Fiber/Free-Space Optics with Open Radio Access Networks Supplements the Coverage of Millimeter-Wave Beamforming for Future 5G and 6G Communication

Highlights

What are the main findings?

• Fiber and FSO technology can be effectively integrated with O-RAN and deployed in practical applications. The advantages of FSO include its ability to transcend physical barriers, thereby enhancing real-world communication capabilities.
• The mmW beamforming capabilities of array antennas can be enhanced by employing reinforcement learning techniques to optimize network resource allocation for signal reception at the receiver end.

What are the implications of the main findings?

• When integrated with fiber and FSO, O-RAN antennas can be strategically deployed in underserved or remote regions that are beyond the reach of traditional mmW beamforming technology. This approach facilitates a more equitable distribution of network resources while effectively reducing the costs associated with future 5G and 6G communication infrastructures.
• Both FSO and O-RAN antennas can be strategically positioned, enabling the adjustment of their locations in response to short-term network demands. This capability enhances the overall flexibility of the network infrastructure.

Abstract

Conceptually, this paper aims to help reduce the communication blind spots originating from the design of millimeter-wave (mmW) beamforming by deploying radio units of an open radio access network (O-RAN) with free-space optics (FSOs) as the backhaul and the fiber-optic link as the fronthaul. At frequencies exceeding 24 GHz, the transmission reach of 5G/6G beamforming is limited to a few hundred meters, and the periphery area of the sector operational range of beamforming introduces a communication blind spot. Using FSOs as the backhaul and a fiber-optic link as the fronthaul, O-RAN empowers the radio unit to extend over greater distances to supplement the communication range that mmW beamforming cannot adequately cover. Notably, O-RAN is a prime example of next-generation wireless networks renowned for their adaptability and open architecture to enhance the cost-effectiveness of this integration. A 200 meter-long FSO link for backhaul and a fiber-optic link of up to 10 km for fronthaul were erected, thereby enabling the reach of communication services from urban centers to suburban and remote rural areas. Furthermore, in the context of beamforming, reinforcement learning (RL) was employed to optimize the error vector magnitude (EVM) by dynamically adjusting the beamforming phase based on the communication user’s location. In summary, the integration of RL-based mmW beamforming with the proposed O-RAN communication setup is operational. It lends scalability and cost-effectiveness to current and future communication infrastructures in urban, peri-urban, and rural areas.

1. Introduction

Fiber-optic technology is an essential part of the Internet and is an area of ongoing interest and development [1]. The rollout of 5G and the upcoming 6G systems, focused on the Internet of Everything (IoE) and related applications, are significant trends in telecommunications [2,3]. These advancements signal a gradual move away from 4G technology. While 5G and 6G can offer faster internet speeds, they are expected to have higher costs and shorter ranges than 4G. Therefore, finding ways to reduce costs and improve coverage to make 5G and 6G more effective and encourage widespread adoption is crucial. Recently, researchers have made important progress in developing systems for 5G new radio (NR). This includes technologies like power-over-fiber (PoF) [4], radio-over-fiber (RoF) [5], wired–wireless convergence [6], investigations on optimization of energy-efficient configurations [3], and optical wireless communication [2,4]. These advancements will help make the rollout of 5G NR and future technologies more cost-effective, efficient, and manageable.

In 4G applications, beamforming can improve performance and signal quality by directing signals to specific devices. However, many see it as an added feature rather than a necessary part of the setup. Without advanced beamforming techniques, the characteristics of low frequencies can still provide good coverage and performance. In contrast, beamforming becomes crucial in mmW communications, which operate at higher frequencies (over 24 GHz). The shorter wavelengths at these frequencies lead to greater signal loss, especially when obstacles or atmospheric conditions are present. Beamforming helps direct signals, improving coverage and reliability for mmW applications like 5G. Machine learning, as a component of artificial intelligence, presents a viable approach to enhancing network performance and optimizing resource allocation [7,8,9,10]. Specifically, RL techniques can be applied to improve mmW beamforming by concentrating on user-specific needs to achieve superior communication outcomes [11]. However, as the steering angle increases, aligning the beam with the target receiver becomes less efficient [12,13]. If the receiver moves or changes its distance from the antenna, the ability to maintain good alignment decreases, which is often unrealistic in real-world situations. It has been found that the EVM rises with larger steering angles, leading to weaker network performance [12,13]. The network may not perform well at the edges of coverage areas served by mmW beamforming. While adding more beamforming could improve coverage, it is not a practical solution. The high costs for the necessary components, along with interference issues from antennas using the same frequency and overlapping coverage areas, make this option impractical. A better solution is to use frequency diversity [14] to place additional antennas around the outer edges of the mmW beamforming area, where coverage gaps are noticeable.

