6DMA-Aided Hybrid Beamforming with Joint Antenna Position and Orientation Optimization

I. INTRODUCTION

• Fixed-antenna hybrid beamforming cannot fully exploit spatial channel variation, while sub-connected arrays trade some design degrees of freedom for lower hardware cost and complexity.Prior work explored movable positions and six-dimensional position-and-rotation adjustment as ways to reconfigure channels.Source: §I · pp. 1–2
• Prior 6DMA work largely assumed fully digital arrays and neglected rotation's effect on polarization, while a polarization-aware movable-antenna model considered only LoS channels.The paper addresses the unexamined combination of polarization-aware 6DMA channels and cost-effective array structures.Source: §I · p. 2
• The proposed scheme jointly optimizes the digital and analog beamformers with sub-array positions and orientations to maximize multi-user sum rate, using an FP-aided alternating-optimization framework.The framework combines the Lagrange multiplier method, manifold optimization, and gradient descent to solve the non-convex problem sub-optimally.Source: §I · pp. 2–3

A. Description of the 6DMA-aided System

• Each sub-connected UPA has M antennas linked through phase shifters to a dedicated RF chain, and its antennas move and rotate collectively within specified regions.UPA centers move within separated planar regions, and rotation is restricted to a limited angular range.Source: A. Description of the 6DMA-aided System · pp. 3–4

B. Channel Model

• The model assumes far-field, quasi-static, slow-fading channels; within the movable region, path angles and amplitudes stay fixed while phases vary with UPA movement.Rotation changes local departure angles, which in turn affects radiation power and polarization losses.Source: B. Channel Model · p. 4
• The proposed channel model incorporates multipath propagation, directional radiation patterns, and polarization matching rather than assuming omnidirectional elements.The model accounts for orientation-dependent changes in local departure angles, radiation power, and polarization loss.Source: B. Channel Model · p. 4

1) Basic Field-Response Model:

• The field-response model combines antenna responses for the transmit paths with a path-response matrix that captures large- and small-scale fading to form the BS-to-user channel.The path-response matrix entries represent response coefficients between transmit and receive paths.Source: 1) Basic Field-Response Model · p. 4

2) Directional Radiation Pattern:

• The model uses a cosine radiation pattern to represent directional antenna power instead of assuming a constant omnidirectional pattern.The text notes that directional patterns are widely adopted in practice.Source: 2) Directional Radiation Pattern · p. 4
• κ characterizes antenna energy-focusing capability, and Gκ normalizes the radiation pattern over the unit sphere.For convenience, the model sets κ = 2 and Gκ = 3/2.Source: 2) Directional Radiation Pattern · p. 5

3) Antenna Polarization:

• The model extends polarization to three dimensions because elevation spread cannot be neglected, and transforms each field direction from the local to the global coordinate system.The field direction is determined by antenna orientation and lies in the tangent plane normal to the propagation direction.Source: 3) Antenna Polarization · pp. 5–6
• Vertically polarized transmit antennas are assumed, setting the two local polarized field components to F~θ=1 and F~φ=0.Source: 3) Antenna Polarization · p. 5
• Receive-antenna polarization also affects received power through transmit–receive mismatch, but its orientation is fixed, so specific receiver expressions are omitted.Source: 3) Antenna Polarization · p. 6

C. Problem Formulation

• The objective is to maximize the sum rate by jointly optimizing UPA positions and orientations, the analog beamformer, and the digital beamformer.Source: C. Problem Formulation · p. 6
• The range R is defined as the Cartesian product of three intervals, R = [αL, αU] × [βL, βU] × [γL, γU], as stated in the cited passage.Source: C. Problem Formulation · p. 6
• Position and orientation are highly coupled, while the reconstructed practical channel model yields a non-concave objective and non-convex constraints.The model accounts for multipath propagation, directional radiation patterns, and antenna polarization; these complexities make optimal solution of the problem challenging.Source: C. Problem Formulation · p. 6

III. ALGORITHM DESIGN

• The fractional-programming problem is decomposed into four alternating updates: the digital beamformer, analog beamformer, and UPA positions and orientations.Source: §III · p. 6
• The digital beamformer is derived in closed form using the Lagrange multiplier method, while manifold optimization handles the analog beamformer’s unit-modulus constraint.Source: §III · p. 6
• Gradient descent sequentially searches UPA positions and orientations, and consistent-objective updates guarantee convergence of the FP-aided AO algorithm.This direct method differs from approaches that derive beamformers by approximating the fully-digital beamformer.Source: §III · p. 6

A. Problem Reformulation Based on FP

• The fractional-programming framework reformulates problem (16) into an equivalent form to simplify the objective.Source: A. Problem Reformulation Based on FP · p. 6
• The equivalent formulation includes slack variables u and v, which are updated optimally when the beamformers and UPA positions and orientations are fixed.Source: A. Problem Reformulation Based on FP · p. 6
• Alternating optimization iteratively updates each variable while fixing the others; a term determined only by u and v is omitted from the remainder of the paper.Source: A. Problem Reformulation Based on FP · pp. 6–7

