A Review of Pointing Modules and Gimbal Systems for Free-Space Optical Communication in Non-Terrestrial Platforms

Abstract

As the world is technologically advancing, the integration of FSO communication in non-terrestrial platforms is transforming the landscape of global connectivity. By enabling high-data-rate inter-satellite links, secure UAV–ground channels, and efficient HAPS backhaul, FSO technology is paving the way for sustainable 6G non-terrestrial networks. However, the stringent requirement for precise line-of-sight (LoS) alignment between the optical transmitter and receivers poses a hindrance in practical deployment. As non-terrestrial missions require continuous movement across the mission area, the platform is subject to vibrations, dynamic motion, and environmental disturbances. This makes maintaining the LoS between the transceivers difficult. While fine-pointing mechanisms such as fast steering mirrors and adaptive optics are effective for microradian angular corrections, they rely heavily on an initial coarse alignment to maintain the LoS. Coarse pointing modules or gimbals serve as the primary mechanical interface for steering and stabilizing the optical beam over wide angular ranges. This survey presents a comprehensive analysis of coarse pointing and gimbal modules that are being used in FSO communication systems for non-terrestrial platforms. The paper classifies gimbal architectures based on actuation type, degrees of freedom, and stabilization strategies. Key design trade-offs are examined, including angular precision, mechanical inertia, bandwidth, and power consumption, which directly impact system responsiveness and tracking accuracy. This paper also highlights emerging trends such as AI-driven pointing prediction and lightweight gimbal design for SWap-constrained platforms. The final part of the paper discusses open challenges and research directions in developing scalable and resilient coarse pointing systems for aerial FSO networks.

1. Introduction

With 6G aiming to extend seamless connectivity beyond terrestrial limits, non-terrestrial platforms including satellites, HAPSs, and UAVs are increasingly relying on free-space optical (FSO) communication to deliver high-capacity, low-latency, and secure links across space–air–ground networks. The rapid increase in applications of unmanned aerial vehicles (UAVs) across commercial, military, and scientific sectors has created a growing demand for high-throughput, low-latency, and secure communication links. Applications such as real-time video surveillance, environmental monitoring, aerial mapping, and autonomous swarming operations require wireless communication systems capable of supporting multi-gigabit data rates and robust connectivity because of the dynamic and interference-prone mission environments [1,2].

Free-space optical (FSO) communication has emerged as a viable solution to conventional radio frequency (RF) systems’ limited spectrum, low data rates, larger beam divergence, and lack of terminal security. FSO communication systems exploit the vast spectrum available in the optical domain, thereby enabling high data transmission capacity. Owing to their inherently narrow beam divergence, these systems offer high spatial confinement of the signal, which not only enables high-capacity links, but also enhances link security against interception and jamming. In terms of size, weight, and power (SWaP), FSO systems are proven to be more compact due to shorter wavelengths in the infrared and visible range. The resulting narrow beam divergence can be described by Equation (1), where $\theta$ is beam divergence, $\lambda$ is the wavelength, and D is aperture [3,4,5].

$$\theta \approx {\displaystyle \frac{1.22\lambda }{D}}$$

(1)

FSO systems benefit from shorter wavelengths in the optical regime, which allow the use of relatively smaller apertures compared to RF systems for achieving the same beam divergence. As a result, the transmit and receive optics are more compact and lightweight than conventional RF systems. This, however, brings the challenge of precise alignment and stabilization mechanisms for FSO systems. Components like collimators, lenses, micro electro-mechanical system (MEMS) mirrors, fast steering mirrors (FSMs), solid-state lasers, and fiber-based components contribute to low system mass. The narrow divergence of optical beams leads to highly directional links, which reduces the energy dispersion and enables lower transmit power requirements.

Further, due to the high directivity of optical beams, sidelobe loss is negligible since most of the transmitted power is concentrated within a narrow divergence angle. These conditions have made FSO systems ideal for SWaP-constrained platforms like UAVs, CubeSats, and microsatellites. However, where smaller wavelengths and narrow beam divergence are a boon for FSO links, there are also certain difficulties in these systems. FSO with non-terrestrial platforms is inherently prone to misalignment due to the continuous platform motion, atmospheric disturbances, and mechanical disturbances caused by non-terrestrial platform movements and vibrations during aerial missions [6]. Therefore, it is essential to maintain a stable optical link and precise alignment between transceivers during aerial missions. Similarly, in the context of emerging 6G non-terrestrial networks, FSO links face unique challenges. While their high capacity and secure narrow-beam communication make them ideal for inter-satellite, satellite-to-ground, and HAPS/UAV backhaul, these same characteristics also necessitate ultra-precise acquisition, tracking, and pointing (ATP). The dynamic motion of non-terrestrial platforms, coupled with atmospheric turbulence and platform jitter, can easily disrupt alignment. In order to guarantee dependable FSO connectivity in upcoming 6G NTN deployments, strong beam steering, AI-driven predictive tracking, and hybrid RF–optical control mechanisms are essential. Figure 1 provides a visual representation of the core challenges that must be overcome during FSO-based communication with non-terrestrial platforms. Precise alignment in these platforms is achieved through a two-stage process that involves coarse pointing and fine pointing. Coarse pointing modules (CPMs) are used to provide wide-range beam steering to bring the receiver within the acquisition field-of-view (FoV), whereas fine pointing systems make high-resolution angular corrections using MEMS mirrors or liquid crystal deflectors [7].

In CPMs, gimbal-based systems are widely used due to their mechanical simplicity, robustness, and ability to support large angular deflection with high repeatability [8]. Figure 2 gives a basic representation of how gimbal modules are used for maintaining the line of sight with aerial platforms and keeping the optical communication link intact. As seen in the respective figure, the gimbal has two kinds of movement, i.e., azimuth and elevation, to achieve the target, where azimuth and elevation correspond to the yaw and pitch movement, respectively. This survey reviews the current state of the art in pointing modules, with a focus on gimbal systems designed for FSO communication in aerial platforms. This paper explores mechanical configurations, actuation techniques, control strategies, and integration methods with onboard sensors and navigation systems of different models that are being used and have been published by researchers. This paper also lists key challenges such as SWaP constraints, stabilization under flight dynamics, and future directions involving AI-enhanced predictive tracking and multi-agent coordination.

2. Existing Technologies and Implementations

This section reviews the state-of-the-art literature, prototypes, and commercial/experimental systems implementing coarse pointing gimbals for FSO communication in non-terrestrial platforms. Here, this comparative analysis highlights performance metrics and trade-offs.

2.1. Literature Survey

Gimbal systems are a critical component in FSO communication systems. Over the past two decades, research in pointing system design has evolved significantly—from conventional mechanical and optical configurations to hybrid opto-electronic architectures integrated with intelligent control algorithms. Some initial investigations established a framework and the feasibility of using mechanical gimbals for alignment in ground-to-aerial FSO links. In [9], the researcher investigated the feasibility of using a mechanical gimbal for alignment and tracking in an FSO communication link between a ground station and an aerial platform. The experiment used a 633 nm helium–neon laser on a gimbal which was mounted on a vibration-isolation optical table. The laser position was recorded with a high-resolution position-sensing photodiode positioned 1.77 m from the source. The data acquisition and gimbal were controlled via a computer, and a total of 14,000 data points were recorded for analysis. The simulation software was used to model the effects of atmospheric turbulence on the transmitted laser beam to study beam divergence and geometric loss with a 4 km altitude. The authors analyzed their gimbal’s repeatability and error, which allowed them to use that error distribution in the alignment and tracking algorithms. As per their analysis, the gimbal’s repeatability for elevation was 0.41 m ($0.{004}^{\circ }$) and for azimuth was 1.24 m ($0.{013}^{\circ }$), with an overall pointing error of 0.3 m ($0.{0032}^{\circ }$). The experimental results showed that a mechanical gimbal is capable of the precision needed for the link. The simulation results indicated that the natural beam divergence in FSO links—even under atmospheric turbulence—can effectively compensate for gimbal repeatability and accuracy errors. This suggests that gimbals can serve as an efficient tool for ATP in ground-to-UAV FSO links. However, the experiments were conducted on a vibration-isolation table, whereas in real-world scenarios, mechanical gimbals are subject to inherent limitations such as friction, inertia, and unmodeled dynamics, all of which can degrade their precision.

Similarly, Giggenbach et al. [10] proposed a free-space experimental laser terminal (FELT) for high-altitude platforms (HAPs), where a motorized periscope was used for beam steering. The periscope was able to rotate on two axes, i.e., azimuth and elevation, to direct the optical beam. The periscope had a clear aperture of 50 mm. It was used to steer the outgoing laser beam and focus the incoming beacon light from the ground station onto a CMOS tracking camera, and then the video signal was processed by a compact vision system, which ran control algorithms to keep the periscope aimed at the ground station’s beacon. The optical ground station (OGS) used a receiver system with a 40 cm aperture diameter, which consisted of a cassegrain telescope and an optical bench. A weather balloon platform was used for trials during different weather conditions. The system was compact, and a hybrid approach for fine pointing was used. The paper did not explicitly mention what mechanism was used for fine pointing. The authors also faced specific problems with the setup, as the weather balloon used in the trial could rotate up to six rounds per minute, and this rotation and limited two degrees of freedom (DoFs) created a strong impact and limitations on the terminals’ ATP system. The FELT payload lacked attitude sensors, and due to this it could not point directly at the ground station. Instead, it had to scan for the OGS beacon along a ring of uncertainty. Alongside this, the authors highlighted a major long-term signal outage due to cloud coverage. While the setup was compact and less complex due to its two DoFs, the setup’s total mass was 18 kg. HAPs, particularly solar-powered fixed-wing aircraft, are designed to be extremely lightweight to conserve energy for long-endurance flights. For example, NASA’s pathfinder-plus aircraft mentioned by the authors had a payload capacity of 67.5 kg [11], but airbus zephyr has a much lower payload capacity of only 5 kg [12]. The weight of HAP payload is a critical factor and depends heavily on the specific platform’s design, power capabilities, and mission.

Subsequent studies extended mechanical gimbals with closed-loop feedback and embedded control algorithms. To support this, authors in [13] developed an auto-alignment transceiver system for maintaining LoS with mobile platforms. The system utilized a mechanical gimbal with servo motors, a position-sensing detector (PSD) to receive the signal, a 40 mW industrial laser module for data transmission, and a computer for coordination. It was a lab-based setup, where the system used a simple proportion algorithm that calculated gimbal coordinates based on the laser spot’s position on the PSD and a pre-determined correction coefficient. This coefficient varied across different zones of the PSD to ensure accurate recentering and prevent overshooting. The system’s efficiency was tested on a mobile unit on a model train track, and it maintained alignment even at angular velocities up to $3.{21}^{\circ }$/s. The gimbal specifications indicated a maximum angular speed of 60°/s, and experimental results further demonstrated that the system was capable of maintaining power above the critical thresholds during operation, thereby validating its overall efficiency.

In [14], the authors developed a high-performance, two-axis gimbal system for free-space laser communications onboard UAVs. Their system aims to achieve affordable, reliable, and secure air-to-air laser communication between two UAVs. The system was designed with several efficiency factors in mind. The gimbal structre is lightweight, with a target mass of 2.3 kg. This custom-designed gimbal offers a ${180}^{\circ }$ FoV in both azimuth and elevation, with increased velocities of up to ${479}^{\circ }$ per second. It is a 24-volt system with integrated motor controllers and a driver. This system also complements a passive vibration-isolation system. The gimbal uses piezoelectric servo motors, a signal amplifier, a motor controller, an onboard flight computer, and a tracking algorithm. The use of piezoelectric servo motors provides several technical advantages, including an extremely high rotational resolution of 0.069 arc seconds, no intrinsic magnetic field, and the ability to eliminate servo dither when static. Their tracking algorithm was developed to aim an airborne laser at a stationary ground station. With known GPS coordinates, the algorithm autonomously calculates an LoS vector in real time using the UAV’s differential GPS (DGPS) and inertial measurement unit (IMU) data along with the ground station’s GPS location. The tracking algorithm is executed on an onboard embedded flight computer running a lightweight Linux distribution, minimizing overhead and maximizing processing power for real-time calculations. But, on the contrary, the current numerical tracking algorithm provides only coarse resolution tracking. To achieve the level of accuracy required for fine pointing, an enhanced tracking method is needed, as the accuracy of the algorithm is dependent on the DGPS and IMU data. The authors have acknowledged this and are investigating a hybrid system that would use machine vision for fine-resolution tracking to augment the existing DGPS/IMU system.

In parallel, the authors in [15] designed a prototype of a compact, lightweight, two-axis gimbal for air-to-air and air-to-ground laser communication with UAVs. The entire structure is made of carbon fiber and magnesium composite to achieve a lightweight design. The total mass of the structure is 3.5 kg. The gimbal incorporates a refractive telescope with a 7.5 cm diameter aperture folded between mirrors, and an FSM for fine pointing. The gimbal uses custom-built servo motors with optical encoders, where the azimuth stage connects via a slip ring, and the elevation stage is equipped with passive optics, supported with custom-built ceramic-on-steel bearings. The design allows for a hemispherical $+{30}^{\circ }$ field of regard and is intended to cover both air-to-ground of $\pm {60}^{\circ }$ cone around nadir and air-to-air of $\pm {20}^{\circ }$ annulus communication requirements. Although the manuscript does not mention which tracking algorithm they are using, their lab-based demonstration registers stable operation and an effective performance in demanding environments. The system is designed to be mechanically robust, with a high first mode of natural frequencies above 400 Hz and the ability to operate in a wide temperature range from $+{60}^{\circ }$C to $-{80}^{\circ }$C. However, the primary challenges lie in the complexity of the mechanical design, assembly, and optical alignment processes, which are compounded by the non-orthogonal structural layout and the multi-mirror optical path configuration. The total cost of the system is also quite high, because high-precision custom built components were used and carbon fiber and magnesium composite are lightweight but expensive to manufacture.