O-RAN is a cost-effective choice for 5G NR applications. It allows network operators to connect hardware and software from different manufacturers easily and flexibly. This flexibility comes from O-RAN’s use of standardized interfaces and protocols, like open application programming interfaces, which help devices and software work together smoothly [15]. O-RAN helps lower costs and speeds up deployment by using readily available standard equipment that works with commercial off-the-shelf (COTS) technologies. Traditional RANs often have high power use, high operating costs, and poor sharing of processing power because the baseband unit (BBU) and remote RU are located together at the base station and link to the core network through backhaul. O-RAN addresses this issue by separating the base station into two parts: the centralized unit (CU) and the distributed unit (DU) at the center and the RU at the far end, which is linked by a fronthaul [3].

Fiber optics can guide light through a collimator, allowing signals to be transmitted over long distances in places without fiber optics, like space, land, and underwater [16,17]. This helps simplify fiber-optic routing. As mentioned earlier, FSO systems can work with O-RAN technology in tests [18]. This shows that non-terrestrial networks are possible and highlights future applications. Previously, FSOs were often used as a backup for fiber optics [19]. However, to manage faster transmission, mobility, and lower maintenance costs, FSOs now play a key role [20,21,22], even though atmospheric turbulence, fog, bad weather, or building movement can disrupt it [23,24,25]. Solutions like power control, diversity techniques, beam steering, and adaptive modulation can help address these issues [26,27,28,29]. Recent studies have shown that FSO communication can still be effective, even in tough weather conditions [30,31].

Figure 1 illustrates the solution’s primary goal: the server data center’s network sent to the repeater building via the FSO link, overcoming the land barrier for use by the office network and beamforming application through the network switch. The mmW beamforming part is installed in the repeater building, mainly supporting the civic center and its outskirts to the suburban neighborhoods, such as household and shop networks. Where mmW beamforming is not effective, O-RAN RUs can be used as fronthaul terminals via fiber-optic links to supplement the coverage. Moreover, the RUs can be extended by fiber-optic links to reach farther distances, such as villages or remote locations where beamforming would be wasteful due to a small population. Historically, FSO communication has been integrated with radio frequency (RF) systems to enhance the capacity and coverage of traditional RF communication [32]. However, these integrations often entail limited beamforming capabilities and exclude O-RAN architectures. Furthermore, the RU antennas associated with O-RAN can be strategically positioned in alignment with network demands through the deployment of long-haul FSOs or optical fiber connections. This strategic placement contributes to improved coverage efficiency and optimizes overall cost-effectiveness.

In this paper, it is emphasized that based on the 200 m FSO link as the backhaul and then processed by the O-RAN, it can undergo a fronthaul of up to 10 km to reach the RUs of the O-RAN and then through the customer premise equipment (CPE) or directly for users for 5G NR communication use. Conceptually, individual RUs of the O-RAN can be strategically positioned in the application area where mmW beamforming antennas cannot effectively cover, such as near the periphery of a sector. The 5G NR frequency band is n79. This exclusive frequency band can transmit data confidentially and is staggered from the proposed beamforming frequency of 28 GHz to avoid co-channel interference. Furthermore, mmW beamforming can optimize the phase of beamforming by RL with different beamforming steering angles for the best communication experience. Hence, conceptually, it is both cost-effective and feasible to complement the communication blindness of RL-based mmW beamforming with FSO link-based O-RAN RUs for future 5G/6G communication applications.

The subsequent sections of this paper are organized as follows: Section 2 examines the architecture and experimental findings related to the incorporation of RL techniques into beamforming. Section 3 presents a discussion of the architectural considerations and experimental results concerning the application of FSOs within the context of the O-RAN system. Finally, Section 4 provides a conclusion to the paper.

