WorksComputer Science
Mutual-Coupling-Aware Joint Precoding and Antenna Positioning for Movable-Antenna MU-MIMO Systems
Abstract
Movable antennas (MAs) improve wireless communications by adapting antenna positions to the propagation environment. However, antenna movement also changes the mutual coupling (MC) among elements of the transmit and receive arrays, which affects the performance of MA-enabled systems. In this paper, we investigate MC-aware joint precoding and antenna positioning for a downlink multiuser multiple-input multiple-output (MU-MIMO) system with MAs at both the base station and users. We establish an MC-aware system model by combining field-response propagation with a circuit-based representation of the transmit and receive arrays. Based on this model, we formulate a weighted sum-rate (WSR) maximization problem that jointly optimizes multi-stream precoders and antenna positions. The problem is challenging because the precoders and positions are coupled through the propagation channels and position-dependent MC matrices. To address this challenge, we reformulate the problem in the antenna-current domain to to efficiently handle the MC matrices and develop a current-domain weighted minimum mean square error (WMMSE) framework for alternating optimization. Furthermore, we handle the nonconvex position updates within the framework by using majorization–minimization (MM) and solve the resulting convex subproblems with a dedicated conic alternating direction method of multipliers (ADMM) algorithm. Simulation results demonstrate that the proposed design improves the WSR over both an MC-aware fixed array and a movable-array design that ignores MC during optimization.
The authors are with the Department of Electronic and Computer Engineering, Hong Kong University of Science and Technology, Hong Kong SAR. Emails: [email protected], [email protected], [email protected], {eesshong, eejzhang, eekhaled}@ust.hk.
Introduction
Sixth-generation (6G) wireless networks are envisioned to support diverse applications, from immersive communication to intelligent connected systems, with increasingly demanding data rate and connectivity requirements [1, 2, 3]. Multiuser multiple-input multiple-output (MU-MIMO) systems use multiple antennas to serve users simultaneously, with precoding at the base station (BS) and receive processing managing interference [4, 5]. Their performance depends on the propagation channels, but conventional arrays have fixed element positions and cannot adapt them to changing channel conditions. Recently, reconfigurable antenna systems, including movable antennas (MAs) and fluid antenna systems (FASs), have emerged as a means of addressing this limitation by adapting antenna positions to the propagation environment [6, 7, 8, 9]. By introducing antenna position as a design dimension, these systems offer greater flexibility in using the available spatial resources. The antenna positions can be configured alongside other communication parameters to improve system performance under different operating conditions.
In an MA system, antenna movement changes both the propagation channels and the electromagnetic interactions within the arrays. Specifically, movement changes the phases of multipath components and reshapes the channels [6, 10], while also changing the element separations and hence the mutual coupling (MC) within the transmit and receive arrays [11, 12]. MA system design therefore needs to account for both effects of antenna movement. In particular, MC affects the effective channels, the transmit power associated with the antenna currents, and the receive noise covariance [11, 13]. Antenna positions that improve the propagation channels may also change these coupling-dependent quantities, so their overall impact on performance cannot be assessed from propagation alone. In a multiuser system, these effects influence both desired signals and interference, and must therefore be considered when adapting antenna positions and precoding to serve the users. This motivates the joint design of precoding and transmit and receive antenna positions for MA-enabled MU-MIMO systems with MC.
Related Works
Existing studies relevant to this problem have explored antenna position optimization in MA-enabled MIMO systems and the effects of MC on communication performance. The former examines how position adaptation can improve communication performance, often together with transceiver design. The latter provides models for relating electromagnetic interactions to communication performance and includes recent work on position-dependent MC in movable arrays.
Optimization of MA-enabled MIMO and MU-MIMO systems:
The field-response model provides a framework for describing how antenna movement changes a multipath channel [6, 10]. Based on this model, the authors of [6] investigated point-to-point MIMO capacity by jointly optimizing the transmit and receive MA positions and the transmit covariance matrix. For MIMO-FAS, New et al. [7] studied joint port selection and transceiver design and analyzed the associated information-theoretic performance. In multiuser communication, antenna positioning can improve spatial separation between users as well as their individual channel strengths. Zhu et al. [14] demonstrated this benefit by optimizing antenna positions and receive combining for multiuser transmission.
Joint design of beamforming and antenna positions has since been studied under different multiuser objectives and system configurations. In particular, Feng et al. [15] employed weighted minimum mean square error (WMMSE) and majorization–minimization (MM) methods to optimize beamforming and MA positions for weighted sum-rate (WSR) maximization. For FAS-assisted MU-MIMO, Liao et al. [16] combined matrix fractional programming with parallel position updates and developed a decentralized implementation to reduce computation time. The authors of [17] incorporated transmit and receive antenna movement into a flexible WMMSE beamforming framework, using sparse optimization to select antenna locations. Beyond instantaneous joint design, Zheng et al. [18] considered a two-timescale approach that adapts positions to channel statistics and beamforming to instantaneous channels. The authors of [19] addressed WSR maximization with near-field propagation, while Irshad et al. [20] evaluated MA arrays in wideband multiuser operation, including hardware impairments and their interaction with user loading. These studies demonstrate the versatility of antenna position adaptation, but their optimization models do not explicitly account for position-dependent MC within the arrays.
Mutual coupling in antenna arrays:
MC has long been studied using circuit-based communication models, which describe antenna arrays as multiport networks and connect their electromagnetic behavior to communication performance [13]. For dense holographic arrays, Yuan et al. [21] examined the effects of MC on spatial degrees of freedom and antenna efficiency, while Pizzo and Lozano [22] analyzed coupling and spatial correlation and showed how antenna patterns can be designed to improve information-theoretic performance. In an MA array, coupling is not determined solely by a fixed deployment: it varies with the antenna positions and therefore changes during position optimization. A minimum-spacing constraint can limit coupling and prevent physical overlap, but it does not generally eliminate coupling for arbitrary antenna types and arrangements. Moreover, treating coupling only as an impairment can overlook the performance attainable by accounting for it in array design.
