The authors introduce hybrid free-space optical and RF wireless links as potential technology for designing next-generation broadband wireless networks.
There is a surge in demand for high-data-rate connectivity among users in metropolitan area networks. This high bandwidth demand is driven by the increasing commercial use of the Internet, private intranets, videoconferencing, and voice over IP. Providing high-speed connections to the fiber optics backbone (last mile problem) is the key challenge in realizing this objective. Wireless technologies appear to be far more attractive than wired technologies due to low deployment costs, fewer regulatory restrictions, and configurable network topology. However, classical radio frequency (RF) bandwidth is limited and cannot fully utilize the high bandwidth offered by the fiber optics backbone. Free space optical wireless (FSOW), on the other hand, is one promising line-of-sight, high-speed, secure wireless technology that can facilitate the realization of next-generation carrier-grade high-reliability backbone and last-mile network access in an inexpensive and timely manner. For instance, the broadband wireless backbone architecture, based on FSOW links, has recently been introduced in the literature [1] [2] [3]. The multigigabit-per-second data carrying capacity of FSOW systems render them the most economically viable solution for point-to-point and point-to-multipoint deployments. Rapid deployment is another important advantage of FSOW systems that facilitates the realization of high-speed mesh-like networks in urban and semi-urban environments. However, the question that arises next is whether the FSOW technology is mature enough to guarantee high-grade link and network availability figures. Unfortunately, there has been much practical evidence in the literature that the performance of FSOW links severely deteriorates under adverse weather conditions [4] [5] [6]. Despite the fact that excessive scattering due to dust particles, heavy rain, or snow may possibly degrade the quality of FSOW links, moderate to dense fog conditions turn out to be the most challenging factor. This is attributed to the fact that tiny fog particles scatter and distort the FSOW signal in conjunction with absorption of the radiated energy significantly, which causes atmospheric attenuation to reach high levels resulting in link failure. In some scenarios, the overall signal attenuation can be as high as 300 dB/km. Experimental results have recently shown that optical wireless links alone cannot achieve required availability figures over long distances and high data rates [4], [5]. Natural movement of buildings is another challenge. The FSOW beams tend to lose alignment due to the relative movement of buildings [7] and therefore may lead to link outages. This problem can be taken care of via:
- Slightly increasing the footprint of the coverage
- Employing auto-beamtracking capabilities in the system
In addition to adverse weather conditions, random air turbulence due to temperature variation in the atmospheric layers can also affect the performance in some cases, but not significantly.
Several methods have been introduced for improving the FSOW link availability. The following two technologies are among those proposed:
- Providing hybrid link protection [4] [5] [6] via an RF (micro-/millimeter-wave) link complementary to FSOW with respect to weather sensitivity
- Scaling down the distance between transmitter-receiver pairs via multihop routing [3], [5], [6]
Our main contribution in this article is to introduce novel architecture and networking algorithms that efficiently exploit the advantages of both technologies. To this end, we first introduce a link monitoring scheme that provides real-time characterization of the link performance under various weather conditions. We introduce a simple yet efficient algorithm for dynamically dcciding the link status based on measured bit error rate (BER) data. Next, we propose a dynamic load switching (DLS) algorithm that partitions the traffic between the two types of links depending on weather conditions, desired link quality, and transmission delay constraints. We present experimental and simulation results that emphasize the significant role DLS plays in utilizing the hybrid architecture and, hence, improving the quality of service (00s). Finally, we shed some light on the trade-off between multihop path availability and delay. This trade-off is attributed to the fact that going over a multihop path, consisting of short-length links, is expected to achieve higher availability than going from source to destination over a long single hop. On the other hand, sending packets over large number of hops increases the end-to-end delay, which in turn leads to QoS degradation for delaysensitive applications. In addition, we provide insights about the relationship between topology discovery and routing criteria on one hand and link availability on the other.
The article is organized as follows. We introduce the hybrid wireless architecture. We devote a section to proposing novel FSOW link mvnitoring and restoring techniques. Afterword, the hybrid wireless testbed and experimrntal results are presnted. Finally, conclustion are drawn.
