High Reliability and Availability in Radio over Fiber Networks

Transcription

1 High Reliability and Availability in Radio over Fiber Networks Gerardo Castañón 1,, Gabriel Campuzano 1, and Ozan Tonguz 2, 1 Center of Electronics and Telecommunications, Tecnológico de Monterrey Eugenio Garza Sada 2501 Sur, Monterrey, N.L., 64849, México gerardo.castanon@itesm.mx. 2 Department of Electrical and Computer Engineering Carnegie Mellon University Pittsburgh, PA , USA tonguz@ece.cmu.edu Corresponding author: gerardo.castanon@itesm.mx. The emerging vehicular applications such as active safety, infotainment, interactive games, and Internet access might dictate very high reliability and availability in future radio over fiber networks. This paper provides guidelines for network architecture and topology selection in future radio over fiber networks in terms of network availability, reliability, and fiber length. We present a comparative analysis of network topologies for Radio over Fiber Networks. In particular, we present results that compare different toplogies such as star, ring, bus, multilevel stars, multilevel rings, multilevel buses, hybrid multilevel star-rings, hybrid multilevel ring-stars, and hybrid multilevel ring-buses in terms of availability, reliability, and fiber length requirements of networks. Protection is also considered and results on the protected versions of the aforementioned topologies are also presented. c 2008 Optical Society of America OCIS codes: ,

2 1. Introduction Increasing interest in road accident prevention, radar vehicular technology, and traffic congestion in intelligent transportation systems has triggered research activities in the information and communication technology sector to develop secure and reliable network infrastructure for such applications [1], [2], [3]. As shown in Figure 1, fiber network topologies investigated in this paper can be used for Radio over Fiber (RoF) services as well as for fiber to the home and fiber to the building services. Due to the vast bandwidth of the fiber, both wireless and wired services in a single fiber access network can co-exist. However, this will require network planning to distribute the bandwidth for both services according to the traffic and demand requirements. Fig. 1. The envisioned radio over fiber network in urban areas. Radio over fiber for vehicular, mobile, broadband wireless computer communications and also network infrastructure can be used for fiber to the building and fiber to the home services. Without proper security and reliability mechanisms, networks will be confined to limited, controlled environments, not fulfilling much of the promise they hold. The limited ability of Vehicular/Mobile/Wireless/Computer networks to thwart failures or attacks makes ensuring 2

3 network availability more important and difficult. Consideration of security, reliability, and availability at the design stage is the best way to ensure successful network deployment. Recently, considerable attention has been paid to the merging of Radio over Fiber technologies with mm-wave band signal distribution [4], [5], [6], [7]. This communication system has great potential to support secure, cost-effective, and high capacity vehicular/mobile/wireless access. For the future provisioning of broadband, interactive and multimedia services over wireless media, current trends in cellular networks are: 1) reduction of the cell size to accommodate more users; and 2) to operate in the microwave/millimeter wave (mm-wave GHz) frequency bands to avoid spectral congestion in lower frequency bands (2.4 or 5 GHz). The larger radio frequency (RF) propagation losses at mm-wave bands reduce the cell size covered by a single base station (BS) and allow an increased frequency reuse factor to improve the spectrum utilization efficiency. On the other hand, this type of network demands a large number of BSs to cover a service area, and cost-effective BSs are key to success in the market place. This requirement has led to the development of system architectures where functions such as signal routing/processing, handover and frequency allocation are carried out at a central control station (CS), rather than at the BS. Furthermore, such a centralized configuration allows sensitive equipment to be located in a safer environment and enables the cost of expensive components to be shared among several BSs. An attractive alternative for linking a CS with several BSs in such a radio network is via an optical fiber network, since fiber has low loss, is immune to Electro-Magnetic Interference, and has broad bandwidth [8], [9]. The transmission of radio signals over fiber, with simple optical-to-electrical conversion, followed by radiation at remote antennas, which are connected to a central CS, has been proposed as a method for minimizing costs. This cost reduction can be brought about due to the following three advantages. First, the remote antenna BS or radio distribution point needs to perform only simple functions, and it is small in size and low in cost. Second, these simple antennas can be installed on the traffic lights or on electric poles avoiding the expensive right of way costs of private owners of buildings where today s antennas need to be installed. Third, the resources provided by the CS can be shared among many antenna BSs. Furthermore, the small cells allow for the use of low RF power at the BSs resulting in increased battery lifetime of mobile/wireless terminals. The distribution of radio signals to and from BSs can be either mm-wave modulated optical signals (RF-over-fiber) [10], [11] or lower frequency subcarriers (IF-over-fiber) [12], [13], [14]. These research activities, fueled by rapid developments in both photonic and mmwave technologies suggesting simple BSs based on Radio over Fiber (RoF) technologies, will be available in the near future. In this paper, we investigate the potential usefulness of multilevel networks from the viewpoint of Radio over Fiber networks. We consider several topology options such as star, ring, bus, multilevel stars, multilevel rings, multilevel buses, 3