Recent work has begun to investigate this opportunity. The authors of [23] incorporated MC into MA array design and employed a hybrid convex and particle swarm optimization method for sum-rate maximization, while Xu et al. [24] optimized MA positions to enhance array directivity. For planar fluid arrays, Zhang et al. [25] developed a current-domain electromagnetic model incorporating coupling and power constraints. The authors of [26] related circuit-based coupling matrices to field-theoretic kernels for movable and fluid antennas on holographic surfaces. Further studies considered coupling-aware beam synthesis through continuous position refinement [27] and sparse port selection [28]. These array and beam synthesis studies highlight the importance of modeling the relationship between port excitation, antenna currents, and radiation when positions or active ports change.
In our previous work [11], we combined field-response and circuit-based models to study MC-aware capacity maximization in point-to-point MA-MIMO systems. That work showed how antenna movement can jointly reshape the propagation channel and coupling matrices. In a multiuser downlink, Xu et al. [12] considered position-dependent coupling and jointly optimized MA positions and beamforming for sum-rate maximization using greedy search, gradient descent, and fractional programming. The point-to-point formulation in [11] does not involve inter-user interference, whereas the multiuser setting in [12] focuses on beamforming and the positions of the transmit antennas.
These studies show that antenna movement can reshape both propagation channels and array coupling. In a multiuser downlink, the same transmit and receive antenna positions therefore influence desired signals and inter-user interference, as well as coupling-dependent transmit power and receive noise. Jointly accounting for these effects changes the position-design problem beyond propagation-channel adaptation. However, joint optimization of multi-stream precoders and antenna positions at both the BS and users under position-dependent MC remains an open problem.
Contributions
In this paper, we investigate MC-aware WSR maximization in a downlink MA-enabled MU-MIMO system and develop an alternating framework for joint precoding and antenna positioning. Our main contributions are as follows.
We establish an MC-aware MU-MIMO model by combining field-response propagation with a circuit-based representation of the transmit and receive arrays. The model captures the coupled effects of movement on propagation channels and array interactions, including the resulting changes in accepted transmit power and receive noise. Based on this model, we formulate a WSR maximization problem that jointly optimizes multi-stream precoders and antenna positions under power, movement region, and minimum-spacing constraints.
To address the position-dependent inverse square roots of the coupling matrices in the effective channels, we develop an equivalent current-domain representation using transmit antenna currents and receive port observations. This representation avoids the matrix inverse square roots while retaining MC in the transmit power and receive noise covariance. The WSR remains non-concave and couples the precoders and positions. We therefore establish an equivalent WMMSE formulation and develop an alternating framework for their joint optimization.
For fixed antenna positions, the precoders can be updated using WMMSE, but the antenna position subproblems remain nonconvex. We therefore develop adaptive phase graph majorization for the position subproblems. We exploit the channel phase and pairwise MC structures to construct quadratic upper bounds for the objective and transmit power. Combining these bounds with inner approximations of the spacing constraints yields convex subproblems. We develop a conic alternating direction
method of multipliers (ADMM) algorithm that exploits their shared structure and closed-form cone projections to efficiently solve the subproblems. The remainder of this paper is organized as follows. Section II present the system model and problem formulation. Section III introduces the proposed optimization framework. Numerical results are provided in Section IV, and conclusions are drawn in Section V.
System Model and Problem Formulation
As illustrated in Figure 1, we consider a downlink MU-MIMO system consisting of one BS and users. The BS is equipped with movable antenna elements, while user is equipped with movable antenna elements. Each element has a fixed orientation and a single feed port, and we use the same index for an element and its associated port. The BS and all users are located in the far-field region of each other. The BS transmits independent data streams to user , where . The transmit antennas can move within a prescribed two-dimensional region . Let denote the position of transmit antenna , and collect the transmit antenna positions as
Figure 1. MA-enabled MU-MIMO downlink system. In the upper panel, gray and colored traces show the highlighted configurations before and after antenna movement. The movement changes path phases () and inter-element spacings (), producing channel shaping and MC variation that jointly affect the WSR. The lower panel gives the circuit-based current-domain representation using transmit antenna currents and receive port observation . The details of this representation are presented in Section II-B.
Similarly, the receive antennas of user can move within a prescribed two-dimensional region . With denoting the position of receive antenna , the receive antenna positions of user are collected as
Field-Response Channel Model
We adopt the field-response model to characterize the physical propagation channel [6], [10]. Let and denote the numbers of transmit and receive propagation paths associated with user , respectively. The projections of the departure and arrival directions onto the local antenna movement planes are denoted by , , and , , respectively. For a projected departure direction , the transmit steering vector is
where is the wavenumber and is the wavelength. Similarly, the receive steering vector of user for a projected arrival direction is
Then, the transmit field-response matrix (FRM) associated with user is constructed by stacking the steering vectors of all departure paths as
Likewise, the receive FRM of user is given as
Let denote the path-response matrix of user , whose -th entry characterizes the complex gain from departure path to arrival path . The propagation channel between the BS and user is then given by
Circuit-Based Communication Model
In addition to modifying the propagation phases, antenna movement changes the MC among elements within each array. Due to these interactions, the current at one antenna port contributes to the voltages at other ports, so the antenna elements must be modeled jointly. To capture this behavior, we adopt a circuit-based model that represents each antenna array as a multiport network described by its port voltages and currents [13]. In this representation, the signal sources are connected to the transmit antenna ports through a matching network, and the source voltages produce currents at these ports. These currents generate electromagnetic waves that propagate through the environment and induce voltages at the receive antenna ports. The receive matching network then transforms these induced voltages into the output signals observed by the receivers.
Specifically, the interactions within the arrays can be characterized by the transmit MC matrix and the receive MC matrix . These matrices represent the normalized mutual radiation resistance of the arrays, with off-diagonal entries describing the mutual interactions between distinct elements. Their entries are modeled as
where and are the coupling kernels that map the distance between two elements to the corresponding coupling coefficient. Their functional forms depend on the antenna type and radiation pattern. In this paper, we consider isotropic antenna elements1, for which the coupling kernels are given by [22]
Given the above coupling kernel (10), the MC matrices and are real symmetric and positive definite whenever the antenna positions are pairwise distinct.