SECTION II
HYBRID WIRELESS ARCHITECTURE
In view of the aforementioned challenges, the current trend is to adopt novel hybrid architectures that exploit the complementary nature of FSOW and RF links with respect to their individual weather sensitivities. As mentioned above, FSOW links arc highly susceptible to dense fog, mist, and dust particles, hut relatively less vulnerable to rain conditions. On the contrary, the performance of RF systems degrades significantly during rain events (especially at frequencies above 10 GHz), but they are less susceptible to dense fog particulates. The hybrid architecture combines both technologies to improve the vverall wireless link andlor network availability. Figures 1 and 2 depict several application scenarios for protected high-availability access and distribution networks based on the RFIFSOW hybrid concept. These scenarios include protection for single and multichannel multiservice links, broadband network extension, wavelength-division multiplexed (WDM) synchronous optical network (SONET) ring protection, and terrestrialiairhorncisatelite link protection architectures.
The FSOW and RF subsystems are placed parallel to each other for 1:l protection. Currcnt state-of-the-art FSOW systems achieve data rates on the order of multiple gigabits per second using single or multiple WDM channels. On the other hand, RF systems operating in the 5.4 GHz, 28 GHz, or 38 GHz licensed frequency bands can sustain up to OC-24 (1200 Mbls) data rates and 2.5 Chis with polarization diversity. During clear sky conditions, RF channels can be used to augment the overall data capacity of FSOW channels to take over a portion or all of the traffic when FSOW cxperiences total outage. System-relatcd concepts and implcmentation details are prcsented in later sections.
SECTION III
ADAPTIVE OPTICAL WIRELESS LINKS
In this section wc investigate candidate techniques for maintaining the quality of FSOW links at an acceptable level under severe weather conditions. The system must not only be able to adapt to changing weather conditions and maintain sustained cud-to-cnd connectivity, but also to meet the desired level of QoS.
SECTION IV
FSOW LINK MONITOR
Thc development of adaptive FSOW links requires implementation of a performance monitoring subsystem that can probe and accurately characterize the performance of the wireless channel in real timc. Essentially, the data recorded by the monitoring subsystem plays a crucial role in governing the dynamic link restoration techniques and, consequently, the link performance metrics. In this section we discuss challenges and proposed techniques to characterize the dynamic behavior of FSOW links in response to time-valying weather conditions. To this end, we go through two phases. First, we introduce candidate metrics for judging whether the performance of an FSOW link is satisfactory or not. Second, we introduce a procedure for mapping the gathered measurements to QoS paramcters.
Performance Mefricr
The performance of wireless links is normally characterized by the signal detection threshold which is defined as a lowcr bound on the signal-lo-noise-ratio (SNR) (or equivalently an upper bound on the BER) at the rcceiver [8]. Therefore, the BER threshold is normally adopted as the QoS parameter for wireless links. In this work, the real-time BER parameter is used to characterize the FSOW channel performance and initiate link restoration techniques whenever adverse weather conditions arise. However, although instantaneous (real-time) BER measurements may be good indicators of the performance of the wireless channel at a particular instant of time, they may not accurately reflect the cause of link deterioration. For instance, link deterioration due to temporary line-of-sight obstructions, which are unavoidable, may erroneously activate expensive link restoration techniques. Thus, the software supporting the adaptive feature of FSOW links must effectively prevent such momentary events and small-scale weather fluctuations from initiating link restoration. This challenge is discussed in more detail in the next section. Furthermore, the FSOW system may utilize multiple environmental and system parameters to define a comprehensive and effective performance metric. For example, received signal strength (RSS), humidity, and temperature, can be recorded and used individually or cooperatively in the decision making process. It is shown later that, unlike optical fibers, the RSS performance parameter alone is not an accurate indicator of FSOW link quality. Average BER turns out to be more accurate in determining FSOW link status.
Sliding Window Averaging
The objective of the proposed algorithm is to eliminate the impact of temporary line-of-site obstructions, since they may unnecessarily activate expensive link restoration algorithms. Accordingly, we propose to smooth out the sudden fluctuations in the measured BER data via a sliding window averaging mechanism.