4 (a) (b) (c) (d) (e) (f) (g) (h) (i) Fig. 2. Topologies a) star, b) ring, c) bus, d) multilevel stars (stars), e) multilevel rings (rings), f) multilevel buses (buses) g) multilevel star-rings, h) multilevel ring-stars, and i) multilevel ring-buses. 4

5 hybrid multilevel star-rings, hybrid multilevel ring-stars, and hybrid multilevel ring-buses. Also, we present results of the protected versions of the aforementioned topologies. We compare topologies in terms of link distance, reliability, and availability of the networks. The remainder of the paper is organized as follows. Section II presents the considered network topologies. Section III details the reliability calculations of the paper. Section IV outlines the methodology used for availability calculations. Section V presents reliability, availability and link distance results and a discussion. Section VI contains the conclusions. 2. Network topologies Figure 2, shows topologies with 25 nodes where Fig. 2a is a star, Fig. 2b is a ring, Fig. 2c is a bus Fig. 2d multilevel stars, Fig. 2e is a multilevel rings, Fig. 2f is a multilevel buses, Fig. 2g is a multilevel star-rings, Fig. 2g is a multilevel ring-stars, and Fig. 2f is a multilevel ring-buses topology. We will consider inexpensive passive network components in order to demultiplex and split the WDM signals. We will assume the use of sub-carriers multiplexed per wavelength. Depending on the number of users per cell (BS) and the required bandwidth, the system is flexible enough to dedicate several sub-carriers in a cell or to dedicate several wavelengths per cell. However, in order to make a fair topology comparison we assume that there is one dedicated wavelength from the CS to any BS. For example in Fig. 2d, the branch starting at node 12 (that connects nodes 15, 16, 20, 21) will need four wavelengths, one per node. The link between nodes 12 and 16 will carry four wavelengths. At node 16 we will need to drop the desired wavelength(s) and send the remaining wavelengths to the other nodes. Therefore the optical add drop multiplexor (OADM) at node 16 should be capable of detecting the desired wavelength(s) using a tunable optical filter to select the desired wavelength or wavelengths in case more users demand service from that BS. In case several wavelengths are dropped, the OADM should be equipped with the respective fixed optical filter and receiver. There are several possible combinations about designing the OADM which are beyond the scope of this paper. If we need more re-configurability for the OADM (for example, the capability of detecting a variable number of wavelengths) the number of components and hence the cost of the OADM will increase [15]. The separation between BSs used in this paper is 120 meters considering that the typical average length of a street block in large cities in the USA is 100 meters. We will assume that each base station is at the corner of every street intersection in a rectangular grid of streets. The reliability and availability calculations consider the reference distance of 120 meters between base stations. Now, the number of wavelengths in a fiber link depends on the number of BSs that a given link is interconnecting to the central station. Assuming each wavelength can serve about 50 wireless users per BS and assuming we are dedicating one wavelength per BS, for the case of the star topology in Fig. 2a every link will carry just one 5