To address similar issues of cost and SWaP, the authors in [16] exerted efforts and used a micro-electro-mechanical system (MEMS)-based modulating retroreflector (MRR) as the communication terminal onboard the UAV. Their design significantly reduces power, size, and weight on the UAV by eliminating the need for a laser transmitter and ATP subsystems on the UAV platform. As the gimbal was manufactured by FLIR motion control systems, based on publicly available data, their pan-tilt gimbal’s weight varies between 2.1 kg and 11.8 kg [17]. The MRR combines a retroreflector with an optical modulator. The retroreflector sends an incoming laser beam back along the same path to its source without any additional pointing requirements on the UAV. In contrast, the ground station design utilizes ATP as a subsystem, where a two-axis gimbal for coarse pointing and an FSM are employed. This gimbal features a high-speed control system with a speed of up to ${100}^{\circ }$/s and a resolution of $0.{0064}^{\circ }$, while the FSM has a maximum angular resolution of $2\mathsf{\mu }$rad. The system uses a beacon-based tracking algorithm, where the laser beacon is employed at the ground station. An IR camera monitors the reflection of the beacon from the MRR to determine the UAV’s exact position. Fine positioning is achieved by correlating the beacon image’s position on the camera’s focal plane with the necessary FSM movement to illuminate the MRR. The system also manages laser power optimally through distance-dependent beam-divergence control. It also uses a polarization discrimination system with a Faraday rotor and polarized beam splitter to create separate optical paths for transmitted and received beams. This helps in minimizing power loss and achieving a measured gain of 7 dB over a standard 50-50 beam splitter. The authors compared two technologies for the MRR, liquid crystals and MEMSs. The liquid crystal approach was lightweight and consumed little power, but its slow response time limited data rates to under 100 kbps. While the MEMS-based modulator was capable of achieving data rates up to 1 Mbps, the MEMS-based modulating retroreflector (MRR) required a driving voltage exceeding 100 V to attain high optical contrast. The integration of the MRR in this system effectively eliminated the need for a complex acquisition, tracking, and pointing (ATP) subsystem onboard the aerial platform and transferred the weight and power requirements of the laser transmitter to the ground station, thereby simplifying the airborne payload architecture. However, it put a significant burden on the ground station that it had to handle by both transmitting and receiving on a single optical axis.

In regard to SWaP, gimbal-less architectures have also emerged as an alternative. The authors in [18] presented a gimbal-less body-pointing architecture, where the laser was rigidly mounted to the Aerocube-7 cubesat’s body while the entire 1.5U cubesat acted as the pointing mechanism. To aim the laser, the spacecraft’s attitude control system (ACS) would reorient the whole satellite. This was made possible by an extremely precise ACS that included miniature star trackers, sun sensors, and reaction wheels. This system allowed the spacecraft to achieve a pointing accuracy of better than 0.005 degrees. Later in [19], the authors launched Aerocube-10A and Aerocube-10B, each a 1.5U cubesat for inter-satellite laser pointing. 10A emitted a laser beacon while 10B used its optical sensor to detect the laser. Both of the cubesats used their advanced ACS. The authors were successful in validating the performance of their gimbal-less body-pointing architecture, which was actuated by reaction wheels. This kind of setup offers a high SWaP efficiency because it eliminates the need for a bulky, power-consuming mechanical gimbal. By integrating the pointing function with the ACS, the satellite can use its existing reaction wheels and torque rods to orient the entire spacecraft for coarse pointing. The star tracker provides precise attitude solutions, which is a key advantage for accurate pointing. But this method is highly susceptible to a single point failure in the ACS. As demonstrated by the software anomaly, if the main ACS processor fails, the FSO communication system becomes completely inoperable, whereas a dedicated gimbal might provide some redundancy in such a scenario.

Researchers have also emphasized analytical modeling and control refinement for mitigating dynamic disturbance during optical transmission. In [20], the researcher focuses on a theoretical framework and analytical model for a spatial beam tracking and data detection module for FSO communication with UAVs. Instead of using a bulky mechanical gimbal, the authors use a quadrant photodetector (QPD) array for optical beam tracking. They employ a maximum likelihood criterion for spatial beam tracking when channel state information is known. For unknown CSI scenarios, authors have proposed a blind channel estimation method that does not require pilot symbols but enhances the bandwidth efficiency. Then, data detection uses those results. However, this method increases computational complexity and can introduce detection delay due to the need for a sufficiently long observation window. The efficiency of the tracking model is assessed by analyzing tracking error probability and bit-error rate (BER) through derived closed-form expressions and Monte-Carlo simulations. Their simulations register that hovering fluctuations severely degrade the system performance, but this can be mitigated by optimizing the detector size and balancing the increased field of view against background noise and a reduced electrical bandwidth. The authors also optimized the length of the observation window, representing a trade-off between performance and system complexity. Extending this work, the researchers in [21] further explored Monte-Carlo simulations and used optimal detector sizing based on UAV motion statistics to develop blind channel estimation algorithms for a higher bandwidth. But there is a compromise between mitigating beam position deviation and accepting undesired background noise. A larger detector size helps to mitigate performance degradation from hovering fluctuations by increasing the receiver’s field-of-view (FoV), but it also increases background noise and reduces the detector’s electrical bandwidth.

In [22], the authors developed a statistical channel model for FSO communication between two multi-rotor UAVs equipped with oscillating mirror terminals. The model accounts for atmospheric loss, turbulence, and pointing error caused by UAV jitter. Pointing error is modeled using normal distributions for azimuth $(\sigma ={0.4}^{\circ })$ and elevation $(\sigma ={0.05}^{\circ })$, with the Hoyt distribution used to capture anisotropic jitter. The framework enables derivation of the joint PDF and evaluation of link performance under realistic UAV vibrations. The oscillating mirror approach is lightweight and SWaP-efficient, making it well-suited for UAV payloads. However, it is also highly vulnerable to vibration and attitude changes, which remain the dominant source of pointing error. In [23], the authors proposed a framework for FSO communication using a Reconfigurable Intelligent Surface (RIS)-equipped UAV, optimizing UAV trajectory and RIS phase shifts to maximize capacity under atmospheric loss and pointing error. They used particle swarm optimization (PSO) for RIS phase control and proximal policy optimization (PPO) for UAV trajectory, achieving a better performance than DF relays and DQN. The results showed that combining PSO and PPO significantly improved average capacity and mitigated fog and pointing error effects. RISs offer energy-efficient passive phase adjustment, but practical challenges remain, including UAV power consumption, processing demands, and sensitivity to atmospheric conditions. The authors of [24,25] have also investigated FSO communication with RIS-equipped UAVs. Further, the authors in [26] designed a two-axis stabilization loop using a linear quadratic Gaussian (LQG) controller with loop transfer recovery, positioning it as a robust alternative to traditional controllers like lead-PI or PID. Unlike PID, which struggles with nonlinearities, LQG handles MIMO systems and incorporates disturbances and sensor noise as stochastic inputs via a Kalman filter. The experimental results showed better stability, with higher gain and phase margins, and improved decoupling between pitch and yaw, reducing cross-coupling effects. The main limitation was that LQG, being a linear method, required simplifying the nonlinear gimbal model by omitting the low-frequency spring constant—a compromise validated as having minimal impact in the frequency domain.

At the same time, the authors in [27] demonstrated a system with a piezoelectric fast steering mirror (PFSM) along with a Hammerstein structure and MPC control. The aim was to achieve nanoradian precision, a higher bandwidth, and better handling of MIMO dynamics, hysteresis, and axis coupling than PID or DIM controllers. In [28], the authors proposed a generalized pointing error model for UAV-based FSO communication, incorporating 3D jitter from the roll, pitch, and yaw of a fixed-wing UAV. The model integrated FSO link design, pointing error modeling, and trajectory optimization, solved using successive convex approximation (SCA) and Dinkelbach methods. Simulations showed that optimizing UAV trajectories with this model improved energy efficiency by up to 11.8% compared to conventional approaches. The model highlights trade-offs, as trajectory solutions may show oscillations under dominant yaw jitter, though they remain feasible. Key assumptions—such as fixed UAV altitude and a single ground station—simplify the problem but limit applicability in dynamic networks. Additionally, reliance on approximations means the algorithm may not always achieve a globally optimal solution. However, these FSM-assisted systems present promising fine-tracking capabilities for hybrid ATP configurations, combining mechanical range with optical precision.

In NASA’s optical communication and sensor demonstration (OCSD) mission, instead of a gimbal, the architecture employed a body-pointing system, where a star tracker and GPS were used as tracking aids. Both of these systems employ body-pointing, which avoids the need for complex gimbals. OCSD was one of the first demonstrations to prove that a cubesat could achieve the pointing accuracy required for laser communications by steering the entire spacecraft body. It was successfully able to demonstrate downlink of up to 200 Mbps [29]. Similarly, in [30], the author proposed and implemented an architecture for terabit-class optical downlinks for NASA’s terabyte infrared delivery (TBIRD) cubesat, where they were able to transmit more than 200 Gbps per pass to ground stations. It was a 6U cubesat, and the same fundamental method was used for its downlink. The optical subassembly included both the transmit and receive apertures and was mounted in a fixed monolithic housing that contained no steering mirrors. To aim the downlink laser, the entire spacecraft bus maneuvered using its reaction wheels. The primary limitation in this setup was in the ground station’s adaptive optics system. The throughput was not limited by atmospheric turbulence, but by internal disturbances within the telescope and back-end optics. These disturbances caused large power fluctuations (>20 dB) in the signal coupled to the fiber. This led to communication dropouts and a measured throughput of 150 Gbps, below the operating rate of 200 Gbps.

Moreover, intelligent trajectory prediction using transformer-based models such as the trajectory embedding transformer (TET) offers new potential for anticipatory gimbal control. In [31], the authors present a novel model involving TET for long-term trajectory prediction, primarily for aircraft. The authors use the automatic dependent surveillance broadcast (ADS-B) system as their data source, specifically the opensky network. This TET model is based on the transformer’s encoder–decoder structure. Since the attention mechanism lacks a sense of sequence, positional encoding is added to the input trajectory data. The encoder’s role is to map a sequence into a vector space that contains learned information about the entire sequence. It uses a multi-head self-attention layer, normalization, and residual connections to capture both long-term and short-term features of the trajectory data. Then, the decoder takes the encoded information from the encoder and combines it with a position-encoded real trajectory to predict the future trajectory. The entire setup is a lab-based experiment. The training and testing are conducted on an ubuntu operating system using an nvidia GeForce RTX 3090 GPU and an intel xeon CPU. The model is shown to be more robust in long-term trajectory prediction. This model demonstrates a stronger ability to learn from features from long-time series data. It reduces average displacement error (ADE) by 8.2% and final displacement error (FDE) by 51.4%. Their transformer model addresses the limitations of recurrent neural networks (RNNs), such as memory loss and vanishing or exploding gradients, when dealing with long time series data. Unlike RNNs, which process data sequentially, the transformer’s attention mechanism allows it to encode trajectory data at once. This enables parallel processing and leads to shorter training and inference time compared to RNN-based models. But, despite its efficiency in parallel processing, the TET model has a drawback in training time due to its large number of model parameters and longer path length from input to output for the same data set. Although the paper primarily focuses on predicting future trajectories, accurate trajectory prediction can significantly enhance the performance of an acquisition, tracking, and pointing (ATP) system. By providing anticipatory positional data, it enables the gimbal mechanism to preemptively adjust its orientation, thereby minimizing the need for reactive corrections and improving overall pointing stability and responsiveness. Similarly, in [32,33], the authors investigate this model, which can improve an ATP system’s speed, efficiency, and precision.

In [34], the author investigates a dual-link communication scheme combining RF and FSO links to support federated learning (FL) tasks. They propose an asynchronous FL model where FSO links are used for high-capacity, low-latency transmission in clear weather, and RF links serve as a fallback in adverse conditions. In [35], the authors propose an adaptive optics (AO) compensation scheme to mitigate the effects of atmospheric turbulence on FSO communication. The system uses a deep learning model called the mixed convolution channel attention prediction network (MCCA-PNet) to predict turbulence from distorted beam intensity patterns. This prediction is then used to correct the beam in real time. The authors investigate this in a lab-based environment, where the beam propagation distance is set to 1000 m. The FL allows the authors to train the MCCA-PNet model without sharing their raw data. The server aggregates model parameters from clients and builds a personalized parameter dictionary based on layer similarity to improve training efficiency and accuracy. The total computation time of the MCCA-PNet-based AO scheme is just 0.06 s with GPU acceleration, making it suitable for real-time applications. This is a major improvement over classic mathematical algorithms. However, the core experiments were conducted in a lab environment using an environmental chamber to simulate turbulence. While this validates the model’s effectiveness, a real-world pilot project would be needed to confirm its performance in a truly open-air, dynamic FSO link with unmodeled atmospheric effects. Also, the system uses a Gaussian beam to measure turbulence, which is separate from the signal beam. This requires additional hardware like a polarization beam splitter and may not be feasible in all FSO system designs, especially those with strict SWaP constraints.