Given the instantaneous BER measurements of an FSOW link recorded every minute over a period of several hours, we noticed that the measured BER values may oscillate between low and highly unacceptable values over a short period of time. The objective of the proposed algorithm is to eliminate the impact of temporary line-of-site obstructions, since they may unnecessarily activate expensive link restoration algorithms. Accordingly, we propose to smooth out the sudden fluctuations in the measured BER data via a sliding window averaging mechanism. Two parameters need to be specified for this algorithm: window.shape and window length (W) selected according to application required QoS. For simplicity, we employ a rectangular window where all samples are given equal weight in the averaging operation. On the other hand, we examine a wide range of window lengths, W = 5–100 min. It is worth noting that the optimum window length, which strikes a balance between filtering out a high rate of temporary link state changes and maintaining necessary information, depends solely on the rate of weather changes and intermittent BER variations. According to the proposed scheme, the system monitors the BER every single minute. However, the link restoration decisions are based on the averaged BER and recur only once in an interval greater than or equal to 
$W$ min.
From Fig. 3, it is straightforward to notice that short window lengths (
$W<5$ min) suffer the problem of high-rate BER changes caused by temporary sudden fluctuations. On the other hand, large window lengths (
$W=100$ min) filter out important information and lead to inaccurate representation of the link status (link outage intervals become much longer than reality). Finally, intermediate window lengths (
$W=5-20$ min) provide the best balance between filtering out high rates of sudden state changes and preserving necessary information.
SECTION V
FSOW LINKRESTORATION
In this section we investigate candidate techniques for FSOW link restoration. To this end, we go through two phases. First, we examine the known techniques of dynamic power and rate control. The transmission power can be increased to mitigate atmospheric attenuation during high attenuation conditions. This can be implemented such that the transmission power increases incrementally in moderate steps to ensure acceptable link quality and energy efficiency. However, boosting the transmitted power is constrained by maximum power, eye safety standards, and health considerations. An alternative approach is to dynamically adapt the data rate of the FSOW link in response to weather conditions. During adverse conditions, the data rates can be reduced in predefined steps to improve/maintain the link QoS. For example, FSOW data rates can be reduced from an optical channel rate of OC-48 to OC-24 and further down to OC-12 with successive increase in the permissible atmospheric attenuation levels, as shown by the FSOW link budget equation in [4], [6]. Second, we examine two novel techniques for FSOW availability enhancement at the link and network levels: dynamic load switching and multihop routing.
Dynamic Load Switching
The hybrid architecture must be designed such that in the event of adversities, it automatically regulates all RF links into a standby mode followed by automatic switching of data traffic from FSOW to RF as soon as the FSOW link experiences outage. This can easily be achieved using requisite hardware and appropriate software. The system hardware can be configured to monitor the performance of the FSOW link in real time. On the other hand, the system software can be customized to map the measurements to QoS parameters such as average BER or average RSS, as described in the previous section. These metrics can then be used to make necessary link restoration decisions depending on the desired QoS.
In this section we present a novel algorithm for utilizing “all-weather” hybrid RF/FSOW wireless links. It partitions the load in the form of incremental load shifting from one link to the other as atmospheric attenuation becomes significant on one of them. The size of these increments certainly affects link utilization and availability; finer increments yield more efficient link utilization and higher availability at the expense of circuit complexity. On the other hand, larger increments simplify circuit design and operation at the expense of poor link utilization. Therefore, the increment size should be chosen carefully in order to balance the aforementioned trade-off. In the simulation study discussed later, we chose the increment size to be 
$R/4$, where 
$R$ is the maximum FSOW link bit rate. Finally, it is worth noting that the DLS algorithm is expected to be activated periodically at a rate that can be adapted depending on the rate of weather changes. It should not be activated too frequently to avoid processing overhead, and at the same time we cannot tolerate link outage due to long periods between decision epochs.
It is worth noting that the DLS algorithm is expected to be activated periodically at a rate that can be adapted depending on the rate of weather changes. It should not be activated too frequently to avoid processing overhead, and at the same time we cannot tolerate link outage due to the long periods between decision epochs.