The above transmit MC matrix describes the radiated power associated with a given antenna current vector through the normalized radiation resistance of the array. The antenna conductors also dissipate power through ohmic loss. With an identical normalized loss resistance at each transmit element, the normalized total resistance matrix of the transmit array includes both contributions and is given by [22]
Here, corresponds to the absence of ohmic loss, and for all . We omit the position arguments below for notational brevity.
Let denote the data symbol vector intended for user . We assume independent Gaussian data streams, with and for . The BS applies the precoding matrix to form the source signal as
In the circuit model, this precoded transmit signal corresponds to the normalized baseband representation of the source voltages. The source signals are applied to the transmit array through a lossless matching network, which is assumed to maintain perfect power matching as the antenna positions change. Under the matching model and normalization adopted in [11], the normalized transmit antenna current is given by
Therefore, the average power accepted by the transmit array is given as [13]
To describe propagation from the transmitter to the receivers, we adopt the unilateral approximation, which neglects the influence of the receive antenna currents on the transmitter [13]. This allows the transmit currents to be determined independently of the receive circuits. In the normalized coordinates, represents the open-circuit voltages induced at the receive antennas, with constant scaling factors absorbed into (7).
At user , we assume that the receive matching network is lossless and maintain receive noise matching as the antenna positions change. Under the adopted circuit model and normalization, this network transforms the induced voltages by [11]. Thus, the received signal at user is given as
where is the normalized baseband representation of the output voltages after receive matching and before digital combining, and denotes the additive noise in this observation. Based on (15), we define the corresponding effective channel as
To account for both background radiation and receiver hardware noise, we model the equivalent noise covariance referred to the receive ports before matching as
where the first term represents correlated noise from an isotropic background field. The second term models receiver hardware noise referred to the same ports, which we assume to be independent across ports. The transformation associated with receive matching acts on both the signal and this equivalent noise. Accordingly, the noise in (15) follows , with
Then, the achievable rate of user is
where is the covariance of interference and noise, given by
Problem Formulation
In this paper, our goal is to maximize the WSR. Since the transmit and receive antenna positions jointly determine the propagation channels and array coupling, which in turn affect the effective channels and receive noise under the accepted-power constraint. Therefore, we need to jointly optimize the multi-stream precoders and antenna positions, leading to the following problem
Here, in (21) denotes the priority weight of user , and in (22) is the maximum power budget at the BS. Constraints (23) and (25) restrict the transmit and receive antenna positions to their prescribed movement regions and , respectively. The parameters and denote the minimum antenna spacings at the BS and user , respectively. Existing studies often impose a minimum spacing of half a wavelength to mitigate MC and neglect its effects in the communication model. In this work, however, MC is explicitly included in the model and its effects are accounted for in the WSR optimization. Therefore, the minimum antenna spacings can be smaller than half a wavelength, provided that physical collisions are avoided. Constraints (24) and (26) enforce these spacing requirements, with the minimum distances determined by the antenna dimensions and mechanical constraints.
Problem (21) is quite challenging to solve because the precoding matrices and antenna positions are coupled in the non-concave WSR objective through both the desired signals and multiuser interference. Moreover, antenna movement changes not only the propagation channels but also the MC matrices, affecting both the effective channels and the receive noise covariances. In particular, the effective channels involve matrix inverse square roots that depend on the antenna positions, which complicates the derivatives required for position optimization. These difficulties motivate the reformulation and alternating optimization framework developed in the next section.
The Proposed Current-Domain WMMSE with Majorization–Minimization Framework
To address the challenges in solving the WSR maximization problem (21), we develop a current-domain WMMSE framework with majorization–minimization (CD-WMMSE-MM). We first reformulate the problem in the current domain and establish an equivalent WMMSE formulation, which provides the basis for alternating optimization of the precoders and antenna positions. For fixed antenna positions, we derive the current-domain precoder update. For the transmit and receive antenna positions, we construct convex subproblems using majorization–minimization and solve them using ADMM.
Problem Reformulation
Directly solving (21) is challenging because the effective channels involve matrix inverse square roots of the coupling matrices, which vary with the antenna positions. To facilitate optimization, we reformulate the problem in the current domain by introducing the current-domain precoding matrix and equivalent receive port observation as
The matrix maps the data streams to the normalized transmit antenna currents, while expresses the receive observation in the coordinates before matching. Since and are positive definite, both transformations are invertible. Then, substituting (27) and (28) into (15), we obtain
where and
Thus, the achievable rate of user in the current domain can be expressed as
where is the interference-plus-noise covariance matrix in the current domain, given by
Moreover, the accepted transmit power in (14) becomes
Although the above current-domain reformulation simplifies the treatment of MC matrices, directly optimizing the WSR remains difficult because the precoders and antenna positions are still coupled within the non-concave log-determinant rate expressions. We therefore use the standard WMMSE identity [29, 30, 31] to replace these expressions with an equivalent formulation in terms of weighted MSEs. Introducing auxiliary receive matrices and positive definite weight matrices , we obtain
where is the MSE matrix of the estimate , given by
with the aggregate transmit current covariance defined as
Summing (34) with weights , we can obtain the weighted MSE objective as
For any fixed precoders and antenna positions, optimizing the auxiliary variables in (34) yields
Therefore, problem (21) can be equivalently reformulated as the following joint WMMSE minimization problem:
In the following, we develop an alternating optimization algorithm to solve problem (34). For fixed antenna positions, the receiver, auxiliary weight, and precoder blocks can be updated using standard WMMSE steps, as detailed in Section III-B. With these matrices fixed, we next perform one MM step for the transmit antenna positions while holding the receive positions fixed, as detailed in Section III-C. We then perform one MM step for each user’s receive antenna positions using the latest transmit positions, as detailed in Section III-D. The receive-position subproblems are independent across users and can be solved in parallel. These position steps complete one iteration of the alternating algorithm.