Similarly, the authors in [36] used federated reinforcement learning (FRL) to optimize the scheduling of laser inter-satellite links (LISLs) in mega-constellations. The authors proposed an asynchronous hierarchical federated learning architecture, where LEO satellites (clients) perform local training using a multi-agent deep reinforcement learning (MADRL) method, while GEO satellites (servers) perform partial aggregation of models from the LEO satellites they cover. It was a simulation-based experiment using an LEO constellation similar to oneweb, with 720 LEO satellites and 3 GEO satellites. The simulation used the aerospace toolbox in MATLAB to model satellite coordinates and movement. The link distance of the LEO–GEO uplink communication distance ranged from 34,593 km to 42,184 km. The MADRL method reduced the average hop count by about two hops and decreased the number of LISLs by over 25%, which led to lower network latency and energy consumption. Although the reported performance improvements were derived from mathematical models and simulated traffic distributions, they do not reflect real-world deployment scenarios. Critical factors such as unpredictable solar weather events, hardware malfunctions, and sensor-induced noise were not incorporated into the simulations. Furthermore, the authors utilized RF links for LEO-to-GEO satellite communication to minimize the hardware overhead associated with acquisition, tracking, and pointing (ATP) mechanisms. This represents a notable limitation, as the model’s effectiveness could potentially be enhanced through the use of free-space pptical (FSO) uplinks, which offer a higher bandwidth and reduced latency.

Table 1. Literature survey summary table.

Reference	Study Type	Link Distance	Turbulence Considered	Other Key Parameters
[9]	Lab experiment + Simulation	1.77 m lab test; 4 km altitude simulated	Yes (simulated beam divergence and turbulence)	633 nm He-Ne laser, gimbal repeatability $0.{0004}^{\circ }$ elevation/$0.{013}^{\circ }$; tested on vibration-isolated table
[10]	Real-world balloon trial	Balloon platform, 50 mm aperture periscope	Limited (cloud outages noted; turbulence not quantified)	Compact FELT payload (18 kg), 2-DoF periscope steering, CMOS camera + Vision tracking
[13]	Lab experiment (model train track)	Short range (in meters)	No	Mechanical gimbal + PSD; simple proportional algorithm; angular velocities up to $3.{21}^{\circ }$/s
[14]	Prototype for UAV (lab tested)	Air-to-Air (targeted km-scale, but not specified)	No	Lightweight (2.3 kg) 2-axis gimbal, piezo servo motors (0.069 arcsec resolution), DGPS/IMU-based tracking.
[16]	Ground-to-UAV experiment (lab + concept)	Indoor prototype; scalable to km	No (beacon assumed ideal)	MEMS-based modulating retroreflector (MRR). gimbal + FSM on ground; beacon-based tracking
[15]	Lab prototype (UAV gimbal)	Indoor demo	No	Compact carbon-fiber/magnesium gimbal, 3.5 kg, refractive telescope + FSM.
[18,19]	Real-world cubesat missions	LEO body pointing (∼500 km)	No explicit turbulence model (space vacuum)	Body pointing using ACS (reaction wheels, star tracker, GPS). Achieved <${0.005}^{\circ }$
[20,21]	Simulation (theoretical model)	UAV hovering link; distances not specified	Implicit; UAV fluctuations modeled statistically	QPD-based tracking; ML-based channel estimation. BER + tracking error analyzed with Monte Carlo simulations
[22]	Simulation (Channel model)	UAV-to-UAV, variable	Yes; pointing error with Hoyt distribution	Statistical model including UAV jitter; oscillating mirror terminals; anisotropic error distributions
[23]	Simulation (RIS-assisted UAV)	UAV–ground; distances not explicit	Yes; atmospheric loss + pointing error modeled	RIS-based optimisation with PSO/PPO; improved capacity and robustness. Energy-efficient passive beamforming
[26]	Lab/Simulation (gimbal control loop)	2-axis gimbal (short range)	No turbulence; focused on control stability	LQG with loop transfer recovery; better margins vs PID.
[29,30]	Real-world cubesat missions	LEO–ground (200–1000 km)	Yes; ground telescope AO turbulence	Body pointing with star trackers and RW. TBIRD achieved >150–200 Gbps downlink
[27]	Lab experiment	Short-range optical bench	No atmospheric turbulence	MPC-controlled piezo FSM; nonradian precision, high bandwidth
[28]	Simulation (UAV model)	UAV–ground (fixed altitude, 3D jitter)	Yes; roll/pitch/yaw jitter modeled statistically	Joint trajectory + pointing optimization; 11.8% energy efficiency gain
[31]	Simulation/Lab training	Aircraft ADS-B data; prediction only	Not turbulence, but trajectory errors modeled	Transformer-based long-term trajectory predictor (GPU trained). ADE reduced by 8.2% FDE by 51%
[34]	Simulation	UAV swarm; distances not explicit	No	Collaborative pointing via federated learning; dual RF/FSO links
[35]	Lab experiment	1 km simulated	Yes; turbulence chamber + Gaussian probe beam	Deep learning AO correction (MCCA-PNet)
[36]	Simulation	LEO–GEO (∼34,593–42,184 km)	No turbulence	Federated reinforcement learning for constellation scheduling; reduce hop count and latency

lists a summary of the literature survey. It registers that most studies focus on simulation with idealized turbulence models, while few real-world validations exist. Lab setups have often neglected turbulence or used simplified emulators. Control algorithms like PID, LQR, and EKF dominate, while advanced controllers like MPC, sliding-mode, H-infinity, and AI-based controllers are rarely tested under experimental turbulence. This shows a clear gap in large-scale outdoor experiments covering diverse turbulence regimes and SWaP-constrained platforms (UAVs, cubesats).

2.2. Motivations and Contributions

One of the primary motivations for this review came from the lack of a consolidated understanding of the trade-offs between gimbal-based and gimbal-less ATP designs in non-terrestrial platforms. Early research primarily focused on the validation and demonstration of the mechanical feasibility of gimbal systems for short-range optical alignment under controlled conditions. But those laboratory setups lacked a key consideration in calculations, namely the dynamic real-world environment. Therefore, some experiments and investigations with these platforms transformed from laboratory setups to real-world flight experiments. This led to the emergence of new constraints in the form of structural vibrations, limited payload capacity or SWaP, and aerodynamic coupling effects. While some investigations cited in the literature survey have demonstrated a milliradian-level pointing accuracy, very few have explored how such precision translates under real-world conditions where coupling between attitude control and optical subsystems becomes non-negligible. This leads to a need to systematically evaluate how design variables such as material composition, actuation type, and control algorithms collectively influence FSO link reliability.

This review is also motivated by the absence of standardized performance benchmarks and cross-platform comparison frameworks. Studies referenced in the literature survey employ varied metrics such as pointing accuracy, response bandwidth, repeatability, and angular range without consistent normalization across different system scales and operational environments. Consequently, identifying the relative maturity or suitability of specific ATP technologies for non-terrestrial platforms such as UAVs, HAPs, or CubeSats becomes challenging. A systematic comparative framework that correlates actuation type, control method, and environmental tolerance with FSO link metrics such as received power, BER, and outage probability is essential for guiding future research and engineering design.

The key contribution of this review paper is that it provides a quite comprehensive synthesis of ATP mechanisms for FSO communication with non-terrestrial platforms. It identifies critical design trade-offs in pointing accuracy, structural complexity, bandwidth, and SWaP, forming a unified taxonomy for evaluating coarse and fine pointing mechanisms. This paper also introduces a forward-looking perspective on the integration of intelligent and data-driven control strategies such as transformer-based trajectory prediction, federated learning-based collaboration, and reinforcement learning-based control optimization with mechanical gimbal systems. This synthesis also establishes a conceptual foundation for next-generation adaptive ATP architectures which can combine mechanical precision, optical agility, and computational intelligence.

Overall, this review paper tries to bridge the gap between traditional optomechanical stabilization methods and modern AI-enhanced control frameworks, offering a multidimensional understanding of how future FSO systems can achieve robust, scalable, and autonomous alignment across dynamic aerial and spaceborne environments.

3. Fundamentals of FSO Communication for Non-Terrestrial Platforms

An FSO communication system transmits a collimated optical laser beam within the near-infrared (2800 nm) to visible light (380 nm) wavelength range from the transmitter aperture to the receiver aperture through the atmospheric medium. The received optical power $P_r$ is governed by the link budget (Equation (2)), as follows:

$$P_r=P_t·T_t·T_r·L_{geo}·L_{atm}·L_{point},$$

(2)

where $P_t$ is transmitted power, $T_t$ and $T_r$ are transmitter and receiver optical efficiencies, $L_{geo}$ is geometric loss (due to beam spreading), $L_{atm}$ is atmospheric loss (due to scattering, and absorption), and $L_{point}$ is pointing loss due to misalignment. Figure 3 depicts a laser traveling from a ground transmitter to a target aerial platform and along the path, visually representing each loss component from Equation (2).

Among these, $L_{point}$ or pointing loss is highly sensitive, particularly with non-terrestrial platforms, as even sub-milliradian angular deviations can lead to significant losses [37]. Therefore, maintaining LoS is fundamental to FSO operation with aerial platforms, and to support this, optical laser beams with divergence angles in the order of a few milliradians, depending on the optics used, are implemented. This restricts the beam footprint to a few centimeters at moderate distances. But any deviation due to aerial platform translation, rotation, or vibration can still cause the receiver to fall outside the beam cone, leading to breakage of the link [7]. The following subsection outlines the different types of non-terrestrial links along with their associated challenges.

3.1. UAV–Ground FSO Link

Generally, UAVs experience continuous six degrees of freedom (DOFs) motion, affected by wind gusts, flight maneuvers, and control lag. These dynamics introduce platform jitter and spatial drift; these hindrances affect the LoS, leading to link outages if not compensated by real-time tracking mechanisms [38]. In addition to this, atmospheric effects such as aerosol scattering, turbulence-induced beam wander, and scintillation add stochastic signal degradation, particularly at altitudes below 2 km [39,40,41].

To address the LoS sensitivity and beam drift issue, a real-time tracking mechanism is needed. Therefore, FSO systems employ ATP subsystems. The ATP system is further divided into a coarse pointing module (CPM) and a fine pointing module (FPM). The CPM is responsible for the acquisition phase, where it brings the ground station into the platform’s FoV and broadly aligns the optical beam toward the aerial platform. CPMs are generally mechanical systems that use servo or BLDC motors and encoders for orientation control and support the fine pointing modules by keeping the UAVs in the FOV. These are effective for coarse pointing but are limited in speed and resolution [42].

The second module, i.e., the FPM, is responsible for achieving precise control of the beam direction using high-bandwidth actuation systems by maintaining link stability and ensuring the optical beam stays within the receiver aperture for high-data-rate transmission [43]. This fine pointing is achieved by using fast steering mirrors or MEMS devices, optical beam deflectors, liquid crystal beam steerers, etc. Also, ATP systems must be lightweight and responsive to be compatible with UAV payload constraints [44].

3.2. HAPS–Ground FSO Links

A HAPS-to-ground FSO link presents a unique engineering puzzle. It combines the atmospheric challenges of a terrestrial link with the platform stability issue of an aerial vehicle. The pointing system must be a master of both domains, i.e., robust enough to handle platform motion and precise enough to cut through atmospheric turbulence. The central challenge for this link is compensating for the unique motion of the HAPS. Unlike a satellite on a predictable orbit or a fast-vibrating UAV, a HAPS exhibits low-frequency and large-amplitude oscillations. Similar to satellites-to-HAPS links, these slow changing stratospheric winds cause the HAPS platform to drift and sway. While the frequency is low, the angular displacement can be significant [44,45].

Furthermore, the FSO link travels from near-vacuum at a 20 km altitude down through the entire atmosphere to the ground. This path is a gauntlet of atmospheric turbulence, with the most severe effects concentrated in the last few kilometers near the ground. So, the beam is affected by beam wandering effects and scintillation effects. The strength of turbulence $\left(C_n^2\right)$ varies significantly with altitude $\left(h\right)$. This is often modeled using the Hufnagel–Valley (H-V) model, which provides a realistic profile of $C_n^2$ versus altitude (Equation (3)), as follows:

$$C_n^2\left(h\right)=0.00594{\left({\displaystyle \frac{v_w}{27}}\right)}^2{\left({10}^{-5}h\right)}^{10}{e}^{-h/100}+2.7\times {10}^{-16}{e}^{-h/1500}+A_0{e}^{-h/100},$$

(3)

where $v_w$ is the upper-atmosphere wind speed in m/s and $A_0$ is the ground-level turbulence strength. This equation shows that turbulence is strongest near the ground and drops off rapidly with altitude [3,46,47].