2. Adaptive Beam Alignment Using RL

Figure 2 depicts the proposed phase adaptive tuning architecture for beamforming, which uses the universal software radio peripheral (USRP) to transmit and receive signals. The first step is to generate the baseband 16-QAM signal after setting the parameters of the modulated signal through a personal computer (PC) using the radio hardware driver software LabVIEW. Subsequently, the baseband signal will be converted to 6 GHz through the USRP by digital–analog conversion, and the bandwidth is 250 kHz. To transmit millimeter-wave signals, the 6 GHz signal is boosted to 28 GHz by a frequency upconverter (UPC) and then transmitted to the other end of the 4 × 4 array antenna through a 4 × 4 array antenna, and then the 28 GHz signal is downconverted to 6 GHz by a frequency downconverter (DC). Then, the signal is transmitted to the USRP receiver, where it is demodulated, converted to the digital signal reduced to the base frequency, and analyzed for signal quality using LabVIEW. The beamforming steering angle setting options are −45°, −30°, −15°, 0°, 15°, 30°, and 45°, and the distance between the beamforming 4 × 4 array antenna Tx and Rx is fixed at 2 m. Figure 3 shows the measured radiation patterns of the 4 × 4 array antenna at different steering angles: (a) 0°, (b) 15° and −45°, (c) 30° and −30°, and (d) 45° and −15°. It can be noticed that the maximum power point of the radiation pattern distribution is consistent with the set steering angle. Hence, the highest power associated with the main lobe is the transmission direction for the communication application. The main lobe’s power already exceeds the side lobe’s power by more than 10 dB to guarantee that the main lobe transmission will not encounter interference.

Collecting EVM values at various phases is necessary for the application of RL-based beamforming with phase-adaptive beam alignment for the set steering angles of −45°, −30°, −15°, 0°, 15°, 30°, and 45°. The RL model, which seeks to predict and modify the beamforming antenna’s phase, is trained using collected data. RL comprises three fundamental components: state, action, and reward. The state denotes the present circumstances or environment, furnishing all necessary information for the agent to make decisions. The action is the choice taken by the agent according to the current state, which determines how the agent interacts with the environment and may lead to different outcomes. The reward is the score that the agent gets from the environment’s feedback signals after action. This score indicates how well the action accomplished the agent’s objective and serves as a basis for making subsequent decisions.

In real-world scenarios, the system relies on trained RL algorithms to anticipate and modify the phase of beamforming, as the precise angle of the user is not always known. Based on past data, these models can quickly determine the optimal phase configuration for situations when the user’s position is uncertain or fluctuating, resulting in improved signal coverage. To run efficiently and provide improved communication quality even in situations when the user’s angle is not instantly known, the system relies on broad patterns in the environment to draw inferences and optimizations instead of real-time user position information. RL is capable of dynamically adapting to changing circumstances in addition to determining the current ideal phase angle. For example, RL can swiftly adapt each beamforming phase to offer the best possible signal strength to each user even when many users’ locations change. This is done by using the knowledge it has gained over time. This approach is more efficient and adaptable to dynamically changing environments than manually testing all possible phase combinations one by one. Thus, while the optimal phase angle can be found manually in static environments, RL can respond more quickly to changes and continuously optimize system performance in real-world applications. In this setting, instead of receiving feedback from the EVM in real time, the RL model is trained offline on previously collected data. The model learns to predict the EVM based on different antenna phase configurations by analyzing these data to optimize its strategy, effectively “knowing” about the EVM because it learns the relationship between the EVM and antenna phase from experimental data.