6 wavelength. This is not the case for the ring topology, in Fig. 2b, where every link is carrying 23 wavelengths. The number of wavelengths at every link is topology dependent and traffic dependent and requires a wavelength assignment mechanism which is beyond the scope of this paper. In this paper we are limiting the reliability and availability analysis to the case of considering the probability that the central node communicates with the other nodes of the network. 3. Reliability Reliability is a key parameter in the design of highly reliable radio over fiber networks. Reliability indicates the quality of the network to transport traffic. Thus, reliability is a probability measure and is defined as the probability that there exists one operative path between a given pair of communicating entities. Network reliability is a classical problem and has been studied extensively by many researchers in the past [16], [17]. In this paper, reliability is computed using the algorithm SYREL [18], [19] which is accurate and has a low computational complexity. SYREL methodology uses minimum hop routing in order to determine the reliability of a network. A wavelength assignment mechanism is not necessary in order to perform an analysis of this type. However, wavelength assignment is an important practical issue in multilevel networks and efficient algorithms that minimize the number of wavelengths and the number of hops to reach destination have been proposed by Nagatsu et al. in [20] and [21]. The results presented in section 5 consider the probability that the central node communicates with the other nodes of the network. Considering Figure 2, the central station is located at node number 12. Reliability is computed in this way because it is assumed that the central station is communicating with all the antennas and all antennas with each other and also the public switched telephone network (PSTN) through the central station. One important feature about the analysis presented in this paper is that it considers the probability of link failure as link length dependent. For example, if the probability a link of length 120 meters does not fail is ρ, then the probability that a link of length 900 meters does not fail is ρ (900/120). This makes the reliability analysis of the topologies length dependent, link failure parameter dependent, and number of paths (source-destination) dependent, which emulates what happens in reality in an effective manner. 4. Availability The availability of service is also an important issue in today s networks. In case of a link or node s component failure, several communication connections will be affected. The outage of the link is not the only conceivable scenario. Each hardware component involved in the transmission path may fail with a certain probability. The transmitted information involve 6

7 a great number of hardware components in a network such as transceivers, multiplexers, switches, amplifiers, which inevitably contribute to the overall availability. In order to guaranty a desired level of communication service availability, the network must be analyzed in an integrated manner. Calculation of availability in this paper is done through the availability maps that are a function of the connection length and of the number of traversed nodes. They provide results to visualize the combined influence of node and link failures on the connection availability in a network. The connection availability is calculated by the following equation [22], [23]: A c = 1 (AP link km D + AP drop + AP add +AP pass through (H 1)(1 P reg ) + AP reg (H 1)P reg ) where AP link km is the link availability penalty per kilometer, D is the connection distance in kilometers, H is the number of hops traversed by the connection, AP add, AP drop, AP pass through, AP reg are the availability penalties due to adding, dropping, passing through, and regeneration node operations, respectively. P reg is the fraction of nodes in which the connection is regenerated. The functions x and x are the ceiling and floor functions, respectively. In general, the availability penalty imposed by a system can be derived from its failure rate (FR) in FIT (1 FIT = failure rate of 1 failure in 10 9 hrs) by the following equation: (1) AP = 10 9 FR MTTR (2) where MTTR is the mean time to repair (in hours). This paper assumes a link failure rate of 310 FIT/km [22]. Here we do not consider Raman-pumped amplifiers and we only consider optical Erbium doped fiber amplifiers (EDFAs). The link MTTR was assumed to be 12 hrs and the node components MTTR is assumed to be 6 hrs [22]. These values may vary from operator to operator; nevertheless, this assumption does not have a strong impact on the quantitative availability results. The architecture of a node is shown in Fig. 3. It is based on passive components splitters/couplers, filters, wavelength blockers, and optical amplifiers to compensate for loss when it is required. Node operations are also shown in Fig. 3. The signal can pass through the node, can be dropped and a new signal can be added. The block diagram used for calculating the availability penalty of each node operation is shown in Fig. 4 which also shows the components involved in case the signal needs to be regenerated. Table 1 contains some common component failure rates used in the design of optical networks [23]. The needed values are substituted in equation 2 to calculate node operations according to Fig. 4 and the penalties calculated are substituted into equation 1 to compute the connection availability. 7

8 Fig. 3. OADM architecture. Node operations are: passing through (solid line), dropping (dashed line), and adding (dashed-dotted line). Fig. 4. OADM block diagrams of the availability penalties. 8

