathbf{a}\right)\mathbf{b}\mathbf{b}^{T}, \tag*{(25)}

where =(a)\overset{(a)}{=} follows from aTΘa\mathbf{a}^{T}\boldsymbol{\Theta}\mathbf{a} being a scalar. Since HTTΘHT\mathbf{H}_{T}^{T}\boldsymbol{\Theta}\mathbf{H}_{T} is symmetric, this corollary follows from Theorem 1. □

Corollary 1 includes all systems with a single transmitting antenna (NT=1N_{T}=1), regardless of the number of receiving antennas. It also includes all systems where all transmitting antennas belong to one device in line-of-sight (LoS) and far-field conditions with the RIS, since HT\mathbf{H}_{T} is then the outer product of the steering vectors of the RIS and of the device. Similarly, for metrics of the form f(HRΘ)f(\mathbf{H}_{R}\boldsymbol{\Theta}), the same holds if rank⁡(HR)=1\operatorname{rank}(\mathbf{H}_{R})=1, e.g., with a single receiving antenna (NR=1N_{R}=1) or a single receiving device in LoS and far-field conditions with the RIS. For metrics that are functions of HRΘHT\mathbf{H}_{R}\boldsymbol{\Theta}\mathbf{H}_{T}, reciprocal RIS performs as non-reciprocal RIS if either HT\mathbf{H}_{T} or HR\mathbf{H}_{R} has rank one. Hence, single-user SISO, SIMO, and MISO systems never benefit from non-reciprocal RIS.

A Necessary and Sufficient Condition in the Presence of Precoding

In this section, we refine the condition in Theorem 1 by accounting for the transmitter precoder.

Consider a transmitted signal x=Fs\mathbf{x}=\mathbf{F}\mathbf{s}, where s∈CK×1\mathbf{s}\in\mathbb{C}^{K\times1} contains K≤NTK\leq N_{T} symbols (data streams) and F∈F⊆CNT×K\mathbf{F}\in\mathcal{F}\subseteq\mathbb{C}^{N_{T}\times K} is the precoder, with F\mathcal{F} being the set of feasible precoders, e.g., satisfying a power constraint. Since the received signal is y=HRFs+n\mathbf{y}=\mathbf{H}_{R}\mathbf{F}\mathbf{s}+\mathbf{n}, we consider metrics that depend on Θ\boldsymbol{\Theta} through ΘHTF\boldsymbol{\Theta}\mathbf{H}_{T}\mathbf{F}. We therefore write them as g(F,ΘHTF)g(\mathbf{F},\boldsymbol{\Theta}\mathbf{H}_{T}\mathbf{F}), where gg is real-valued and can also depend on F\mathbf{F} directly, e.g., through HRF\mathbf{H}_{R}\mathbf{F}. We assume that the maximum of the metric over U×F\mathcal{U}\times\mathcal{F} is attained, which holds, e.g., when gg is continuous and F\mathcal{F} is compact.

Definition 2. We say that reciprocal RIS performs as non-reciprocal RIS in the presence of precoding when there exists a pair (Θ⋆,F⋆)∈S×F(\boldsymbol{\Theta}^{\star},\mathbf{F}^{\star})\in\mathcal{S}\times\mathcal{F} such that

g(F⋆,Θ⋆HTF⋆)=max⁡Θ∈U,F∈Fg(F,ΘHTF).(26)g(\mathbf{F}^{\star},\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}\mathbf{F}^{\star})=\max_{\boldsymbol{\Theta}\in\mathcal{U},\mathbf{F}\in\mathcal{F}}g(\mathbf{F},\boldsymbol{\Theta}\mathbf{H}_{T}\mathbf{F}). \tag*{(26)}

Note that a reciprocal RIS now only needs to reproduce the matrix Θ⋆HTF⋆\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}\mathbf{F}^{\star} produced by an optimal pair of non-reciprocal RIS Θ⋆\boldsymbol{\Theta}^{\star} and precoder F⋆\mathbf{F}^{\star}, rather than the whole matrix Θ⋆HT\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}. This leads to the following necessary and sufficient condition, which is weaker than the one in Theorem 1.