The Q-learning algorithm is one type of off-policy RL technique that allows an agent to learn from behaviors based on an alternative policy. It continuously adjusts Q-values, or state–action pairs, in order to improve the agent’s decision-making abilities over time. In this context, the ε-greedy algorithm is used to choose the next course of action, striking a balance between exploitation and exploration [33]. The impact of future rewards is taken into account using a discount factor of 0.88, which reduces their ability to affect present actions. In essence, Q-learning chooses the course of action that maximizes the Q-table by updating the value function. The agent is provided with input on the EVM and the associated variable phases of the array antenna to optimize the EVM. Until the optimal EVM performance is reached, the agent iteratively modifies the array antenna’s variable phase over a predetermined number of steps. Thus, the performance of the system may be optimized by the adaptive beam alignment approach that has been suggested, which can independently adjust the phase in response to the EVM input from the RL agent. The beamforming Q-tables for five actions in the phase state that correspond to steering angles of 45°, 30°, 15°, 0°, −15°, −30°, and −45° are displayed in Figure 4a–g. The regions in the figure that are indicated in dark red have higher Q-values, meaning that the RL agent favors choosing the movements in these areas. Conversely, the dark blue regions denote lower Q-values, suggesting that the RL agent would rather not function in these regions. On the other hand, the system takes one measurement of the EVM each time the beamforming antenna’s phase is changed, and it reports whether the EVM grows or shrinks. If the EVM becomes larger, it means the signal quality has deteriorated, and the system will give a negative reward score and vice versa, as depicted in Figure 2.

Figure 5a–g illustrate the actions that the agent can take to achieve optimal convergence within a specific time period, which represents the time of a single interaction between the agent and the environment and is used to adjust the phase of the beamforming over consecutive time periods to optimize its performance. Therefore, the time period helps the RL algorithm to adjust the phase change strategy at different time intervals. At different beamforming angles of 45°, 30°, 15°, 0°, −15°, −30°, and −45°, there were 50 random phase settings employed in the initial stage, and the optimal phase convergence and EVM values obtained after RL operation were 300° and 8.54%, 105° and 7.61%, 240° and 7.55%, 0° and 7.38%, 120° and 7.3%, 240° and 7.53%, and 60° and 8.86%, respectively. These specific phases were derived through the utilization of the pertinent Q-tables depicted in Figure 4. It can be noted that the phase summation of the symmetric angles of the beamforming steering angles at ±45°, ±30°, and ±15° are 360°, 345°, and 360°, respectively, which are close to the optimal phase angle of 0° (which is also equal to 360°) for a 0° steering angle (no steering). Furthermore, although EVM values below 12.5% of the 3GPP EVM standard for 16-QAM signals can be attainable at all of these beamforming steering angles, the EVMs obtained deteriorated significantly when the steering angle was aligned with the edge of the sector, as described previously.

3. O-RAN with 200 m FSO Link

3.1. FSO Setup and Background

In the proposed scheme, 200 m of the FSO link are arranged utilizing pairs of commercially available FSO devices, as illustrated in the upper section of Figure 6. It should be clarified that the commercial FSO equipment can reach a transmission distance of at least 3 km, but to facilitate the setup and collection of experimental data in the urban area, it is only installed at a length of 200 m. In this work, 1330 nm small form-factor pluggable (SFP) transceivers are used as the barrel hoop for the network switches and the FSO devices, and the FSOs automatically convert the 1330 nm beam to a 1550 nm beam, which has lower propagation attenuation and has less impact on the human eye.

3.2. Preoperation and Measurement of 200 m FSO Link Erection

To ensure the normal operation of the FSO link and its related covenant, a basic test was conducted after the FSOs were set up. Since FSO devices support 1 Gbps and 2.5 Gbps transmission rates, the received power of the FSOs is an important indicator to ensure proper operation. Figure 7a shows the throughput of the FSO device at 1 Gbps and 2.5 Gbps for different power levels, which represents the lowest power that the FSO device requires at 1 Gbps and 2.5 Gbps at about −25 dBm and −20 dBm. The measurement of throughput was conducted using third-party speed measurement software (Iperf3) via a PC. Iperf3 is a tool for measuring network performance over a local network. It creates a data stream between a client and a server. Users can change settings like the buffer length and window size to help monitor and improve network performance. The configurations can impact the throughput measured during testing, enabling users to simulate various network conditions and assess how these changes influence performance. Iperf3 shows information on bandwidth and other important metrics. To conduct a test, two computers are set up: one as the client and the other as the server. The client sends requests to the server, which then transmits data streams to the client through a wireless optical communication system, completing the transmission test. The signal-to-noise ratio (SNR) is the reason why the attenuation of 2.5 Gbps transmission is less acceptable than that of 1 Gbps transmission. Higher receiving power is needed for a higher SNR, assuming that the background noise remains constant. As a result, 2.5 Gbps requires a higher SNR than 1 Gbps, and it naturally requires a higher receiving power than 1 Gbps. In addition, when there is an obstacle in the FSO link path for a certain time to block the passage of the beam, it will also affect the normal operation of the FSOs, and Figure 7a shows the effect of different masking times (MTs) on the FSO transmission. When the MT is below 30 s, the recovery time (RT) of FSO transmission is within 10 s, while when the MT is above 40 s, the RT of FSO transmission will be greatly increased or even unrecoverable, so this operation should be avoided.