3.3. Satellite–Ground FSO Links

Satellite-to-ground FSO links are fundamentally dominated by the challenge of transmitting a laser through the Earth’s turbulent and unpredictable atmosphere. While the satellite itself is a stable and predictable platform, the atmospheric channel is a chaotic medium that distorts and degrades the optical signal. Atmosphere is not a uniform medium and is composed of pockets of air, or eddies, with varying temperatures and pressures, which lead to different refractive indices [3]. The strength of the atmospheric turbulence can be characterized using the refractive index structure parameter $C_n^2$ [48,49]. This lack of uniformity leads to rapid fluctuations in the intensity of the received laser signal, analogous to the tinkling of stars. The turbulence cells act like random lenses, leading to focusing and defocusing parts of the beam and creating bright and dark spots at the receiver. These fluctuations cause deep signal fades which further lead to burst errors. The magnitude of these fluctuations is described by the scintillation index $\left({\sigma }_I^2\right)$, which is the normalized variance of the signal intensity in Equation (4) [50,51], as follows:

$${\sigma }_I^2={\displaystyle \frac{\langle I^2\rangle -{\langle I\rangle }^2}{{\langle I\rangle }^2}},$$

(4)

where $\langle I\rangle$ is the mean intensity. For weak turbulence conditions, this is often approximated by the rytov variance $\left({\sigma }_R^2\right)$. Turbulence also has large-scale turbulent eddies which act like random prisms, deflecting the entire beam from its intended path. This effect is called beam wander. This effect is the tilt component of the wavefront distortion. To correct this, an FSM in the transmitter’s pointing module is specifically used.

3.4. Satellite–HAPS FSO Links

Satellite-to-HAPS FSO links area unique hybrid scenario, where there is a mixture of challenges of long-distance communication with the complexities of an atmospheric platform. Therefore, the poinitng system for such a link should be a master of two worlds, i.e., the vacuum of space and the stratosphere. A HAPS is neither a satellite nor a fast-moving UAV. It operates at an altitude of 20 km and is designed to remain relatively stationary over a specific location. However, it is subject to stratospheric winds, causing slow but persistent drift and sway. This introduces a unique low-frequency disturbance that the pointing module must counteract. The pointing control system on the HAPS must have a sufficient control bandwidth to cancel out these motions to maintain alignment with the satellite. For a link from a GEO satellite to a HAPS, the distance is 35,780 km. Due to this distance and the difference in environmental conditions, the link is inherently asymmetric, especially in a GEO to HAPS scenario. A satellite, particularly a GEO, can host a larger, more powerful, and heavier optical terminal, while the HAPS payload is significantly constrained by weight and power. This means HAPSs will have a smaller aperture, lower transmit power, and a more lightweight pointing gimbal [52].

3.5. Inter-Satellite FSO Links

LEO–LEO inter-satellite FSO links face unique challenges due to the dynamic environment. LEO satellites orbit the earth at approximately 7.8 km/s. The relative veolcity betweeen two LEO satellites can be up to twice this value, or nearly 15.6 km/s. This high relative motion introduces significant pointing challenges like doppler shift and point-ahead angle [53,54].

Doppler Shift: The high relative velocity causes a significant doppler shift in the frequency of the received laser light. The doppler shift $\left(\mathsf{\Delta }f\right)$ can be calculated as follows in Equation (5):

$$\mathsf{\Delta }f={\displaystyle \frac{v_{rel}}{\lambda }}$$

(5)

where $v_{rel}$ is the relative velocity between the satellites and $\lambda$ is the wavelength of the laser [55]. This frequency shift must be compensated by the receiver’s tracking system to maintain the lock of the signal. While this is more of a receiver challenge than a pointing one, it is a direct consequence of the high relative velocities that the pointing system must accommodate [54].

Point-Ahead Angle: Due to the finite speed of light, the transmitting satellite must aim the laser beam not at the current position of the receiving satellite, but at its anticipated position when the light arrives. This is known as the point ahead-angle. The point-ahead angle $\left(\alpha \right)$ can be calculated as follows in Equation (6):

$$\alpha ={\displaystyle \frac{2v_{rel}}{c}},$$

(6)

where $v_{rel}$ is the relative velocity component perpendicular to the LoS and c is the speed of light [54,56]. Given the high relative velocities, this point-ahead angle can be substantial and is constantly changing as the satellites move in their orbits. The ATP system must continuously calculate and apply this correction with extreme precision. An error in the point-ahead calculation can directly translate into a pointing error [57].

Now, to keep the narrow laser beam centered on the distant receiver, the pointing accuracy should be in microradians, which is equivalent to arcseconds. Therefore, the overall pointing error is a combination of quasi-static errors and high-frequency jitter. Quasi-static errors are slowly varying errors which can be generally compensated by the satellite’s attitude determination and control system (ADCS). These also include the attitude determination and thermal distortion inaccuracies of the satellite structure. Alongside this, jitter is the high-frequency vibrations of the satellite bus, which can be caused by the operation of reaction wheels, thrusters, and other mechanical components. These vibrations lead to rapid steering of the laser beam off the target. To counter these situations, two-stage control systems, i.e., CPA and FPA modules, are used. Here, the CPA module is often controlled by the satellite’s ADCS, while FPA uses the high-bandwidth FSMs [58].

Alongside these, other significant challenges include acquisition and solar conjunction. Because before communication can begin, the two satellites must first acquire each other in space, since the initial position uncertainty of each satellite can be much larger than the laser beam’s divergence.

To address this, the transmitting satellite may scan a predefined uncertainty cone with a wider beacon beam or use an RF-assisted system until the receiving satellite detects the signal [59]. After acquisition, communication can still be disrupted if the sun is near the LoS, as its radiation may saturate the detectors [60,61].

LEO–GEO FSO links represent one of the most demanding applications for pointing systems. One of the defining challenges of a LEO–GEO link is the vast distance between the targets. In this link, the minimum distance is up to 33,786 km. This distance has a profound impact on the link budget due to geometric spreading of the laser beam. Even with a highly collimated beam, diffraction leads to divergence of the laser beam over distance. The diameter of the laser spot $\left(D_{spot}\right)$ at the receiver is given by Equation (7), as follows:

$$D_{spot}\approx D_{tx}+\theta L,$$

(7)

where $D_{tx}$ is the diameter of the transmitter aperture, $\theta$ is the full beam divergence angle in radians, and L is the link distance [62]. In this link, $\theta L$ is the dominant term, as a typical diffraction-limited beam divergence of 20 microradians over a 33,786 km path results in a spot size over 720 m in diameter [63]. The second challenge in LEO–GEO links is the sub-microradian pointing accuracy requirement. Because the received power is inversely proportional to the square of the pointing error, even a minuscule pointing error can lead to a catastrophic loss of signal. For a Gaussian beam, the power loss due to a radial pointing error $\left(\epsilon \right)$ is given by Equation (8), as follows:

$$L_p=exp(-2{\displaystyle \frac{{\epsilon }^2}{w^2}}),$$

(8)

where w is the beam waist radius at the receiver. This system requires a sub-microradian pointing accuracy, and the pointing error budget must account for all sources of error, including thermal distortions in the satellite structure, sensor noise, and residual jitter from the FSM. This challenge is similar to LEO–LEO links in type but exceedingly different in magnitude [64]. Similar to LEO–LEO link, the process of acquisition is conceptually the same, but the uncertainty cone is much larger, and, therefore, the time required to scan it and establish a link is significantly longer and more complex. Furthermore, while a GEO satellite remains relatively stationary, an LEO satellite travels at 7.8 km/s, resulting in a large, rapidly varying point-ahead angle that must be calculated continuously. The required point-ahead angle $\alpha$ is dependent on the relative velocity vectors of the two spacecraft and can be calculated using Equation (9) [65], as follows:

$$\alpha \approx {\displaystyle \frac{|v_{LEO}-v_{GEO}|}{c}}.$$

(9)

After this, solar conjunction is a major challenge for LEO–GEO optical links. When solar conjunction happens, the sun’s overwhelming brightness completely floods the sensitive light detector on the receiving satellite [66].

Table 2. Comparative summary table.

Feature	UAV–Ground Links	HAPS–Ground Links	Satellite–Ground Links	Satellite–HAPS Links	Inter-Satellite Links
Primary Pointing Challenge	Robust stabilization against high-frequency platform vibrations.	Long duration stability against platform vibrations.	Compensating for intense atmospheric turbulence and accuracy over vast distance.	Asymmetric link with a quasi-stationary aerial platform.	High relative velocity, maintaining lock during fast passes, and extreme absolute accuracy over vast distances.
Link Distance	<10 km	20–50 km	500–2000 km (LEO)	500–36,000 km	1000–8000 km (LEO), minimum 33,786 km (GEO)
Turbulence Path	Severe, often with the entire path in the thickest, most turbulent air.	Severe, path traversing from stratospheric layer to densest part of the atmosphere.	Entire atmosphere with severe turbulence near the ground.	Minimal (only thin stratospheric layer).	None (Operates in vacuum).
Platform Vibration	High-frequency and -amplitude vibrations from engines/rotors.	Low-frequency, high-altitude HAPS platform sway.	Predictable orbital motion, ground station is stable.	High-frequency satellite jitter and low-frequency HAPS sway.	High-frequency jitter at LEO, low amplitude from reaction wheels.
Key Complexity	Rejecting aggressive platform motion in a cluttered, dynamic environment.	Maintaining a continuous link for days/weeks from a drifting platform.	Weather dependency like cloud blockage.	Balancing space-grade precision with atmospheric link adaptability.	Point-ahead calculation at high angular rates, extreme link budget, acquisition time, and pointing loss.

provides a comparative summary table of all the non-terrestrial platform FSO communication links.

4. Coarse vs. Fine Pointing in Non-Terrestrial FSO Links

Maintaining a stable LoS in FSO-based communication links with non-terrestrial platforms is challenging because of the dynamic environments during aerial missions. These kinds of challenges require a precision alignment system capable of tracking both rapid and large-scale angular displacements. To achieve this, a two-stage ATP strategy is typically used by researchers and engineers, i.e., coarse pointing for wide-range mechanical alignment and fine pointing for high-accuracy beam correction. These stages operate synergistically to ensure that the transmitted laser beam remains within the FOV of the platform, despite its continuous motion and environmental disturbances. Figure 4 represents a basic geometrical representation of a gimbal system consisting of both coarse and fine pointing systems. Figure 5a and Figure 5b display a close-up image of the CPM and an MEMS mirror (which is used for fine pointing), respectively.

4.1. Coarse Pointing: Mechanical Beam Acquisition

As their name suggests, CPMs are responsible for coarse pointing, i.e., gross angular positioning of the transmitter and receiver optics. CPMs generally consist of mechanically actuated systems such as gimbals or pan-tilt mounts with one-DOF, two-DOF, or three-DOF configurations. This system’s operational angular range is around tens of degrees, ensuring that the laser beam is directed toward the target’s general location, even under large initial pointing uncertainties due to UAV position errors, drift, or high-speed maneuvers. The coarse pointing objective is given in Equation (10), as follows:

$$|{\theta }_c-{\theta }_t|<{\theta }_{FoV}.$$

(10)

Here, ${\theta }_c$ is the coarse pointing angle, ${\theta }_t$ is the target’s angular displacement relative to boresight, and ${\theta }_{FoV}$ is the acquisition field-of-view of the fine pointing module. If the error exceeds this, the fine-pointing loop cannot engage. Alongside this, the mechanical resolution $\mathsf{\Delta }{\theta }_c$ of coarse pointing is limited by actuator resolution and backlash is given in Equation (11), as follows:

$$\mathsf{\Delta }{\theta }_c={\displaystyle \frac{{360}^{\circ }}{N·r}},$$

(11)

where N is the steps per revolution (stepper or servo resolution) and r is the gear reduction ratio. In CPMs, gimbals typically have angular resolutions in the range of 0.1–1 milliradian and slew rates around 10–${100}^{\circ }$/s, which are sufficient for acquiring moving UAV targets [6,14].

Depending upon application, system complexity, and available feedback sources, CPMs work either in open-loop or closed-loop. Open-loop gimbals work on predefined inputs based on global positioning system (GPS) sensors or inertial navigation system (INS) sensors. These gimbals move to a calculated position without real-time correction, which makes their mechanisms simpler, faster, and easier to implement. But they are exceedingly unreliable and susceptible to error buildup due to GPS/INS drift, platform motion, etc. On the contrary, closed-loop gimbal systems work with real-time feedback. They use sensors like inertial measurement units (IMUs), beacon signal (from the other terminal), and angle-of-arrival (AOA) as feedback sources, which help them continuously correct their orientation to minimize angular error. The actual gimbal angles, i.e., azimuth and elevation angles, are measured using encoders. Then, errors are calculated based on the measured and desired angles and then fed into a controller like PID, then the controller drives the motor of the gimbal to minimize this error [67,68].