In Fig. 4, we illustrate the operation of the proposed algorithm. The FSOW link is configured as the primary link and the RF link(s) as a backup that should take over part (or all) of the load whenever necessary. Furthermore, we assume that for each FSOW link there is one (or a set) of RF backup link(s) able to accommodate the maximum transmission rate 
$R$ (i.e., full load) carried by the FSOW link if needed. For instance, for the OC-12 FSOW link supporting 
$R =622$ Mb/s rate, we would need four RF backup links each operating at 
$R/4=155$ Mb/s. Throughout this article, we refer to the set of RF backup links as one RF link in order to simplify notations. Thus, we assume that initially all traffic is transmitted over the FSOW link (i.e. the FSOW bit rate is 
$R_{\rm FSOW}=R$ and the RF bit rate 
$R_{\rm RF}=0$. As illustrated in the previous section, the system records BER on both links every minute. To eliminate unnecessary link state alterations, the DLS algorithm averages the actual atmospheric attenuation over a rectangular window of length 
$W$. Afterward, the algorithm computes the permissible atmospheric attenuation on both links, as illustrated in [5], [6]. The last step is to compare actual to permissible atmospheric attenuation for each link in order to identify the dynamic status of both links and partition the traffic accordingly. Load switching can be realized using a switch controlled by a feedback mechanism discussed in [2].
Multihop Routing
In this section we extend the single link availability study, discussed in the previous section, to multihop networks. More specifically, we focus on improving the availability of FSOW networks connecting several establishments in an urban environment. A mesh-like design that includes long links with RF protection [2] provides the desired redundancy and allows multihop routing over short-length links via the following techniques:
- Reducing the distance between the transmitter and the receiver
- Modifying the routing criteria in order to take link availability into consideration
It is evident that going from source to destination over multihop paths consisting of short-length links would achieve higher end-to-end availability than going from source to destination over a long single hop. This is attributed to the fact that shorter FSOW links are less prone to outage due to hefty built-in margins and thus can continue to function satisfactorily even when longer FSOW links experience outage. However, sending packets over a large number of hops would increase the end-to-end delay, which in turn leads to QoS degradation for real-time applications. Therefore, there is a fundamental trade-off between end-to-end path availability and end-to-end path delay. Our main objective in this section is to shed some light on this tradeoff and point out guidelines for striking a balance between availability and delays depending on the traffic type and QoS constraints.
Shortest path (SP) routing is widely used in wired as well as multihop wireless networks [9]. It relays packets through the shortest path, where the link metric could be physical length, transmission delay, or load, depending on the QoS parameter of interest. For optical wireless networks, we introduce a new criterion for routing information: link availability. More specifically, we focus on how to make packet routing decisions under severe weather conditions where many links suffer outage. In addition, the routing algorithm should be designed to cope with temporary line-of-site obstructions causing intermittent failure of the optical wireless links. Thus, the impact of FSOW link availability on the following aspects of the routing problem are investigated.
Topology Discovery
The objective of this phase is to define, under adverse weather conditions, the network connectivity (topology). This depends solely on the status of various links and the transmission parameters associated with active links (bit rate, power, etc.). To this end, the topology discovery algorithm goes through three phases in an attempt to restore broken links. First, it boosts the transmitted power to the maximum value that does not violate eye safety standards or cause potential health risks. If this fails to restore the link, the algorithm attempts to reduce the link bit rate since losing a connection completely is certainly much worse than communicating at a lower speed for a short period of time. As the last resort, the DLS algorithm is activated to restore the disrupted link. The output of this phase includes the following information:
- State of each link
- Link transmission parameters (bit rate and power)
- Atmospheric attenuation over each link
This information is to be fed to the routing algorithm in order to make the best decision depending on the QoS requirements and weather conditions.