Moreover, measurements were carried out on sunny, cloudy, and rainy days to ascertain the impact of weather conditions on the FSO transmission. The measurement scene, which is the same as the test scene in Figure 7a, is depicted in the upper half of Figure 6. As shown in Figure 6, the integrated technology building (ITB) is where the transmitter of FSO transmission is located, and the everlight building (EB) is where the receiver of FSO transmission is located. The data sampling was done every minute, and the results reveal that the throughput was similar and close to 2.5 Gbps for sunny, cloudy, and rainy days, as shown in Figure 7b. The FSO system exhibited considerable transmission attenuation during rainy weather conditions. Conversely, on sunny days, one would anticipate minimal transmission attenuation; however, this expected outcome was not consistently observed due to higher wind speeds recorded during such weather. While variations in weather can introduce discrepancies in the received power of FSO transmissions, it has been empirically demonstrated that these fluctuations do not significantly impact the overall throughput. This assertion is supported by the substantial levels of received power, as corroborated by the test results in Figure 7a. Hence, FSOs can function normally on sunny, cloudy, and rainy days. In hefty rain or strong winds, the transmission can be enhanced using the previously mentioned methods [26,27,28,29].

To assess network performance parameters, the following experiments were undertaken, disregarding variations in weather conditions. Figure 8 illustrates the outcomes of the physical network connection. The physical network can provide actual connections for subsequent O-RAN users, and it should be clarified that the campus network traffic limit is 100 Mbps. The ITB is the network source routed through the switch to enable the PC to conduct network performance measurements and is transmitted to the FSO device for subsequent transmission. The transmission then proceeds to the EB and is routed through the switch for the PC to perform network performance measurements. Evidently, the average values of uplink (UL) throughput, downlink (DL) throughput, and ping measured at the ITB before the 200 m FSO transmission at 2-min intervals were 92.051 Mbps, 89.527 Mbps, and 4.63 ms, respectively. The average values at the EB post-FSO transmission, on the other hand, were 92.165 Mbps, 88.785 Mbps, and 5.367 ms for uplink throughput, downlink throughput, and ping per 2-min sampling. This approximate measurement result indicates that the 200 m FSO link had a minimal impact on network performance.

3.3. The 200 m FSO Link Backhaul for O-RAN with 4 m Fronthaul for Indoor Network

After the successful establishment of FSO normal operation, the FSO link can be integrated into the O-RAN system. As depicted in Figure 6, the servo data center network is routed through a switch to the FSO, which then transmits across the physical barrier to the receiving FSO. Subsequently, the receiving FSO connects to the O-CU and O-DU sides of the RAN via a switch. The repeater building’s O-CU and O-DU sides can facilitate the signal transition to the O-RU side through fiber-optic fronthaul. The O-RU can directly serve mobile phone users or be converted to Wi-Fi by CPE for cellular phones or other electronic devices. Notably, direct communication between the O-RU and cell phone users necessitates a proprietary 5G chip, whereas electronic device users via CPE do not require this proprietary chip. Incidentally, the frequency band used by O-RAN is n79 (4.8 GHz to 4.9 GHz), the bandwidth is 100 MHz, and the downlink and uplink ratio is DSUUU (2:3). Based on this setup, the maximum uplink throughput is 181 Mbps, and the maximum downlink throughput is 410 Mbps, as provided by the supplier of O-RAN equipment. Similarly, it should be noted that the experiment was conducted on a university campus, and due to network traffic limitations, the uplink and downlink throughput limit was 100 Mbps, as shown in Figure 8. Figure 9 shows the network performance measurement of the O-RU via 4 m fiber-optic fronthaul directly to cell phone users. There were four scenarios of the distance between the O-RU antenna and the mobile phone: 10 m, 20 m, 30 m, and 40 m (measured within an indoor corridor). The O-RU antenna was about 2 m above the ground, the cell phone was about 1 m above the ground, and the sampling interval was once every two minutes. From Figure 9a, it can be seen that, irrespective of the distance between the cell phone and the O-RU antenna, the downlink throughput stayed around 80 to 90 Mbps. However, the uplink throughput was smaller and more variable than the downlink throughput, which is mainly attributed to the multipath interference generated by the non-open space, the smaller packets processed by the uplink than the downlink, and the upper limit of the transmission power of the mobile device. In addition, there were also antenna directivity issues, such as a significant height difference between the antenna and the cell phone, which indirectly led to more path loss and multipath effects interference, whereas downlinks are usually designed to be resilient against interference and fading and are less susceptible to antenna directivity issues. In Figure 9b, the performance of jitter (J) and ping are shown. Ping is the average latency, jitter is a measure of the variation in the time it takes for a packet to arrive, and jitter reflects the consistency of the latency. The ping stayed around 20 to 30 ms, while the jitter stayed around 0 to 11 ms. These indicators demonstrate that the network transmission is functioning well, and the values are consistent with the tests conducted by the supplier without the FSO link.