PID controllers are a very common method to control these dynamic systems. Equation (12) represents PID control, where $e\left(t\right)={\theta }_d\left(t\right)-{\theta }_m\left(t\right)$ is the angular pointing error, ${\theta }_d\left(t\right)$ is the desired azimuth/elevation angle, ${\theta }_m\left(t\right)$ is the measured gimbal orientation from encoders, and $u\left(t\right)$ is the control input to the motors [67].

$$u\left(t\right)=K_{pe}\left(t\right)+K_I{\int }_0^{t}e\left(\tau \right)d\tau +K_D{\displaystyle \frac{de\left(t\right)}{dt}}$$

(12)

Equation (13) represents the gimbal dynamics ($2^{n}d$-order approximation), where J is the inertia of the gimbal and B is the damping coefficient [67].

$$J{\ddot{\theta }}_m+B{\dot{\theta }}_m=u\left(t\right)$$

(13)

Combining Equations (12) and (13) yields the closed-loop second-order system given by Equation (14) [67], as follows:

$$J\ddot{e}+B\dot{e}+K_{p}e+K_I\int e+K_D\dot{e}=0$$

(14)

This equation models how the gimbal responds to disturbances (like UAV vibrations, wind gusts, or mechanical jitter). It shows how mechanical inertia $\left(J\right)$, friction $\left(B\right)$, and control parameters (PID gains) interact to minimize pointing error. Without such stabilization, the narrow optical beam would quickly lose alignment due to motion and disturbances of the aerial platforms. Here, $K_{p}e$ corrects instantaneous pointing error, $K_I\int e$ eliminates steady-state error (keeps beam centered on UAV/ground station over time), and $K_D\dot{e}$ provides damping and stability, countering fast UAV vibrations and reducing overshoot.

In addition to GPSs, IMUs, and servo gimbals, star trackers are also being investigated as coarse pointing sensors in non-terrestrial FSO links. These optical units help in determining absolute attitude by imaging star fields and matching them with onboard catalogs. Traditionally, star trackers are a spacecraft technique, but they are also used with high-altitude platforms like UAVs in a miniaturized form. This system supplies high-precision orientation data without GPSs and significantly reduces initial pointing uncertainty, which is quite critical when pointing a narrow optical beam at long distances [69,70,71]. During the coarse acquisition phase, systems typically operate in open-loop (feedforward) mode, and the transmitter is aimed solely based on the estimated attitude and position (via GNSS, IMU, or star tracker), without feedback. This open-loop acquisition can be represented a followss:

$${\theta }_{cmd}\left(t\right)=f(GPS\left(t\right),INS\left(t\right),StarTracker\left(t\right)).$$

(15)

where ${\theta }_{cmd}\left(t\right)$ denotes the azimuth/elevation command angles derived from navigation sensors. Since this stage is feedforward, the instantaneous pointing error is simply

$${\theta }_{err}\left(t\right)={\theta }_{target}\left(t\right)-{\theta }_{cmd}\left(t\right).$$

(16)

Only when $|{\theta }_{cmd}\left(t\right)|<{\theta }_{FoV}$ according to Equation (10) is the beam approximately aligned, and a closed-loop pointing and tracking stage takes over (using beacon signals, FSMs, quadrant photo detectors (QPDs), position-sensitive devices (PSDs), etc.) to precisely aim the beam on an optical detector. These kinds of hybrid two-stage ATP architectures, which follow open-loop and closed-loop strategies, are well established in small satellite and aerial laser communication systems [72,73]. Notably, in [16], the authors described an FSO communication system with UAVs where coarse pointing is controlled by a GPS-aided gimbal, then preceded by tip/tilt fine tracking, and AeroCube-7 relies on a high-performance star tracker for coarse alignment before any fine pointing stage [70].

4.2. Fine Pointing: High-Precision Optical Stabilization

The second stage of pointing is fine pointing; once the coarse system places the beam within the FOV, fine pointing mechanisms come into play. These systems typically comprise fast steering mirrors, piezo-actuated mirrors, MEMS tilting platforms, liquid crystal beam deflectors, and electro-optic or acousto-optic beam deflectors. Like CPMs, fine pointing mechanisms have limited angular ranges. Their angular ranges lie in the sub-degree regime, but operate at much higher bandwidths (typically > 1 kHz) to suppress high-frequency jitter and small-scale misalignment caused by vibrations of non-terrestrial platforms or atmospheric turbulence [7].

Fine pointing systems are closed-loop control systems; therefore, they are able to maintain the alignment of an optical beam more precisely with a remote receiver, despite such vibration, drift, or atmospheric turbulence. This fine pointing loop involves real-time sensing, actuation, and correction to keep the beam tightly aligned. First, the sensor detects deviation or jitter in the incoming or outgoing optical beam. These sensors could be a quadrant photodiode (QPD), position-sensitive detector (PSD), camera-based tracking sensor, or beacon-based optical sensor. Then, the sensor output is processed to determine the pointing error and calculate the angular offset between the desired and actual beam direction. After calculating the offset, a proportional integral derivative (PID) or adaptive controller interprets the error signal and computes the correction. Some systems use kalman filters for predictive smoothing or sensor fusion. Then, the interpreted signal is sent to an FSM or MEMS capable of high-speed angular adjustments, which then applies fine angular corrections to steer the beam accurately. The updated position is continuously sensed again, creating a feedback loop to constantly minimize the pointing error [74]. In Figure 6, a basic block diagram of a closed-loop fine pointing module, which is used in gimbals, is shown.

The fine pointing loop maintains the following:

$$|{\theta }_f\left(t\right)|\le {\theta }_{div}/2,$$

(17)

where ${\theta }_f\left(t\right)$ is the instantaneous angular deviation due to disturbances and ${\theta }_{div}$ is the laser beam divergence. Since beam divergence is typically <2 mrad, even sub-milliradian deviations can degrade link performance.

The optical pointing loss factor $L_p$ due to pointing error ${\theta }_e$, where ${\theta }_{div}$ is divergence angle, is approximated by the following [40]:

$$L_p\left({\theta }_e\right)=exp(-{\displaystyle \frac{2{\theta }_e^2}{{\theta }_{div}^2}}).$$

(18)

The exponential loss relation in Equation (18) represents the importance of maintaining tight angular control within a small fraction of the divergence angle [40]. The

Table 3. Difference between coarse pointing modules and fine pointing modules.

Feature	Coarse Pointing	Fine Pointing
Range	10–${180}^{\circ }$	<$1^{\circ }$
Resolution	0.1–1 mrad	<0.01 mrad
Bandwidth	1–50 Hz	1–10 kHz
Actuation	Gimbals, Pan-Tilt Units	FSMs, MEMS Mirrors, LCBDs
Response to	Platform motion, link acquistion	Jitter, turbulence, fine correction
Sensors	GPS, IMU, Camera, Star Trackers	QPD, PSD, Beam Position Detectors
Control	PID, Kalman, Model-Predictive	Adaptive, LQG, Fast PID

briefly differentiates coarse pointing modules and fine pointing modules based on features like range, resolution, bandwidth, actuation, response, sensors used, and control algorithms used.

4.3. Need for Combined Operation

Both coarse and fine pointing are critical in the accurate and precise pointing of the optical beam and maintaining LOS with the aerial platform. Both mechanisms are interdependent. While a coarse system cannot track high-frequency jitter due to mechanical inertia, fine pointing systems lack the angular range to correct large displacements from flight dynamics. They both operate under a hierarchical control system, where the outer loop, or coarse pointer, tracks the positional changes of non-terrestrial platforms based on either GPS/IMU or visual feedback, and the inner loop, or fine pointer, suppresses rapid jitter and beam wander. Both systems work together in a nested feedback loop, often with kalman or complementary filtering for optimal disturbance rejection and control resource allocation [44].

These systems are exceedingly dependent on sensor fusion and adaptive control to maintain alignment. Fine pointing systems often utilize a quadrant photodiode (QPD) or a position-sensitive detector (PSD) to generate error signals. Equation (19) represents the calculation of error signal ${\theta }_{err}\left(t\right)$, as follows:

$${\theta }_{err}\left(t\right)={\displaystyle \frac{x\left(t\right)}{f}}$$

(19)

Here, $x\left(t\right)$ is the lateral displacement of the beam spot on the detector and f is the focal length of the tracking lens. The system uses this information to correct the beam in a closed-loop manner using actuators, while the coarse pointing is updated periodically based on motion estimates from GPSs or vision-based systems.

5. Gimbal Architectures and Technologies for Non-Terrestrial FSO Communication

Gimbal subsystems are a critical part of FSO communication systems, especially for UAV platforms. These systems form the core mechanical structure of the coarse pointing subsystem, enabling macroscopic steering of the optical beam to maintain LOS alignment under dynamic UAV motion. Their role is particularly crucial during the acquisition phase and to ensure the beam remains within the fine pointing system’s capture range during communication. A well-designed gimbal must have a better accuracy, inertia, speed, and robustness, with strict SWaP constraints.

5.1. Classification of Gimbal Architectures

Gimbal architectures are classified by their degrees of freedom (DoFs) and actuation mechanisms.

Table 4. Classification of gimbal achitectures based on DoF.

Type	DoF	Motion Axes	Application	Power Consumption	Pointing Accuracy	Pros	Cons
1-axis Gimbal	1 DoF	Pitch or Yaw	Fixed-wing UAVs with constrained payload, small high-altitude balloons	Low	Moderate	Simple, low-cost	Limited motion, less precise
2-axis Gimbal	2 DoFs	Pitch + Yaw	Multirotors, high-speed beam alignment, airships, stratospheric UAVs	Medium	High	Better Alignment	Medium complexity
3-axis Gimbal	2 DoFs	Pitch+Yaw+ Roll	Advanced UAVs needing roll compensation, satellites, spacecraft, payloads, high-altitude reconnaissance platforms	High	Very High	Full compensation precise	High copmplexity, heavy
CubeSat ADCS	Reaction wheels + 2-DoF micro gimbal	Full-body attitude control + fine az/el stage	Cubesats, small satellites (OCSD, AeroCube-7, TBIRD)	Medium	High (sub-mrad)	Low-SWaP integration with ADCS; compact	Limited torque, relies on ADCS stability
OPA	Electronics fine steering	Beam steering without mechanics	CubeSats, future smallsat terminals	Low	Very high (micro radian level)	No moving parts, fast response	Limited angular range, optical loss, emerging tech
Hexapod	6 DoFs	Translational + Rotational	HAPs, stratospheric airships, vibration isolation platforms	High	High (sub-mrad)	6-DoF correction, robust to turbulence	Bulky, heavy, high power demand

mentions the basic classification of gimbal architectures based on DoFs. The simplest gimbal has one DoF, which allows for rotation about a single axis, typically pitch (elevation) or yaw (azimuth). This gimbal is used when relative movement is restricted in one plane. One-DoF gimbals are used with aerial platforms that have constrained trajectories, for example, UAVs constantly heading with a varying altitude; therefore, the gimbal has to take care of pitch changes only when stabilizing a laser or detector for LoS. As these gimbals have a single DoF, they fail when the target is moving in three dimensions and, therefore, require the non-terrestrial platforms to compensate for azimuth adjustments through flight control. These can be open-loop or closed-loop control mechanisms using feedback from IMUs or angle encoders [42].

Gimbals with two DoFs provide independent rotation about pitch (elevation) and yaw (azimuth). These are the most commonly used gimbal types in UAV-based FSO systems, because they allow the terminal to track any target within a hemisphere. They are also used in HAPSs or balloon-borne systems, where the platform is relatively stable but still subject to wind gusts and platform oscillations. Two DoFs allow the gimbal to track the movement of the ground stations or other UAVs in full three dimensions; therefore, LoS acquisition and maintenance are simpler than one DoF in dynamic FSO links. These are generally implemented in a closed loop with PID controllers per axis, angle encoders, and IMU or optical beacon feedback. They can also be combined with an FSM for fine pointing inside the gimbal. These gimbals are sufficient for all practical aerial FSO links where non-terrestrial platforms have reasonable roll stabilization, and are much easier to implement than full three-DoF gimbals [43].

Three-DoF gimbals just add rotation along the roll axis in addition to azimuth and elevation, which enables full 3D stabilization and alignment, including roll compensation, which is critical during aggressive maneuvers of non-terrestrial platforms. They are more commonly used in CubeSat and small satellite missions where sub-arcsecond pointing is required for laser communication, imaging payloads, or precision tracking. These are favorable for fixed-wing aircraft during banking turns and UAVs in turbulent or high-acceleration environments, as they can compensate for UAV roll, pitch, and yaw independently, therefore leading to improved pointing stability during high dynamics. Because of the three DoFs, three independent controllers are required, each synchronized to eliminate cross-axis coupling. They are often implemented with inertial feedback, gyroscopes, and kalman filters. But these are bulkier and heavier, which can affect payload constraints in the case of UAVs, and have higher power consumption and complexity [7,75]. Figure 7 shows the geometrical representation of different DoF gimbal systems based on xyz plane.

Beyond conventional 1–3-DoF gimbal architectures, cubesats and small satellites often employ alternative coarse pointing solutions integrated into their attitude determination and control systems (ADCSs). In such systems, coarse pointing is handled by reaction wheels and magnetorquers, while fine pointing is achieved with either a micro-gimbal azimuth–elevation terminal or an electronics beam steering approach such as an optical phased array (OPA). Reaction wheel-based cubesat terminals, as demonstrated in the OCSD [29] and AeroCube-7 [18] missions, achieved coarse pointing stability sufficient to bring the optical beam within the fine steering capture range. Also, for HAPSs and stratospheric airships, mechanically complex gimbal systems are impractical due to structural flexibility and wind-induced oscillations. Therefore, hexapod architectures like the stewart platform have been explored for coarse pointing and vibration isolation. The hexapods are based on six DoFs with correction capability and robustness to airframe motion, allowing them to maintain sub-mrad stability during long-endurance stratospheric flights.

