A Performance Comparison of Routing Protocols for Maritime Wireless ...

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2008 proceedings.

A Performance Comparison of Routing Protocols for Maritime Wireless Mesh Networks Peng-Yong Konga , Haiguang Wanga , Yu Gea , Chee-Wei Anga , Su Wena , Jaya Shankar Pathmasuntharama , Ming-Tuo Zhoub , and Hoang Vinh Dienb a Institute for Infocomm Research, Agency for Science, Technology & Research (A*STAR), Singapore. b Wireless Comm. Lab., National Institute of Information & Communications Technology (NICT), S’pore Office.

Abstract—We envisage coverage extension of the high bit rate terrestrial communication networks to the ships to reduce the cost in maritime communications. The coverage extension is achieved by forming a mobile wireless mesh network amongst neighboring ships, marine beacons and buoys. The wireless mesh network will be connected to the terrestrial networks across multiple hops via land stations at shore. In such a multi-hop wireless network, routing protocol is important. This paper compares the performance of three existing routing protocols in a maritime communication environment. The three routing protocols are OLSR, AODV and AOMDV. The performance comparison is done via simulation. In the simulation, the wireless mesh network is formed using WiMAX mesh MAC protocol. Also, the random sea surface movement and maritime communication link characteristic are simulated. From the simulation results, we found that OLSR is not as efficient as AODV and AOMDV. Also, compared to AODV, the performance of AOMDV is less affected by sea condition. Index Terms—wireless mesh networks, maritime communication, WiMAX, OLSR, AODV, AOMDV.

I. I NTRODUCTION In these days, an international call from home will set you back by as little as US$0.03 per minute thanks to the advances in technologies that allows voice packets to be carried over the toll-free Internet. On the other hand, a similar telephone call on board a ship would be 30 times more expensive. This is because the costly satellite links have to be used for communications from ships where the less expensive conventional terrestrial communication links are not available at sea. This strongly motivates a project called TRITON, which aims to extend the coverage of the high bit rate terrestrial communications to the ships to reduce the cost in maritime communications. This idea is similar to [1]. In TRITON, the coverage extension is achieved by forming a mobile wireless mesh network amongst neighboring ships, marine beacons and buoys. The wireless mesh network will be connected to the terrestrial networks across multiple hops via land stations at shore. Generally, mobile wireless mesh network is a special case of Mobile Ad Hoc Network (MANET). Recently, VANET (Vehicular Ad Hoc Network) has been actively researched for the purpose of improving road safety and efficiency by increasing the horizon of drivers and on-board devices [2]. VANET is different from a traditional MANET because of its unique characteristics given as follows: (a) Vehicular mobility: Vehicle movement is completely different

from the random movement commonly assumed in a typical MANET scenario. This is because car movement is restricted by the road directions and traffic regulations. For example, cars that travel in a highway move along an original trajectory at high speeds. (b) Geographically constrained topology: Consider each car as a network node, roads limit the network topology to some straight lines with small-angle curves. The topology may change slightly when cars leave and enter the road. The node entrance and departure rate depends on the environment, the speed as well as on the driver’s destination. (b) Rich power supply: In traditional MANET, nodes are power limited and their lifetime depends on their batteries. On the contrary, cars can provide continuous power to their computing and communication devices. As a result, network protocols do not have to account for methodologies that try to prolong the battery life. (d) Readily available position information: Increasingly, cars are equipped with accurate positioning systems such as the Global Positioning System (GPS) and GALILEO, well integrated with electronic maps. the enlarged area

rt ore po Singap

limit

shipping lane direction

Fig. 1.

shipping lane

An illustration of the directed shipping lane off Singapore

Let us consider each ship as a vehicle. Then, the mobile wireless mesh network that helps in extending coverage of terrestrial networks to the sea is also a VANET. This is true because ships and cars share some similar characteristics as follows: (a) In busy narrow navigation channels, such as the Strait of Malacca, the Strait of Dover, the Strait of Singapore and the Turkish Straits, ship movement is restricted by shipping lane directions and traffic regulations. This is illustrated in Fig. 1. Ships can only make a turn at some designated areas. Within a shipping lane, ships are bounded by

2170 1525-3511/08/$25.00 ©2008 IEEE


p(x = λi ) = a × bx , and

(1)

land station westbound shipping lane eastbound shipping lane

ship

20km

II. S IMULATION S ETTINGS To represent a narrow navigation channel with traffic separation scheme by the International Maritime Organization (IMO), we use a generic network topology as illustrated in Fig. 2. In the figure, there are two parallel shipping lanes each with a width of 20 km, and the land station is located 10 km from the shore. The ship mobility is characterized by the inter-arrival time λi of a new ship i to the network depicted in Fig. 2 and the ship’s constant speed vi . Here, λi is the time difference between the arrival of ship i − 1 and ship i in a given shipping lane. In the simulation, λi and vi are random variables generated using the following probability density functions.

