Aerial Reconfigurable Intelligent Surfaces

• ARISs are aerially deployed reconfigurable surfaces using UAVs, HAPs, or satellites to manipulate signals in 3D space with passive or active reflection.
• Studies show that increasing the number of reflecting elements can reduce required transmit power by up to 8.2 dB, enhancing energy efficiency.
• ARIS designs balance improved line-of-sight and panoramic coverage with challenges like UAV payload limits, control overhead, and channel nonstationarity.

Aerial reconfigurable intelligent surfaces (ARISs) are reconfigurable intelligent surfaces mounted on aerial platforms and used as airborne passive or nearly passive reflecting nodes that manipulate propagation in three-dimensional space. In the literature, ARISs appear as UAV-mounted RIS, mobile RIS, aerial-RIS, RIS-UAV, flight-assisted smart surfaces, and, in earlier survey terminology, aerial reconfigurable smart surfaces; they are studied in blocked terrestrial links, RIS-assisted non-terrestrial networks, UAV swarm-enabled distributed surfaces, low-altitude ISAC, aerial backhaul, vehicular mmWave support, and secure satellite systems (Do et al., 2021, Devoti et al., 2022, Ye et al., 2021). Relative to terrestrial RISs, ARISs are motivated by better deployment flexibility, richer line-of-sight opportunity, and panoramic or full-angle reflection; relative to active relays or aerial base stations, they are motivated by lower payload and power consumption, although those advantages are coupled to channel nonstationarity, control overhead, payload limits, and platform instability (Abdalla et al., 2020).

1. Terminology, scope, and architectural variants

Within the RIS-assisted NTN taxonomy, the literature distinguishes terrestrial RISs (TRISs) from aerial RISs (ARISs) and further classifies deployments into RIS-assisted A2G/G2A, ARIS-assisted G2G, and RIS-assisted A2A communications (Ye et al., 2021). In that taxonomy, ARIS denotes an RIS mounted on an aerial platform and acting as an intermediate reflection layer between end devices. The carrier platform is not unique: ARIS-related work explicitly considers UAVs, HAPs, fixed-wing aircraft, tethered balloons, and LEO satellites (Devoti et al., 2022, Alfattani et al., 2020).

The airborne platform can host the RIS in multiple physical forms. Earlier architecture papers discuss coating the outer surface of a balloon or aircraft, mounting a separate horizontal surface under a rotary-wing UAV, or integrating the RIS into a broader control stack consisting of a ground processing unit, a flight-control unit, and an RSS/RIS control unit (Alfattani et al., 2020). More recent system papers use an airborne RIS node with a 3D position, fixed or moving altitude, or a swarm of UAVs each carrying a moderate-size RIS (Do et al., 2021, Shang et al., 2021).

ARIS is not a single architectural class. A narrow passive formulation treats the aerial node as a passive reflector that only applies controllable phase shifts. A broader class includes distributed or hierarchical realizations such as UAV swarm-enabled ARIS (SARIS), where multiple UAVs collectively realize a larger reflecting aperture and more flexible spatial deployment (Shang et al., 2021). A separate sensing line uses the acronym ARIS for active reconfigurable intelligent surface, specifically an active RIS mounted on a UAV to support SAR imaging; in that context, “ARIS” denotes active reflection and amplification rather than only aerial placement (Sun et al., 2024). This suggests that the acronym is context-sensitive, whereas the underlying encyclopedia topic is best understood as aerially deployed reconfigurable intelligent surfaces in passive, nearly passive, distributed, or active forms.

The literature also distinguishes ARISs from terrestrial RISs and active aerial relays through geometry and hardware burden. ARISs are repeatedly described as providing 3D deployment flexibility, reliable air-to-ground links, panoramic full-angle reflection, and a higher probability of creating LoS connections, while avoiding the bulky, heavy, power-hungry active antennas and transceiver chains of aerial BS or relay architectures (Shang et al., 2021, Devoti et al., 2022). At the same time, single-platform ARIS is limited by UAV payload and battery capacity, which motivates multi-UAV cooperative surfaces and large-scale continuous-density formulations (Shang et al., 2021, Luo et al., 12 Sep 2025).

2. Signal models, channel models, and propagation assumptions

A canonical ARIS communication model appears in a blocked ground link assisted only by an airborne RIS node $R$ with $N$ reflecting elements (Do et al., 2021). The phase-shift matrix is

$$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$

where $\phi_r \in [0,2\pi)$ is the phase shift at element $r$ and $\kappa\in(0,1]$ is a fixed reflection amplitude coefficient. For transmitted symbol $s$ with $\mathbb{E}[|s|^2]=1$, the received signal is

$$y = \sqrt{P_S} \sum_{r=1}^N [\mathbf{h}_{SR}]_r \kappa e^{j \phi_r} [\mathbf{h}_{RD}]_r s + w_D,$$

with $w_D\sim\mathcal{CN}(0,\sigma^2)$. Under ideal passive beamforming, the coherent phase design is

$N$0

so the effective cascaded channel becomes a coherent sum of nonnegative source–RIS and RIS–destination amplitude products (Do et al., 2021).

