Inter-vehicle mobile ad hoc network for road transport systems B.S. Sharif, P.T. Blythe, S.M. Almajnooni and C.C. Tsimenidis Abstract: Access to information services while on the move is becoming increasingly prevalent within transport systems. Whereas Internet access is now common place in trains, it still remains a challenge for vehicles, particularly when travelling through high speed motorways. Motorway vehicles equipped with wireless communication nodes form an ad hoc network have been examined by which data can be exchanged among them without the need for a pre-installed infrastructure. The main challenge with such an infrastructure-less network is developing communications and protocols that can deliver robust and reliable ad hoc communications between vehicles, when the relative speed between vehicles that can be extremely high under opposite traffic conditions. To address this opposite direction effect, a solution has been presened by minimising the effect of opposite traffic on routing packets. Firstly, a router direction index is introduced to enhance the performance of ad hoc on demand distance vector protocol in updating its routing table and secondly, a new queue priority mechanism is proposed which is based on crosslayer collaboration. Simulations were performed for an ad hoc network consisting of 200 vehicles driving with speeds between 90 and 120 km/h on a two-way motorway for different traffic loads sent through a Gateway adjacent to the motorway. The results obtained demonstrate a performance increase in the average data goodput and less routing overhead for the proposed solution.
1
Introduction
As future intelligent infrastructure will bring together, and connect through wireless communications, the individual, vehicles and infrastructure, it is critical that robust communications protocols are developed. Moreover, road networks currently are not widely fitted with wireless infrastructure (although the recent M3 trials by the Highways Agency and the ‘intelligent corridor’ trials in Newcastle (UK) both are trialling such infrastructure). The use of such mobile ad hoc networks for communications between vehicles where the appropriate infrastructure is not yet available is an issue. Mobile ad hoc networks (MANETs) are self-organised mobile networks in which nodes exchange data without the need for an underlying infrastructure. In the road transport domain, schemes which are fully infrastructure-less and those which use a combination of fixed (infrastructure) devices and mobile devices fitted to vehicles and other moving objects are of significant interest to the ITS community, as they have the potential to deliver a ‘connected environment’ where individuals, vehicles and infrastructure can co-exist and cooperate, thus delivering more knowledge about the transport environment, the state of the network and who indeed is travelling or wishes to travel. This may offer benefits in terms of real-time management, optimisation of transport systems, intelligent design and the use # The Institution of Engineering and Technology 2007 doi:10.1049/iet-its:20070001 Paper first received 11th January 2007 B.S. Sharif, S.M. Almajnooni and C.C. Tsimenidis are with the School of Electrical, Electronic and Computer Engineering, Merz Court, Newcastle University, NE1 7RU, UK P.T. Blythe is with the Transport Operations Research Group, School of Civil Engineering and Geosciences, Cassie Building, Newcastle University, NE1 7RU, UK E-mail:
[email protected] IET Intell. Transp. Syst., 2007, 1, (1), pp. 47 –56
of such systems for innovative road charging and possibly carbon trading schemes as well as through the CVHS (cooperative vehicle and highway systems) for safety and control applications. For this paper, the emphasis on the research is in one element of this ‘big picture’ for future ITS, the ad hoc communications between vehicles to enable the passage of data using vehicles as mobile delivery devices. In such networks each node can be designated as source, destination or data router between source and destination nodes. However, the deployment of MANETs presents several challenges that include route finding, packet collision, lack of power and band width and medium access. Routing protocols can be divided into four classes: proactive (also known as table driven), reactive (also known as on-demand driven), hybrid and position-based routing protocols. In proactive routing algorithms (e.g. destination sequenced distance vector (DSDV) [1]), nodes keep up-to-date information about all paths to all nodes in the network even if these routes are currently not in use. Routing messages are sent frequently as well as when any change in routing states is detected. This routing of data is kept in the routing table of each node. These routing messages consume the network’s bandwidth, however, they keep the network well connected. In these algorithms, a few data packets are delivered but with a slight delay as compared to other algorithms. However, using these algorithms in highly dynamic networks consumes almost all the bandwidth, hence, proactive protocols are not ideal in highly dynamic networks such as motorway-based networks. In reactive routing algorithms (e.g. ad hoc on-demand distance vector (AODV) [2] and dynamic source routing (DSR) [3]) to avoid excess message routing, nodes maintain only the active routes in their routing tables and routing messages are broadcast only when a desirable route is not in the routing table or is not active. If there is no active route, data packets have to wait at the node until a route finding process is performed and an active route is available. This 47