• The digital-beamformer design subproblem is expressed after the power constraint is rewritten, with the analog beamformer irrelevant to constraint (16c).Source: B. DBF Design · p. 7
• The Lagrange multiplier method gives a closed-form solution to the convex quadratic digital-beamformer design problem.Source: B. DBF Design · p. 7
• The multiplier is zero in the stated power-feasible case and otherwise is determined by bisection.Source: B. DBF Design · p. 7

• The analog-beamformer subproblem is difficult because its unit-modulus constraint is non-convex.Source: C. ABF Design · p. 7
• The authors select manifold optimization over penalty or majorization-maximization methods, whose results depend heavily on initialization.Source: C. ABF Design · p. 7
• The unit-modulus feasible set is a Riemannian submanifold formed by the product of MN complex circles.Source: C. ABF Design · p. 7
• The manifold-optimization workflow uses a Riemannian gradient, Polak–Ribiere search directions, retraction, and Armijo backtracking.Source: C. ABF Design · pp. 7–8
• Algorithm 1 terminates when the Riemannian gradient approaches zero, which the authors identify as confirming convergence.Source: C. ABF Design · pp. 7–8

• Each sub-array’s central position and orientation are updated sequentially along gradient directions while the other sub-arrays’ positions and orientations remain fixed.Source: D. Antenna Position and Orientation… · p. 8
• Position and orientation gradients are derived using simplified channel-model parameters that include antenna radiation-pattern and polarization-matching power gains.Source: D. Antenna Position and Orientation… · p. 8
• Because gradient descent is step-size sensitive, the step size is repeatedly halved until the update is feasible and does not decrease the objective.Source: D. Antenna Position and Orientation… · p. 9
• The overall algorithm is reported to converge because its objective is non-decreasing across iterations and has a finite optimal value.Source: D. Antenna Position and Orientation… · p. 9

IV. NUMERICAL RESULTS

• Algorithm 2’s sum rate increases over iterations until reaching a constant for the tested power budgets.The passage presents this behavior as validating the convergence analysis.Source: §IV · p. 9, Figure 3 · p. 10
• At P = 20 dBm, 6DMA improves sum rate over sub-connected-FA by 135%, fully-connected-FA by 127%, fully-digital-FA by 118%, fully-digital-MA with flexible position by 4.05%, and fully-digital-MA with flexible orientation by 1.19%.The proposed scheme also improves over the unpolarized 6DMA scheme by about 352% at P = 20 dBm.Source: §IV · p. 10, Figure 4 · p. 10
• As the movable-region size increases, 6DMA sum rate rises and then plateaus; with a sufficiently large region, it outperforms all other schemes.The passage attributes the plateau to channel-gain periodicity and spatial correlation, which mean further expansion adds no spatial DoFs.Source: §IV · p. 10, Figure 5 · p. 10
• With sufficiently wide rotatable-angle ranges, 6DMA achieves the highest sum rate among the schemes, although the simulations’ limited rotation ranges do not reach the sum-rate upper bound.Rotation adjusts radiation and polarization directions and antenna positions within each UPA.Source: §IV · p. 10, Figure 6 · p. 11
• Without rotation, sub-connected-MA with flexible position improves performance over fully-digital-FA by 169%.The passage states that antenna movability alone can compensate for the performance loss of a sub-connected structure.Source: §IV · p. 11, Figure 6 · p. 11
• As the number of paths increases, all schemes’ sum rates rise, and 6DMA remains much higher than the other schemes.The passage links equal-power multipath fading to greater spatial diversity and DoFs for 6DMA to exploit.Source: §IV · p. 11, Figure 7 · p. 11
• Across different user counts, sub-connected-6DMA consistently outperforms the five baselines.The passage attributes the result to creating balanced multi-user channels, which can reduce channel-gain correlation and mitigate interference.Source: §IV · p. 11, Figure 8 · p. 11

• The sub-connected 6DMA scheme outperforms FA- and MA-based benchmarks and, under specific practical conditions, fully-digital schemes with less antenna-movement flexibility.Source: §V · p. 11
• The paper jointly optimizes digital and analog beamformers and UPA positions and orientations to maximize sum rate under a practical 6DMA channel model.Source: §V · p. 11
• An FP-aided alternating-optimization framework decomposes the non-convex problem into four subproblems solved using Lagrange multipliers, manifold optimization, and gradient descent.Source: §V · p. 11
• A baseline optimizing beamformers and UPA configurations with an unpolarized channel model incurs notable performance degradation, highlighting polarization’s role in orientation optimization.Source: §V · p. 11