9 Table 1. COMPONENT FAILURE RATES. Component Symbol Failure Rate (FIT) Number of wavelengths W Number of inputs N MUX/DEMUX MUX 25 * W EDFA EDFA 2850 Optical Switch 1 OSW1 21 * W * W/4 Optical Switch 2 OSW2 21 * 2 * 2N Coupler 1 COUP1 50 Coupler 2 COUP2 25* W/4 Coupler 3 COUP3 25*(N-1) Tunable Transmitter TTx 745 Fix Transmitter FTx 186 Tunable Receiver TRx 470 Fix Receiver FRx 70 Digital Switch 1 DSW1 875* W Digital Switch 2 DSW2 875* W *N Wavelength Blocker WB 50 * W 5. Results 5.A. Reliability results Figure 5, shows results of the network reliability for cases with 25 nodes (a), 49 nodes (b), 121 nodes (c), and 441 nodes (d). Figure 6 shows results for the case when 1+1 protection is used in the topologies with 25 nodes (a), 49 nodes (b), 121 nodes (c), and 441 nodes (d), respectively. We are assuming a 1+1 protection scheme because it is the scheme that increases network availability. Under 1+1, a copy of data signal is transmitted simultaneously on a working and a protection channel. At the receiver side, the receiver can make a decision to accept which copy of signal based on the signal quality. In contrast to 1+1, 1:1 sends a copy of signal on a working channel only, while the protection channel is reserved for future use in case the working channel fails. Only when a failure occurs on the working channel, a preempt operation is carried out to take over the protection channel to carry the traffic from the failed working channel. In general, 1:1 has better efficiency in the protection capacity usage; however, it needs an additional action to switch over the traffic from a working channel to a protection channel, thereby affecting restoration speed and therefore the availability of the network. For the 1+1 protection case we assumed redundant disjoint fiber links are installed. The link disjoint path protection is the one where the working link and the protection link are physically separated from each other. This is a protection usually used in long distance terrestrial networks where high reliability and availability is required. The physical separation of the working path and protection path may be implemented by installing them in separate ducts, if the fiber is installed underground or installing the fiber 9

10 in separate electric poles if the fiber is installed over ground. The impact of considering this in the reliability and availability analysis is that the link failure of the working path is an independent variable with respect to the link failure variable of the protection path Reliability 0.7 Reliability Probability link is connected (a) Probability link is connected (c) Reliability Probability link is connected (b) Following symbols also apply to Figs. (a), (b), and (c ). Star Ring Bus Stars Rings Buses Star-rings Ring-stars Ring-buses Probability link is connected (d) Fig. 5. Reliability against probability link is connected for star, ring, bus, multilevel-stars, multilevel-rings, multilevel-buses, star-rings, ring-stars, and ring-buses topologies considering: (a) 25 nodes, (b) 49 nodes, (c) 121 nodes, and (d) 441 nodes. To compute the results in Figures 5 and 6 we assumed link failure only. The probability a link of length L (meters) is working, i.e. the probability a link of length L remains without failure ρ, has a range from 0.99 to It is important to note that here denotes the basic length unit. In this paper, we consider the separation L between BSs equals to 120 m, assuming that 100 m is the average length of a typical block in a USA city as shown in Figure 1. Therefore, the probability a link of length 900 m is connected is equal to ρ (900m/L) and the probability of link failure is thus (1 ρ (900m/L) ). 10

11 Probability link is connected 1 (a) Reliability Probability link is connected (c) Reliability 0.93 Probabilty link is connected Star Ring Bus Stars Rings Buses Star-rings Ring-stars Ring-buses (b) Symbols also apply to Figs. (a), (b), and (c ) Probabilty link is connected (d) Fig protection reliability results against probability link is connected for star, ring, bus, multilevel-stars, multilevel-rings, multilevel-buses, starrings, ring-stars, and ring-buses topologies considering: (a) 25 nodes, (b) 49 nodes, (c) 121 nodes, and (d) 441 nodes. 11

