WorksEngineering
Movable Beyond-Diagonal RIS for ISAC: Optimal Placement and Pareto Boundary
Abstract
This letter investigates integrated sensing and communication using a movable beyond-diagonal reconfigurable intelligent surface (BD-RIS). Under a rank-one line-of-sight BS–RIS channel, we derive the exact communication–sensing Pareto boundary of a fully-connected BD-RIS in terms of user–target channel correlation. For line-of-sight RIS links, maximizing this correlation yields the optimal placement. We characterize the minimum track aperture that eliminates the illumination trade-off and show that a movable diagonal RIS then attains the same optimum. Below this threshold, a dynamic program offers certified placement accuracy. Numerical results examine hardware connectivity, angular mismatch, and Rician fading.
Introduction
INTEGRATED sensing and communication (ISAC) lets one waveform and one aperture serve users and probe targets [1]. Beyond-diagonal reconfigurable intelligent surfaces (BD-RISs) interconnect their elements through a reconfigurable impedance network, which generalizes the diagonal phase-shift model to symmetric unitary scattering matrices [2], [3]. BD-RIS-aided ISAC outperforms diagonal RIS in sensing SNR and throughput [4], [5], but with elements at fixed positions and through alternating optimization.
A fixed surface, however, faces a limitation that no choice of scattering matrix can remove. When the BS reaches the surface over a line-of-sight link, this link has rank one, and the BS can only scale the incident field, not shape it. The surface must then serve the user and the target from a single field. With spacing, the steering vectors toward two well-separated directions are nearly orthogonal, and their correlation decays as the surface grows. In this case each function can be favored only at the expense of the other. Interconnections widen the set of reflected fields but do not change this geometry, and adding elements makes it worse. Moving the elements changes the geometry itself, which raises two questions this letter answers: where should the elements go, and how much hardware do we still need once they are placed? The second question also has a cost dimension: a fully connected (FC) BD-RIS needs tunable impedances, whereas a diagonal RIS needs only .
The element position is a second degree of freedom [6]. RISs with movable elements improve ISAC [7], [8]. Xu et al. [9] combined movement with BD interconnections and found numerically, for the multiuser sum rate, that moving and interconnecting each win in a different regime. For ISAC, it is not yet known how the BD-RIS elements should be placed, nor whether moving and interconnecting complement or replace each other.
This letter answers these questions for a movable BD-RIS that serves one user and illuminates one target. Our contributions are:
We show that an FC BD-RIS turns the surface into a field synthesizer, so the Pareto boundary takes the two-direction form known from MISO beamforming [10] and is governed only by the user–target correlation . For group-connected and diagonal RISs, the supported part of the boundary follows from an exact two-dimensional search.
Under LoS, placements with larger Pareto-dominate those with smaller , so maximizing is optimal for the FC BD-RIS. The boundary reduces to the single-function optimum iff the aperture exceeds , using a lattice that places a grating lobe on the target; there, a movable diagonal RIS is also optimal. Below , a dynamic program (DP) with a certified accuracy bound and a closed-form clustered lattice give the placement.
Simulations validate the analysis, compare against a fixed sparse array of the same aperture under geometry mismatch, and evaluate Rician fading with an absolute user-SNR requirement.
Notation: , , denote conjugate, transpose and Hermitian transpose; is the subvector of indexed by ; .
C.K. Sheemar and S. Chatzinotas are with the Interdisciplinary Centre for Security, Reliability and Trust (SnT), University of Luxembourg, L-1855 Luxembourg (e-mail: [email protected]; [email protected]). S. Solanki is with the National Institute of Technology Warangal, TS 506004, India (e-mail: [email protected]).