Theorem 2. Reciprocal RIS performs as non-reciprocal RIS in the presence of precoding if and only if there exists an optimal pair of non-reciprocal RIS Θ⋆∈U\boldsymbol{\Theta}^{\star}\in\mathcal{U} and precoder F⋆∈F\mathbf{F}^{\star}\in\mathcal{F} for which F⋆THTTΘ⋆HTF⋆\mathbf{F}^{\star T}\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}\mathbf{F}^{\star} is symmetric.

Proof. Necessity. If (25) holds, (Θ⋆,F⋆)∈S×F(\boldsymbol{\Theta}^{\star},\mathbf{F}^{\star})\in\mathcal{S}\times\mathcal{F} is an optimal pair for the non-reciprocal RIS problem (since S⊆U\mathcal{S}\subseteq\mathcal{U}), and F⋆THTTΘ⋆HTF⋆\mathbf{F}^{\star T}\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}\mathbf{F}^{\star} is symmetric because Θ⋆\boldsymbol{\Theta}^{\star} is.

Sufficiency. Assume there exists an optimal pair (ΘU⋆,F⋆)∈U×F(\boldsymbol{\Theta}^{\star}_{U},\mathbf{F}^{\star})\in\mathcal{U}\times\mathcal{F} of the non-reciprocal RIS problem such that F⋆THTTΘU⋆HTF⋆\mathbf{F}^{\star T}\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T}\mathbf{F}^{\star} is symmetric. We show that a reciprocal RIS can produce the same ΘHTF⋆\boldsymbol{\Theta}\mathbf{H}_{T}\mathbf{F}^{\star}, and therefore achieve optimal performance with the same precoder. To this end, consider X=HTF⋆\mathbf{X}=\mathbf{H}_{T}\mathbf{F}^{\star} and Y=ΘU⋆HTF⋆\mathbf{Y}=\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T}\mathbf{F}^{\star}. Since ΘU⋆\boldsymbol{\Theta}^{\star}_{U} is unitary, we have YHY=XHX\mathbf{Y}^{H}\mathbf{Y}=\mathbf{X}^{H}\mathbf{X}. In addition, XTY=F⋆THTTΘU⋆HTF⋆\mathbf{X}^{T}\mathbf{Y}=\mathbf{F}^{\star T}\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}_{U}\mathbf{H}_{T}\mathbf{F}^{\star} is symmetric. Therefore, by Lemma 2, there exists a reciprocal RIS ΘS⋆∈S\boldsymbol{\Theta}^{\star}_{S} \in\mathcal{S} such that ΘS⋆HTF⋆=ΘJ⋆HTF⋆\boldsymbol{\Theta}^{\star}_{S}\mathbf{H}_{T}\mathbf{F}^{\star} = \boldsymbol{\Theta}^{\star}_{J}\mathbf{H}_{T}\mathbf{F}^{\star}. The pair (ΘS⋆,F⋆)(\boldsymbol{\Theta}^{\star}_{S},\mathbf{F}^{\star}) is feasible for the reciprocal RIS problem and achieves optimal performance.

The condition in Theorem 2 involves the K×KK \times K round-trip channel seen by the transmitted streams, i.e., from the KK streams, via the RIS, back to the KK streams, as shown in Figure 1(c). This is weaker than the condition in Theorem 1, since F⋆TMF⋆\mathbf{F}^{\star T}\mathbf{M}\mathbf{F}^{\star} is symmetric whenever M\mathbf{M} is symmetric, but not vice versa. Thus, a reciprocal RIS only needs to preserve the reciprocity of the round-trip channel along the directions actually used for transmission. Similarly, if the receiving antennas use a combiner G⋆∈CK×NR\mathbf{G}^{\star} \in\mathbb{C}^{K \times N_R}, Theorem 2 holds with G⋆HRΘ⋆HRTG⋆T\mathbf{G}^{\star H}\mathbf{R}\boldsymbol{\Theta}^{\star}\mathbf{H}_{R}^{T}\mathbf{G}^{\star T} in place of F⋆THTTΘ⋆HTF⋆\mathbf{F}^{\star T}\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}\mathbf{F}^{\star}.