72km 10km 20km

a maximum and minimum speeds. (b) Consider each ship as a network node, the shipping lanes have the same effect of roads in limiting the network topology. (c) Similar to cars, there is also no power limitation to network nodes that are installed on board ships. (d) Ships are similarly equipped with GPS and GALILEO to achieve accurate positioning, and are required to regularly announce their positions through the Automatic Identification System (AIS). While terrestrial VANET for cars is actively researched, the findings are not directly applicable to a ship-to-ship VANET. This is partly because ships have a different mobility pattern than cars. For example, compared to a car, a 30,000 tons ship takes a much longer time to make a turn, as well as accelerate and decelerate to a new steady state speed. Also, different from the static earth ground, sea surface is rough and random. The random movements of sea surface can lead to time-varying received signal quality as well as wave occlusions that break communication links. Given the challenging maritime environment, this paper studies through simulation the performance of several existing routing protocols over a WiMAX mesh network. Specifically, we compare the performance of Optimized Link State Routing (OLSR) [3], Ad Hoc On Demand Distance Vector Routing [4] and Ad Hoc On Demand Multipath Distance Vector Routing (AOMDV) [5] over a WiMAX mesh network in a maritime communication environment. We choose OLSR and AODV because they are respectively the proactive and reactive routing protocols developed by the IETF MANET Chapter. AOMDV is chosen because it is an improvement to AODV by adding the capability of identifying multiple paths in a single route discovery process. The additional paths can be used as backup paths where backup routing has been shown to be useful in improving packet delivery ratio ([6], [7]). In the literature, there are existing works on routing protocol comparison for MANET and VANET. However, these works are not for maritime communication environment. Also, they do not compare OLSR, AODV and AOMDV in a single study. The rest of this paper is organized as follows. Next section describes in detail the simulation settings. Section III presents the simulation results and analysis. The paper ends with concluding remarks in Section IV.

Fig. 2. Generic topology for a maritime communication network at a narrow navigation channel

0 x≤0 2 2 c (2) √ e−(x−µ) /(2σ ) otherwise, 2πσ 2 where a, b, and c are factors to ensure best-fit between the functions and empirical data. The empirical data on ship mobility is provided by the Maritime and Port Authority of Singapore (MPA). Based on the empirical data [8], a = 0.292064, b = 0.998286, c = 4.763146, µ = 10.383780 and σ = 5.488477 for the Westbound shipping lane. For the Eastbound shipping lane, a = 0.284495, b = 0.998324, c = 4.354804, µ = 10.800865 and σ = 5.189037. With the ship mobility, distance between two ships, and distance between ships and the land station are in the range of km. We assume the use of WiMAX technology and careful planning of antenna parameters to provide a radio range of 35 km between the land station and ships. Similarly, the radio range between two ships is 20 km. With the adoption of WiMAX, each ship and the land station is a network node that operates the WiMAX mesh MAC (medium access control) protocol. The MAC protocol has been defined in IEEE 802.16-2004 standard [9]. For an accurate simulation, we have implemented in QualNet the WiMAX mesh protocol with a MAC frame length of 10ms at 5.5Mbps raw data rate. The data rate is achieved with OFDM 512 FFT, QPSK 1/2 over 16MHz channel bandwidth. In the implemented MAC protocol, each MAC frame is subdivided into a control sub-frame and a data sub-frame [10]. There are two types of control sub-frames, namely scheduling sub-frame and network sub-frame. In the scheduling subframe, nodes send request to the receiver node to transmit in the data sub-frame as well as receive grant from the receiver node indicating which time slots in data sub-frame to use for transmission. In the network sub-frame, nodes broadcast control messages related to its perceived network configuration. Each MAC frame has only either one of the two types of control sub-frame. There is only one network control sub-frame after every 9 consecutive scheduling control sub-frames. To accurately simulate a maritime communication environment, we have also modeled the radio propagation impairment due to strong reflection from sea surface and random sea surface movement ([11], [12]). In general, the different sea conditions can be represented in terms of Pierson-Moskowitz’s sea states [13]. Table I gives the conditions for sea state 3 and

2171

p(x = vi ) =


wave direction wave height

ship effective height

node

PHY

wave length wave period

ship size ship elevation angle

MAC

random sea surface movement

ship mobility ship initial location

ship direction

higher protocol layer

ship speed ship current location

higher protocol layer

MAC node

communication traffic

link distance antenna orientation and height

channel model

received power

transmission power

Fig. 3.