12 The reliability parameter of Figures 5 and 6 is the probability that the central node communicates with the other nodes of the network. Recall that in Figure 2, the central node is node number 12. Reliability is computed in this way because it is assumed that the central node is the central station communicating with all the antennas and all the antennas with each other and also the public switched telephone network (PSTN) through the central station. Reliability of the wireless part (antenna-wireless entity and the other way around) is not considered in the analysis. It is important to note that the multilevel-rings topology referred to in Figures 5 and 6 as rings provides the best reliability performance for networks with 25, 49, 121, and 441 nodes without protection. When 1+1 protection is used the topology that yields the best reliability is also the multilevel-rings topology and the second best is the multilevel ring-stars topology. The reason for this is the fact that the number of hops and link length from the central node to all other nodes in the network increases with the number of nodes in the network, which is the case for all other topologies except the multilevel-rings topology. This increase has an impact on the number of links the signal has to pass through. Considering that the probability a link is connected depends on the link length, the reliability has a tendency to decrease when the average number of hops in the network increases and also when the distance from the BS to the nodes increases, which happens in all other topologies except in the multilevel-rings topology (rings). Figure 7 shows fiber distance of topologies similar to the topologies shown in Figure 2. Figure 7 shows results for networks without protection with 25, 49, 121, 441, and 1089 nodes. If 1+1 protection is considered, the amount of fiber length will double. It is important to note that results in Fig. 7 show the total amount of fiber length in the network and not the maximum distance from the CS (central node) to the farthest BS. The networks were designed using the same logic of interconnection of the nodes as in Fig. 2. Note that the star topology needs more link length than the other topologies and results for the 1089 nodes network are not included in the figure due to the fact that that will require a huge fiber length. Also note that the topology with the lowest link length values is the ring topology; however, its reliability performance is the second best for a network with 25 nodes and becomes worse when the number of nodes increases, as shown in Figures 5 (b-c). The ring-stars topology provides a higher reliability performance than the other topologies except for the multilevel-rings and also the fiber distance is comparable to the multilevel-rings topology when a large number of nodes were considered. These features make the protected ring-stars topology the second best candidate for radio over fiber networks where thousands of antennas need to be connected by a central station. Also, this ring-stars topology is interesting from the point of view that it is relatively easy to install and may grow easily when a new BS node is required in a certain location. 12

13 Fig. 7. Link distance in kilometers against topologies for networks without protection. 5.B. Reliability results of special ring topologies It is also important to analyze the ring topologies with special features and to understand how those features affect the reliability performance. The special features analyzed in this paper are the number of nodes per ring and the number of paths between a given sourcedestination pair. Figure 8 shows ring topologies with different features. Figure 8(a) shows rings with four nodes for the outer rings and no shared nodes. Note that in this topology the number of paths to reach some nodes of the outer rings is four. Also, note that the number of paths to reach any of the nodes of the rings that interconnect the outer nodes with the central BS is two. Figure 8 (b) is similar to Fig. 8(a) with six nodes instead of four. Figure 8 (c) shows a topology with six nodes and one node shared between the adjacent outer rings. Figure 8 (d) shows a topology with six nodes where there are two nodes that are shared between pairs of outer adjacent rings. The sharing of nodes between rings brings the benefit of increasing the number of paths between a source-destination pair considerably. The reason to compare the topologies of Figure 8 is to optimize the number of nodes per ring, fiber length, and the number of nodes shared per ring, since these parameters affect the reliability performance and also have an impact on the fiber length used in the network. Because of this, it may happen that the multilevel-rings network with few nodes per ring may yield a high reliability at the expense of increasing the total network fiber length, as we 13

14 (a) (b) (c) (d) Fig. 8. Topologies a) rings with 4 nodes, b) rings with 6 nodes, c) rings with 6 nodes and one node shared, d) rings with 6 nodes and 2 nodes shared between adjacent rings. 14

15 will show below. Another important question regarding the topologies shown in Figure 8 is how the number of paths between a source-destination pair affects the reliability parameter performance and how many paths are enough to obtain a cost effective reliability. Increasing the number of shared nodes also increases the number of paths between source-destination pairs in the network. Fig. 9. Reliability results of special topologies of rings with 4 and 6 nodes and 0, 1, and 2 shared nodes. One way to improve the reliability is to reduce the number of hops and link length to reach the nodes in a multilevel-rings network. In this analysis, we are making the reliability parameter link length dependent because we are considering the probability that a link is on is given as ρ (L d/l min ). Because of this, it may happen that the multilevel-rings network with few nodes per ring may yield a high reliability at the expense of increasing the total network fiber length as we will show below. Another important question regarding the topologies shown in Figure 8 is how the number of paths between a source-destination pair affects the reliability parameter performance and how many paths are enough to obtain a cost effective reliability. These two aspects of the ring topologies can be observed in Figure 9. More specifically, Fig. 9 shows results of reliability versus the connection probability ρ for topologies with 4 and 6 nodes per ring. To obtain these results we considered topologies similar to those shown in Fig. 8 (a-d). The number of nodes in the topologies used to obtain Fig. 9 is 121 nodes. 15

16 Observe that the smaller the number of nodes per ring the greater the reliability. Also, Fig. 9 shows results considering different number of nodes shared per ring. Increasing the number of shared nodes also increases the number of paths between source-destination pairs in the network. Note that the greater the number of nodes shared the greater the reliability. However, results indicate that sharing two nodes do not improve significantly the reliability compared to the case when one node is shared with adjacent rings. Note that in addition to reducing the number of nodes per ring, a good way to improve reliability is by increasing the number of paths between a source-destination pair. This is the well-known benefit of diversity in communications and networking. However, this benefit comes at the expense of increasing the fiber link length. This is because reducing the number of nodes per ring and increasing the number of shared nodes increases the fiber length in the network. Also, it is important to mention that in order to increase the number of paths, the nodes that interconnect adjacent rings need to be equipped with a greater switching capability, and also a greater number of inputs and outputs, which again increases the cost. 5.C. Availability results Results of average network availability are shown in Fig. 10 for networks without protection and with 441 nodes. For the case of the multilevel-rings we used 10 nodes per internal ring and 10 nodes per outer ring and for the case of the multilevel ring-stars we used 10 nodes per internal ring and 10 nodes per outer star. Also, for the case of multilevel ring-buses we used 10 nodes per internal ring and 5 nodes per outer bus. The internal rings are the ones that interconnect the CS with one node of the outer rings/stars/buses. Note in Fig. 10 that the star topology yields the highest availability performance at the expense of requiring a large fiber length. The reason for this is that the star topology only requires one hop to reach every node of the network. The signal has to pass through the CS node, fiber link, and the BS node. In the case of the star topology the signal passes through a relatively few number of network components. Due to the fact that availability performance depends on the number of components traversed by the signal, it is evident that the fewer the number of components traversed by the signal, the better the availability performance is. Note that the availability penalty produced by fiber links is practically negligible due to the fact that fiber lengths (CS to the BSs) are small for metropolitan networks. Also, the lower the number of hops to reach destination, the better the availability performance is. Note that the multilevel stars topology yields the second best availability performance, this is due to the fact that at most three hops are required to reach destination. In order to make a complete topology comparison we have to consider the reliability performance, the availability performance and the fiber length of the topologies. Table 2 presents a comparison of all the topologies analyzed in this paper for networks with

17 Fig. 10. Availability of star, multilevel stars, multilevel rings, multilevel buses, multilevel star-rings, multilevel ring-stars, and multilevel ring-buses, topologies for networks with 441 nodes without protection. nodes. We classified the topologies assigning values from 1 to 9, where 1 refers to the best performance and 9 to the worst performance. The last column of Table 2 shows the overall score of the topologies considering the combined effect of the reliability, availability and fiber length scores. Note that for this comparison we assigned the same weight to the three parameters considered. Observe that the topologies with better overall score are multilevelrings with an overall score of 11 and multilevel ring-stars with an overall score of 12. Star and multilevel stars topologies get worse total scores than multilevel-rings and multilevel ring-stars because both networks require a large amount of fiber installed and also they do not perform well in terms of reliability. It is interesting to see that most of these results corroborate the predictions of the earlier work done by Tonguz and Falcone in early 1990 s (see, e.g., [16] and [17]). In order to provide more insight into the network availability performance we present results of network availability versus number of hops in Fig. 11. Results are for the multilevelrings and for the multilevel ring-stars topologies with 121, 441, and 1089 nodes for the unprotected and protected schemes. Fig. 11 presents results of availability against number of hops. For the multilevel-rings topologies we used 10 nodes per internal ring and 10 nodes per outer ring and for the multilevel ring-stars topologies we used 10 nodes per internal ring and 10 nodes per outer star. The internal rings are the ones that interconnect the CS with one 17

18 Protection Nodes No protection 441 Nodes 1089 Nodes 121 Nodes 441 Nodes 1089 Nodes Number of Hops (a) Protection No protection Nodes 441 Nodes 1089 Nodes 121 Nodes 441 Nodes 1089 Nodes Number of Hops (b) Fig. 11. Availability against number of hops for a) multilevel-rings and b) multilevel ring-stars topologies for networks with and without protection with 121 nodes, 441 nodes, and 1089 nodes. 18