A notable case where the condition of Theorem 2 always holds is single-stream transmission, as stated in the following.

Corollary 2. If single-stream transmission (K=1K = 1) is optimal with non-reciprocal RIS, then reciprocal RIS performs as non-reciprocal RIS.

Proof. Since F⋆THTTΘ⋆HTF⋆\mathbf{F}^{\star T}\mathbf{H}_{T}^{T}\boldsymbol{\Theta}^{\star}\mathbf{H}_{T}\mathbf{F}^{\star} is a scalar for K=1K = 1, and therefore symmetric, this corollary follows from Theorem 2.

Corollary 2 extends Corollary 1 to systems where HT\mathbf{H}_{T} and HR\mathbf{H}_{R} have arbitrary rank but only one stream is transmitted. This includes, for example, single-user multiple-input multiple-output (MIMO) systems at low signal-to-noise ratio (SNR) and multi-user MISO systems maximizing the sum-power, where transmission over the dominant channel eigenmode is optimal. In addition, this also includes multi-user MISO systems maximizing the sum-rate at low SNR, where only the strongest user is served. In these systems, reciprocal RIS performs as non-reciprocal RIS, consistent with the optimal reciprocal RIS solutions derived for single-stream systems in [19, 20].

Conclusion

We have answered the question of whether reciprocal RIS can perform as non-reciprocal RIS. This happens if and only if there exists an optimal non-reciprocal RIS that makes the round-trip channel from the transmitting antennas, via the RIS, back to the transmitting antennas symmetric. When the precoder is taken into account, this condition only needs to hold for the round-trip channel seen by the transmitted streams. As a result, reciprocal RIS performs as non-reciprocal RIS when the channel from the transmitting antennas to the RIS has rank one or when single-stream transmission is optimal. In these systems, the additional hardware complexity of non-reciprocal RIS brings no performance gain.

Conversely, non-reciprocal RIS is beneficial when optimal performance requires breaking the reciprocity of the round-trip channel. The clearest example is a full-duplex system, where a base station simultaneously receives from an uplink user and transmits to a downlink user located in a different direction, both via the RIS [12, 13, 14, 15]. Here, a reciprocal RIS forces the channel from the uplink user to the base station to be equal to the reverse one, while optimal performance requires the former to be strong and the latter weak, since the base station signal should be reflected towards the downlink user. More generally, non-reciprocal RIS is expected to be beneficial when numerous streams are transmitted through well-conditioned channels [6], e.g., when the RIS serves multiple transmitters and receivers in different directions, or in rich scattering environments.