System Model
In Figure 1, a BS with antennas serves a single-antenna user and senses a point target through an -element BD-RIS, since the direct links are blocked. The elements slide along a linear track of length , with positions in

Figure 1. A movable BD-RIS serves a user and illuminates a target whose direct links to the BS are blocked; the target echo returns to the BS over the same surface.
where is the minimum spacing. Let . The BS–RIS link is LoS, , where is the BS steering vector. The RIS–target link is , and the RIS–user link is unless stated otherwise. We assume ; otherwise the two channels are collinear for every placement and no trade-off exists.
The scattering matrix is lossless and reciprocal, and , and is realized by a real symmetric reactance matrix as [2]. A group-connected BD-RIS splits the elements into groups of size , making block diagonal; is the FC BD-RIS and the diagonal RIS. Unlike [9], where interconnected elements move as one unit, every element moves individually and the lossless network is assumed to follow it1.
The BS transmits with covariance , under the power budget . The user receives , . By reciprocity, the target echo at the BS is
with reflection coefficient and .
Lemma 1 (Transmit design): For and any and , a single stream with maximum-ratio transmission (MRT), , maximizes both the user SNR and the echo SNR.
Proof: Write , where carries the user data and dedicated sensing streams. The user receives the useful power and the interference , while the echo energy in (2) is proportional to . Hence , and with equality iff .
Lemma 1 is a direct consequence of the rank-one BS–RIS link; it shows that all spatial processing happens at the surface. The incident field is , with . With the reflected field , the user SNR and the target illumination are
Since for , the echo becomes . Combining with over known snapshots gives SNR. For each placement , we study the achievable region , where is the feasible set of for the given architecture, its Pareto boundary, and the placement that is best for it.
Analysis
Reachable Fields and the Pareto Boundary
Lemma 2: For a group-connected BD-RIS, . In particular, an FC BD-RIS reaches every with , and a diagonal RIS every with .
Proof: Necessity follows from the unitarity of each block. For sufficiency, take one block and drop the index. Choose with and , e.g., with and . The pairs and have the same Gram matrix, so a unitary exists with and . Then is symmetric and unitary, and .
Lemma 2 concerns ideal scattering matrices. The reactance is finite iff is not an eigenvalue of . When it is, e.g., for , the field is approached with growing reactances. Since (3) depends on only up to a common phase, can be rotated, and is free on the orthogonal complement.
Let and , so that and , where and are the single-function maxima. Define
Theorem 1: For an FC BD-RIS and arbitrary with , the Pareto boundary of is
attained by , with , and from Lemma 2. Formula (5) also holds for , with , and for , where the boundary is the single point attained by .
Proof: For , and write , with of unit norm. Then with of unit norm. Then , with equality iff , i.e., . Points with are dominated by .
This is the two-direction structure of the MISO Pareto boundary [10]. The new element is Lemma 2, which makes a passive FC BD-RIS behave like a fully digital transmitter.
Corollary 1: (a) For and a requirement , the maximum illumination is , which equals iff . (b) is convex, hence a weighted-sum boundary point of and [11], and . Both increase with .
Proof: (a) means in (5), and increases with on . (b) The maximum is the largest eigenvalue of , whose nonzero eigenvalues are those of a Gram matrix with trace and determinant .
For group-connected and diagonal RISs, Lemma 2 turns the weighted sum into over a product of spheres, with .
Proposition 1: For any group partition,
and a maximizer is with , where any with may be used if . Since has two real degrees of freedom up to a common phase, every supported point of the boundary follows from a two-dimensional search.
Proof. . For fixed , the Cauchy–Schwarz inequality on each block gives , with equality at the stated . Exchanging the two maximizations gives (6).
When the region is nonconvex, as for the diagonal RIS in Figure 2, its unsupported points are computed numerically: we maximize under over the same family with local refinement, and check the result against random-restart SQP over all phases. We make no global-optimality claim for these points.

Figure 2. Pareto boundaries for two movable apertures and the fixed array (). Thick lines: FC (Theorem 1); thin dashed: diagonal (Proposition 1), whose largest gap to FC is 0.16, 0.04 and 0.003; markers: direct optimization of ; green: .