For aerial propagation, several channel regimes recur. One analytical ARIS paper models each hop by composite fading consisting of Nakagami-$N$1 small-scale fading and inverse-Gamma large-scale shadowing, with channel amplitude

$N$2

and shows that outage is highly sensitive to both $N$3 and the inverse-Gamma shadowing parameter $N$4 (Do et al., 2021). Backhaul-oriented ARIS designs instead assume LoS-dominated source–RIS and RIS–UAV channels with free-space losses and steering-vector models, which is natural when the RIS is mounted on a high-altitude aerial platform (Jeon et al., 2021). IoT freshness work with a UAV-mounted RIS assumes Rician fading with dominant LoS on both IoTD-to-UAV and UAV-to-BS hops, while only the UAV altitude is dynamically controlled online (Samir et al., 2020).

The signal model becomes more elaborate when the RIS is active. In UAV-mounted active-RIS-assisted SAR, the $N$5-th coefficient at slow time $N$6 is

$N$7

with amplitude $N$8 and phase $N$9, collected in

$$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$0

because the RIS not only phase-shifts but also amplifies the incident signal (Sun et al., 2024).

At much larger scales, discrete UAV-by-UAV placement can be replaced by a mean-field / continuous-density model. In that setting, a large ARIS swarm is represented by a spatial density $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$1 over a feasible region $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$2, with the effective BS-to-user channel written as an integral over the ARIS field rather than a sum over explicit UAV locations (Luo et al., 12 Sep 2025). This replaces combinatorial deployment by a functional optimization problem and is specifically motivated by large-scale ARIS swarms.

Across these models, one persistent distinction separates ARIS from terrestrial RIS: aerial geometry directly enters the channel law through altitude, angular support, LoS/NLoS probability, and motion-induced variation. Several survey papers therefore emphasize that free-space-only modeling is too optimistic for aerial deployment and that practical A2G/G2A analysis must include shadowing, misalignment, or mobility effects (Do et al., 2021, Ye et al., 2021).

3. Performance analysis, scaling laws, and engineering trends

Analytical ARIS performance work is dominated by outage, rate, energy efficiency, and link-budget studies. For a delay-limited blocked terrestrial link assisted by a static airborne RIS, the outage probability is derived through a sequence of approximations: moment-matching of product distributions, Gaussian–Laguerre quadrature to obtain a mixture-Gamma form, Laplace-transform analysis of the sum over RIS elements, and conversion of the resulting CDF into an outage expression (Do et al., 2021). The reported result is a tight approximate closed-form outage expression, and the same paper shows that fading and shadowing conditions have strong impact on outage, that ARIS achieves higher energy efficiency as the number of reflecting elements increases, and that the ARIS-aided system outperforms conventional relaying schemes such as HD-DF, HD-VG-AF, FD-AF, and FD-DF under the stated assumptions (Do et al., 2021).

The number of reflecting elements is a central scaling variable. For the outage target

$$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$3

the required $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$4 drops by 8.2 dB when $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$5 increases from $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$6 to $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$7, and the same study interprets this as higher energy efficiency through lower required transmit SNR (Do et al., 2021). In swarm-enabled ARIS, the aperture argument is even more explicit: the paper states that single-UAV ARIS received power scales on the order of $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$8, whereas a cooperative swarm with $$\mathbf{\Psi}=\kappa \,\mathrm{diag}\!\left(e^{j\phi_1},\ldots,e^{j\phi_N}\right),$$9 UAVs and $\phi_r \in [0,2\pi)$0 elements per UAV can scale on the order of $\phi_r \in [0,2\pi)$1 under coherent combining (Shang et al., 2021). Aerial backhaul analysis similarly identifies a peak passive beamforming gain of $\phi_r \in [0,2\pi)$2, while warning that beamwidth shrinks with array size; the half-power beamwidth in the sin-AoD domain is approximated by

$\phi_r \in [0,2\pi)$3

which explains why full-array operation may fail when multiple UAV-BSs are too widely spread in angle (Jeon et al., 2021).

System-level evaluations report the same qualitative trend in different environments. In a nomadic UAV scenario with a 100-element squared RIS, an adaptive mobility-aware control architecture incurs a reconfiguration overhead cost of about $\phi_r \in [0,2\pi)$4 to $\phi_r \in [0,2\pi)$5 while keeping data-rate degradation below $\phi_r \in [0,2\pi)$6 over UAV speeds from 5 to 50 km/h (Devoti et al., 2022). In a campus-scale vehicular simulation at 28 GHz, a central base station covers only 26% of the whole track without RIS; seven static RISs make comprehensive coverage achievable, while additional UAV-mounted RIS support yields path losses even below 127 dB and, in the V2V NLoS analysis, combined static plus UAV-RIS deployment limits the maximum path loss to 120 dB (Heimann et al., 2021).

