---
title: 'Rotatable Antenna (RA): Design & Optimization'
url: https://www.emergentmind.com/topics/rotatable-antenna-ra
type: topic
---

# Rotatable Antenna (RA): Design & Optimization

Rotatable antenna (RA) denotes a directional antenna architecture whose boresight direction can be reconfigured in three-dimensional space while the antenna position remains fixed. Across recent arXiv work, RA is defined through orientation control rather than translational motion: the antenna or array rotates its radiation pattern toward desired directions by mechanical, electronic, or hybrid means, thereby exposing additional spatial degrees-of-freedom (DoFs) for communication and sensing without requiring full six-dimensional movement [2505.16828, 2411.08411]. In this literature, RA appears both as a single servo-driven directional antenna and as an array or subarray architecture with per-element or per-panel orientation variables, with applications ranging from uplink reception and secure transmission to mixed-field communications, cell-free networking, multicast, cognitive radio, covert communication, sensing, and mobile edge computing [2502.17097, 2502.21036].

## 1. Conceptual foundations and relation to adjacent antenna paradigms

RA is introduced in the recent literature as a middle design point between conventional fixed directional antennas and more general movable-antenna architectures. A fixed directional antenna has a fixed position and a fixed boresight after deployment; an RA keeps the position fixed but makes the boresight variable. In this sense, RA changes the angular response of the antenna rather than its location, so the added DoF is orientation control rather than translational mobility [2411.08411].

This distinction is explicit in comparisons with fluid antenna systems (FAS), movable antennas (MA), and six-dimensional movable antennas (6DMA). FAS and MA are described as exploiting positional flexibility, while 6DMA extends flexibility to both 3D position and 3D rotation. RA retains the rotational part only and is therefore framed as a streamlined, lower-cost, and more compact alternative to full 6DMA, while still allowing directional-gain adaptation to user or target geometry [2505.16828]. Several papers also stress that RA is not merely a synonym for beamforming: beamforming optimizes complex excitation weights over a fixed element orientation, whereas RA changes the element-wise directional pattern itself.

The steering mechanism depends on the embodiment. The radar-sensing hardware demonstration uses a single directional antenna physically rotated by a servo, and the paper explicitly notes that it does not provide a formal comparison against electronically steerable phased arrays or analog beamforming arrays; its steering mechanism is mechanical boresight alignment rather than electronic phase control [2502.21036]. At the same time, survey work treats electronic and hybrid RA embodiments as part of the same architectural family, so RA is better understood as orientation reconfigurability, not as a single actuation method [2505.16828].

## 2. Architectures and physical implementations

The hardware design space reported in the literature is broad. Survey work classifies RA implementations into antenna-wise mechanically driven RA, electronically driven RA, hybrid mechanical-electronic RA, array-level rotation, dual-scale rotation, and hybrid deployments in which RAs are co-deployed with fixed sector antennas [2505.16828]. Mechanical embodiments physically rotate the radiator, for example with servo motors or MEMS actuators. Electronic embodiments keep the antenna body fixed while redirecting the effective main lobe through multi-feed switching, tunable parasitic elements, or tunable materials such as liquid crystal. Hybrid embodiments combine wide-angle mechanical steering with faster electronic fine adjustment.

Prototype papers make these abstractions concrete. One visual-recognition-guided RA prototype integrates a directional antenna, a two-dimensional digital servo, and a microcontroller, and couples them to a camera, a PC, and a USRP-based transceiver. The directional antenna has gain **10 dBi** and beamwidth **60°**. The camera provides RGB images to a PC, which uses **YOLO** for detection, **DeepSORT** for tracking, and a **PID steering algorithm** to drive the servo through pulse-width commands. In that prototype, azimuth and zenith are both controllable, although the reported experiment fixes the receiver zenith at \(\theta=0^\circ\) [2502.17097].

A second prototype uses radar sensing rather than vision guidance. Its transmitter integrates a **laser radar module**, a **servo module**, and a **USRP module**; the receiver uses a **USRP with an omni-directional dipole antenna**. The radar operates by **time-of-flight (TOF)**, scans **360°** by default, outputs **light intensity, distance, and two-dimensional (2D) angle of arrival (AoA)** over a **serial port**, and runs at **10 Hz**. Because the raw AoA fluctuates, the system averages the AoA information over **one-second intervals** before passing it to an **STM32**-based servo controller that generates **PWM** signals and applies **PID** tracking. That prototype is limited to horizontal deflection because the radar provides only 2D AoA data, so zenith-angle changes are excluded [2502.21036].