A framework for maritime communication network simulation

sea state 6 that are used in the simulation study. In short, sea state 3 has a smoother sea surface movement compared to sea state 6. A more rough sea condition is expected to cause a bigger variation in link quality. At the MAC layer, a received packet may be lost and discarded due to its low signal-tonoise-and-interference ratio (SINR), which in turn is affected by the propagation impairments and sea condition as illustrated in Fig. 3.

receiving the TC, each node can build a topology table consists of entries each indicates the last hop to reach a destination node. Based on the topology table, a routing table can be proactively built. The performance of OLSR depends on the sending intervals of HELLO and TC. With a long interval, OLSR will take more time to detect the failure and re-establish new route. As a result, data packet forwarded through the broken route will be lost. On the other hand, a faster adaptation to changes with a shorter interval is achieved at the expense of increased routing control overhead. Further, excessive routing control messages implies some of the HELLO and TC will be lost due to congestion, but this may be interpreted wrongly by OLSR as degrading link quality. This wrong interpretation may cause performance degradation. The default HELLO and TC intervals are 2s and 6s, respectively. After a series of evaluation, we find the optimal HELLO and TC intervals are 10s and 30s (c.f Table. II), respectively for our simulations. These values are used hereafter. TABLE II E FFECT OF HELLO AND TC INTERVALS ON OLSR P ERFORMANCE

HELLO interval TC interval Initial packet delay1 2s 6s 49.4527s 6s 18s 0.36159s 10s 30s 0.28390s 16s 48s 23.2753s 20s 60s 19.4381s 1 Initial packet delay is the time difference between the start time of traffic source and the time of the first packet arrival at the destination.

TABLE I P IERSON -M OSKOWITZ S EA S TATE TABLE [13]

Sea State State 3 6

Wind Speed (knots) 14 27

Significant Wave Height, W (meter) 1.0667 4.2670

Average Wave Period, T (second) 3.5 7.5

Average Wave Length, L (meter) 14.0201 56.0805

A node has to wait to transmit its packet in the particular time slots in the data sub-frame as indicated in the grants it has received in control sub-frame. In our simulation, there is a capacity to buffer 200 packets in the MAC layer awaiting for transmission. In the MAC layer, routing control messages are treated as normal packets but with a strict priority compared to data packets. Also, each routing control message requires a separate request and grant, while a single request-grant exchange for data packets can last for as long as the traffic flow lifetime. As a proactive routing protocol, OLSR periodically sends two types of routing control messages, namely HELLO and TC (Topology Control). HELLO contains one-hop neighborhood information, and are broadcast to one-hop neighbors so that each node can build its own one-hop and two-hop neighborhood information. Based on the neighborhood information, each node can then select a set of one-hop neighbor nodes called Multi-Point Relay (MPR) through which the node can reach all nodes within its two-hop neighborhood. The selection of MPR is announced in HELLO. As such, each node can know its MPR selectors, i.e., nodes that have selected it as their MPR. Each node announces its MPR selectors by sending TC which is forwarded by MPR throughout the network. Upon

Different from OLSR that proactively builds the routing table for all destinations in the network, AODV setup a route between two nodes only when a traffic flow has been started between the nodes. Thus, AODV is a reactive routing protocol. In AODV, the route is setup and maintained only for the lifetime of the flow. At the start of the flow, the source node floods the network with route request which will be rebroadcast by the intermediate nodes. With such a rebroadcast, the route request will eventually reach the destination node if there is a route between them. Upon receiving the first route request, the destination node will send a unicast route reply along the path traversed by the received route request. When the route reply reaches the source node, a route is considered setup and data packets can be forwarded. To limit routing overhead, an intermediate node will rebroadcast a route request only if it is not a duplicate copy, and the Time-To-Live (TTL) value in the IP header of the route request is not zero after being decreased by 1. Thus, the source node can control how far the route request will propagate by setting the value of TTL. The source node may gradually increase the TTL value in successive retransmissions of a same route request if the previous attempt does not return a route reply. The TTL value used for a fresh route request is called initial TTL. In subsequent retransmission, TTL value increases at a step size of 2 until it equals the TTL threshold after which TTL is set to the network diameter. In our simulation, initial TTL

2172


3

Initial packet delay (second)

2

1.5

1

0.5

0

3

6

Sea condition (sea state)

Fig. 4.