Current-Domain Precoder Update
In this step, we update for fixed antenna positions and . First, we minimize the WMMSE objective in (37) with respect to the receive matrices and MSE weights . For each user , the receive matrix subproblem is convex quadratic for fixed , while the MSE weight matrix subproblem minimizes over for fixed . Their first-order optimality condition yields the closed-form updates with given 2 as
where , and is (35) evaluated at and . With , , we define
Then, the current-domain precoder block becomes
Obviously, problem (39) is convex and satisfies Slater’s condition for , we can obtain its optimal solutions by solving the KKT conditions. Specifically, we introduce the Lagrange multiplier for the constraint (40), and then obtain the stationarity condition as
Since , premultiplying (49) by and collecting factors gives
Then, according to (50), we can obtain the current-domain precoder update as
where denotes the Moore–Penrose pseudoinverse. To determine , let the accepted transmit power associated with this update be
Based on the primal feasibility and complementary slackness conditions, the optimal multiplier is the smallest nonnegative value satisfying the power constraint (48), i.e.,
Since is nonincreasing in , we can determine by bisection search. However, each bisection iteration evaluates (51) at a different trial multiplier. A direct implementation therefore requires repeatedly factorizing the multiplier-dependent matrix in this expression. To avoid these repeated factorizations, we use the Cholesky factorization of as
and then compute the eigendecomposition as
Concatenate , where , and define
Using these factors, the concatenated current-domain precoder can be expressed equivalently for as
and the accepted transmit power in (52) becomes
Here, terms with are interpreted as zero. With this eigendecomposition, each bisection trial requires only evaluating the scalar power expression (58), avoiding the direct evaluation of the precoder update (51) and thereby reducing the computational cost.
Transmit Antenna Position Update
In this step, we update while holding the receive antenna positions and the matrices fixed. Thus, , whereas the receive noise covariances remain constant. The fixed current covariance is . For notation brevity, we define
Substituting (35) into (37) and omitting terms independent of both transmit and receive antenna positions, we obtain
The accepted transmit power in (33) is
The transmit antenna position subproblem of (34) is therefore
Problem (63) is hard to solve because of its nonconvex objective and constraints. To address this difficulty, we combine an MM upper bound of the objective [32], [33] with convex inner approximations of these constraints. At each update, we minimize the objective upper bound over an inner feasible set containing the current antenna positions.
Let stack the candidate transmit antenna positions, and let contain their positions before the MM step. Using as the expansion point, we define the displacement , so that the current positions correspond to . Since is fixed during the update, optimizing is equivalent to optimizing . Once the MM step yields a displacement , we recover the updated positions as .
To obtain a convex surrogate objective, we construct a quadratic upper bound in . For a twice continuously differentiable scalar function , if upper bounds its Hessian along each candidate segment from to , Taylor’s theorem gives
Following (64), we first compute the gradient of at and then construct a Hessian upper bound over the candidate update region.
Objective Gradient and MM Upper Bound
The transmit positions affect in (61) through the channel matrices . We first express the differential of with respect to the channel matrices as
For brevity, define the coefficient of the channel variation as
To compute the gradient of with respect to the transmit positions, we next compute the partial derivatives of the channel matrices with respect to the antenna coordinates. Moving transmit antenna affects only the -th column of . For its -coordinate, where , this partial derivative is
where is the -th basis vector, , and is diagonal with entries . Combining this channel derivative with the differential of gives the -th entry of the channel gradient block
Then, evaluating the gradient blocks in (68) at and stacking them in antenna order gives the objective gradient as
Next, we construct a Hessian upper bound. We first rewrite the channel-dependent part of (61) in terms of as
where and . Since each entry of is a complex exponential of a linear position phase, (70) has the real harmonic representation
where indexes the individual nonzero position-dependent terms obtained by expanding (70) and is the sum of constant terms independent of . For each term , and denote its complex coefficient and phase vector, respectively. For a term arising from the quadratic part, the corresponding coefficient and phase vector are
respectively, and for a term arising from the linear part, the corresponding coefficient and phase vector are
respectively. For each harmonic , the Hessian is
The position dependence of this Hessian lies in the scalar cosine term. We bound that term over the phases allowed by the aperture constraints, assuming compact convex apertures. Partition into blocks corresponding to the antenna positions, and define
Then, the phases over the aperture are contained in
The maximum of over this interval therefore bounds the scalar in (75). Since , a nonnegative upper bound on this scalar gives a positive semidefinite upper bound on the harmonic’s Hessian. The following proposition sums these bounds and gives the resulting MM upper bound.
Proposition 1. For in (71), let and . Then, for every , the Hessian satisfies
Consequently, for every such that , we have
Proof. Please refer to Appendix A.
Hence, combining the bound (78) in Proposition 1 with the objective gradient in (69), we obtain the MM upper bound, where , as
Power Gradient and Convex Inner Approximation
The accepted transmit power in (62) depends on the transmit positions through the pairwise distances in , making the power constraint generally nonconvex. To address this, we construct a convex upper bound on the power function and impose the power limit on this bound, yielding a convex inner approximation of the constraint.
Let denote the distance in (8), and let . For distinct antenna positions, the gradient of w.r.t. antenna is computed as
where . Evaluating these gradient blocks at and stacking them in antenna order gives .
Next, we bound the power Hessian over the candidate update region. We replace the nonconvex minimum-spacing constraints with their tangent inner approximations. Let and impose
This constraint implies the original spacing constraint by convexity of the squared norm. Let and . By the Cauchy–Schwarz inequality, the tangent and aperture constraints bound each pairwise distance as
Each pairwise Hessian contribution has radial and tangential eigenvalues and , respectively. The following proposition bounds these eigenvalues over and combines the resulting pairwise bounds.
Proposition 2. For every edge , define
and
For every such that the pairwise distances along the entire segment from to lie in their respective intervals ,
Proof. Please refer to Appendix B.