19 Table 2. COMPARISON OF DIFFERENT TOPOLOGIES FOR 441 NODES NETWORK. LOWER THE VALUE THE BETTER PERFORMANCE. Topology Reliability Availability Fiber length Overall score Star Ring Bus Multilevel stars Multilevel rings Multilevel buses Multilevel star-rings Multilevel ring-stars Multilevel ring-buses node of the outer rings/stars. Note that in a square node location multilevel-rings network with 121 nodes, the maximum number of hops is 9. If the network contains 441 or 1089 nodes the maximum number of hops is 10. Observe also that for the multilevel ring-stars topology the maximum number of hops is 6 independently of the number of nodes in the network. The multilevel ring-stars topology has the advantage of requiring a lower number of hops to reach the nodes compared to the multilevel-rings topology. Also, the multilevel ring-stars topology has the advantage that it is relatively easy to grow when a new BS node is required in a certain location. Availability results are an average of the availability of nodes that have the same number of hops to and from the CS. We quantify the availability parameter from the CS to all BSs and then we compute the availability average among all the BSs that have the same number of hops from the CS. Note that the availability for the unprotected case becomes worse with the number of hops and the availability is lower than the five nines standard for both topologies. If the network is protected with the 1+1 scheme the availability improves considerably and results are greater (i. e., better) than the five nines ( ) standard of network availability for both topologies. Future research should also consider the wireless service reliability and availability contribution. An integrated analysis of reliability and availability of the fiber network and the wireless network, will give more detailed information of how both networks will perform. 6. Conclusion The contribution of this paper is to provide guidelines for topology selection for Radio over Fiber networks. Reliability, availability, and fiber length results of networks with 25, 49, 121, 441 and 1089 nodes, show that, if one assumes that reliability, availability and fiber length are equally important, then unprotected multilevel-rings and multilevel ring-stars topologies 19

20 are the ones with the best overall performance in our analysis. Multilevel-rings topology can provide higher reliability performance than the other topologies considered in the analysis. Also, when 1+1 link disjoint protection is considered the multilevel-rings topology has a better performance than the other topology options. The second best option, in terms of reliability, is the multilevel ring-stars topology for radio over fiber networks where thousands of antennas need to be connected by a central station. However, results presented show that the star and multilevel stars topology yields slightly better availability results than the other topology options. This is due to the fact that the star and multilevel stars topologies have the advantage of a lower number of hops to reach the nodes compared to the multilevel-rings and multilevel ring-stars topologies. A lower number of hops improves the availability parameter due to the fact that the signal has to pass through a lower number of OADMs and therefore through a lower number of components. However, fiber length required for star and multilevel stars topologies is much greater than the fiber length (hence cost) required for multilevel rings and multilevel ring-stars. Considering the three parameters in the comparison, it turns out that the multilevel rings and multilevel ring-stars achieve the best overall scores. Also, multilevel ring-stars topology is interesting from the point of view that it is relatively easy to grow when a new BS node is required in a certain location. The network availability results presented show that in order to fulfill the five nines availability standard the networks must have a protection scheme. The 1+1 link disjoint protection scheme analysed in this paper yields availability results superior to the five nines standard. Other protection schemes may be analysed using the methodology presented in this paper in order to compare availability and cost reduction of a specific protection scheme. For the safety applications in urban areas, car manufacturers and Departments of Transportation in USA and other countries may resort to Vehicular-to-Infrastructure (V2I) and Infrastructure-to-Vehicular (I2V) communications in the foreseeable future. The results reported in this paper will be instrumental in such safety applications for vehicular ad-hoc networks in the future where the availability and reliability of exiting fiber-based access networks are critical. References 1. N. Wisitpongphan, F. Bai, P. Mudalige, V. Sadekar, and O. Tonguz, Routing in sparse Vehicular Ad Hoc wireless networks, IEEE J. on Selected Areas in Communications, vol. 25, no. 8, pp , Oct N. Wisitpongphan, O. Tonguz, J. Parikh, F. Bai, P. Mudalige, and V. Sadekar, Broadcast storm mitigation techniques in vehicular Ad Hoc networks, IEEE Wireless Communications Magazine,, vol. 14, no. 6, pp , December O. Tonguz, N. Wisitpongphan, F. Bai, P. Mudalige, and V. Sadekar, Broadcasting in 20