Line of Sight: Positions Act Only Through
Theorem 2: Under LoS, , and do not depend on , and
independently of the BS–RIS geometry. For an FC BD-RIS, if , every point of is weakly Pareto-dominated by a point of . Hence every criterion that is nondecreasing in and is maximized by
Proof. All entries have constant modulus, so the norms are fixed and (7) follows from (4). Every point of is dominated by a point of (5), and for the point of (5) is dominated by the point of the boundary for .
Dominance cannot be replaced by set inclusion: for , is the segment , which does not contain .
Theorem 3: Let . Then iff for all , and
attained by . At this placement, the Pareto boundary of the FC BD-RIS reduces to the single point : the user and the target are served as if each were alone.
Proof. Equality in the triangle inequality for Eq. (7) holds iff all phases coincide modulo . Since , , so ordered lattice points are feasible iff their span fits in . With , Theorem 1 gives .
Proposition 2 (Clustered lattice): Let divide with . Placing clusters of elements, spaced within each cluster, at the lattice points achieves
with .
Hence for all , and recovers Theorem 3.
Proof. The sum in (7) factorizes into the cluster array factor, which gives (10), and the lattice factor, which has unit modulus. Clusters do not overlap because .
Corollary 2 (Diagonal RIS): Under LoS, , so a diagonal RIS realizes every constant-modulus . (a) The endpoints of (5), and , have constant modulus, so a diagonal RIS attains them for any . (b) For , a movable diagonal RIS on the lattice attains , and interconnections bring no further gain.
For , the diagonal region depends on all the phases , not only on : for , and , the placements and both give , but their best values of are 1.25 and 1.457. The placement (8) is therefore optimal for the FC BD-RIS only. For diagonal and group-connected RISs below , we use it as an achievable design.
Corollary 3 (Minimum aperture): Let . The FC BD-RIS meets without any sensing loss iff . Moreover, , with equality for , and is nondecreasing because for .
Remark 1 (Position errors): For i.i.d. errors on , Jensen’s inequality gives . For and , the factor is , so the lattice is robust to small positioning errors.
Placement for D<D and Beyond LoS
For a smaller aperture, (8) is nonconvex. Using ,
For fixed , the inner problem is separable apart from the ordering constraints in (1). On a grid of step with and , the recursion
evaluated with running maxima, solves the inner grid problem exactly in operations.
Algorithm 1 Movable BD-RIS ISAC design
Input: Angles , aperture , requirement
Placement: If , set (Theorem 3); otherwise solve (11) by the DP (12). For non-LoS channels, refine by coordinate ascent.
Transmitter: MRT toward the RIS (Lemma 1).
Field: FC: (Theorem 1); group-connected or diagonal: search over in (6).
Scattering matrix: Synthesize and blockwise by Lemma 2.
Proposition 3 (Accuracy): Let be the best returned by (12) over equispaced values of . Then .
Proof. Rounding every position of a continuous optimizer down to the grid keeps it in , because , and changes each phase by at most . The grid optimum therefore loses at most in . For the grid optimum , some lies within of , and the DP at returns with .
We use with local refinement around the best , so the factor is 0.9995. For , the analytical lattice of Theorem 3 is used directly.
For non-LoS channels, Theorem 1, Corollary 1 and Proposition 1 still hold, but depends on . For an absolute requirement , Corollary 1(a) with gives a closed-form objective in and . If , the user is served at . We improve this objective by exact grid updates of each between its neighbours, initialized at the lattice, accepting only strict improvements. The search is monotone and terminates, but it is local. Algorithm 1 summarizes the design.
Complexity: Step 1 costs , step 3 costs for the FC BD-RIS and for search points otherwise, and step 4 costs over all groups. Unlike [4], [5], [7], [8], no alternating optimization is needed.
Simulation Results
In this section, we asses the perforamnce of the proposed design. Table I lists the simulation parameters. Under LoS, and do not depend on , so we report metrics normalized by these single-function maxima; they hold for any transmit power , number of BS antennas , path loss and reflection coefficient . Since SNRs , the sensing SNR loss is dB.