References

  1. [1]Q. Wu, S. Zhang, B. Zheng, C. You, and R. Zhang, “Intelligent reflecting surface-aided wireless communications: A tutorial,” IEEE Trans. Commun., vol. 69, no. 5, pp. 3313–3351, 2021.arxiv.org/abs/2007.02759
  2. [2]S. Zhang and R. Zhang, “Capacity characterization for intelligent reflecting surface aided MIMO communication,” IEEE J. Sel. Areas Commun., vol. 38, no. 8, pp. 1823–1838, 2020.arxiv.org/abs/1910.01573
  3. [3]D. M. Pozar, Microwave Engineering. John wiley & sons, 2011.
  4. [4]H. Li, M. Nerini, S. Shen, and B. Clerckx, “A tutorial on beyond-diagonal reconfigurable intelligent surfaces: Modeling, architectures, system design and optimizations, and applications,” IEEE Commun. Surv. Tutor., vol. 28, pp. 4086–4126, 2026.arxiv.org/abs/2505.16504
  5. [5]S. Shen, B. Clerckx, and R. Murch, “Modeling and architecture design of reconfigurable intelligent surfaces using scattering parameter network analysis,” IEEE Trans. Wireless Commun., vol. 21, no. 2, pp. 1229–1243, 2022.DOI
  6. [6]J. V. Alegría, J. Thunberg, and O. Edfors, “Channel orthogonalization with reconfigurable surfaces: General models, theoretical limits, and effective configuration,” IEEE Trans. Wireless Commun., vol. 24, no. 6, pp. 5179–5195, 2025.
  7. [7]G. Bartoli, A. Abrardo, N. Decarli, D. Dardari, and M. Di Renzo, “Spatial multiplexing in near field MIMO channels with reconfigurable intelligent surfaces,” IET Signal Process., vol. 17, no. 3, p. e12195, 2023.arxiv.org/abs/2212.11057
  8. [8]J. Xu, H. Wang, R. Liu, J. A. Nossek, and A. L. Swindlehurst, “Non-reciprocal reconfigurable intelligent surfaces,” IEEE Wireless Commun. Lett., vol. 14, no. 10, pp. 3329–3333, 2025.arxiv.org/abs/2411.15617
  9. [9]H. Wang, Z. Han, and A. L. Swindlehurst, “Channel reciprocity attacks using intelligent surfaces with non-diagonal phase shifts,” IEEE Open J. Commun. Soc., vol. 5, pp. 1469–1485, 2024.arxiv.org/abs/2309.11665
  10. [10]H. Wang, J. Hu, J. Xu, Y. Ju, and A. L. Swindlehurst, “DRL-based mitigation against nonreciprocal RIS-aided channel reciprocity attacks,” in ICC 2026 - IEEE International Conference on Communications, 2026.DOI
  11. [11]H. Wang, J. Hu, J. Xu, Y. Ju, and A. L. Swindlehurst, “Nonreciprocal RIS-aided covert channel reciprocity attacks and countermeasures,” arXiv preprint arXiv:2509.11117, 2025.arxiv.org/abs/2509.11117
  12. [12]H. Li and B. Clerckx, “Non-reciprocal beyond diagonal RIS: Multiport network models and performance benefits in full-duplex systems,” IEEE Trans. Commun., vol. 73, no. 11, pp. 12211–12234, 2025.arxiv.org/abs/2411.04370
  13. [13]Z. Liu, H. Li, and B. Clerckx, “Non-reciprocal beyond diagonal RIS: Sum-rate maximization in full-duplex communications,” IEEE Trans. Commun., vol. 74, pp. 5169–5183, 2026.arxiv.org/abs/2411.18523
  14. [14]Z. Liu and B. Clerckx, “A secure full-duplex wireless circulator enabled by non-reciprocal beyond-diagonal RIS,” IEEE J. Sel. Areas Commun., vol. 44, pp. 3750–3765, 2026.arxiv.org/abs/2507.23381
  15. [15]Z. Liu, W. Shin, and B. Clerckx, “Enabling full-duplex LEO satellite systems with non-reciprocal BD-RIS-assisted beamforming,” IEEE Trans. Wireless Commun., vol. 25, pp. 15459–15473, 2026.arxiv.org/abs/2511.22525
  16. [16]Z. Wu, M. Nerini, and B. Clerckx, “RIS optimization and scaling laws in multi-operator systems: Is quadratic scaling achievable?” arXiv preprint arXiv:2605.01389, 2026.arxiv.org/abs/2605.01389
  17. [17]R. A. Horn and C. R. Johnson, Matrix analysis. Cambridge University Press, 2012.
  18. [18]Z. Wu and B. Clerckx, “Beyond-diagonal RIS in multiuser MIMO: Graph theoretic modeling and optimal architectures with low complexity,” IEEE Trans. Inf. Theory, vol. 71, no. 11, pp. 8506–8523, 2025.arxiv.org/abs/2502.16509
  19. [19]M. Nerini, S. Shen, and B. Clerckx, “Closed-form global optimization of beyond diagonal reconfigurable intelligent surfaces,” IEEE Trans. Wireless Commun., vol. 23, no. 2, pp. 1037–1051, 2024.arxiv.org/abs/2211.06117
  20. [20]I. Santamaria, M. Soleymani, E. Jorswieck, and J. Gutiérrez, “SNR maximization in beyond diagonal RIS-assisted single and multiple antenna links,” IEEE Signal Process. Lett., vol. 30, pp. 923–926, 2023.DOI

Paper details

vlnc:W_sxg95jvg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents