Open Access
Add to BibSonomy
Abstract
Free-space optics (FSO) convey an enormous potential for ultra-high-capacity seamless fiber-wireless transmission in 5G and beyond communication systems. However, for its practical exploitation in future deployments, FSO still requires the development of very high-precision and robust optical beam alignment. In this paper, we propose two different methods to achieve tight, precise alignment between a pair of FSO transceivers, using a gimbal-based setup. For scenarios where there is no information about the system, a black-box artificial intelligence (AI)-based method resorting to particle swarm optimization (PSO) is presented, enabling to autonomously align the system with a success rate above 96%, converging from a blind starting position. Alternatively, for scenarios with partial information about the FSO system, we propose a tailored custom algorithm, with a success rate of 92%, but with a ∼4 × reduction on the alignment time. The automatic alignment is then validated in a 5G-like fiber-FSO scenario, transmitting a 16 × 400 MHz signal and achieving a maximum bit-rate of 30 Gbps. Moreover, we propose the implementation of a fail-safe mechanism with a backup FSO receiver, thereby providing an extra degree of robustness towards temporary events of strong degradation on the FSO channel or line-of-sight (LOS) interruption.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Telecommunication systems are currently facing a capacity overload associated with the development/deployment of the fifth-generation (5G) and beyond-5G (B5G) technologies. The new generations standards are already being established by the 3rd generation partnership project (3GPP), with the famous 5G triangle specifying enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable and low-latency communications (uRLLC) [1]. To comply with the increased data-rate requirements, 3GPP has specified mm-Wave communications with transmission frequencies between 24 GHz and 53 GHz [2]. Furthermore, recent studies are already addressing the migration of radio frequency (RF) communications to the THz-band to cope with future generations [3,4]. These emerging challenges are forcing traditional radio access networks (RANs) to evolve in order to fulfill all these novel requirements [5].
Within this context, free-space optics (FSO) are rapidly attracting the research interest as an alternative fronthaul technology capable of meeting these demands [6], thanks to its high bandwidth, license-free spectrum, ease of deployment and improved security [7]. While recent proof-of-concept studies have already demonstrated the feasibility of 5G-over-FSO transmission [8,9], the practical deployment of high-capacity FSO systems [10,11] still faces numerous challenges, namely in terms of high-precision beam alignment [12], which becomes even more critical in seamless fiber-FSO systems using passive fiber collimators [13].
In recent years, in order to mitigate the impact of pointing errors, several acquire, tracking and pointing (ATP) mechanisms have been developed to increase the robustness of FSO links. Depending on the equipment used, these ATP mechanisms can be broadly classified as gimbal-based, mirror-based, adaptive optics (AO) and liquid crystal solutions [14]. Moreover, hybrid ATP methods have been proposed, integrating the different advantages of each mechanism [15].
Despite their limited steering speed and precision, gimbal-based systems provide the advantage of a wide field-of-view (FoV), which is a key limiting factor in its counterpart technologies [13,14]. Nevertheless, recent advances in high precision electro-mechanical systems may allow to effectively answer the needs of FSO systems. Furthermore, the wide FoV makes gimbal-based systems ideal for MIMO-FSO systems, enabling them to switch between receivers that can be several meters apart [16].
Given the aforementioned advantages of FSO systems, the scientific community has been focused on trying to mitigate the key impairments that can compromise this technology. In [17], the authors show a remarkable 11 mm misalignment tolerance using a custom made 2-D photodetector (PD) array. This high misalignment tolerance is obtained at the cost of a bandwidth trade-off, where the −3 dB point is below 5 GHz. Indeed, using a PD-based receiver considerably increases the robustness of the link to the impact of pointing errors and sub-optimal angle of arrival (AoA). However, it also imposes a set of constraints to the FSO system due to the requirement of an optical-to-electrical (O/E) conversion at the receiving end of the FSO link, thus setting an ultimate bandwidth bottleneck for the system and requiring an additional electrical-to-optical conversion whenever the signal has to be relayed over optical fiber to a remote final receiver.
In contrast, in seamless fiber-FSO links, the free-space signal is directly collimated into the fiber core, and therefore, no bandwidth limitations are imposed by the FSO receiver. Removing the O/E conversion at the receiver, seamless fiber-FSO has the potential to be integrated in any type of optical communications, supporting signals that can be multiplexed in the polarization, wavelength or amplitude [18]. Moreover, this type of connection enables full-duplex communication without requiring an extra Tx-Rx collimator pair. These advantages come at the cost of increased pointing errors and AoA sensitivity. Tackling this latter problem, in [13], the authors propose an FSO transceiver based on quadrant detectors (QD) and a fine pointing mirror (FPM) to track changes in the beam AoA and steer it to the fiber core. With this FSO transceiver, the authors are able to show the reliability of a 1 km FSO link that is subjected to atmospheric turbulence. In other work [19], seamless FSO is employed to establish a link inside a datacenter using a combination between a 2-axis gimbal and a spatial light modulator (SLM) to steer the beam to the fiber core, and a camera to keep track of the beam position. In [15] the authors present an FSO transceiver that can automatically align itself with an equal receiver pair. A first coarse alignment is obtained based on GPS positioning, and afterwards, there is an extra "alignment-link" composed of a near-infrared (NIR) camera and a 650 nm laser. Finally, in the main link, a mirror is used to couple the free-space signal into a fiber array, imposing coupling losses of 16 dB. It is worth noticing that these previous works always rely on some kind of external active aid to assist in the steering and tracking of the beam, thus increasing the FSO-transceiver footprint and power consumption. Furthermore, alignment from a blind position is not approached, and while classical optimization algorithms might be adequate choices in systems with well-defined operation boundaries, the problem of aligning a newly deployed FSO link from scratch, without any a-priori knowledge from the system, poses a daunting challenge in terms of convergence robustness and speed. In that sense, the recent progresses made on artificial intelligence (AI) algorithms applied to optical communications [20] might provide a disruptive solution to this critical problem, which currently severely limits the practical applicability of high-capacity FSO systems.
In this work, we consider a centralized RAN (C-RAN) paradigm, where the central unit (CU) is connected to the remote unit (RU) through a fiber-FSO fronthaul via radio-over-fiber (RoF) transmission, similarly to our previously proposed network architecture published in [8]. The FSO section of the fronthaul link is aimed at application scenarios in which standard optical fiber cannot be deployed (e.g. due to harsh terrain conditions or licensing issues) or as a temporary high-capacity wireless solution after an abrupt fiber link disruption (e.g. due to invasive construction work or natural disasters).
In this paper, we build upon this fronthaul architecture adding new key capabilities such as adaptive FSO receiver discovery and beam alignment. Additionally, we present a fail-safe mechanism, enabling automatic optical beam steering between two RUs. The proposed C-RAN architecture featuring the aforementioned capabilities is depicted in Fig. 1. In order to discover and automatically align the FSO-link we propose a novel AI-driven method, which in a black-box way, tackles the problem of fully-blind beam alignment in newly deployed 5G-FSO systems. A complementary method is also presented, taking a process of gray-box modelling. This method is composed of a 3-stage greedy algorithm, which, assuming partial knowledge of the system, is able to provide fast alignment. It is worth emphasizing that, both proposed algorithms rely only on a gimbal-based ATP mechanism, avoiding the typical problems encountered in the majority of related works that require extra hardware and/or a dedicated alignment link. Therefore, our proposed solution provides a novel efficient methodology for blind FSO beam alignment with improved energy efficiency and potentially reduced cost and footprint. The obtained results show the transmission of a $16\times 400$ MHz subcarrier multiplexing (SCM) signal over the proposed architecture with efficient beam steering to prevent link-failure.
2. Proposed ATP methods
In order to automatically align the FSO transceivers we used the setup depicted in Fig. 2 to implement two different alignment methods. At the transmitter, we use a continuous-wave (CW) laser with an output power of 4 dBm. In a seamless way, the laser output signal is connected to the Tx fiber collimator, which sends the optical signal to free-space. After travelling a distance of 3 m, the optical signal is received by a similar fiber collimator, which focus the optical signal directly into the fiber core. Finally, the received power is measured using a high-precision power-meter (HP - 81531A). All fiber collimators (Thorlabs F810APC-1550) are characterized by a 24 mm lens diameter, 0.0017$^\circ$ divergence angle, and 0.24 numerical aperture. Additionally, we have two high-precision stepper motors (Thorlabs ZST206) attached to the Tx collimator, which allow to control the optical beam position in the $y-$ and $x-$axis. The stepper motors support a minimum incremental motion of the lead screw of $\Delta s \simeq 1$ $\mu$m in standard operation mode, or $\Delta s \simeq 0.5$ nm in high-precision mode, i.e. using micro-steps, which enable to divide each regular step into 2048 sub-steps [21]. Since the height of the collimator mount (Thorlabs KM100) is 44.4 mm, the maximum precision for beam adjustment over a given distance $d$ is given by $\Delta x = d \cdot \Delta s / (44.4\times 10^{-3})$. Setting a minimum alignment precision of 0.1 mm, it turns out that the maximum supported distances are about 5 m and 10 km, when operating in standard or high-precision modes, respectively. Since the distance between fiber collimators in the experimental setup has been limited to 3 m (due to laboratory space constraints), the stepper motors have been only operated in the standard precision mode, i.e. without resorting to the usage of micro-steps. Nevertheless, it becomes clear that the high-precision micro-steps feature will play a key role in practical outdoor deployment scenarios, enabling to cover a wide range of useful distances for the majority of commercial FSO applications. In order to implement the adaptive beam alignment, the received power value is processed by a computer, which uses that information to move the motors and thereby align the FSO transceivers. In the next subsections, a detailed description of the proposed black-box and gray-box models is provided.
Fig. 2. Experimental setup used for automatic beam alignment from a blind starting position, it is also depicted the colormap of power received at the Rx for the different beam focus position and the power penalty of beam misalignment.
2.1 Black-box alignment: AI-based algorithm
With a more general perspective, we started by implementing an AI-based algorithm to perform the FSO-link alignment. In this stage, we choose to use particle-swarm optimization (PSO) to find the beam position that returns the highest received power.
The PSO algorithm was first proposed by Kennedy and Eberhart [22], with the purpose of optimizing nonlinear cost-functions and training artificial neural networks (ANN). PSO belongs to the group of evolutionary algorithms and relies on swarm intelligence between different particles to obtain the global minimum of a function. An overview of this algorithm flow is thoroughly exposed in [23]. In a brief note, for optimizing a problem composed of $N_{\textrm {var}}$ variables with $N_p$ particles, PSO starts by defining a position and a velocity vector for each particle, $X(t)$ and $V(t)$, respectively. These vectors are defined as [23]:
andFig. 3. Example of PSO convergence in the problem at hands, for a scenario where $N_p=300$ particles are considered.
After these preliminary tests with the PSO algorithm, we started by analyzing the optimal swarm size that allows minimizing the convergence time. We performed tests with swarm sizes varying from 5 to 70, whose results are depicted in Fig. 4. From the analysis of this figure we verify that the faster convergence time of approximately 18 minutes is obtained with 15 particles. Therefore, $N_p=15$ will be considered as the default swarm size for all subsequent experimental trials in this work. Note that the convergence time reported in Fig. 4 is highly limited by the offline prototype paradigm of the experiment, which includes utterly inefficient hardware and software interfaces (e.g. MATLAB-based processing in a general-purpose CPU). These convergence times should then only be interpreted in relative terms, whereas the absolute convergence times are expected to be reduced by orders of magnitude when implemented in an hardware-optimized real-time platform.
2.2 Gray-box alignment: multi-stage greedy algorithm
In the previous subsection, we have introduced an AI-aided fully-blind approach to perform beam alignment in a newly installed FSO link. The proposed methodology does not resort to any information from the system, such as the number of local minima, the width of the optimum alignment point and gradient of the optimization problem. Although such a black-box approach provides a more robust and system-independent convergence, the required time for beam alignment might be compromised by the blind optimization paradigm.
As opposed to this black-box approach, let us now address a different methodology in which some partial knowledge of the FSO system is taken into account, leading to a semi-blind gray-box model. Namely, it is assumed that the alignment problem is characterized by a well-behaved global minimum, and that the optimization gradient is circularly symmetric around it. In such conditions, an algorithm following always the higher power path (greedy) can be considered for a faster optimization of the beam alignment problem. In addition, it is assumed that a-priori calibration data is available for the receiving collimator, characterizing its received power (or conversely, attenuation) as a function of the $x$ and $y$ spatial coordinates of the stepper motors, as shown in the inset of Fig. 2. The availability of this calibration data, which might be provided by the manufacturer in a practical application, allows the implementation of adaptive step-size control over the optimization algorithm, thus enabling to minimize the convergence time.
Following this reasoning, we developed a tailored 3-stage algorithm to reduce the time of alignment. The division of the algorithm into stages is based on the use of predefined power threshold values, $P_{\mathrm {th},n}$, that define the criteria to advance from stage $n$ to stage $n+1$. The algorithm mechanism is exposed in Fig. 5, and can be described as follows:
- • Stage 1 - Optical Power Sniffer: In the first stage, the main goal is to move the optical beam into a position in which some measurable optical power is detected at the receiving collimator. Since in the starting point, the optical beam might be pointed completely outside the lens of the receiving collimator, the more frequently measured power value corresponds to the power-meter sensitivity (approx. $-73$ dBm), which provides no information for which path to follow. Due to this lack of information, we designed an algorithm that randomly chooses a 10-step path. After collecting the power value in each path position, the beam is returned to the position that yields the highest power and iterates a new 10-step random path, as depicted in the blue shadowed area in Fig. 5. If at any time a power higher than the predefined $P_{\mathrm {th},1}$ threshold (−70 dBm in our setup) is found, then the algorithm advances to stage 2.
- • Stage 2 - Coarse Alignment: The second stage is shown in the green shadowed area of Fig. 5. Its main goal is to provide a first coarse beam alignment, whose convergence is more critical due to the low measured optical powers, and therefore requires a robust optimization algorithm. Since a measurable optical power has been provided by the previous stage, the optimization no longer needs to take random steps. However, the $x/y$ direction that should be taken is still unknown. Therefore, we propose the use of a robust optimization approach, based on a growing matrix algorithm, as exposed by the circled numbers in stage 2 of Fig. 5. When an optical power higher than the predefined threshold, $P_{\mathrm {th},2}$, is achieved (−60 dBm in our setup), the alignment procedure advances to its last stage of optimization.
- • Stage 3 - Fine Alignment: Finally, in the third stage of the proposed optimization procedure, a fine and fast beam alignment is pursued through the use of a greedy algorithm, as shown in the red shadowed area in Fig. 5. Thanks to the starting optical power provided by the previous stage ($P>P_{\mathrm {th},2}$), it can be assumed that the algorithm is now in a region with a well-behaved gradient of optimization, which enables the use of a fast greedy approach without compromising the likelihood of convergence.
This alignment is then performed by a typical greedy neighbour search with the neighbour position being dynamically calculated in each iteration. The optimization step-size, $\mu _{\textrm {greedy}}$ (which defines the search-space size), is adaptively assigned on each iteration, $i$, based on the measured received power, $P_{\textrm {rec}}$, as:
where $\alpha$ and $\beta$ are constants that are optimized to minimize the alignment time. Finally, when the target optimization threshold, $P_{\mathrm {th},2}$, is achieved (−1 dBm in our setup) the optical beam alignment is concluded.
Fig. 5. Flowchart of the 3 Stage Greedy Algorithm (Gray-Box Alignment), where the stages are highlighted with different colors.
An example of the functioning of this algorithm is illustrated in Fig. 6, where the search path is depicted on top of a previously extracted contour map of received optical power in each position. As can be observed, the initial random search is the stage that requires a higher number of iterations, due to the lack of information when the beam is far from its optimum point. After finding some residual power the following two stages converge in just a few iterations.
Fig. 6. Example of proposed algorithm convergence in the problem at hands, where the 3-Stages are represented with different colors, following the code : Stage 1 ($\color{green}{\circ}$), Stage 2 ($\color{purple}{*}$) and Stage 3 ($\color{red}{\rm x}$).
3. Experimental validation of the proposed ATP methods
In order to validate the proposed algorithms, we have again used the experimental setup of Fig. 2. In this scenario, the stepper motors are utilized to perform two different functions: i) first they are used to deviate the beam to a random position outside the Rx collimator, thereby generating the so-called phenomena of pointing errors, ii) second they have the function of automatic aligning the beam to its optimal point. To guarantee the statistical significance of the experimental analysis, the automatic alignment procedure is independently performed 300 times for each algorithm. The randomly generated pointing errors and the output power achieved by each algorithm are depicted in Fig. 7. The PSO results are represented in the blue-shadowed area, whereas the results with the 3-stage greedy algorithm are inside the red-shadowed area. From the bar charts, it can be concluded that the PSO algorithm is able to successfully align the beam from a blind position in 96% of the experimental trials, while the 3-stage greedy algorithm has achieved a success rate of 92%. However, it can be observed that all registered events of non-convergence are associated with the inability of stage 1 to converge to the defined threshold, $P_{\mathrm {th},1}$, within the maximum allowed time for this stage (25 minutes), while the individual success rate of stages 2 and 3 is 100%.
Fig. 7. Results obtained for the automatic alignment procedure with the PSO and the proposed 3-Stage greedy algorithm.
Let us now compare the average alignment times required by each algorithm, as shown in Fig. 8. It can be seen that the 3-stage greedy algorithm provides a faster convergence to the maximum optical power (represented in Fig. 8 by the normalized value of 0 dB), with a mean alignment time around 6 min. Indeed, as can be seen in the inset of Fig. 7, most of the convergence time required by the 3-stage greedy algorithm is spent on stages 1 and 2, which are not yet greedy-based. Once the power threshold to initiate stage 3 is achieved, the greedy algorithm is able to very quickly converge to the optimum alignment point. Instead, the PSO algorithm tends to follow a more progressive convergence path, as shown by the square blue markers in Fig. 8. Indeed, this is the expected behavior: since the PSO implements a black-box alignment procedure that has no information about the optimization problem, it can not benefit from the assumptions utilized in the greedy algorithm. Consequently, its average convergence time of $\sim$22 min is roughly $4\times$ higher than that enabled by the 3-stage greedy algorithm. Note that, although the PSO algorithm is able to converge in a much lower number of iterations, as shown by the insets of Fig. 7, each PSO iteration implies a swarm-size of 15 particles, and therefore it typically takes longer than the corresponding iteration with the 3-stage greedy algorithm. Nevertheless, despite its longer convergence times, the PSO algorithm provides an universal and robust solution for optical beam alignment, without requiring any a-priori information from the system.
As a final note, it is worth re-emphasizing that the convergence times hereby reported should be interpreted in relative terms, while the absolute values of alignment time are expected to be significantly reduced when implemented in real-time, in a field deployment. Also note that the reported alignment procedure corresponds to the challenging condition of a new FSO link deployment, in which the starting point of alignment is completely unknown (i.e. the initial power feedback is null). After this initial beam alignment procedure, the continuous fine-tuning of the system to counteract dynamic pointing errors can be performed with negligible delay.
4. Impact and mitigation of pointing errors on a wideband optical signal
After the comprehensive analysis of reliability and convergence time associated with the proposed optical beam alignment methods, let us now address the impact and mitigation of pointing errors on an actual optical transmission system including fiber and FSO links.
Figure 9 depicts the experimental fiber+FSO setup, including adaptive beam alignment and steering between the transmitter and receiver collimators. In addition to the alignment setup we have an arbitrary waveform generator (AWG) with 16 GSa/s that is transmitting a signal composed of the multiplexing of 16 subcarriers, each one with 400 MHz bandwidth, resembling the highest standardized capacity for 5G-NR in frequency-range 2 (FR2) to be supported by mm-wave frequencies in the range of 24.25 GHz and 52.6 GHz [2]. In order to avoid the bandwidth constraints associated with the direct radio-over-fiber transmission of mm-wave frequencies, in this work we consider that the signal is first down/upconverted to an intermediate frequency (IF) of 3.3 GHz and then propagated over the fiber-FSO system (IF-over-fiber/FSO). In our downlink setup, this IF is generated through digital upconversion and the resulting RF signal is amplified and fed to a directly modulated distributed feedback laser (DFB) emitting at 1545 nm with an output power of 0 dBm and a 3 dB RF bandwidth of 15 GHz. The laser is followed by an erbium-doped fiber amplifier (EDFA) with a 5 dB noise figure to yield a total transmitted optical power of 6 dBm. Inserting this gain in the system assures that the optical signal arrives at the PIN with an optical power close to the optimal one. The optical signal is then collimated and sent to a free-space travelling distance of 3 m, before being received by one of the Rx collimators. At the receiver-side, an extra fiber collimator is added to enable and test a fail-safe mechanism scenario. An optical switch controls which receiver collimator is connected to the fiber link. The signal then travels through 11 km of standard single-mode fiber (SSMF) with an attenuation of 0.2 dB/km, introducing a total propagation loss of 2.2 dB, plus approximately 1 dB attenuation on the fiber connectors. Afterwards, the signal is divided into two branches with a coupling ratio of 90/10, resulting in an additional 0.5 dB attenuation. The O/E conversion is performed by a InGaAs PIN with a responsivity of 0.88 A/W and a 3 dB bandwidth of 11 GHz. Afterwards the signal is amplified by a low noise transimpedance amplifier. The electrical signal is then sampled and digitized by an oscilloscope with 8 GHz bandwidth and 12 GSa/s sampling-rate. Finally, offline digital signal processing (DSP) is performed, including carrier-phase estimation and an 2$\times$2 least-mean square equalization (51 taps). The system performance is assessed through the evaluation of the average error vector magnitude (EVM) of the multi-carrier received signal. In this architecture the Tx motors have an additional function, consisting in responding to a link-failure, i.e. if a sudden drop of optical power is identified, suggesting an FSO link disruption, the motors are used to switch between Rx collimators. After failure detection, the control algorithm uses the stepper motors to steer the beam to the other collimator, and to prevent any misalignment introduced by mechanical steering, automatic alignment is again applied to refine the beam position.
Fig. 9. Experimental setup used for transmission of 16$\times$400 MHz 64QAM carriers, with automatic beam alignment and fail-recovery capability.
4.1 Performance evaluation with non-ideal link alignment
The first test consisted in evaluating the impact of link misalignment in the system performance. In this stage, the motors are utilized to change the beam focus position and synchronous EVM measurements are taken to estimate the penalty induced by misalignment. Figure 10(a) depicts the impact of pointing errors in the received EVM in a B2B scenario and with 11 km SSMF. As expected, the EVM surface shows an approximate Gaussian shape, similarly to the optical power budget characterization performed in Fig. 2. The very high sensitivity of the system to the impact of pointing errors becomes evident: a 1 dB worsening of the EVM is incurred by an average beam misalignment of only 0.3 mm, as shown by the projected isobar in Fig. 10(a). After propagation over 11 km SSMF, it is worth noting that the sensitivity to beam misalignment is roughly halved, tolerating a pointing error of approximately 0.6 mm to yield the same 1 dB EVM penalty reference. This shows that the tolerance to beam misalignment strongly depends on the required performance level: while the minimum EVM achieved in B2B is about −24 dB (or equivalently, 6%), the corresponding minimum EVM after 11 km SSMF transmission is about −14.4 dB (or equivalently, 19%) due to the combined effect of laser chirp and chromatic dispersion. Overall, these results show that the impact of beam misalignment is partially overshadowed by other practical limitations in realistic optical transport networks, thereby conferring an inherent level of robustness to pointing errors.
Fig. 10. Performance measured with a $16\times 400$ MHz signal in function of the position of the beam in the receiver collimator, for an optical back-to-back scenario and with 11 km SMF: a) average EVM obtained, b) corresponding bitrate achieved.
Figure 10(b) translates the measured EVM values into achievable bitrate. It is worth noticing that for correctly translating the EVM of each carrier into bitrate, we need to assure that each carrier has a modulation format that respects the 3GPP EVM requirements (Table 1) [24]. Therefore, this bitrate is calculated taking into account that the modulation format in every subcarrier meets the EVM requirements exposed in Table 1. This figure shows that the propagation of the signal over fiber has a considerable impact on the system, decreasing the achievable bitrate from 30 Gbps (B2B) to 14 Gbps (11 km SSMF).
Table 1. 3GPP limits for each modulation format
4.2 Automatic link protection switching
We proceed by testing the fail-safe mechanism. At an initial stage, we used the aforementioned 3-stage greedy algorithm to discover the position of both Rx collimators. It is worth noticing that the motors change the tilt of the Tx collimator instead of changing its actual position, so the range covered by the optical beam is proportional to the link length. In our scenario, since the FSO link length is 3 m, the maximum inter-collimator distance for safe operation is $\sim$30 cm in the $x$ and $y$ axis. Nevertheless, if the link distance is increased to 1 km in practical applications, the receivers can be separated further apart by approximately 100 m, while using the exact same alignment apparatus. We would like to re-instate that the results here presented are obtained for a proof-of-concept system. In this scenario, the FSO receivers are too close to guarantee spatial decorrelation between links, and the only way to have different link performances is by inducing perturbations, hereby introduced in the form of pointing errors. In a real-world deployment, with increased distance between receivers, the links will tend to experience different free-space channels, including different atmospheric turbulence and blockage probability. In such a scenario, the same fail-safe mechanism can be used not only to prevent failure due to pointing errors but also to counteract other channel impairments that might impose a catastrophic degradation on one of the channels. With the positions of both receivers known, we proceed by testing the automatic beam steering between receivers when a link failure is detected. Due to the inherent variance of FSO power measurements, a no-failure tolerance of 1.5 dB is considered. When a link failure is detected, the stepper motors steer the beam to the backup collimator, using its pre-stored position.
In order to emulate a more practical system, we introduce an intentional deviation in the beam steering, progressively offsetting the ideal collimator position by 1 mm on each switching event. The results of this test are depicted in Fig. 11, where 4 failure events are emulated over time. It can be observed that the steering process takes approximately 3 s. The impact of the intentional systematic 1 mm error on the switching process can be clearly observed during the recovery from the last two failure events (2 mm and 3 mm deviations), where the beam arrives at a point where the optical power is still below the 1.5 dB tolerance, and therefore the automatic alignment algorithm is temporarily reactivated in order fine-tune the alignment. Aided by the spatial information collected during the discovery phase, the PSO algorithm is now able to realign the beam into its perfect position in less than 10 s. As previously highlighted, these switching times can be significantly reduced by optimizing the system’s hardware and software for real-time operation, which however is outside of the scope of this paper.
5. Conclusions
Addressing the challenges imposed by 5G and B5G technologies, we proposed a fiber-FSO RAN architecture with automatic beam alignment and a fail-safe mechanism. In order to precisely align the FSO-link, we implemented two different approaches. Resorting to a black-box concept, we implemented an AI-based alignment system (PSO), which was shown to enable a blind and robust operation, adequate for novel FSO deployments where the link and component characteristics are largely unknown. Alternatively, for scenarios where partial link information is given, we propose a tailored 3-stage greedy algorithm, enabling to cut down the beam alignment time by roughly a factor of 4, while requiring only the basic assumption of a well-defined global minimum for the alignment problem.
Using a gimbal-based ATP mechanism, we demonstrated the transmission of a $16\times 400$ MHz SCM signal over a $1\times 2$ FSO system with adaptive beam alignment obtaining a bitrate of 30 Gbps on B2B and 14 Gbps in a scenario with 11 km SSMF. In addition, an automatic optical path switching for link failure recovery, is also demonstrated. After the initial blind discovery of the receiver collimator positions, a switching time of 3 s is demonstrated, enabling physical layer 1+1 protection in the event of the catastrophic FSO link degradation or failure.
Funding
“la Caixa” Foundation (LCF/BQ/PR20/11770015, 100010434); Fundação para a Ciência e a Tecnologia (2020.07521.BD, CENTRO-01-0145- FEDER-022141, UIDB/EEA/50008/2020).
Acknowledgments
The authors would like to acknowledge the valuable contributions and insightful discussions with Daniel Albergaria and Abel Lorences-Riesgo.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. H. Tullberg, P. Popovski, Z. Li, M. A. Uusitalo, A. Hoglund, O. Bulakci, M. Fallgren, and J. F. Monserrat, “The METIS 5G system concept: Meeting the 5G requirements,” IEEE Commun. Mag. 54(12), 132–139 (2016). [CrossRef]
2. 3GPP, “Release 15,” Technical Specification (TS) 38.306, 3rd Generation Partnership Project (3GPP) (2019). Version 15.7.0.
3. T. Kawanishi, “THz and photonic seamless communications,” J. Lightwave Technol. 37(7), 1671–1679 (2019). [CrossRef]
4. V. Petrov, A. Pyattaev, D. Moltchanov, and Y. Koucheryavy, “Terahertz band communications: Applications, research challenges, and standardization activities,” in 2016 8th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT), (IEEE, 2016).
5. I. A. Alimi, A. L. Teixeira, and P. P. Monteiro, “Toward an efficient C-RAN optical fronthaul for the future networks: A tutorial on technologies, requirements, challenges, and solutions,” IEEE Commun. Surv. Tutorials 20(1), 708–769 (2018). [CrossRef]
6. F. Demers, H. Yanikomeroglu, and M. St-Hilaire, “A survey of opportunities for free space optics in next generation cellular networks,” in 2011 Ninth Annual Communication Networks and Services Research Conference, (IEEE, 2011).
7. M. Matsumoto, K. Kazaura, P. Dat, A. Shah, K. Omae, T. Suzuki, K. Wakamori, T. Higashino, K. Tsukamoto, and S. Komaki, “An alternative access technology for next generation networks based on full-optical wireless communication links,” in 2008 First ITU-T Kaleidoscope Academic Conference - Innovations in NGN: Future Network and Services, (IEEE, 2008).
8. A. O. Mufutau, F. P. Guiomar, M. A. Fernandes, A. Lorences-Riesgo, A. Oliveira, and P. P. Monteiro, “Demonstration of a hybrid optical fiber-wireless 5g fronthaul coexisting with end-to-end 4G networks,” J. Opt. Commun. Netw. 12(3), 72 (2020). [CrossRef]
9. M. Alzenad, M. Z. Shakir, H. Yanikomeroglu, and M.-S. Alouini, “FSO-based vertical backhaul/fronthaul framework for 5G+ wireless networks,” IEEE Commun. Mag. 56(1), 218–224 (2018). [CrossRef]
10. A. Lorences-Riesgo, F. P. Guiomar, A. N. Sousa, A. L. Teixeira, N. J. Muga, M. C. R. Medeiros, and P. P. Monteiro, “200 G outdoor free-space-optics link using a single-photodiode receiver,” J. Lightwave Technol. 38(2), 394–400 (2020). [CrossRef]
11. F. P. Guiomar, A. Lorences-Riesgo, D. Ranzal, F. Rocco, A. N. Sousa, M. A. Fernandes, B. T. Brandao, A. Carena, A. Teixeira, M. C. R. Medeiros, and P. P. Monteiro, “Adaptive probabilistic shaped modulation for high-capacity free-space optical links,” J. Lightwave Technol. 38(23), 6529–6541 (2020). [CrossRef]
12. A. A. Farid and S. Hranilovic, “Outage capacity optimization for free-space optical links with pointing errors,” J. Lightwave Technol. 25(7), 1702–1710 (2007). [CrossRef]
13. P. T. Dat, A. Bekkali, K. Kazaura, K. Wakamori, and M. Matsumoto, “A universal platform for ubiquitous wireless communications using radio over FSO system,” J. Lightwave Technol. 28(16), 2258–2267 (2010). [CrossRef]
14. Y. Kaymak, R. Rojas-Cessa, J. Feng, N. Ansari, M. Zhou, and T. Zhang, “A survey on acquisition, tracking, and pointing mechanisms for mobile free-space optical communications,” IEEE Commun. Surv. Tutorials 20(2), 1104–1123 (2018). [CrossRef]
15. M. M. Abadi, M. A. Cox, R. E. Alsaigh, S. Viola, A. Forbes, and M. P. J. Lavery, “A space division multiplexed free-space-optical communication system that can auto-locate and fully self align with a remote transceiver,” Sci. Rep. 9(1), 19687 (2019). [CrossRef]
16. M. K. Al-Akkoumi, H. Refai, and J. James J. Sluss, “A tracking system for mobile FSO,” in Free-Space Laser Communication Technologies XX, S. Mecherle, ed. (SPIE, 2008).
17. T. Umezawa, Y. Yoshida, A. Kanno, A. Matsumoto, K. Akahane, N. Yamamoto, and T. Kawanishi, “FSO receiver with high optical alignment robustness using high-speed 2D-PDA and space diversity technique,” J. Lightwave Technol. 39(4), 1040–1047 (2021). [CrossRef]
18. G. Parca, A. Shahpari, V. Carrozzo, G. M. T. Beleffi, and A. L. J. Teixeira, “Optical wireless transmission at 1.6-Tbit/s (16×100 Gbit/s) for next-generation convergent urban infrastructures,” Opt. Eng. 52(11), 116102 (2013). [CrossRef]
19. F. Liu, M. Chen, W. Jiang, X. Jin, and Z. Xu, “Effective auto-alignment and tracking of transceivers for visible-light communication in data centres,” in Broadband Access Communication Technologies XIII, B. B. Dingel, K. Tsukamoto, and S. Mikroulis, eds. (SPIE, 2019).
20. F. Musumeci, C. Rottondi, A. Nag, I. Macaluso, D. Zibar, M. Ruffini, and M. Tornatore, “An overview on application of machine learning techniques in optical networks,” IEEE Commun. Surv. Tutorials 21(2), 1383–1408 (2019). [CrossRef]
21. “ZST206 - 6 mm Travel, Stepper Motorized Actuator, 1/4$^{\prime\prime}$-80 Thread,” https://www.thorlabs.com/thorproduct.cfm?partnumber=ZST206.
22. J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceedings of ICNN’95 - International Conference on Neural Networks, (IEEE, 1995).
23. B. S. G. de Almeida and V. C. Leite, “Particle swarm optimization: A powerful technique for solving engineering problems,” in Swarm Intelligence - Recent Advances, New Perspectives and Applications, (IntechOpen, 2019).
24. 3GPP, “Release 15,” Technical Specification (TS) 38.104, 3rd Generation Partnership Project (3GPP) (2018). Version 15.3.0.