21 VANET, Proc. IEEE INFOCOM MOVE Workshop, Anchorage, Alaska, USA, May B. Masella and X. Zhang, A novel single wavelength balanced system for Radio over Fiber links, IEEE Journal of Lightwave Technology, vol. 18, pp , January Tae-Sik Cho and K. Kim, Effect of third-order intermodulation on Radio-Over-Fiber systems by a dual-electrode Mach Zehnder modulator with ODSB and OSSB signals, IEEE Journal of Lightwave Technology, vol. 24, pp , May H. Ogawa, D. Polifko, and S. Banba, Millimeter-wave fiber optics systems for personal radio communication, IEEE Trans. Microwave Theory Tech., vol. 40, pp , December R. Braun, G. Grosskopf, D. Rohde, and F. Schmidt, Low-phase-noise millimeter-wave generation at 64 GHz and data transmission using optical sideband injection locking, IEEE Photon. Technol. Lett., vol. 10, pp , May O.K. Tonguz and H. Jung, Personal communications access networks using subcarrier multiplexed optical links, Joint Special Issue of the IEEE Journal of Lightwave Technology and IEEE Journal on Selected Areas in Communications, on Multiwavelength Optical Technology and Networks, vol. 14, no. 6, pp , June H. Jung and O. K. Tonguz, Coax/Fiber hybrid access network for microcellular personal communication systems, IEICE Transactions on Communications, Special Issue on Optical Access Networks toward Life Enhancement, vol. E79-B, no. 7, pp , July K. Kitayama, A. Stohr, T. Kuri, R. Heinzelmann, D. Jaager, and Y. Takahashi, An approach to single optical component antenna base stations for broad-band millimeterwave fiber-radio access systems, IEEE Trans. Microwave Theory Tech., vol. 48, pp , December G. H. Smith, D. Novak, and C. Lim, A millimeter-wave full-duplex fiber-radio star-tree architecture incorporating WDM and SCM, IEEE Photon. Technol. Lett., vol. 10, pp , November G. H. Smith and D. Novak, Broadband millimeter-wave fiber-radio network incorporating remote up/down conversion, Microwave Symposium Digest, IEEE MTT-S, vol. 3, pp , June K. Kitayama and R. A. Grifin, Optical downconversion from millimeter-wave to IF- Band over 50-km-long optical fiber link using an electroabsorption modulator, IEEE Photon. Technol. Lett., vol. 11, pp , February U. Gliese, S. Norskow, and T. N. Nielsen, Chromatic dispersion in fiber-optic microwave and millimeter-wave links, IEEE Trans. Microwave Theory Tech., vol. 44, pp , October

22 15. Gerardo Castañón, Preferred WDM packet switched router architecture and method for generating the same, USA Patents US6,810,211, B1, no. 2, pp. 2-14, October O. K. Tonguz and K. A. Falcone, Fiber-Optic Interconnection of local area networks: physical limitations of topologies, IEEE/OSA J. of Lightwave Technology, vol. 11, no. 5/6, pp , May/June O. K. Tonguz and K. A. Falcone, Cost function considerations in hierarchical growth of lightwave networks, IEEE Photonics Technology Letters, vol. 6, no. 2, pp , Feb S. Hariri and C. S. Raghavendra, SYREL: A symbolic reliability algorithm based on path and cutset methods, IEEE Transactions on Computers, vol. C-36, pp , October C. S. Raghavendra, V. K. Prasanna Kumar, and S. Hariri, Reliability analysis in distributed systems, IEEE Transactions on Computers, vol. 37, pp , March N. Nagatsu, S. Okamoto, and K. Sato Optical Path Cross-Connect System Scale Evaluation Using Path Accommodation Design for Restricted Wavelength Multiplexing, IEEE Journal on Selected Areas in Communications, vol. 14, pp , June N. Nagatsu, A. Watanabe, S. Okamoto, and K. Sato Architectural Analysis of Multiple Fiber Ring Networks Employing Optical Paths, IEEE Journal of Lightwave Technology, vol. 15, pp , October D. A. A. Mello, D. A. Schupke, and H. Waldman, A matrix-based analytical approach to connection unavailability estimation in shared backup path protection, IEEE Photonics Technology Letters, vol. 9, pp , September D. A. A. Mello, D. A. Schupket, M. Scheffelt, and H. Waldman, Availability maps for connections in WDM optical networks, Proceedings of the 5th International Workshop on Design of Reliable Communication Networks (DRCN), vol. 1, pp , October