Routing Criteria
Given the set of available links along with their bit rates and link qualities (measured by the atmospheric attenuation), we attempt to answer the question of how to route a packet from source to destination. As pointed out earlier, the speed of wireless links can be traded for their quality; that is, the bit rate of an optical wireless link can be reduced in order to increase permissible atmospheric attenuation and hence improve its quality [2]. Under the proposed scheme, the network is expected to consist of a number of heterogeneous links ranging from high-speed vulnerable links to low-speed reliable links. A major challenge is how to route information packets efficiently in this heterogeneous environment. For example, loss-tolerant real-time traffic should be transferred over high-speed paths, while delay-tolerant data traffic should be transferred over reliable paths. Accordingly, the proposed routing algorithm plugs the link quality criterion, along with the classical link delay and link load criteria, into the link metric depending on the QoS requirements. More specifically, for real-time traffic more weight should be given to minimizing link delays, while for data traffic more weight should be given to minimizing link attenuation. In addition, the interaction between the DLS decision and the routing decision should be examined. For instance, link failures might be tolerated if there is an alternate route, via other available links, that leads from source to destination. More precisely, the problem of path availability could be solved via either DLS, alternate path routing, or both techniques.
SECTION VI
PERFORMANCE EVALUATION AND DISCUSSION
In this section we present performance evaluation results of an experimental hybrid testbed. In addition, we discuss simulation results of the DLS algorithm.
SECTION VII
FIELD EXPERIMENTAL SETUP
Figure 5 shows a schematic diagram of the field experimental testbed shown in Fig. 1b. A general-purpose computer is used to control the wireless channel monitoring subsystem, traffic switch, BER tester, and link data rate controller. The FSOW transceivers are capable of carrying up to OC-12 data rates (622 Mb/s). The multichannel RF transceivers operate at 38 GHz and are capable of carrying OC-3 data rates on each channel. During clear sky conditions, the FSOW and RF subsystems can provide BER performance on the order of 10−13 and 10−12, respectively, for the ~500 m link length. A pair of optical fibers connects the FSOW transceivers to the BER test (BERT) control panel in the laboratory network center.
As part of the hybrid testbed, two wireless channel monitoring subsystems are developed and compared:
- Real-time RSS
- Real-time BER
The RSS-based system involves monitoring and recording real-time RSS values at the receiver end of the system. The experimental results from the RSS- and BER-based monitoring subsystems are compared to ascertain their credibility in judging the time-varying channel conditions. Figures 6a and b present typical BER/RSS measured performance results for comparison.
The instantaneous RSS and BER values were recorded over 24- and 12-hour periods. From Fig. 6a, it is easily noticed that there is a strong correlation between the real-time recorded values of RSS and BER. During the first six hours, a relatively low RSS level translates very well to the corresponding high BER values. In the next four hours, as the RSS tends to improve slightly, a corresponding slight improvement in the BER values can also be noticed. Finally, during the last two hours (dense fog event), the correlation between extremely low RSS and high BER values can be noticed again. On the other hand, Fig. 6b shows a distinct lack of correlation between the RSS and BER values due to fog scattering and other atmospheric effects. During the first few hours, the RSS level is fairly constant or fluctuates in a very narrow range. Based on the observations from Fig. 6a, one would expect the BER statistics to show a similar trend (i.e., stay fairly constant or fluctuate slightly around the mean value). On the contrary, for the first hour the BER remains high and then abruptly drops to zero (error-free performance). The error-free performance continues for a few hours, during which the RSS values increase significantly. Again, one would expect continued error-free performance since higher power is being received at the FSOW receiver [2]. However, the BER fluctuates significantly despite the consistently high RSS levels during that interval. Thus, we conclude that the correlation between real-time recorded RSS and BER values may vary over time for FSOW links. RSS-based monitoring systems are more favorable due to ease of implementation and integration into the hybrid architecture than other possible schemes. However, unlike optical fiber transmission, RSS is not an accurate indicator of FSOW link performance, and hence cannot be considered a reliable criterion for activating link restoration schemes.
SECTION VIII
EXPERIMENTAL AND SIMULATION RESULTS
The lack of consistency between real-time RSS values and FSOW system performance necessitated the development of a BER-based monitoring subsystem. The real-time BER statistics measured by the BERT data controller were recorded and processed by a computer running a comprehensive connectivity management and link restoration software algorithm. A rectangular window size of 5 min was chosen for the calculation of the desired performance metric.