Hence, combining Proposition 2 with the power gradient gives the convex power upper bound over the update domain defined below:
where . Finally, combining the objective upper bound (80), the power upper bound (87), and the position update domain (82), we obtain the convex surrogate of problem (63) as
where
Now, problem (79) is a convex optimization problem and can be solved using standard convex optimization methods [34]. To efficiently solve this problem, we develop a conic ADMM method in Section III-E.
Receive Antenna Position Update
After the transmit MM step, we hold the updated transmit positions and the matrices fixed and perform one MM step for each user’s receive positions. For user , and both vary with . This position block affects only the -th summand of (61), which we denote by
The receive-position subproblem for user is
where the constraints apply to user . Let and . Define the receive displacement as . The accepted transmit power is independent of the receive positions, and the subproblems are mutually independent for fixed transmit positions.
The objective and minimum-spacing constraints make (92) nonconvex. As in the transmit update, we construct an MM upper bound on the objective and a convex inner approximation of the spacing constraints. Write , where the channel and noise terms consist of the first two terms and the last term of (91), respectively.
All channel quantities below use the updated transmit positions . In particular, we recompute in (66) using the channel after the transmit-position update. For the channel-gradient calculation, let be diagonal with entries and let . The derivative of the -th row of with respect to the -coordinate of receive antenna is
The corresponding channel-gradient block is
Substituting (17) into the noise term of (91), we apply the same differentiation as in (81) to obtain its receive-position gradient blocks, using , the kernel , and the coefficients
The hardware-noise term is position independent and contributes zero gradient. Evaluating the channel and noise gradients at and stacking their antenna blocks gives and , respectively. Adding these two contributions gives the objective gradient
We next construct Hessian upper bounds over the convex receive update domain , obtained from (90) using the receive positions, aperture , and minimum spacing .
To bound the channel Hessian, we rewrite the channel-dependent objective using the fixed transmit FRM . Let and define
The channel-dependent part of (91) can then be written as
The channel term in (99) has the same harmonic structure as (70). Its coefficients and phase vectors follow from the transmit expressions by replacing with . Following Proposition 1, we obtain the channel Hessian upper bound over the receive aperture. For the noise term, the pairwise construction in Proposition 2, using , the coefficients in (95), and the corresponding receive-distance intervals from (83), gives .
The Hessian of is the sum of the channel and noise Hessians, so their upper bounds add. We therefore set
Let . Combining the channel and noise bounds gives the MM upper bound of the entire receive objective:
Minimizing this upper bound over the receive update domain gives
The receive-position subproblems are convex and can be solved in parallel.
ADMM-Based Solution to Position Subproblems
The transmit and receive position subproblems share a common conic structure. We exploit this structure to solve them using ADMM. We present the formulation and updates for the transmit displacement , and the receive counterpart follows from the corresponding coefficients and constraints. To express the transmit subproblem in conic form, we represent the spacing and aperture constraints using affine inequalities and, for disk apertures, second-order cone (SOC) constraints. The SOC is defined as . The remaining transmit power constraint admits a rotated SOC representation. In particular, factoring gives
where . We collect these constraints into
where and is the product of the corresponding cones. For the receive subproblem, the cone product contains only the aperture and spacing blocks.
To solve (104), we introduce and apply two-block ADMM with a scaled dual variable [35]. Let index the ADMM iterations for the fixed surrogate subproblem at outer iteration . With penalty , define
A nonzero null direction of the aperture-constraint blocks would leave all aperture constraints unchanged, contradicting boundedness of the apertures. Hence has full column rank and . We obtain the updates
The product structure of allows the projection to be evaluated separately for each cone, with closed-form projections onto the orthant, SOC, and rotated SOC. The resulting displacement is then used to update the transmit antenna positions, and the receive positions are updated in the same manner.
Overall, the alternating optimization procedure, i.e., CD-WMMSE-MM method, is summarized in Algorithm 1.
Convergence and Complexity Analysis
With a feasible initialization and exact subproblem solutions, the WMMSE updates and the MM-based transmit and receive position updates preserve feasibility and do not increase the WMMSE objective. By the equivalence between the WMMSE and WSR formulations, the WSR is non-decreasing over the outer iterations. Since the WSR is bounded above, the objective value of Algorithm 1 converges. For the precoder update, the eigendecomposition-based computation costs , where and is the number of bisection trials. For a fixed ADMM penalty, the transmit and receive position solves have an aggregate cost of , where and are their iteration counts.
Algorithm 1 The proposed CD-WMMSE-MM method
Input: Feasible , , and
Initialize using (27); set
repeat
Update using (57) with scalar bisection on
Refresh using the updated
Construct and solve the transmit-position subproblem (79) by ADMM to update
Using the updated transmit positions and channels, construct and solve the receive-position subproblems (102) by ADMM in parallel to update
Set
until The relative WSR improvement satisfies the prescribed tolerance
Recover by
Output: , ,
Numerical Results
In this section, we evaluate the WSR performance of the MC-aware joint precoding and antenna positioning design in MA-enabled MU-MIMO systems.
Simulation Setup
We consider a 28 GHz narrowband downlink, for which the wavelength is approximately 10.714 mm. The default system consists of users, transmit antennas, and receive antennas per user. Two data streams are sent to every user, i.e., , and all user weights are set to one. The transmit antennas move in a rectangular aperture, whereas the receive antennas of every user move in a rectangular aperture. The minimum spacing at the BS and users is .
The physical channel follows the clustered Rician structure used in [11]. In addition to the LoS component, three scattering clusters with eight subpaths per cluster are generated, giving 25 paired departure/arrival paths for each user. The Rician factor is 0 dB, the log-normal shadowing standard deviation is 4 dB, and user distances are drawn between 100 m and 300 m. The nominal departure angles of the four users are . Both the transmit and receive MC matrices use the isotropic-element kernel [22]. The normalized transmit ohmic-loss coefficient is . The total isolated-port noise power is dBm; a 5 dB noise figure is used to divide it into the correlated background-noise and independent receiver-hardware-noise components in (17).