A distinct architectural line addresses scaling cost. The cross-linked rotatable antenna array replaces independently actuated elements with a row-column coupling mechanism: a motor on each vertical track controls one rotation axis collectively, and a motor on each horizontal track controls the other. For an \(M\times N\) array, the paper states that a traditional independently rotatable array needs at least \(2MN\) motors, whereas the cross-linked architecture uses \(M+N\) motors. A panel-level variant reduces the motor count further to \(\frac{M+N}{\sqrt{Q_b}}\) when each panel contains \(Q_b\) antennas [2601.04862].

## 3. Modeling principles and orientation-dependent channels

The dominant mathematical abstraction is the unit pointing vector. Different papers parameterize it with slightly different angle conventions, but the common structure is a 3D unit vector representing the boresight direction. One representative model writes
\[
\mathbf{f}_k=
\big[\sin\theta_{z,k}\cos\phi_{a,k},\ \sin\theta_{z,k}\sin\phi_{a,k},\ \cos\theta_{z,k}\big]^T,
\]
with zenith and azimuth variables constrained by a maximum allowable zenith deflection [2603.15313]. Another writes the boresight using eccentric and azimuth angles relative to a reference \(x\)-axis, again with a bounded eccentric angle [2411.08411]. In optimization formulations, these angle variables are often replaced by the pointing vectors themselves to avoid nonconvex trigonometric parameterizations.

The essential geometric quantity is the misalignment between boresight and propagation direction. For a user or target direction vector \(\mathbf{q}\), papers typically define
\[
\cos(\epsilon)=\mathbf{f}^T\mathbf{q},
\]
so \(\epsilon\) is the angular mismatch between where the antenna points and where the signal arrives from or propagates to. The element directional gain is then modeled as a cosine-power law. Two closely related forms recur in the literature:
\[
G_{k,m}=G_0\cos^{p}(\epsilon_{k,m})
\]
and
\[
G_e(\epsilon,\varphi)=
\begin{cases}
G_0\cos^{2p}(\epsilon), & \epsilon\in[0,\frac{\pi}{2}),\ \varphi\in[0,2\pi),\\
0, & \text{otherwise}.
\end{cases}
\]
The first appears in MEC-oriented array reception models, and the second in secure communication, cell-free, multicast, covert, and cognitive-radio formulations [2603.15313, 2504.10473]. In both cases, larger \(p\) means stronger directivity and therefore a sharper dependence on alignment.

Because gain depends on \(\mathbf{f}^T\mathbf{q}\), the channel becomes orientation dependent at the element level. Representative channel models write the scalar coefficient as path loss times orientation-dependent directional gain times small-scale fading, for example through terms of the form
\[
h_{k,n}(\mathbf{f}_n)\propto (\mathbf{f}_n^T\mathbf{q}_{k,n})^p
\]
or through \(\sqrt{G_0\cos^{2p}(\epsilon)}\) factors inside the channel amplitude [2411.08411, 2504.10473]. This is the formal expression of the RA idea: rotation does not just redirect radiation visually; it changes the channel coefficients that downstream beamforming or combining must operate on.

A more complete electromagnetic treatment appears in polarization-aware work. There, each antenna is rotated by a full matrix \(\mathbf{R}_m\in \mathrm{SO}(3)\), so the boresight \(\mathbf{e}_{B,m}\), horizontal polarization direction \(\mathbf{e}_{H,m}\), and vertical polarization direction \(\mathbf{e}_{V,m}\) all rotate together as a rigid body. This work argues that conventional rotatable-antenna models often optimize only boresight direction and implicitly ignore the fact that mechanical rotation also rotates the radiated polarization basis [2603.01166]. Within that framework, RA becomes not only a gain-steering mechanism but also a polarization-orientation mechanism.