Initial packet delay

indication of the route discovery latency. As expected, OLSR has the lowest initial packet delay because it has proactively setup routes even before the traffic sources start generating packets. This low initial packet delay comes with a high cost. Fig. 5 shows the significantly higher routing control overhead incurred in OLSR. The routing overhead is measured in terms of routing packets (1 message = 1 packet). In the simulations, OLSR has to periodically transmit HELLO and TC messages from 0-th second despite data packets will only be generated from 300-th second. If we consider only the routing overhead from 300-th onwards, OLSR still has multiple times higher overhead. Specifically, OLSR overhead within 300-th and 600th second is 8.7 times and 3.0 times higher than that of AODV at sea state 3 and sea state 6, respectively. For the same period, OLSR overhead is 12.2 times and 8.4 times higher than that of AOMDV at sea state 3 and sea state 6, respectively. 8000

III. S IMULATION R ESULTS To compare the performance of different routing protocols, the topology as depicted in Fig. 2 is divided into fives nonoverlapping regions. In each simulation, 1 ship is randomly selected from each region as a CBR traffic source. So, there are 5 CBR sources in a simulation and their destination is the land station. Each traffic source generates fixed size 40 bytes packets at the interval of 0.02s. Each simulation lasts for 600s but all CBR sources start generating packets only from 300-th second. The simulation is repeated 5 times by randomly selecting different ships as the traffic sources from each region. The results presented here are average values of the 5 simulations. The initial packet delays for sea state 3 and sea state 6 are shown in Fig. 4. Initial packet delay is the time difference between the start time of traffic source and the time of the first packet arrival at the destination. This metric gives an

OLSR AODV AOMDV

2.5

Overhead (total number of routing packets)

= 3, TTL threshold = 7, and network diameter = 10. In our simulation, we have turned off the AODV HELLO broadcast because it results in excessive routing control overhead. The AODV buffer size is 200 packets. In AODV, the source node is notified when a broken route is detected. The source node has to re-establish a new route by repeating the same route discovery process described above. During the route re-establishment, data packets are buffered. This may significantly increase the packet delay depending on how fast a new route can be found. This problem is less severe in AOMDV which can identify in a single route discovery process, up to 3 routes between each sourcedestination pair. Hence, when a route is broken, AOMDV can quickly switch to other backup routes without performing route discovery. Practically, AOMDV is an improvement of AODV and has inherited the route request-reply mechanism with a few modifications. First, the hopcount field in AODV routing table has been replaced by the advertised hopcount. For a node, the advertised hopcount represents the maximum value of hopcount among multiple routes that it has to the destination. A node will only accept alternative route with a lower hopcount compared to the advertised hopcount. Second, a field called first hop is added to route request to ensure link disjointness. All nodes that hear the route request from source node directly have to provide its IP address as the first hop in the route request. At the intermediate node, duplicate copies of route request are not discarded but their first hop is checked. The duplicate route request will be forwarded if its first hop has not been seen. Third, instead of replying only to the first received route request, AOMDV sends up to 3 route replies, one to each route request received from different neighbors of the destination node. Similar to AODV, an AOMDV route that has been setup but not used in the last expiry interval is removed. If the expiry interval is too short, AOMDV may find no backup route when the current route breaks only after the expiry interval. In our simulation, the expiry interval is set to 1000s since we have a large radio range. Identical to AODV, the AOMDV buffer size is 200 packets.

OLSR AODV AOMDV

7000 6000 5000 4000 3000 2000 1000 0

3

6


Fig. 5.