The reference SNR is defined at a distance of 200 m, with zero shadowing and before array and beamforming gains. For a requested reference SNR , the accepted-power budget is selected as
where is the path loss at the reference distance. Thus, the instantaneous received SNRs still vary with user distance, shadowing, and small-scale fading.
We compare the proposed MC-aware joint precoding and antenna positioning design with the following baselines.
MC-aware fixed array: The transmit and receive antennas are fixed on uniform grids. We consider both conventional half-wavelength-spaced arrays and compact arrays with grid spacings chosen to satisfy the aperture and minimum-spacing constraints. For each fixed layout, we optimize the precoders and receive matrices using WMMSE [29, 30], with MC explicitly accounted for.
MC-ignored movable array: The transmit and receive antenna positions are jointly optimized without accounting for MC. The resulting positions and current-domain precoders are evaluated under the MC-aware model.
The default maximum number of outer iterations is 200. The relative WSR stopping tolerance is . The absolute and relative stopping tolerances of the inner ADMM solver are and , respectively, and at most 30,000 ADMM iterations are allowed per position update. Unless otherwise stated, each operating point is averaged over 1,000 independent channel realizations.
Convergence Behavior
Figure 2 illustrates the WSR trajectories of CD-WMMSE-MM for one representative channel realization at 0 dB and . All curves are recorded for 200 outer iterations without early stopping. They exhibit a sharp increase during the first few iterations, followed by progressively smaller improvements as the transceiver variables and antenna positions are refined. The WSR is nondecreasing for every array size, and the larger arrays attain higher WSR values in this realization.
Figure 2. Convergence of CD-WMMSE-MM for different numbers of transmit antennas under a representative channel realization at 0 dB reference SNR.
WSR Versus Reference SNR
Figure 3 shows the average WSR versus the reference SNR. The MC-aware joint design consistently outperforms both baselines over the entire range from dB to 10 dB. At 0 dB, it achieves 9.173 bit/s/Hz, compared with 7.552 bit/s/Hz for the conventional half-wavelength fixed array and 8.302 bit/s/Hz for the movable design that ignores MC. These correspond to relative gains of 21.5% and 10.5%, respectively. The absolute gain over the fixed array increases from 1.021 bit/s/Hz at −10 dB to 1.913 bit/s/Hz at 10 dB. Meanwhile, the persistent gap to the MC-ignored design demonstrates that position optimization based only on the field-response channel cannot fully exploit the movable apertures when closely spaced antenna elements are coupled.
Figure 3. Average WSR versus reference SNR.
| Fixed | TX-only | RX-only | Joint | ||
| 16 | 4 | 10.997 | 11.070 | 12.044 | 12.215 |
| 16 | 8 | 14.285 | 14.370 | 15.432 | 15.656 |
| 20 | 4 | 11.685 | 11.690 | 12.788 | 12.803 |
| 20 | 8 | 15.289 | 15.295 | 16.504 | 16.521 |
| 24 | 4 | 11.925 | 11.928 | 13.045 | 13.056 |
| 24 | 8 | 15.610 | 15.614 | 16.864 | 16.875 |
Table 1. AVERAGE WSR (BIT/S/HZ) FOR ANTENNA-MOVEMENT CONFIGURATIONS AT 0 dB REFERENCE SNR.
Ablation of Transmit- and Receive-Side Antenna Movement
Table I compares joint position optimization with the fixed-array, TX-only, and RX-only designs for and at 0 dB. The joint design achieves the highest average WSR in all six settings.
Among the single-block variants, RX-only position optimization captures most of the gain from antenna movement, whereas TX-only position optimization remains close to the fixed feasible grid. At , joint TX–RX position optimization still improves upon RX-only position optimization by 1.4% for both user loads. The incremental gain decreases to 0.06%–0.12% for , although the joint design remains the best in every setting. Thus, receive-side movement provides most of the antenna-movement gain for the larger compact transmit arrays considered here, while jointly optimizing both position blocks yields the highest WSR in these settings.
Impact of the Transmit and Receive Apertures
Figure 4 separately varies the transmit- and receive-aperture scales, with the other aperture held at its nominal size. The same feasible initial antenna positions are used across aperture scales. The fixed-array baseline remains at 8.251 bit/s/Hz because its antenna positions do not change.

Figure 4. (a) WSR performance w.r.t. different transmit aperture scale. (b) WSR performance w.r.t. different receive aperture scale. Average WSR versus transmit- and receive-aperture scales at 0 dB reference SNR.
In Fig. 4??, enlarging the transmit aperture scale from 0.8 to 2 increases the WSR of the MC-aware joint design from 9.595 to 9.982 bit/s/Hz, with most of the gain already realized by scale 1.5. In contrast, Fig. 4?? shows a steadier increase from 9.689 to 10.232 bit/s/Hz as the receive aperture is enlarged. At scale 2, the MC-aware joint design outperforms the fixed grid by 21.0% in the transmit-aperture sweep and by 24.0% in the receive-aperture sweep. The MC-ignored design benefits from a larger feasible region but remains well below the MC-aware joint design throughout both sweeps. These results indicate that the nominal compact apertures already capture a large fraction of the available gain from antenna movement, while additional receive-side space remains particularly useful for jointly improving the desired channel and the coupling-dependent noise covariance.
Scaling With the Number of Transmit Antennas
Fig. 5 studies the effect of increasing under two aperture policies. In Fig. 5??, the physical transmit aperture is kept fixed. The WSR of the MC-aware joint design increases from 9.154 bit/s/Hz at to 13.056 bit/s/Hz at , but the incremental gain gradually decreases as the aperture becomes crowded. The MC-ignored design saturates much earlier: its WSR peaks around and remains near 9 bit/s/Hz for and 24. Consequently, the gain of the MC-aware joint design over the MC-ignored design widens from 10.4% at to 44.9% at . This widening gap highlights the increasing importance of accounting for coupling in the design when more antenna elements are packed into the fixed aperture.