Figure 10 shows the O-RU antenna via 4 m fronthaul to the CPE, followed by the CPE feeds the network to the cell phone. The distance between the CPE and the O-RU antenna was arranged in three scenarios—0 m, 20 m, and 40 m—while the distance between the cell phone and the CPE was arranged in four scenarios—5 m, 10 m, 15 m, and 20—so there were a total of 12 measurement scenarios. Moreover, the O-RU antenna distance was about 2 m from the ground, and the CPE distance was about 0.5 m from the ground. The sampling was also done every two minutes, and the measurements were also conducted in the indoor corridor. Figure 10a illustrates the uplink throughput performance, and it can be found that it was lower than the case of the O-RU antenna directly connected to the cell phone, which is because the position of the CPE was not perfectly matched with the pointing of the antenna. Additionally, due to the multipath effect, the throughput varied but mostly stayed around 30 to 40 Mbps. The uplink throughput significantly degraded when the O-RU was far from the CPE and the CPE was distant from the cell phone; this signal degradation caused by the longer distance is expected. In the downlink throughput performance, as shown in Figure 10b, when the distance between the CPE and O-RU antenna was 0 m, the result mirrors that of the case of the O-RU antenna directly connected to the cell phone. However, when the distance between the CPE and O-RU antenna was 20 m and 40 m, the performance of downlink throughput was weakened due to signal degradation. The outcomes of jitter and ping, as depicted in Figure 10c,d, exhibit striking similarity to those observed when the cell phone was directly connected to the O-RU antenna. An exception can be noted in the ping, which demonstrated a marginal increase when the CPE was positioned 40 m away from the O-RU antenna and the cell phone was situated 20 m from the CPE. Since the CPE was 40 m away from the O-RAN, the ping was slightly larger, which made the corresponding jitter abnormal.