Routing control overhead

The higher overhead in OLSR has negative effect on its other performance metrics. Fig. 6, Fig. 7 and Fig. 8 show that OLSR has the highest average and maximum packet delay, as well as the lowest packet delivery ratio. The high control overhead may have a larger impact on OLSR than the other two routing protocols because OLSR routes are highly node-

2173


8

joint since many nodes may share a same node as their MPRs. This results in highly congested hotspot such as node 22 at time 150-th second as illustrated in Fig. 9. Note that the link connecting node 22 to the land station is broken at 200-th second (c.f. Fig. 9 where Node 1 is the land station as depicted in Fig. 2). Consequently, all up stream nodes that route through node 22 lost their paths. When the paths are restored at time 300-th second, a new hot spot is formed at node 26. Also, we have observed in the simulations that OLSR suffers from frequent route changes since there are always multiple routes with equal hopcount to the land station. When a routing table re-computation is triggered by the break of other route, OLSR may end up dropping the current good route and switch to another route with equal hopcount. After switching to another route, the MAC layer has to perform resource reservation through the request-grant mechanism described in Section II. This leads to higher packet delay.

Maximum packet delay (second)

6 5 4 3 2 1 0

3

6


Fig. 7.

OLSR AODV AOMDV

0.25

Maximum packet delay

0.2

OLSR AODV AOMDV

0.15

1

0.1

Packet delivery ratio

Average packet delay (second)

OLSR AODV AOMDV

because, AOMDV can identify multiple routes in a single route discovery process and thus, reduces the number of necessary route discovery. This is indicated by the observation in our simulations that AOMDV has avoided route discovery as many as 6 times in sea state 3 and 18 times in sea state 6, through the use of backup routes.

0.35

0.3

7

0.05

0

3

6


Fig. 6.

Average packet delay

OLSR consistently shows a worst performance in sea state 6 compared to sea state 3. In sea state 6, a ship is subjected to a larger scale of movement. This affects link quality and causes route break. With HELLO and TC intervals set at 10s and 30s, respectively; OLSR needs a long time for a new route to be built if the existing route is broken. The slow route recovery leads to the performance degradation in sea state 6 compared to sea state 3. From the results, we can see that the changes in sea condition have more impact on OLSR than in the other two routing protocols. With the issues above, we conclude that OLSR is not suitable for maritime communication network, and focus on AODV and AOMDV hereafter. In Fig. 4, AODV has a lower initial packet delay compared to AOMDV. This is because, in AODV, intermediate nodes rebroadcast less route request and thus reducing the congestion during the discovery process. Less congestion means faster forwarding of route request and route reply, and thus faster setup of a route. In AOMDV, intermediate nodes may rebroadcast more route request because a duplicate route request may also be rebroadcast if its first hop has not be seen. Despite the higher route request traffic, as illustrated in Fig. 5, AOMDV still has a lower control overhead compared to AODV. This is

0.8

0.6

0.4

0.2

0

3

6


Fig. 8.

Packet delivery ratio

In Fig. 6 and Fig. 7, AODV and AOMDV has similar performance at sea state 3. At sea state 6, AOMDV has a lower packet delay compared to AODV. This is due to AOMDV’s capability to rapidly switch to readily available backup routes when the current route is broken. For example, in a particular AOMDV simulation for sea state 6, node 42 detects a link break at time 593-th second and switches its next hop to node 28. As a result, delay for the affected packet is only 16ms compared to several seconds required to complete a new route discovery process. Despite the superior performance of AOMDV in terms of packet delay, we do not see a significant difference in packet delivery ratio as compared to AODV. This is mainly because both AODV and AOMDV has large enough buffer size to hold 200 packets during the route discovery process. Similar to OLSR, we observe that both AODV and AOMDV

2174


Route graph at simulation time = 150 s


0.3

0.3 1

1

0.25

0.25

0.2

0.2

0.15

0.1

0.05

0 120

26 11 13 1719 3 12 16 79 8 18 28 24 22 5 14 20 23 6 10 15 21 4 39 40 48 42 37 47 44 49 4546 41 38 52 5150 43 120.1

120.2

120.3

120.4

120.5

25

Latitude

Latitude

2

2729

0.1 34

31

36 0.05 3332 35 120.6

26 11 13 1719 12 16 79 8 18 28 24 22 5 14 20 23 6 10 15 21 4 39 40 48 42 37 47 44 49 4546 41 38 52 5150 43

0 120

120.7

25

3

0.15

120.1

120.2

Longitude

120.3

120.4

120.5

2729

34

31

36 3332 35 120.6



0.3

0.3

1

1

0.25

0.25

0.2

0.1

0.05

0 120

26 11 13 1719 3 12 16 79 8 18 28 24 22 5 20 14 23 6 10 15 21 4 39 40 48 42 37 47 44 49 4546 41 38 52 5150 43 120.1

120.2

120.3 120.4 Longitude

Fig. 9.