In Fig. 5??, both linear dimensions of the transmit aperture are scaled by , so that the aperture area per transmit antenna remains constant. All three designs then continue to improve with . At , the proposed, fixed-grid, and MC-ignored designs achieve 14.712, 13.466, and 12.132 bit/s/Hz, respectively. Comparing the two panels shows that saturation at large under a fixed aperture is consistent with the effects of denser antenna packing and stronger MC. MC-aware position optimization mitigates this effect, while enlarging the physical aperture provides an additional gain.
Scaling With the Number of Users
Fig. 6 reports the average WSR for , with and the nominal departure directions uniformly spread from −60° to 60°. The aggregate WSR increases with for all designs. The MC-aware joint design remains above both baselines, and its advantage over the MC-ignored movable design grows from 25.3% at to 33.6% at .
Conclusion
This paper investigated joint precoding and antenna positioning for MA-enabled MU-MIMO systems with position-dependent MC. We combined field-response and circuit-based models and developed a current-domain WMMSE framework with MM-based position updates and a conic ADMM solver. Numerical results demonstrated the WSR benefits of the proposed design over both MC-aware fixed arrays and movable arrays designed without considering MC. The
Figure 5. (a) Fixed transmit aperture. (b) Fixed transmit-antenna density. Average WSR versus the number of transmit antennas at 0 dB reference SNR. At fixed antenna density, both transmit-aperture dimensions scale as .
Figure 6. Average WSR versus the number of users at 0 dB reference SNR with .
current-domain formulation also clarifies the different roles of transmit and receive antenna positioning: transmit positions affect the power required to generate the antenna currents, while receive positions affect the noise covariance, alongside their respective effects on propagation. The impact of these coupling-dependent quantities is governed by the transmit-current covariance and receive processing. Consequently, the antenna positions should be optimized together with the precoders and receive processing, accounting for their combined effects on desired signals, interference, transmit power, and receive noise.
Appendix
Proof of Proposition 1
For any , the bounds in (76) place the phase in . The definition of therefore bounds the scalar coefficient in (75) from above, and, since , we obtain . Summing these inequalities over gives the Hessian bound in (78). Moreover, convexity of ensures that the line segment from to any feasible lies entirely in , so the same Hessian bound holds throughout the segment. Applying this bound to the integral remainder in Taylor’s theorem bounds the remainder by , where the factor follows from . Substituting this bound into the Taylor expansion gives (79). This completes the proof.
Proof of Proposition 2
Because is real symmetric and is Hermitian, the accepted transmit power can be written as
Differentiating the edge potentials gives the gradient in (81). For one edge, the Hessian with respect to , with and , is
Its radial and tangential eigenvalues are and , respectively, so (84) gives
The contribution of this edge to the stacked position Hessian is
Because , summing the edge bounds gives whenever all pairwise distances lie in their respective intervals . The distance condition in Proposition 2 ensures that this Hessian bound holds throughout the update segment, so Taylor’s theorem with integral remainder gives (86). This completes the proof.
References
References
- [1]K. B. Letaief, W. Chen, Y. Shi, J. Zhang, and Y.-J. A. Zhang, “The roadmap to 6G: AI empowered wireless networks,” IEEE Commun. Mag., vol. 57, no. 8, pp. 84–90, Aug. 2019.arxiv.org/abs/1904.11686
- [2]W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems: Applications, trends, technologies, and open research problems,” IEEE Network, vol. 34, no. 3, pp. 134–142, 2020.arxiv.org/abs/1902.10265
- [3]X. Lin, J. Zhang, Y. Liu, and J. Kim, Eds., Fundamentals of 6G communications and networking. Cham, Switzerland: Springer, 2024.DOI
- [4]E. G. Larsson, O. Edfors, F. Tufvesson, and T. L. Marzetta, “Massive MIMO for next generation wireless systems,” IEEE Commun. Mag., vol. 52, no. 2, pp. 186–195, Feb. 2014.DOI
- [5]L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, “An overview of massive MIMO: Benefits and challenges,” IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp. 742–758, Oct. 2014.DOI
- [6]W. Ma, L. Zhu, and R. Zhang, “MIMO capacity characterization for movable antenna systems,” IEEE Trans. Wireless Commun., vol. 23, no. 4, pp. 3392–3407, Apr. 2024.arxiv.org/abs/2210.05396
- [7]W. K. New, K.-K. Wong, H. Xu, K.-F. Tong, and C.-B. Chae, “An information-theoretic characterization of MIMO-FAS: Optimization, diversity-multiplexing tradeoff and q-outage capacity,” IEEE Trans. Wireless Commun., vol. 23, no. 6, pp. 5541–5556, 2024.
- [8]E. Tang, W. Guo, H. Se, S. Song, J. Zhang, and K. B. Letaief, “A unified two-stage generative diffusion framework for channel estimation and port selection in multiuser MIMO-FAS,” arXiv preprint arXiv:2605.29679, 2026.arxiv.org/abs/2605.29679
- [9]Z. Wang, L. Zheng, Y. Shi, K. Huang, and K. B. Letaief, “Rethinking channel estimation for fluid antenna systems: A physics-informed operator learning,” arXiv preprint arXiv:2610.36787, 2026.arxiv.org/abs/2610.36787
- [10]L. Zhu, W. Ma, and R. Zhang, “Modeling, estimation, and performance analysis for movable antenna enabled wireless communications,” IEEE Trans. Wireless Commun., vol. 23, no. 6, pp. 6234–6250, Jun. 2024.arxiv.org/abs/2210.05325
- [11]T. Liao, W. Guo, J. Qian, Z. Wang, S. Song, J. Zhang, and K. B. Letaief, “Rethinking mutual coupling in movable antenna MIMO systems: Modeling and optimization,” arXiv preprint arXiv:2604.26282, 2026.arxiv.org/abs/2604.26282
- [12]W. Xu, L. Zhu, W. Ma, A. Liu, and R. Zhang, “Movable antenna enhanced multiuser communications with mutual coupling,” IEEE Commun. Lett., vol. 30, pp. 1954–1958, 2026.DOI
- [13]M. T. Iravanchy and J. A. Nossek, “Toward a circuit theory of communication,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 57, no. 7, pp. 1663–1683, Jul. 2010.