In terms of axis coupling, gimbals are divided into the following two types: mechanically decoupled and nested gimbals. In mechanically decoupled gimbals, each axis has its mechanical frame, i.e., the system consists of two physically separated rotational stages. These systems are easier to model and control because there is less interference between axes. These gimbals are typically larger and are scalable for larger payloads, and also require more precise alignment between stages. They are generally implemented in large ground stations. In nested gimbals, each axis is nested inside another, forming a compact stack, and in the case of three DoFs, there is also a third ring for roll. These systems are quite compact and lightweight, which makes them easy to deploy on UAV platforms and suitable for platforms with a limited payload capacity. But their inverse kinematics and control algorithms make them more complex, and mechanics may limit rotation ranges [64,76].

Table 5. Difference between mechanically decoupled and nested gimbals.

Feature	Mechanically Decoupled	Nested Gimbal
Axis Interference	Minimal	High
Size	Large	Compact
Complexity of Control	Simple	Complex (due to coupling)
Cable Routing	Complicated	Easy
Scalability	High	Limited by nesting constraints
Suitability for UAV	Low to Medium	High
Rotation Range	Wide (independent)	Limited by internal structure

lists the main differences between the two systems.

5.2. Actuation Technologies

The actuation technologies used in gimbal systems, both mechanically decoupled and nested, are based on the required size, weight, power efficiency, performance, and accuracy. Generally, mechanically decoupled gimbals, for example, gimbals for a ground station, use stepper motors with encoders, DC servo motors (for high torque), harmonic drives (for large payloads), or direct drive motors (for zero backlash and high-speed response). In nested gimbals, mostly BLDC motors with encoders, servo motors + planetary or harmonic gearing, or piezo actuators are used [76].

Table 6. Actuation technologies.

Actuation Type	Principle	Use Case	Pros	Cons	Applications
Brushed DC Motors	Electro-mechanical torque via brushes	Legacy or cost-sensitive	Simple, low cost	Wear and tear, “EMI”, less precise	Low-cost UAV gimbals, legacy payloads
Brushless DC Motors (BLDC)	Electronic commutation, magnetic torque	UAV gimbals, fine pointing	Compact, efficient, long life	Requires ESC/FOC drivers	UAV lasercom terminals, small-satellite gimbals
Stepper Motors	Electromagnetic discrete steps	Coarse pointing systems	Open-loop control possible, precise steps	Can miss steps under high load	Coarse alignment in satellite optical terminals
Voice Coil Actuators	Linear magnetic field interaction	Fast stabilization (fine)	Fast response, frictionless	Small travel range, needs feedback	Fine beam steering for FSO ATP subsystems
Piezo-electric Actuators	Crystal deformation via voltage	Sub-microradian fine tracking	Ultra-precision	Very small range, expensive	Fine beam steering mirrors in optical communication
Harmonic Drive (w/Servo Motor)	Flex-spline torque gear system	Nested gimbal needing precision	Zero backlash, high torque density	Complex control, cost	High-precision satellite gimbals for GEO/LEO optical links
Magnetic Torquers	Magnetic interaction with Earth’s field	Space-based altitude control	No moving parts	Very low torque, slow	Satellite coarse atitude control before gimbal pointing
Reaction Wheels	Conservation of angular momentum	Satellite gimbal platforms	Precise torque control	Large, slow, high power consumption	Satellite fine pointing and stabilization for optical links
Gyroscopes (Control Moment Gyros-CMG)	Rotating mass generating torque	Spacecraft fine attitude control	High torque, precise attitude control	Complex, heavy, high power demand	High-capacity satellites needing agile optical beam poiniting
MEMS Actuators (Micro-mirrors, scanners)	Electrostatic/thermal deformation at micro-scale	Beam steering, fine pointing	Compact, low power, high speed	Limited angular range, fragile	CubeSat FSO ATP systems, miniature terminals
Shape Memory Alloys (SMAs)	Phase change deformation via heating	Deployable or low-power actuation	Simple, lightweight, low power	Slow response, hyteresis	CubeSat deployable optics, secondary pointing
Electrostatic Actuators	Coulomb force-based actuation	MEMS-level fine actuation	Ultra-low power, fast response	Tiny forces, limited travel	MEMS micomirrors for optical communication beam steering
Magnetostrictive Actuators	Strain from magnetic domain alignment	Ultra-fine motion systems	Precise, high bandwidth	Expensive, limited adoption	Niche optical ATP systems requiring ultra-stability

represents the different actuation technologies used in CPMs and FPMs [77,78,79,80,81].

Table 7. Actuation technologies comparison.

Actuation Type	Pointing Accuracy/Repeatability	Slew Rate/Bandwidht/ Response Time	Angular Range	SWaP	Lifetime	Environmental Tolerance
Brushed DC Motor	Low–Medium; for coarse positions, arc minutes/degrees, brushes create micro noise, limited repeatablity	Low–Medium; moderate response, limited high bandwidth control due to commutation and brush dynamics	Large (full turn via gearing)	Medium (cheap, moderate mass, brushes consume power)	Lower; brush wear limits life, maintenance required	Moderate; ok in benign air; poor for vacuum (brush outgassing/wear) and long-duration space use
BLDC	Medium–High; sub-arc-sec to arc-sec achievable with precision encoders/direct-drive designs, smooth torque reduces jitter	Medium–High; good dynamic response, high continuous torque; may have cogging/torque ripple if not optimized	Large (with bearing/gimbal)	Medium; good torque density, favorable SWaP for UAVs/space when optimized	High; no brushes so long life, reliability depends on bearings/ electronics	Good; can be used in airborne systems, space-qualified BLDC variants exist but require radiation/ thermal design
Stepper Motor	Medium; step resolution, therefore precise open-loop positioning but risk of missed steps under load, closed-loop microstepping improves repeatability	Low–Medium; strong at low speeds, poorer at high speeds (resonance/noise)	Large (full-turn with steps)	Medium; simple electronics but power drawn can be high (holding current)	Medium; robust mechanically but step loss under overload, heating issues	Moderate; commonly used in ground/air, space use requires careful de-energized holding strategies
Voice-Coil Actuator (VCA)	High for fine stages (sub-microradian to microradian when used on small mirrors/FSMs), very low hysteresis	Very High; excellent bandwidth (hundreds Hz to kHz) and smooth continuous force, ideal for jitter suppression	Small to Medium typically limited stroke (mm), small angular deflection when driving mirror	Low–Medium; compact and relatively lightweight for fine optics, power depends on duty	High; no mechanical contact wear (magnetic drive) but heating can limit continuous use	Good in atmosphere; for space must consider magnetics and thermal design, many FSMs with VCAs are space-qualified (with care)
Piezoelectric Actuators	Very High; sub-microradians to sub arc sec or micro stages, excellent for nanoradian-level fine pointing in short ranges	Very High; bandwidth upto kHz or higher (fast small motion response)	Very Small; limited stroke (micrometer to sub-milimeter) so angular deflection is tiny, fine corrections only	Very Low; but medium for power (drive electronics require high power)	High for many cycles but can suffer from creep, hysteresis and depoling over long time/high temp if not managed	Good for vacuum but temperature sensitivity and long term stability must be managed; space MEMS/piezo use exists
Harmonic Drive	High near zero-backlash, excellent repeatability (used in precision gimbals and space actuators)	Low–Medium; gear reduction reduces bandwidth; good for precise slow moves, not high bandwidth	Large (full motion depending on motor)	Medium–High adds mass/volume but provides high torque in compact package	High; robust gear life but lubrication/wear is a consideration in long missions	Good for atmosphere, space gears require vacuum-compatible lubrication and careful qualification
Magnetic Troquers	Very Low for fine pointing (used fpr coarse attitude control only; torque small relative to reaction wheels)	Low; slow actuation in seconds and low bandwidth	N/A actuates body attitude not local gimbal	Very Low SWaP; lightweight, low power for coarse control	High simple solid state coils, long life	Excellent for LEO momentum management, but ineffective far from Earth (weak field)
Reaction Wheels	High (used for spacecraft coarse and fine attitude control; can support arc-sec or better LoS when coupled with control filters)	Low–Medium; wheel spin control has limited bandwidth, but can be used with FSMs for high-bandwidth corrections	N/A (body control)	Medium–High; substantial mass and power depending on required angular momentum	Medium bearing and motors wears; micro-vibration from RWAs can degrade pointing unless mitigated	Very good in space, but reaction wheel induces jitter must be mitigated for microradians pointing
Control Moment Gyroscopes	High for large torque agile pointing (used for fast slewing/precise reorientation), not used for tiny local beam steering but for body maneuvers	Medium–High; can produce large torques quickly (but control singularities exist)	N/A (body control)	High; heavy and power hungry for significant torque	High but mechanical complexity and singularity management are issues; maintenance is impossible in space	Space Grade; widely used in large spacecraft but not suited for small SWaP platforms
MEMS Actuators	Very high for small angles (microradians to sub micro-radians)	Very High bandwidth often in kHz; excellent for fine jitter suppression	Very Small typical angular ranges $\pm 0.1^{\circ }$ to $\pm 6^{\circ }$ depending on design	Very Low; tiny size, and power, suitable for cubesats.	Variable many cycles possible but fatigue, dielectric charging, or radiation effects can limit life, qualification needed for space	Good in vacuum if space-qualified; radiation and thermal cycling can affect some MEMS types needs qualification
Shape Memory Alloys	Low slow, coarse motion, low precision	Very Low, slow thermal actuation (seconds to minutes)	Medium (depends on linkage)	Very low weight, but low efficiency too (heating required)	Medium–Low; fatigue from cycles, hysteresis and drift are common	Moderate; thermal and repeated cycling impact life; not ideal for continuous pointing
Electrostatic Actuators	High at MEMS scale but limited force/travel	Very High, but at micro scale (microseconds to milliseconds)	Very Small (micrometer displacements, small angular ranges)	Very Low (excellent for micro terminals)	Variable; dielectric charging and stiction can be failure modes; vacuum performance good if designed properly	Good for vacuum if space qualified, careful design to avoid charging and stiction
Magneto strictive Actuators	Very High (ultra fine motion possible in niche systems) high bandwidth and precision	High; good bandwidth for small strokes	Small (limited stroke)	Medium can be compact but require magnetic materials and drive currents	High; solid state, few moving parts, but specialized materials may fatigue under extremes	Good if engineered for environment, limited adoption in mainstream FSO due to cost/complexity

compares all the actuation technologies mentioned in Table 6 on the basis of pointing accuracy, slew rate, angular range, SWaP, lifetime, and environmental tolerance [82,83,84,85,86]. Then,

Table 8. Application suitability: motor type Vs FSO platform.

Platform	Requirements (SWaP, Vibration, Environment, Lifetime)	Suitable Motor/Gimbal Types
Small UAVs/Multirotor	Very tight constraints on weight, power, size; high vibration; need for moderate pointing accuracy; fast slewing; possibly intermittent operations.	Lightweight BLDC + small direct-drive motors; possibly voice coil/arc-segment for fine axes; minimal gear reduction.
HAPSs	Somewhat more forgiving on weight; but still constrained; environmental temperature extremes; long duration; potential atmospheric turbulence partly mitigated by altitude.	Medium-torque BLDC or servo motors; possibly gear assisted for large optics; voice coil for fine pointing; redundancy for reliability.
CubeSats/Small Satellites	Very strict SWaP; vacuum, radiation; long lifetime; limited thermal dissipation; orbital perturbations (nutation, jitter) from reaction wheels, internal mechanisms.	Micro gimbals or MEMS actuators for fine pointing; perhaps small brushless servos for coarse pointing; minimal gear to reduce backlash; use of high-precision encoders.
Large Satellites/GEO Telescopes/Ground Station Tracking	Large payloads; high angular resolution; need for high stability; environmental conditions vary; long missions.	High-torque servos/BLDC with gear reductions; voice coil fine steer mirrors/tip-tilt stages for jitter; large, robust gimbals.
Inter-Satellite/Deep Space FSO Links	Very long missions; high pointing accuracy; large distances therefore small pointing errors magnified; extreme environmental conditions (radiation, thermal swings).	High-precision BLDC/servo with fine steering mirrors (possibly MEMS or piezoelectric); perhaps redundant systems; minimal moving mass; low power consumption; high durability.

compiles information about which motor type will be suitable or is being used with respective platforms [80,81,84,85,86].

6. Control Algorithms Used in Pointing

Coarse pointing modules in FSO-based communication with non-terrestrial platforms require robust control strategies that align the gimbal-mounted beam despite dynamic motion, environmental disturbances, and sensor noise; therefore, they effectively rely on control algorithms. This section examines proportional integral derivative (PID), linear quadratic regulator (LQR), and hybrid approaches like PID and kalman or LQR and kalman.