The dynamic rate control scheme for link restoration has been examined at a fixed BER threshold level of 10−7 [2]. However, the FSOW link performance deteriorates significantly during moderate to heavy fog and mist conditions. Despite the fact that power and rate control can restore the FSOW link and mitigate moderate performance deteriorations, the traffic must be dynamically switched from FSOW to RF during extreme weather cases. An automatic feedback-controlled optical switch is used to carry out the traffic partitioning task. The switching decision is based on the average BER calculated from realtime BER values recorded by the link monitoring subsystem. During this experiment, the QoS threshold was preset to 10−5 under the assumption that channel coding can easily improve the QoS performance to less than 10−9. The experimental behavior of the hybrid architecture is shown in Fig. 7.
It is evident from the figure that the performance of FSOW link starts deteriorating due to the gradual buildup of fog. The traffic continues to flow through the FSOW link despite performance degradation until the threshold of 10−5 is breached. As soon as the predefined QoS threshold level is reached, the data traffic is promptly switched over to the RF link (complete switchover). The control signal and corresponding event A indicate this transition. Error-free data transmission during event A in the figure clearly shows that RF link is not affected by fog. Furthermore, note that as fog fades away, the performance of the FSOW link improves and the link is automatically restored. During this event (between A and B), although the RF link has better QoS, the data traffic is switched back to FSOW to facilitate the possibility of data transmission at higher rates and capacity. During event B, when dense fog rolls in again, the data traffic is again routed through the RF for a few minutes before automatic switch back to FSOW. Thus, Fig. 7 successfully demonstrates the effectiveness of load switching on top of the hybrid architecture. Next, we demonstrate simulation results that reflect the behavior of incremental load switching. Figure 8 shows the dynamic behavior of the actual and permissible atmospheric attenuation measured over time. The permissible atmospheric attenuation of FSOW at maximum bit rate (622 Mb/s) is given by 43.5 dB for the simulated link parameters [6]. The role of DLS appears at the 13th hour, where the FSOW experiences outage. Thus, the transmission rate over the FSOW link is reduced to 3R/4 via switching a quarter of the load to the RF link. This in turn increases the permissible attenuation of the FSOW to 44.749 dB and hence restores it. When weather conditions improve and the attenuation over the FSOW link shrinks, the full load is switched back to the FSOW in order to resume operation at maximum rate. Finally, it is worth noting that the advantages of partial load switching compared to complete switchover prevail under practical scenarios where the FSOW bit rate is greater than the RF bit rate. This is attributed to the fact that when the FSOW link fails, partial load switching utilizes both links in order to support the FSOW bit rate. On the other hand, complete switchover would utilize only the RF link under a reduced bit rate.
Propagation measurements to evaluate the performance of FSOW, alone and as a part of the hybrid architecture, have been consistently performed at our research facility. Table 1 presents some of the consolidated performance statistics recorded over a period of four months.
The data in the available hours column represent the number of hours during which the FSOW link operated within the desired QoS requirements irrespective of prevailing weather conditions. The outage hours statistics indicate the number of hours for which the traffic was routed through the RF link when the performance of the FSOW link deteriorated below acceptable QoS levels (BER < 10–6). It can be clearly seen that the hybrid architecture significantly improves the overall link availability statistics from 68.4, 88.2, 96.5, and 89.1 percent to as high as 99.998 percent.
In this article we present design challenges and potential solutions for carrier-grade availability over FSOW links during adverse weather conditions. We demonstrate a link monitoring scheme that accurately reflects the performance of optical wireless links under various weather conditions. We noticed that, unlike optical fibers, RSS alone does not provide accurate characterization of FSOW link performance. Therefore, we relied on both the RSSI and average BER measurements as the criteria for link adaptation. In addition, we introduced two novel link restoration schemes that efficiently utilize the hybrid architecture:
- Dynamic load switching
- Multihop routing
The extensive long term results, recorded from the experiments during extreme weather conditions, validate the hybrid architecture concept and prove the significant DLS role in achieving sustained high-speed connectivity.
We notice that DLS improves the FSOW link availability considerably. Finally, we present an elaborate field testbed based on the hybrid architecture and various link adaptation techniques. Extensive long-term results, recorded from the experiments during extreme weather conditions, validate the hybrid architecture concept and prove the significant role of DLS in achieving sustained high-speed connectivity.
Acknowledgment
The authors would like to acknowledge the partial support of the DARPA NGI Program for this project.