Large-scale anti-jamming ARIS work extends performance analysis to adversarial settings. There, the key result is not only improved sum-rate but also a structural finding: the jammer’s optimal strategy follows a proximity-directivity trade-off, and the ARIS deployment follows a spatial water-filling principle that concentrates density in high-gain regions while avoiding interference-prone areas (Luo et al., 12 Sep 2025). Secure satellite ARIS systems show similar conditional gains: more subsurfaces and more reflecting elements improve the secure multicast sum rate, but the paper explicitly reports diminishing returns because stronger reflection can also improve the eavesdropper’s channel (Wang et al., 28 Aug 2025).

4. Optimization, learning, and control architectures

ARIS optimization is almost always joint optimization. The variables that recur across the literature are platform position or trajectory, altitude, RIS phase shifts or reflection coefficients, active beamforming, scheduling, association, and, in large swarms, spatial density (Abdalla et al., 2020, Wang et al., 13 Nov 2025). The resulting problems are typically mixed-integer, non-convex, and strongly coupled across communication and mobility time scales.

A protocol-level template appears in RIS-assisted aerial-terrestrial communications via multi-task learning. Each frame consists of a Negotiation Phase (NP) and a Communication Phase (CP); the NP contains Synchronization, Channel estimation, and Optimization, after which the RIS configuration remains unchanged within the frame (Cao et al., 2021). That work is not a direct airborne-RIS deployment paper, but it supplies a control architecture that transfers naturally to ARIS because airborne operation also requires repeated CSI refresh, configuration, and data transmission under short coherence times.

For mobile ARIS outage prediction, analytical averaging over 3D motion becomes intractable, and a feed-forward DNN is trained directly on outage probability. The reported implementation uses 13 neurons at the input layer, 5 hidden layers with 128 neurons each, ReLU in the hidden layers, a linear output layer, Keras and TensorFlow, and $\phi_r \in [0,2\pi)$7 Monte Carlo-generated samples split into 80% / 10% / 10% training, validation, and test sets (Do et al., 2021). The validation MSE converges after about 30 epochs and falls below $\phi_r \in [0,2\pi)$8, supporting the use of learned surrogates when exact mobile ARIS analysis is analytically infeasible (Do et al., 2021).

Learning is also central when the uncertainty is not only geometric but stochastic or adversarial. For IoT freshness with unknown device activation patterns, the ARIS control problem is formulated as a model-free DRL task and solved with proximal policy optimization (PPO), where the agent chooses IoTD scheduling and UAV altitude commands while the RIS phases are set analytically for coherent combining (Samir et al., 2020). For large-scale anti-jamming transmission, the communication-side optimization alternates between ZF + water-filling for BS beamforming, Riemannian manifold optimization for unit-modulus RIS phase functions, and variational optimization for ARIS density control; the density update is implemented through a threshold law of the form

$\phi_r \in [0,2\pi)$9

with $r$0 chosen to satisfy the UAV budget (Luo et al., 12 Sep 2025).

Other application-specific formulations use different algorithmic stacks. Secure multi-beam satellite communications decompose the joint problem over satellite beamforming, ARIS reflection, ARIS-group association, and ARIS deployment through block coordinate descent, with SDR/SDP for beamforming and passive reflection and penalty + SCA for association and deployment (Wang et al., 28 Aug 2025). Active UAV-mounted RIS SAR imaging uses fractional programming (FP) and majorization minimization (MM) to maximize receive SNR under ARIS maximum-power and per-element amplification constraints (Sun et al., 2024). Survey papers consistently list alternating optimization, SCA, SDR, fractional programming, difference-of-convex programming, quadratic transform, and machine learning as the dominant ARIS design methodologies (Ye et al., 2021).

5. Application domains

ARIS applications span both communications and sensing. A recurrent communication scenario is emergency or remote-area recovery, where the direct source–destination path is absent or unreliable and an airborne RIS provides the only reflected route (Do et al., 2021). Aerial backhaul to temporary UAV-BSs is a more structured version of the same idea: a high-altitude ARIS reflects the backhaul signal from a ground source toward multiple UAV-BSs in an urban hotspot or disaster deployment, with the explicit aim of improving energy efficiency through lower required source transmit power (Jeon et al., 2021). Conceptual studies generalize this logic to coverage extension, spectrum sharing, physical-layer security, SWIPT, massive access, mmWave, THz, VLC, and index modulation, always emphasizing the joint role of geometry control by UAV mobility and channel shaping by RIS phase control (Abdalla et al., 2020).