Pareto boundaries: Figure 2 compares and with the fixed array. The FC curves are exact by Theorem 1. As a consistency check, the markers optimize directly by L-BFGS from random starts, without using Lemma 2. The FC markers agree with (5) to within . The diagonal markers (best of 150 starts) lie on the supported points from Proposition 1. The fixed array has nearly orthogonal channels (); movement raises to 0.39 and 0.81, pushing the boundary outward. For , the Pareto boundary of both the FC and the diagonal RIS is the corner (green star), while the achievable region is the green dotted segment, as
| Parameter | Symbol | Value |
| RIS elements | 16 (32 in Figs. 4, 3b, 3c) | |
| Minimum spacing | ||
| Carrier frequency | 28 GHz ( cm) | |
| BS–RIS angle of arrival | ||
| User / target angle | ([] in Fig. 3b) | |
| Spatial-frequency offset | 0.516 | |
| Lattice spacing | ||
| Trade-off-free aperture | (60.1 for ) | |
| Fixed baseline | — | -spaced array |
| DP grid / phase search | / 96 | |
| User SNR loss tolerance | 1 dB, 3 dB | |
| Group size | 4 | |
| Rician RIS–user link (Fig. 3c) | ||
| Rician factor | to 25 dB | |
| Scattered paths | 5, angles uniform in | |
| User requirement | 3 dB below LoS maximum | |
| Position step / realizations | — | / 100 |
Table 1. Simulation parameters.
Theorem 3 predicts. The diagonal boundary touches the FC boundary at its endpoints (Corollary 2); for the fixed array it is nonconvex. The figure thus separates the two degrees of freedom: interconnection matters only at fixed positions, where the largest gap between the FC and diagonal boundaries is 0.16, whereas movement shrinks this gap to 0.04 and 0.003 at and and removes it at , while enlarging the region of both architectures far more.
Moving vs. interconnecting: Fig. 3a shows the sensing SNR loss when the user may lose 1 dB or 3 dB of SNR. The fixed FC BD-RIS loses 12.7 and 5.5 dB; the fixed diagonal RIS (not shown) loses 17.1 and 8.5 dB. With movement, the FC loss vanishes once (Corollary 3), at and . With the same placement, the diagonal RIS matches the FC BD-RIS from these apertures on and trails by at most 0.61 and 0.29 dB for . A group-connected RIS with stays within 0.14 dB of the FC BD-RIS.
Geometry mismatch: In Fig. 3b, the aperture is fixed at for the nominal angle, and the user must keep its full SNR (, so ). The comparison isolates the value of movability at equal aperture: a fixed sparse array on the nominal lattice is optimal at , but away from it its illumination stays below 0.53 () and 0.03 ( Fig. 3. Sensing performance of the proposed architecture: (a) impact of aperture size and user SNR requirements; (b) target-angle mismatch; (c) robustness under Rician fading.
Figure 3a. Sensing SNR loss versus aperture for user SNR losses of 1 dB (solid) and 3 dB (dashed), .
Figure 3b. Normalized illumination at full user SNR versus target angle, at equal aperture designed for .
Figure 3c. Average sensing SNR loss versus the Rician factor of the RIS–user link (absolute user requirement 3 dB below the LoS maximum, 100 realizations, 95% confidence intervals).
Figure 4. Maximum correlation versus aperture for ; dotted lines mark (Theorem 3), green circles Proposition 2.
32), close to the array. The movable surface, re-placed by the DP for each , keeps for , where , and above when the target moves toward the user and .