6.1. Coarse Pointing Control Algorithms

6.1.1. Proportional Integral Derivative (PID)

PID controllers are the most widely used control algorithm in gimbal systems due to their simplicity, ease of implementation, and effectiveness. They control the process variable in the system, which, in the case of gimbals, is the pointing angles. This algorithm efficiently calculates the error/feedback between a desired reference, for example, a target angle or position, and the measured output, for example, gimbal orientation, and then applies the offset action based on proportional (P), integral (I), and derivative (D) terms, where P reacts to current error, I addresses accumulated past errors, and D predicts future errors based on the rate of change. Equation (20) represents the mathematical equation of the PID control system in gimbals, where ${\theta }_{err}\left(t\right)$ is the angular pointing error. The first term in the equation is P, which generates a torque proportional to the instantaneous pointing error. The second term is I, which cancels steady-state biases (e.g., small misalignment, gravity torques, and friction). It must be protected against windup when actuators saturate or during target loss. The third term is D, which adds viscous-like damping, countering the aerial platform’s jitter and structural modes. It improves settling time and reduces overshoot. The derivative action is noise-sensitive and is, therefore, usually implemented with a first-order filtered differentiator.

$$u_{PID}\left(t\right)=K_p{\theta }_{err}\left(t\right)+K_i\int {\theta }_{err}\left(t\right)dt+K_d{\displaystyle \frac{d{\theta }_{err}\left(t\right)}{dt}}.$$

(20)

In gimbal systems, PID is often applied in a cascaded loop structure, where an outer-loop PID controller stabilizes the platform’s attitude or position error correction, while an inner-loop PID or PI controller controls the motor’s speed or current. Figure 8 illustrates the cascaded loop structure, where the outer position loop (Summing Point 1) outputs a velocity command to the inner velocity loop (Summing Point 2), which controls the motor torque.

Although PIDs are robust and easy to tune, in the case of gimbals for UAVs, they struggle with fast dynamics or non-linearities caused by the aerial platform’s motion or aerodynamic disturbances, especially when there are saturation limits, payload shifts, or sensor drift. Despite this, they remain popular in low-cost and medium-performance applications, particularly for two-DOF or three-DOF nested gimbals. For HAPSs and stratospheric airships, PID loops are often used as outer control layers to correct slow drifts caused by stratospheric winds, while inner loops rely on more advanced controllers for precision. On the contrary, cubesats rarely rely on PIDs due to their need for sub-arcsecond pointing, but PID loops are still applied for actuator current or wheel speed regulation [87,88]. In [89], the researcher has implemented a self-tuning PID controller for a two-axis gimbal, enhanced with fuzzy logic and PSO-based gain. Microcontrollers like STM32 or ARM cortex-M, typically executing at loop rates of hundreds to thousands of Hz, are suitable for PIDs. PID tuning requires manual touch or heuristics, and adaptive methods like fuzzy-PID can mitigate this.

6.1.2. Linear Quadratic Regulator (LQR)

LQR is a state-space control technique that minimizes a quadratic cost function involving system states and control inputs. It is a more optimal and mathematically rigorous alternative to PIDs, and is ideal for systems that are modeled by linear differential equations. In UAV gimbals, the state vector might include angular position, angular velocity, and motor torque. LQR achieves optimal control by minimizing a quadratic cost function (Equation (21)) [90], as follows:

$$J={\int }_0^{\infty }(x^{T}Qx+u^{T}Ru)dt,whereu=-Kx.$$

(21)

Here, x is the state vector, u is the control input, and Q and R are weighting matrices. LQR achieves better trade-offs between performance and control effort compared to PIDs. It is more valuable in closed-loop coarse pointing, where UAV dynamics and platform disturbances affect gimbal performance. But its performance decreases with modeling in accuracy and nonlinear regimes. It is also less robust to uncertainties unless combined with estimation techniques like the kalman filter. HAPSs and airships are more exposed to large but slow deviations, and LQR is advantageous for minimizing steady control effort over long durations. Cubesats also benefit from LQR when combined with reaction wheels or magnetic torquers, as the quadratic cost framework balances fine pointing precision with limited onboard power resources [87,91].

As LQR demands a linear model and computation of the gain matrix via solving a Riccati equation, it requires embedded platforms with efficient linear algebra, like ARM cortex-M4 or M7, which can handle LQR loops at tens to hundreds of Hz. LQR’s gains tuning can be automated via performance metrics. Ref. [92] is not an example of a gimbal but a flying wing UAV control that employed LQR with full-state observers to stabilize attitude under disturbances, demonstrating a stronger robustness than PID regulation.

6.1.3. Extended Kalman Filter (EKF)

This kalman-based filter is critical for integrating IMU, GPS, and visual data to form accurate estimates for coarse pointing guidance. It fuses nonlinear, noisy measurements to estimate accurate orientation and angular velocity, which are then sent as feedback control. The EKF linearizes the system dynamics around the current estimate and simultaneously updates state estimates, as shown in Equation (22).

$$x_{k+1}=f(x_k,u_k)+w_k,y_k=h\left(x_k\right)+v_k.$$

(22)

Here, $w_k,v_k$ are process and measurement noise and $f,h$ are nonlinear dynamics and measurement models. In [93], the researcher fused IMU, LiDAR SLAM, and visual odometry via an extended kalman filter (EKF) in GPS-denied indoor conditions, validating precise state estimation as suitable for coarse pointing control. In terms of computation, EKF requires moderate computation at an update rate between 100 and 500 Hz and data from sensors such as IMU, GPS, camera, or LiDAR. These systems often run on embedded linux or high-performance microcontrollers with floating-point support. EKF-based estimation is also valuable for cubesats, where fusing star tracker and gyroscope data ensures accurate attitude determination in orbit. Similarly, in HAPPs and stratospheric airships, EKFs are employed to fuse GPS, barometric, and IMU data to mitigate slow drifts and wind-induced offsets, thereby stabilizing long-duration FSO links [87,91].

6.1.4. Hybrid Approaches PID and Kalman/LQR and Kalman

Hybrid control algorithms like PID + kalman filter and LQR + kalman filter offer a more robust and efficient solution by combining the strengths of classical and optimal control methods with real-time state estimation and noise rejection.

The PID + kalman configuration is used because it has the simplicity of PIDs and the effectiveness of the kalman filter. In this, PID controllers handle real-time actuation for coarse or fine pointing, while the kalman filter estimates accurate angular velocities and orientations from noisy IMU sensor measurements. These sensor measurements allow the PID loop to respond based on the filtered signals, which, in turn, reduces overshoot and improves beam stability. This approach is suitable for systems where model precision is limited [94].

In contrast, the LQR + kalman approach offers a better performance for well-modeled systems. As LQR minimizes a quadratic cost function, balancing control effort and tracking error, the kalman filter estimates the system’s state under noisy conditions. Together, they enable optimal control in dynamic environments [95].

The author in [96] described coarse–fine composite control combining a PID and kalman filtering to provide rapid initial alignment with accurate state estimation. These hybrid systems typically run in cascaded loops, where an outer kalman filter feeds an inner control loop, either a PID or LQR, which requires multi-threaded or co-processor-capable microcontroller stacks. For HAPSs and airships, hybrid PID–kalman controllers provide resilience against wind gusts and platform oscillations. However, cubesats often employ LQR–kalman combinations, where state estimation from star trackers and gyros feeds into optimal control laws for precise optical beam alignment. These approaches emphasize long-term stability and robustness over rapid response, contrasting with UAV-based implementations [87,97].

6.2. Fine Pointing Control Algorithms

6.2.1. Adaptive PID

Adaptive PIDs or fast PIDs are employed in gimbal systems for fine pointing. An adaptive PID controller dynamically adjusts its proportional $\left(K_p\right)$, integral $\left(K_i\right)$, and derivative $\left(K_d\right)$ (refer to Equation (20)) gains to optimize the system’s performance in real time. This is crucial for non-terrestrial FSO platforms that experience variable disturbances such as atmospheric turbulence and mechanical vibrations. Techniques like fuzzy logic, neural networks, and machine learning are used to implement this kind of adaptability. For example, a fuzzy-PID controller uses a set of fuzzy rules to tune PID gains based on the error signal, as detailed in research on aerial vehicle gimbals [98,99].

In this kind of control algorithm (refer to Equation (20)), $\left(K_p\right)$, $\left(K_i\right)$, and $\left(K_d\right)$ are not constant but are functions of the system’s state or error, allowing the controller to respond more effectively to dynamic environments. In these terms, the derivative is highly susceptible to measurement noise, which can lead to erroneous control actions. To mitigate this, a derivative filter is used, which is a low-pass filter that smooths the input to the derivative term [100,101]. This also prevents high-frequency noise from being amplified. The transfer function for a second-order filter $G_f\left(s\right)$ can be written as Equation (23), as follows:

$$G_f\left(s\right)={\displaystyle \frac{1}{1+T_{f}s+\frac{T_f^2}{2}{s}^2}}.$$

(23)

Here, $T_f$ is the filter time constant, which is tuned to balance noise reduction with the responsiveness of the controller [102,103,104].

6.2.2. Model Predictive Control (MPC)

Model predictive control (MPC) offers a crucial solution for predictive jitter suppression in fine pointing. MPC bears the capability to predict future states and optimize control inputs over a defined time horizon. This predictive capability is crucial for proactively compensating for disturbances like satellite vibrations and atmospheric turbulence, which cause pointing errors or jitter [105]. The core of MPC is its cost function J, which is minimized to find the optimal control action. This cost function can be expressed as follows in Equation (24) [106,107]:

$$J=\sum _{k=0}^{N-1}\left(x_k^{T}Q{x}_k+u_k^{T}R{u}_k\right)+x_N^{T}P{x}_N.$$

(24)

Here, $x_k$ represents the system state (for e.g., pointing error), $u_k$ is the control input (gimbal motor command), and N is the prediction horizon. The matrices Q, R, and P are weighting factors that penalize state deviations and control effort. By minimizing this function to system dynamics and constraints, this control algorithm provides optimal control commands that suppress jitter in the pointing module [108,109].

6.2.3. Linear Quadratic Gaussian (LQG)

Linear quadratic Gaussian (LQG) control is an ideal control method for fine pointing, where sensor noise and system jitter are significant issues. This method ingeniously combines a kalman filter with LQR [110]. First, the kalman filter estimates the true state, like the precise angle and angular velocity of the gimbal system, from noisy sensor measurements. During this process, it uses a system model to predict the state and then updates this prediction based on the actual measurement, providing an optimal state estimate $\widehat{x}$. Then, the LQR takes $\widehat{x}$ as an input and calculates the optimal control input u, to minimize the cost function, which balances pointing accuracy against control effort. The control law is a simple state feedback gain K, which can be represented as follows in Equation (25):

$$u=-K\widehat{x}.$$

(25)

By using this, the LQR can generate smooth and precise control signals that can effectively suppress jitter without being mislead by measurement noise [111].

6.2.4. Sliding Mode Control (SMC)

The sliding mode control (SMC) strategy is vital for maintaining pointing accuracy on dynamic non-terrestrial systems. This algorithm provides an exceptional robustness against matched uncertainties and external disturbances. It forces the system’s state trajectory to slide along a predefined surface, which is known as the sliding surface. This makes the system insensitive to disturbances. A sliding surface $s\left(x\right)$ is defined as a linear combination of the state error, e, as follows in Equation (26) [112,113]:

$$s\left(x\right)={({\displaystyle \frac{d}{dt}}+\lambda )}^{n}e,$$

(26)

where $\lambda$ is a positive constant and n is the system order. The control law is designed to ensure $s\left(x\right)=0$, effectively canceling disturbances. This makes it highly effective for rejecting unpredictable forces like wind gusts acting on a gimbal [114,115].

6.2.5. AI/ML-Based Predictive Control

AI/ML-based predictive control leverages methods like neural networks and reinforcement learning (RL) to achieve superior fine pointing in FSO gimbals on non-terrestrial platforms. A neural network can be trained on flight data to predict the future position of a platform, $x_{t+1}$, based on its current and past states $x_t,x_{t-1},\dots$, as follows in Equation (27):

$$x_{t+1}=f_{NN}(x_t,x_{t-1},\dots \dots ,u_t),$$

(27)

where $u_t$ is the control input and $f_{NN}$ is the learned nonlinear function represented by the network. This prediction allows the controller to compensate for movement before it occurs [98,116].

Furthermore, RL agents can learn an optimal control policy, $\pi$, by interacting with the environment. The agent aims to maximize a cumulative reward, R, which is designed to penalize pointing errors. The policy maps observed states like gimbal position and target location directly to optimal control actions $(a_t=\pi \left(s_t\right))$, enabling highly adaptive and predictive jitter suppression without needing an explicit system model [43,117,118].

Table 9. Control algorithm.