- [14]L. Zhu, W. Ma, B. Ning, and R. Zhang, “Movable-antenna enhanced multiuser communication via antenna position optimization,” IEEE Trans. Wireless Commun., vol. 23, no. 7, pp. 7214–7229, Jul. 2024.arxiv.org/abs/2302.06978
- [15]B. Feng, Y. Wu, X.-G. Xia, and C. Xiao, “Weighted sum-rate maximization for movable antenna-enhanced wireless networks,” IEEE Wireless Commun. Lett., vol. 13, no. 6, pp. 1770–1774, 2024.arxiv.org/abs/2404.09734
- [16]T. Liao, W. Guo, H. He, S. Song, J. Zhang, and K. B. Letaief, “Joint beamforming and antenna position optimization for fluid antenna-assisted MU-MIMO networks,” IEEE J. Sel. Areas Commun., vol. 44, pp. 1209–1226, 2026.arxiv.org/abs/2503.04040
- [17]S. Yang, Z. Wan, Y. Xiu, B. Ning, Y. Li, Y. Liu, and C. Yuen, “Flexible WMMSE beamforming for MU-MIMO movable antenna communications,” IEEE Trans. Signal Process., vol. 73, pp. 4479–4491, 2025.arxiv.org/abs/2503.17718
- [18]Z. Zheng, Q. Wu, W. Chen, and G. Hu, “Two-timescale design for movable antenna-enabled multiuser MIMO systems,” IEEE Trans. Commun., vol. 73, no. 11, pp. 10554–10571, 2025.DOI
- [19]X. Pi, L. Zhu, H. Mao, X. Xiao, X.-G. Xia, and R. Zhang, “Movable antenna enabled near-field MU-MIMO communication,” IEEE Wireless Commun. Lett., vol. 14, no. 10, pp. 3319–3323, 2025.DOI
- [20]A. Irshad, E. Björnson, A. Kosasih, and V. Petrov, “Performance evaluation of movable antenna arrays in wideband multi-user MIMO systems,” IEEE Trans. Commun., 2026, early access, doi: 10.1109/TCOMM.2026.3745149.DOI
- [21]S. S. A. Yuan, X. Chen, C. Huang, and W. E. I. Sha, “Effects of mutual coupling on degree of freedom and antenna efficiency in holographic MIMO communications,” IEEE Open J. Antennas Propag., vol. 4, pp. 237–244, 2023.DOI
- [22]A. Pizzo and A. Lozano, “Mutual coupling in holographic MIMO: Physical modeling and information-theoretic analysis,” IEEE J. Sel. Areas Inf. Theory, vol. 6, pp. 111–126, 2025.arxiv.org/abs/2502.10209
- [23]J. Zhu, F. Han, Y. Guo, and Y. Wang, “Mutual coupling-enhanced movable antenna arrays: Breaking minimum-spacing constraints,” IEEE Commun. Lett., vol. 30, pp. 672–676, 2026.DOI
- [24]W. Xu, L. Zhu, W. Ma, A. Liu, and R. Zhang, “Directivity enhancement of movable antenna arrays with mutual coupling,” IEEE Wireless Commun. Lett., vol. 15, pp. 2759–2763, 2026.
- [25]Z. Zhang, Y. Wu, H. Jiang, Z. Chen, and Z. Zhang, “Electromagnetic-aware fluid antenna array,” arXiv preprint arXiv:2607.21375, 2026.arxiv.org/abs/2607.21375
- [26]G. Iacovelli, C. K. Sheemar, and S. Chatzinotas, “Mutual-coupling-aware movable and fluid antennas on holographic surfaces: A wavenumber-domain circuit-field unification,” arXiv preprint arXiv:2609.04968, 2026.
- [27]J. Xu, J. Dang, and Z. Zhang, “Off-grid position optimization under mutual coupling in fluid antenna arrays,” arXiv preprint arXiv:2608.15603, 2026.arxiv.org/abs/2608.15603
- [28]J. Xu, H. Liang, Z. Zhang, and J. Dang, “Sparse port selection under mutual coupling in fluid antenna arrays,” arXiv preprint arXiv:2608.15599, 2026.arxiv.org/abs/2608.15599
- [29]S. S. Christensen, R. Agarwal, E. De Carvalho, and J. M. Cioffi, “Weighted sum-rate maximization using weighted MMSE for MIMO beamforming design,” IEEE Trans. Wireless Commun., vol. 7, no. 12, pp. 4792–4799, Dec. 2008.
- [30]Q. Shi, M. Razaviyayn, Z.-Q. Luo, and C. He, “An iteratively weighted MMSE approach to distributed sum-utility maximization for a MIMO interfering broadcast channel,” IEEE Trans. Signal Process., vol. 59, no. 9, pp. 4331–4340, Sep. 2011.
- [31]X. Zhao, S. Lu, Q. Shi, and Z.-Q. Luo, “Rethinking WMMSE: Can its complexity scale linearly with the number of BS antennas?” IEEE Trans. Signal Process., vol. 71, pp. 433–446, 2023.DOI
- [32]D. R. Hunter and K. Lange, “A tutorial on MM algorithms,” The American Statistician, vol. 58, no. 1, pp. 30–37, Feb. 2004.DOI
- [33]Y. Sun, P. Babu, and D. P. Palomar, “Majorization-minimization algorithms in signal processing, communications, and machine learning,” IEEE Trans. Signal Process., vol. 65, no. 3, pp. 794–816, Feb. 2017.DOI
- [34]S. Boyd and L. Vandenberghe, Convex optimization. Cambridge, U.K.: Cambridge Univ. Press, 2004.
- [35]S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimization and statistical learning via the alternating direction method of multipliers,” Found. Trends Mach. Learn., vol. 3, no. 1, pp. 1–122, 2011.