120.5

25

Latitude

Latitude

0.2

0.15

120.7

Longitude

2729

0.1 34

31

36

0.05

52 5150

3332 35 120.6

26 11 13 1719 12 16 79 8 18 28 24 22 5 20 14 23 6 10 15 21 4 39 40 48 42 37 47 44 49

0 120

120.7

25

3 0.15

120.1

4546 41 120.3 120.4 Longitude

34

31

36 3332

38

43 120.2

2729

120.5

35 120.6

120.7

OLSR routing tree. As depicted in Fig. 2, Node 1 is the land station.

have consistent worst performances in sea state 6 as compared to sea state 3. This is reasonable because sea state 6 is a more challenging environment with more frequent route breaks. Also, we observe that AODV is affected more by the changing sea condition as compared to AOMDV. As such, we conclude that AOMDV is a better routing protocol for maritime communication networks. IV. C ONCLUSIONS We compared the performance of three routing protocols over WiMAX mesh maritime communication environment. We found that OLSR has the lowest initial packet delay, suggesting OLSR may have the smallest route discovery latency due to its proactive nature. However, OLSR is not as good as AODV and AOMDV in all other performance metrics. Especially, OLSR has a much higher routing overhead and its performance is highly affected by changing sea condition. Thus, OLSR is not a suitable routing protocol for a maritime communication network. AODV has a lower initial packet delay compared to AOMDV. However, AODV is not as good as AOMDV in terms of routing control overhead and packet delay. Also, the performance of AODV is not as robust as that of AOMDV against higher sea state. The more robust and efficient performance of AOMDV is due to its capability in setting up multiple routes in a single route discovery process, and ability to switch to backup route when the current route is broken. We

conclude that, compared to OLSR and AODV, AOMDV is the best routing protocol for maritime communication networks. R EFERENCES [1] V. Friderikos, K. Papadaki, M. Dohler, A. Gkelias and H. Agvhami, “Linked Waters”, IEE Communications Engineer, April 2005. [2] M. T.-Moreno, M. Killat and H. Hartenstein, “The challenges of robust inter-vehicle communications”, IEEE VTC, Sept. 2005. [3] T. Clausen, and P. Jacquet, “Optimized Link State Routing Protocol (OLSR)”, IETF RFC 3626, Oct. 2003. [4] C. E. Perkins, E. M. Royer, and S. R. Das, “Ad Hoc On Demand Distance Vector (AODV) Routing”, IETF RFC 3561, July 2003. [5] M. K. Marina, and S. R. Das, “On Demand Multipath Distance Vector Routing in Ad Hoc Networks”, IEEE ICNP, 2001. [6] H. Wang, P. Y. Kong, and W. Seah, “A robust and energy efficient routing scheme for wireless sensor networks”, IEEE ICDCS, July 2006. [7] H. Wang, W. Seah and P. Y. Kong, “Maximizing End-to-end Reliability of Routing with Redundant Path by Optimal Link Layer Scheduling”, IEEE WCNC, March 2007. [8] J. S. Pathmasuntharan, P. Y. Kong, J. Jurianto, Y. Ge, M. Zhou, and R. Miura, “High Speed Maritime Ship-to-Ship/Shore Mesh Networks”, ITST, 2007. [9] IEEE Stand. for Local and metropolitan area networks, Part 16: Air Interface for Fixed Broadband Wireless Access Sys., IEEE 802.16-2004. [10] Y. Zhang, K. S. Tan, P. Y. Kong, J. Zheng, and M. Fujise, “IEEE 802.16 WiMAX Mesh Networking”, Wireless Mesh Networking: Architectures, Protocols and Standards, Dec. 2006. [11] S. Wen, P. Y. Kong, J. Shankar, H. Wang, Y. Ge, C. W. Ang, “A Novel Framework to Simulate Maritime Wireless Communication Networks”, IEEE/MTS OCEANS, 2007. [12] J. Jurianto, S. K. Hazra, S. H. Toh, W. M. L. Tan, J. S. Pathmasuntharam, and M. Fujise, “Path Loss Measurements in Sea Port for WiMAX”, IEEE WCNC, March 2007. [13] [online] http://www.eustis.army.mil/weather

2175