In IoT systems, ARIS is used as a passive relay to improve Age of Information rather than only rate. A UAV-mounted RIS reflects status updates from randomly active IoTDs to a BS, and the control problem jointly chooses altitude, scheduling, and RIS phase shifts to minimize expected sum AoI (Samir et al., 2020). This suggests a broader interpretation of ARIS: it is not merely a coverage-extension device, but a controllable freshness or latency mechanism when relay delay, channel reliability, and scheduler choice interact.

Vehicular communications provide a different deployment logic. There the UAV-mounted RIS is a dynamic extension of the smart radio environment, complementing static building-mounted RISs. It is used for on-demand blockage bypass around corners, crash-site support, traffic congestion hotspots, and V2V continuity in NLoS street segments (Heimann et al., 2021). The low-altitude ISAC literature broadens the role further by explicitly including UAV-mounted IRS for cooperative sensing and an Aerial IRS architecture under “UAV as aerial node,” where the RIS assists both communication and sensing, improves echo SNR, and can reduce CRB for position and velocity estimation (Wang et al., 13 Nov 2025).

ARIS also extends into satellite and non-terrestrial systems. Secure satellite communications deploy multiple ARISs to enhance intended multicast groups while constraining eavesdropper rates through joint optimization of reflection, association, and deployment (Wang et al., 28 Aug 2025). In the broader RIS-assisted NTN taxonomy, ARISs support ARIS-assisted G2G, RIS-assisted A2G/G2A, and RIS-assisted A2A, including UAV, HAP, and satellite platforms (Ye et al., 2021). A separate but increasingly important sensing branch uses a UAV-mounted active RIS to turn a stationary radar into a SAR imager: the moving RIS creates a virtual synthetic aperture and a high-quality virtual LoS propagation path, and the RIS coefficients are optimized to improve receive SNR and image quality (Sun et al., 2024).

6. Constraints, misconceptions, and open directions

ARIS papers are unusually explicit about their assumptions. The outage-analysis baseline assumes no direct $r$1 link, single-user single-antenna terminals, perfect phase alignment, ideal RIS reflection with fixed $r$2, independent channels across elements, AWGN-only reception, no interference, no explicit channel-estimation overhead, no explicit UAV power-consumption model for propulsion or hovering, and no explicit Doppler or time-correlation model in the mobile case (Do et al., 2021). Secure satellite ARIS design assumes perfect CSI and sufficiently stable hovering platforms (Wang et al., 28 Aug 2025). Large-scale anti-jamming ARIS assumes fixed UAV altitude, neglects direct BS-user and jammer-user links, and adopts a continuous-density abstraction that is most faithful when the swarm is large (Luo et al., 12 Sep 2025). These are clean analytical baselines rather than full deployment models.

A common misconception is that ARIS is simply a terrestrial RIS moved into the sky. The architectural literature states the opposite: practical performance depends on an enhanced control architecture that couples mobility sensing, onboard intelligence, in-network intelligence, beam adaptation, and reconfiguration rate (Devoti et al., 2022). Another misconception is that larger altitude or larger RIS size is always better. The surveyed results repeatedly show a tradeoff: higher altitude improves LoS probability but increases path loss, and more reflecting elements improve coherent gain but narrow the beam, increase payload or control burden, and can yield diminishing secrecy or energy-efficiency returns when geometry is poorly optimized (Jeon et al., 2021, Wang et al., 28 Aug 2025).

The unresolved problems are correspondingly practical. Channel estimation remains difficult because cascaded channel dimension can scale as $r$3 for a single-user SARIS system and $r$4 for a multi-user system, while aerial LoS visibility aggravates pilot contamination (Shang et al., 2021). Control signaling and RIS configuration overhead are stressed in conceptual ARIS papers, which propose mitigations such as a few low-power active sensors embedded in the RIS, machine learning for channel estimation or control, and even UAV swarms for distributed computing and more reliable control links (Abdalla et al., 2020). Payload, battery, size, and number of elements remain tightly coupled variables, and several papers explicitly note that propulsion cost, airframe effects, calibration under UAV vibration, synchronization, and real testbeds remain open (Devoti et al., 2022, Abdalla et al., 2020).

The current research trajectory suggests two simultaneous directions. One is greater realism: hardware-aware ARIS models with imperfect CSI, phase quantization, mechanical perturbation, and propulsion-aware objectives. The other is greater scale: multi-ARIS, swarm-based, and mean-field formulations for secure, anti-jamming, or ISAC-centric systems. A plausible implication is that future ARIS systems will be designed less as isolated reflective panels and more as components of a broader reconfigurable airspace that jointly optimizes aerial placement, passive wave control, active beamforming, sensing, and mobility (Wang et al., 13 Nov 2025).