3.4. The 200 m FSO Link Backhaul for O-RAN with 10 km Fronthaul for Outdoor Network

The results depicted in Figure 11 illustrate the network performance evaluation of the O-RU antenna over a 10 km fiber-optic fronthaul, directly supplying the cellular network or via CPE. In comparison to the schematic in Figure 6, the actual configuration in Figure 11 entails the transmission of the network from the FSO device at the ITB to the FSOs at the EB through a switch, followed by forwarding to the O-RAN and fronthaul to the O-RU antenna using a 10 km fiber. The O-RU antenna at the EB can establish connectivity with the cell phone or CPE at the ITB balcony through a 200 m transmission accessibility field for cell phone connectivity. In addition, in the test where the cell phone connects to the network through the CPE with sampling every minute, the distance between the cell phone and the CPE was 1 m, and the CPE was more than 2 m above the ground to avoid being blocked by the balcony wall and aligned with the O-RU antenna. Based on the downlink throughput, it is apparent that whether the cellular device was directly linked to the O-RU antenna or connected to the O-RU antenna via the CPE, the outcomes closely resemble those of the fronthaul when utilizing only a 4 m optical fiber. Similarly, the jitter and ping measurements demonstrate strong similarity to the fronthaul scenario using only 4 m of fiber. Furthermore, the uplink throughput of the O-RU antenna directly connected via cell phone was notably inferior than the throughput of the 4 m fronthaul scheme due to the path loss and lower power of the cell phone. The uplink throughput of the O-RU antenna via CPE was better than that of the 4 m fronthaul scheme and also better than that of the cell phone directly connected to the O-RU antenna, which is due to the matching of the pointing of the O-RU antenna with the CPE and the higher power of the CPE. It is worth noting that the measurements shown in Figure 11 illustrate that the variability in throughput was significantly reduced when transmitting through the open-air field compared to an indoor corridor. This reduction is due to the significant decrease in the impact of multipath effects. A single data measurement aligns with the values displayed in Figure 11 for the 10 km fronthaul solution scenario; this measurement was taken when the cell phone and CPE were positioned near the O-RU antenna without FSO transmission. To test the network performance limit, the same measurement structure as the results in Figure 11 was used, except that the distance between the cell phone and CPE was extended to 20 m and 40 m for single-data measurement. It was found that the 20 m and 40 m cases match the results in Figure 11, except that the uplink throughput in the 40 m case dropped to 45.7 Mbps.

The measurement results and configuration aspects discussed in Section 3 indicate that the influence of FSOs as a backhaul and 10 km fiber as a fronthaul on network transmission appears to be minimal. The performance of the O-RAN is contingent upon its specific configuration [35]. The measurement outcomes of this study align closely with the test data provided by the O-RAN equipment vendor, and the internal configuration of the O-RAN is identical. The observed degradation in network performance can primarily be attributed to the potential misalignment of antennas and user devices, the multipath effect arising from the constraints of the measurement environment, and the operational range of user devices, such as CPE. Furthermore, while the beamforming configuration presented in this study operates independently of the O-RAN antenna unit, recent research has indicated the potential viability of integrating O-RAN antennas with beamforming technology [36,37]. This integration has the capacity to enhance network performance reliability and optimize resource allocation, thereby improving the cost-effectiveness of upcoming network applications. Notably, the 10 km fiber-optic fronthaul incorporated in this framework could be substituted with FSOs, given the open architecture of O-RAN. Consequently, this substitution would significantly benefit the future provisioning requirements of networks, allowing for a more flexible layout configuration and achieving an elevated level of resource allocation.

4. Conclusions

In this paper, a future 5G/6G solution integrating FSO, O-RAN, and mmW beamforming has been proposed. The FSO medium not only achieves transmission performance similar to that of physical fiber but also has the mobility to facilitate fast setup and overcome physical obstacles on the ground, thus helping to extend the end-use applications of O-RAN. In the aspect of mmW beamforming, the use of RL in the beamforming system can optimize the EVM by dynamically adjusting the beamforming phase according to the location of the communicating user. Nevertheless, beamforming at frequencies above 24 GHz has a maximum transmission distance of several hundred meters, leading to communication blind spots at the edges of the sector’s operating range of beamforming supply. O-RAN’s O-RU antenna can be placed at the edge of mmW beamforming sectors, such as mmW beamforming blind spots, by fiber-optic fronthaul. Moreover, the frequency used for O-RAN can be staggered with the frequency used for mmW beamforming to avoid signal interference. In addition, the O-RU antennas can be relayed by fiber-optic fronthaul up to 10 km to extend from urban areas to less populated suburban or rural areas instead of installing mmW beamforming in less populated areas, which is not cost-effective. According to the measurement results, with 200 m FSO backhaul and 10 km fiber fronthaul, the cellular network performance at 200 m from the O-RU antenna of O-RAN is comparable to that without a 200 m FSO link backhaul and 10 km fiber fronthaul. As a result, the adaptive and open architecture of O-RAN improves the cost-effectiveness of integration with FSOs and enables communication services to reach remote rural and suburban areas, as well as urban areas. In summary, the integration of FSOs with RL-based mmW beamforming and O-RAN communication configurations can bring scalability and cost-effectiveness to current and future communication infrastructures in urban, suburban, and rural areas.