Algorithm	Typical Use	Bandwidth	Accuracy	Complexity	Hardware/Software Needs	Energy Efficiency	Resistance to Noise	Adaptability Across Platforms
PID	Coarse and fine (fast PID variants used for FSMs)	Medium–High (inner loops up to ∼1 kHz possible)	Medium (good for well behaved linear regions; sensitive to noise and nonlinearities)	Low (simple to design; manual/heuristic tuning)	Low-end MCU (STM32) + sensors/encoders; filtering recommended	High (low compute)	Low–Medium (derivative term is noise sensitive; needs filtering)	Good for UAVs and HAPSs after tuning; less ideal for sub-arcsecond satellite needs
LQR	Coarse (outer loop), reaction wheel attitude control	Low–Medium (10–100 Hz)	Medium for well modeled linear systems	Medium (requires linear model, Riccati solve)	MCU/small embedded CPU with linear algebra	Medium (optimal control reduces excessive actuation)	Medium (model mismatch reduces robustness unless combined with estimators)	Good for HAPSs, UAVs; cubesats benefit with careful modeling
LQG	Fine and coarse (noisy sensors; requires estimator)	Medium	Medium–High (estimation + optimal control)	High (state model + observer + gains)	MCU/embedded CPU + sensor fusion (IMU, encoders)	Medium	Medium (kalman helps with noisy measurements)	Good across platform if models available
EKF	Supporting role for coarse/fine (sensor fusion)	Medium (100–500 Hz)	High (fuses IMU/GPS/vision; reduces drift)	High (nonlinear models, tuning)	MCU + IMU/GPS/camera	Medium	Medium (robust to sensor noise if modeled)	Essential for GNSS - denied/ vision-assisted modes
MPC	Fine/predictive pointing (jitter supression, trajectory foresight)	Medium–Low (depends on horizon; heavier compute)	High (handles constraints, multivariate coupling)	High (online optimization; horizon tuning)	Embedded CPU/co-processor/ real-time QP solver	Low–Medium (compute intensive)	Medium (explicit constraint handling; can mitigate disturbances if predicted)	High potential for UAV/HAPSs; satellites require lighter-weight or offline MPC
Adaptive PID	Coarse & fine where dynamics vary	Medium–High	Medium (adapts to changing dynamics)	Medium–High (adaptive law or fuzzy rules)	MCU with modest compute; adaptive routines	Medium (some copmute, but less than MPC)	Medium (adapts to changing noise/disturbance statistics)	Very good for UAVs and HAPSs; useful when payload or environment changes
SMC	Fine/coarse when robustness is required (disturbances)	High (can be implemented high-rate)	High (robust to matched uncertainties)	Medium–High (chattering mitigation needed)	MCU; may require filters/observers	Medium	High (robust to disturbances, model errors)	Good for turbulent UAV/HAPS scenarios; needs careful design for sensitive optics
AI/ML predictive controllers	Predictive fine/coarse pointing (trajectory prediction, pre-steer)	Depends (inference fast; training costly)	Potentially high (if trained on representative data)	High (data collection, training, validation)	MCU + lightweight NN runtime (or onboard CPU on larger platforms)	Medium–Low (inference efficient; training offline)	Varies (can be brittle to data; requires robust training)	High potential for UAV swarms and HAPSs; for satellites, needs rigorous validation
Hybrid Schemes	Integrated coarse; fine stacks	Medium–High (combined strengths)	High (best disturbance rejection and estimation)	High (integrates estimators + controller)	Multi-core MCU/embedded CPU + sensors	Medium	High (combines estimation + control robustness)	Flexible across UAV/HAPSs/ satellites with appropriate partitioning

lists all the mentioned control algorithms based on their requirements, features, and adaptability across platforms.

7. Future Directions and Open Research Questions

As FSO links mature across non-terrestrial platforms (satellites, HAPSs, and UAVs), emerging trends such as terminal miniaturization, AI-based predictive ATP, optical phased-array/MEMS beam steering, and cooperative multi-platform coordination (e.g., UAV swarms with satellite backhaul) signal a strong pathway to scalable, secure 6G NTN deployments. These advances promise higher link availability and capacity with lower SWaP, enabling resilient space–air–ground networks for next-generation services. This section explores these directions with scientific depth and also highlights key open research challenges.

Miniaturization is critical for integrating FSO communication systems into micro- and nano-UAVs, which operate under stringent size, weight, and power (SWaP) constraints. In such platforms, ATP mechanisms must be not only compact and lightweight, but also highly reliable and accurate. Passive solutions such as retroreflectors or other non-mechanical modules offer ultra-low SWaP footprints, but they inherently lack the capability for active beam steering, which is essential for maintaining precise optical alignment in dynamic flight conditions. Therefore, recent advances in fine steering technologies, such as micro-electromechanical system (MEMS)-based fast steering mirrors (FSMs) or LiDAR-inspired mechanisms, demonstrate promising capabilities, with units achieving a sub-microradian resolution over angular ranges of $\pm 6^{\circ }$, and they also weigh less than 40 g. Despite these advancements, a major challenge persists in integrating such mechanisms with miniaturized gimbals, particularly those under 500 g, which can maintain the required stiffness and structural stability during UAV aerial missions. Also, some notable efforts have been made in the development of compact gimbal assemblies for high-altitude pseudo-satellites like the olympic HAPS platform and cube-satellite payloads, illustrating the feasibility of lightweight FSO tracking modules. But further miniaturization will require a co-optimization approach, targeting enhancements in both actuator torque density and mechanical stiffness to ensure performance without compromising SWaP limitations [119,120,121].

Now, coarse pointing relies on reactive sensor feedback, sensors like GPS/IMU inputs, encoders, or visual cues. These inputs introduce latency in the pointing and tracking system. In contrast, predictive pointing schemes proactively forecast the future position of the target to pre-steer the beam, thereby significantly reducing alignment delay. In [122], the researcher introduced an LSTM-based recurrent neural network (LRNet) that learns from prior UAV trajectory data to predict subsequent UAV positions. Although their study focused on RF beamforming, the same concept can readily be extended to FSO systems, where fine angular precision is critical. In another paper [123], the researcher explored multi-agent deep reinforcement learning (MARL) to jointly optimize UAV trajectories and relay link selection in hybrid RF/FSO networks. This approach helped in achieving a nearly two times higher end-to-end throughput and $2.25\times$ greater energy efficiency. The mathematical insight shared in Equation (28) shows that predictive pointing reduces error delay $\delta t$ by anticipating motion.

$${\theta }_{cmd}(t+\mathsf{\Delta }t)=\theta \left(t\right)+\dot{\theta }\left(t\right)\mathsf{\Delta }t+{\displaystyle \frac{1}{2}}\ddot{\theta }\left(t\right)\mathsf{\Delta }{t}^2$$

(28)

A promising direction is the integration of predictive control and high-bandwidth fine-steering hardware to actively suppress jitter of the system. Recent works on MPC tailored for PFSM mirrors shows large gains in high-frequency trajectory tracking and hysteresis compensation. This makes MPC a practical candidate for inner-loop jitter suppression [27]. Parallel advances in MEMS/FSM control and piezo-actuated FSM system integration demonstrate that compact, low-SWaP actuators can now support higher-loop rates if paired with model-aware controllers. This enables sub-milliradian stability on UAV and small-satellite platforms [124,125]. At the algorithmic level, robust and nonlinear strategies like sliding mode and H-infinity require attention, as they have shown affirmative results in guarding against wind gusts, actuator faults, and model uncertainty [126,127].

Also, in future, the integration of transformer-based trajectory prediction and FL with ATP systems presents a significant opportunity for FSO communication with non-terrestrial platforms like UAVs, HAPSs, and satellites. By leveraging transformer models, ATP systems will be able to perform proactive, anticipatory pointing, using exceedingly accurate predictions of platform movement to preemptively align the laser beam. This will help in reducing latency and will also enhance link stability, particularly in dynamic mega-constellations. FL offers a scalable solution for collaborative learning among a swarm of non-terrestrial platforms. Each platform can train a local model for trajectory prediction or pointing optimization, and then share only the model updates with a central server or a cluster heads. This approach will not only protect data privacy, but will also reduce communication overhead. This will also enable the swarm to collectively improve its performance without a single point of failure. Such a system can be made more efficient by selectively uploading only the most critical model layers, which will help in conserving power and bandwidth [31,32,35,36,128].

Another promising avenue for investigation is UAV swarms. As UAV swarms have become a viable solution, a mission requires cooperative actions like relay networks and mesh communication. Therefore, coarse pointing control can evolve to support multi-agent coordination and seamless beam pointing. A variety of swarm formation–control frameworks have been proposed, like sliding mode control augmented with artificial potential functions, which provide robust inter-UAV positioning, collision avoidance, and maintenance of formation integrity [129,130]. Also, distributed nonlinear model predictive control schemes like those using consensus-based and directed graph formulation can enable decentralized trajectory planning and inter-UAV collision avoidance under dynamic constraints [131,132]. Within RIS-equipped swarm architectures, the optimization of collective phase-shift settings and UAV trajectories aims to minimize aggregate pointing error across the formation, enabling coordinated FSO pointing with high precision under atmospheric disturbances [23]. Equations (29) and (30) represent the mathematical constructs of consensus constraints that enforce alignment and the swarm optimization objective, respectively.

$${\theta }_i-{\theta }_j\le {\delta }_{ij},\forall (i,j)\in neighbourgraph$$

(29)

$$min{\Sigma }_i{L}_p^i+\lambda {\Sigma }_{i,j}\left|\right|{\theta }_i-{\theta }_j{\left|\right|}^2$$

(30)

where $L_p^i$ is the pointing loss term for UAV i and $\lambda$ penalizes misalignment between neighbors.

RF–optical hybrid integration has already been explored in satellite–ground links and UAV swarm coordination, showing improvements in both link reliability and handover efficiency. Alongside this, the hybrid RF–optical control mechanism is also an important emerging paradigm. In this approach, RF beams with wide divergence are first used for coarse acquisition and initial alignment between aerial platforms and ground stations. Once coarse pointing is achieved, the system transitions to a narrow-beam optical link for fine pointing and tracking. This layered acquisition approach minimizes mechanical gimbal burden, reduces complexity, and enhances system reliability in turbulent environments. Figure 9 illustrates those two stages of a hybrid RF–FSO acquisition process.

Future UAV, HAPS, and cubesat networks are expected to adopt such dual-mode terminals, potentially combining mmWave/Thz RF for acquisition with optical phased arrays or MEMS-based optics for data transport. Some open questions remain around joint resource allocation, cross-layer optimization, and hardware miniaturization of dual-mode payloads under SWap constraints [5,64,133,134].

8. Conclusions

This paper explored the state of the art in coarse and fine pointing modules, especially gimbal architectures for FSO systems deployed on and for non-terrestrial platforms. FSO technology offers unparalleled data rates, immunity to electromagnetic interference, and compactness, making it a strong candidate for high-throughput aerial communication. However, narrow beam divergence, which is inherent to optical links, makes it more susceptibleto misalignment, and, therefore, makes robust pointing and tracking subsystems more crucial for FSO systems.

The paper first established a technical foundation by exploring the fundamentals of FSO communication in aerial platforms, emphasizing how the dynamics of non-terrestrial platforms introduce new challenges for maintaining a stable LoS. Then it provided a detailed differentiation between coarse and fine pointing systems for FSO, including their respective roles in beam acquisition and tracking. Coarse pointing is typically achieved using gimbal systems and provides initial alignment to bring the target within the field of view (FoV) of the fine pointing systems, which then handle micro-level corrections. Through mathematical models and optical equations, the respective sections demonstrated the complementary nature of these two subsystems and their joint importance in achieving high pointing accuracy and system reliability.

The paper reviews gimbal architectures, including nested-frame, stabilized-platform, and hybrid designs. It analyzes trade-offs in payload capacity, pointing precision, stabilization effectiveness, and power efficiency. The literature survey compares more than fifteen recent systems across angular resolution, response time, control bandwidth, and payload-integration metrics, providing a clear map of current capabilities.

On the algorithmic front, the paper examined common control strategies such as PIDs, LQR, kalman filters, and advanced sensor fusion with GPS/INS, and vision systems. Each approach presents different trade-offs in responsiveness, stability, and computational complexity. A recent work implemented hybrid PID–Kalman and LQR–EKF controllers, which displayed more enhanced disturbance rejection and tracking fidelity in gimbal systems. Several control algorithms were tied directly to recent UAV pointing systems, illustrating real-world relevance and performance results. Hardware platforms ranging from low to high computational power required for respective control algorithms were also discussed, concerning specific requirements.

The paper also addressed the pressing challenges encountered in the coarse pointing of non-terrestrial platforms, including platform vibrations, mechanical constraints, wind gusts, and flight path instability. These factors introduce dynamic uncertainties that degrade pointing accuracy, especially in lightweight drone platforms. Research efforts involving vibration isolation and adaptive filtering have shown promising results in mitigating these challenges.

At the end, the paper explored future directions related to the miniaturization of gimbal platforms, AI-based predictive tracking models, and integration with UAV swarming protocols. These innovations look affirmative in redefining the scalability, autonomy, and resilience of FSO links in next-generation aerial networks. Open research challenges remain in multi-axis stabilization under turbulent flow, long-duration LOS stability under high UAV mobility, and resource-optimized control strategies that can balance precision and energy efficiency.

From a system design perspective, the most effective stack for building a high-performance CPM for FSO-based communication with aerial platforms should combine high-torque density BLDC motors with dual-axis direct-drive gimbal architectures to minimize backlash and mechanical delays. Control-wise, a hybrid approach combining MPC for trajectory foresight and EKF for sensor fusion offers a superior tracking accuracy in dynamic aerial conditions. Incorporating adaptive control logic ensures resilience in dynamic aerial conditions, like changing payloads and external disturbances like wind gusts. For sensory inputs, a tight integration of GPS/INS with visual inertial odometry (VIO) enables accurate localization even under GNSS-challenged environments. To future-proof the system, onboard deployment of lightweight AI models, especially LSTM or transformer-based trajectory predictors, can preemptively adjust pointing, reducing response lag during fast maneuvers. Looking ahead, further research must focus on graph-based swarm pointing coordination and modular software architectures that allow plug-and-play integration of future sensors and AI models. This cohesive and reliable stack sets the foundation for agile, scalable, and autonomous FSO networks operating across complex aerial domains. In conclusion, CPMs or gimbals are the backbone of reliable FSO communication on non-terrestrial platforms for 6G communication. Building on recent progress, the continued convergence of lightweight materials, adaptive control, AI, and system integration will set the course for this critical technology. As aerial platforms evolve into more autonomous and collaborative systems, the demand for resilient, low-power, and ultra-precise pointing modules will continue to rise.