## 4. Communication system formulations and network-level roles

Recent work treats RA orientation as a first-class optimization variable, usually coupled with beamforming, power allocation, scheduling, or computing-resource allocation. The resulting formulations cover a wide range of wireless system objectives.

| Domain | Representative objective | Representative papers |
|---|---|---|
| Uplink reception | Maximize minimum SINR by jointly optimizing receive beamforming and RA deflection angles | [2411.08411] |
| Security and spectrum control | Maximize secrecy rate, covert rate, or secondary-link SINR under leakage/interference constraints | [2504.10473], [2603.11716], [2509.25656] |
| Cooperative and fairness-oriented networking | Maximize sum rate or minimum SINR via AP-user association, multicast beamforming, and RA boresight control | [2512.04742], [2603.26388] |
| Mixed-field communications | Maximize near-field or network sum rate while mitigating near-field and mixed-field interference by subarray or array rotation | [2509.04865], [2604.05565] |
| Mobile edge computing | Maximize weighted sum computation rate or minimize maximum computation latency by jointly optimizing RA orientations and resource variables | [2603.15313], [2603.16275] |

Across these scenarios, the dominant algorithmic template is **alternating optimization (AO)**. Within AO, the literature combines **successive convex approximation (SCA)**, **fractional programming (FP)**, **semidefinite relaxation (SDR)**, **generalized Rayleigh quotient** beamforming, **second-order cone programming (SOCP)**, **minimum mean square error (MMSE)** combining, **zero forcing (ZF)**, **Karush-Kuhn-Tucker (KKT)** analysis, **bisection**, and **particle swarm optimization (PSO)**, with **CVX** appearing repeatedly as the convex-solver back end [2504.10473, 2603.26388, 2509.04865]. This repeated structure reflects a common difficulty: RA variables enter channel expressions nonlinearly, are coupled to digital beamforming, and are usually subject to unit-norm and angular-range constraints.

The applications also reveal several distinct operating regimes. Some papers assume per-element independent 3D rotation; others rotate one array or subarray as a rigid body. Some use RA at a transmitter to strengthen desired links and suppress leakage; others use RA at a receiver or base station to shape effective receive channels. In mixed-field FR3/upper-mid-band work, RA rotation modifies both linear phase terms and quadratic Fresnel terms, thereby suppressing not only ordinary angular interference but also the more specific overlap between near-field spherical-wave responses and far-field planar-wave beams [2509.04865, 2604.05565].

## 5. Sensing, control loops, and experimental validation

RA is studied not only as a communication architecture but also as a sensing and integrated sensing-and-communication resource. Survey work attributes three major sensing benefits to RA: **resolution enhancement**, **coverage expansion**, and **multi-target and multi-dimensional sensing**. The same survey also argues that RA is useful for **integrated sensing and communication (ISAC)** because different antennas can be pointed toward communication users and sensing targets separately, reducing directional conflicts between the two functions [2505.16828].

Two hardware demonstrations show how sensing and control can be embedded in the RA loop. The radar-guided prototype closes a sensing-control-communication chain: the transmitter-side laser radar detects the receiver location, estimates 2D AoA, averages that estimate over one-second intervals, sends the processed direction to an STM32/PID servo controller, and then transmits through a **vertically polarized directional antenna** with **maximum gain of 10 dBi** and **60° beamwidth**. The reported indoor experiment uses **16QAM**, **5.8 GHz**, **100 KHz** bandwidth, **0.5 Mbps** data rate, **10 dBm** transmit power, and **4 meters** TX-RX separation. At receiver azimuth \(\frac{\pi}{3}\), the fixed-antenna system receives **\(-87\) dBm** while the RA system reaches **\(-80\) dBm**, corresponding to an SNR improvement of approximately **7 dB** because the noise level is approximately **\(-95\) dBm** [2502.21036].

The vision-guided prototype uses a camera and PC to transform direction finding into an image-recognition problem. The camera delivers RGB images; the PC applies **YOLO** for detection and **DeepSORT** for tracking; a **PID steering algorithm** then drives the two-dimensional digital servo. Its reported communication settings are **5.8 GHz**, **16-QAM**, **10 dBm** transmit power, and **2 Mbps** transmission rate. Over receiver azimuth angles from \(-\pi/2\) to \(+\pi/2\), the RA maintains received power roughly around **\(-58\) to \(-60\) dBm** from visual reading of the figure, while the fixed antenna drops to around **\(-68\) to \(-70\) dBm** at larger azimuth offsets, giving about **8 to 12 dB improvement** and a substantially flatter power-versus-angle profile [2502.17097].