Rician fading: In Fig. 3c, the RIS–user link has scattered paths at angles uniform in , and the user requires an absolute SNR 3 dB below its LoS maximum. Results are averaged over 100 realizations with 95% confidence intervals. The LoS lattice of Theorem 3 loses 1.1 dB at a Rician factor of dB and less than 0.2 dB from 0 dB on. Coordinatewise re-optimization reduces the loss to at most 0.05 dB (FC) and 0.14 dB (diagonal) for both , while the fixed array loses 4.7–5.5 dB. The requirement cannot be met in at most 3% of the realizations with the re-optimized placement and 7% with the fixed array; in those cases the user is served at its maximum.
Phase transition: Fig. 4 plots for . The DP curves approach 1 as , where the analytical lattice gives exactly . At the DP attains for both , whereas the best uniform spacing stays at 0.08 () and 0.04 (). The plateau is the clustered lattice of Proposition 2 with (circles): the DP returns pairs of elements on the lattice and attains at (29.6 for ).
Conclusion
For a movable BD-RIS serving one user and illuminating one target, the communication–sensing Pareto boundary is available in closed form and is governed by a single quantity, the correlation between the user and target channels at the surface. Under LoS, the placement that maximizes this correlation is optimal for the FC BD-RIS. A lattice of spacing steers a grating lobe at the target and reduces the Pareto boundary to the single-function optimum once ; there, a movable diagonal RIS is also optimal. Beyond this aperture, a DP with certified accuracy and a clustered lattice give the placement, and the diagonal RIS remains within a fraction of a decibel of the FC BD-RIS at moderate apertures. The gains hold at equal aperture under geometry mismatch and, after local re-optimization, under Rician fading.
References
- [1]F. Liu, Y. Cui, C. Masouros, J. Xu, T. X. Han, Y. C. Eldar, and S. Buzzi, “Integrated sensing and communications: Toward dual-functional wireless networks for 6G and beyond,” IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1728–1767, Jun. 2022.DOI
- [2]S. Chen, 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, Feb. 2022.
- [3]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, Feb. 2024.arxiv.org/abs/2211.06117
- [4]T. Esmaeilbeig, K. V. Mishra, and M. Soltanalian, “Beyond diagonal RIS: Key to next-generation integrated sensing and communications?” IEEE Signal Process. Lett., vol. 32, pp. 216–220, 2025.arxiv.org/abs/2402.14157
- [5]Z. Liu, Y. Liu, S. Shen, Q. Wu, and Q. Shi, “Enhancing ISAC network throughput using beyond diagonal RIS,” IEEE Wireless Commun. Lett., vol. 13, no. 6, pp. 1670–1674, Jun. 2024.
- [6]L. Zhu, W. Ma, and R. Zhang, “Movable antennas for wireless communication: Opportunities and challenges,” IEEE Commun. Mag., vol. 62, no. 6, pp. 114–120, Jun. 2024.arxiv.org/abs/2306.02331
- [7]J. Ye, P. Zhang, X.-P. Li, L. Huang, and Y. Liu, “Joint beamforming and position optimization for fluid RIS-aided ISAC systems,” IEEE Trans. Commun., vol. 74, pp. 114 422–114 438, 2026.arxiv.org/abs/2501.13339
- [8]X. Peng, Q. Tao, Y. L. Guan, and X. Chen, “Exploiting movable elements of intelligent reflecting surface for enhancement of integrated sensing and communication,” IEEE Trans. Wireless Commun., vol. 25, pp. 1717–1733, 2026.arxiv.org/abs/2507.23296
- [9]S. Xu, M. Nerini, and B. Clerckx, “Movable beyond-diagonal reconfigurable intelligent surfaces: MOK, interconnecting, or both?” IEEE Wireless Commun. Lett., vol. 15, pp. 2144–2148, 2026.
- [10]E. A. Jorswieck, E. G. Larsson, and D. Danev, “Complete characterization of the Pareto boundary for the MISO interference channel,” IEEE Trans. Signal Process., vol. 56, no. 10, pp. 5292–5296, Oct. 2008.DOI
- [11]R. A. Horn and C. R. Johnson, Topics in Matrix Analysis. Cambridge, U.K.: Cambridge Univ. Press, 1991.