Survey-level validation broadens the picture beyond single-link prototypes. In the communication case study, a **4-user** system with a **\(4\times 4\)** RA UPA at **2.4 GHz** shows **max-min SINR** gains of up to **2 dB over random boresight** and **3.2 dB over fixed boresight**, with substantial gains still present for \(\theta_{\max}\le \pi/10\). In the sensing case study, a **3-target** radar system with the same **\(4\times 4\)** RA UPA produces a **spatial power spectrum** with sharper peaks at target directions and lower sidelobes than random and fixed boresight baselines [2505.16828].

## 6. Physical couplings, limitations, and research directions

Despite the breadth of applications, the literature is explicit about practical constraints. Mechanical rotation is slower than electronic beam steering, and prototype papers often leave critical actuator parameters unspecified. The radar-guided demonstration does not report servo model, settling time, or rotation speed, and its need to average AoA over one second to stabilize estimates implies quasi-periodic rather than instantaneous control [2502.21036]. The vision-guided prototype likewise reports no exact numerical values for maximum azimuth or zenith range, angular resolution, servo response time, settling time, or maximum angular velocity [2502.17097].

Several optimization papers distinguish idealized from practical rotation timescales. In RA-assisted MEC, the **dynamic rotating scenario** assumes that RAs can reorient within each TDMA slot and yields a closed-form optimal pointing vector; the paper explicitly treats this as an upper-bound case and notes that such dynamic reorientation is physically challenging because mechanical rotation is slower than slot switching. The **static rotating scenario**, in which all RAs maintain a unified orientation over the frame, is presented as the more practical model [2603.15313]. This distinction recurs implicitly elsewhere whenever rotation is optimized continuously but actuation dynamics are omitted.

A second limitation concerns information assumptions. Secure communication, cell-free communication, and cognitive-radio formulations commonly assume **global CSI** or complete knowledge of desired and undesired channels at the controller, including eavesdropper or primary-receiver channels in some cases [2504.10473, 2512.04742, 2509.25656]. Such assumptions are analytically convenient, but they push the burden onto channel estimation, sensing, or localization subsystems. Survey work therefore identifies **rotational scanning scheduling**, **channel estimation/sensing**, **boresight direction optimization**, and **antenna configuration** as the four main RA design challenges [2505.16828].

A third limitation is the prevalence of continuous relaxations. Many formulations relax the unit-norm pointing constraint from \(\|\mathbf f_n\|=1\) to \(\|\mathbf f_n\|\le 1\), solve a convex surrogate, and then normalize the result; this appears in secure communication, multicast, cell-free, and uplink fairness formulations [2504.10473, 2603.26388, 2411.08411]. This does not invalidate the optimization framework, but it does mean that many reported solutions are explicitly high-quality suboptimal or stationary solutions rather than global optima.

Finally, recent work shows that boresight control alone can be physically incomplete. The polarization-aware study argues that full 3D mechanical rotation also rotates the antenna’s polarization basis, so rotation-only or boresight-only models can incur polarization mismatch. In its simplified single-user LoS analysis, ignoring roll can cost up to **3 dB** relative to the best roll alignment, and one numerical example yields about **15.1 dB** total gain for optimized rotation over a fixed scheme, decomposed into **12 dB** from misalignment recovery and **3 dB** from polarization improvement [2603.01166]. This result sharpens the interpretation of RA: orientation adaptation can affect not only directional gain but also projection loss, polarization direction alignment, and polarization state matching.

Taken together, these studies depict RA as a technically coherent but still developing antenna paradigm. The central premise—orientation as an optimization variable—is stable across the literature. What remains open is how to realize that premise under finite actuator speed, limited sensing and CSI, hardware coupling, discrete orientation states, calibration burden, and full electromagnetic effects, while preserving the low-complexity and low-cost motivations that originally distinguished RA from more general movable-antenna architectures [2505.16828, 2601.04862].

Source: https://www.emergentmind.com/topics/rotatable-antenna-ra