causes some data delay, especially when initiating the route. However, these algorithms save network bandwidth by limiting the broadcasting of routing messages to the necessary routes only, hence, more data packets find their chances to the destinations. Hybrid routing algorithms such as the Zone Routing Protocol (ZRP) [4] are a mixture of both proactive and reactive protocols. Nodes that use these algorithms keep up-to-date information about some nodes in the network (either because they are nearer or belong to the same group) and act reactively with the others. In intervehicular communications, based on the ad hoc networking principle, the network is usually one-dimensional (chain network), thus, large source-to-destination routes are more likely to occur. This is one of the reasons for the utilisation of AODV in this work. This is further supported by the results presented in the works of Gwalani et al. [5] and Charles et al. [6], which show that AODV outperforms DSR, especially in highly dynamic networks. Another reason is that the Internet Engineering Task Force (IETF) has identified AODV as a representative of reactive protocols [7]. Data delivery between vehicles can be classified into two modes: PUSH and PULL [8]. Broadcasting road information (traffic jams or emergency warnings for example) through vehicles is carried out in the PUSH mode, whereas exchanging data between specified requester and specified replier (vehicle-to-vehicle or vehicle-wirednetwork) is performed in PULL mode. The PULL mode is the subject of this work presented in this paper. Fleetnet [9, 10] Virtual Immersive Communications (VICOM) [11] and Mobile Metropolitan Ad hoc Network (MobileMAN) [12] are examples of ambitious projects towards deploying road transport related MANETs [13].
2 2.1
Related work Inter-vehicular related work
TrafficView is one of the first systems developed to provide traffic flow information to drivers dynamically. The basic design of TrafficView and its algorithms are introduced in the work of Nadeem et al. [8]. GPS data, and road traffic conditions are integrated on a map to provide drivers with information in an ad hoc manner. An initial prototype of TrafficView reported in the work of Dashtinezhad [14], consists of a PC, GPS receiver and wireless network card. Local vehicle position is obtained from the GPS receiver, and these data are integrated with data received from other vehicles to display vehicle position with respect to other vehicles and, hence, road conditions beyond the driver’s vision can be observed. Another experiment conducted using two vehicles communicating in three different traffic scenarios (urban, suburban and motorway scenarios) was reported in the work of Singh et al. [15] to assess wireless network performance. The experiment results, in general, show that throughput and signal-to-noise ratio are inversely proportional to the distance between the two vehicles and the relative speed between them. A Self Organizing Traffic Information System (SOTIS) was introduced in the work of Wischoff et al. [16]. The system was suggested as a replacement to the conventional Traffic and Travel Information (TTI) system. The TTI system works in a centralised way and consists of sensors deployed beside the road to measure current traffic density and send the data to a centralised station. The data are then analysed and broadcast to vehicles on the road. In the SOTIS system, monitoring and processing tasks are carried out by the 48
vehicles individually. Each vehicle measures its own traffic parameters (position, velocity and soon) locally with the aid of a GPS receiver. By exchanging the individual information between vehicles and analysing all the information received locally by each vehicle, the traffic information can be determined. The analysed data are then distributed by each vehicle to other vehicles on the road periodically. An adaptive broadcast scheme for an efficient SOTIS system is presented in the work of Wischhof et al. [17]. It depends on the adaptation of the intertransmission interval of the analysed traffic information. A vehicle compares the received information with what is already available in its database. If there is a significant difference between the two sets of information, it decreases the inter-transmission interval, however, if the received information is not much different from that which is already available in its database, it increases the intertransmission interval. Simulation results of the comparison between the presented scheme and strictly periodic broadcasts show a decrease in average error of information while it consumes less bandwidth. 2.2
Routing protocol related work
The node’s relative speed is an important metric in comparing between different mobility models. In the work of Yenliang et al. [18], it was shown that nodes in bidirectional mobility models have less correlation among each other, except for neighbouring nodes travelling in the same direction. Performance comparison of AODV and DSR protocols with different node movements and network loads was introduced in the work of Perkins et al. [19]. A random node movement in a rectangular topology was used in this work, and it was reported that AODV outperforms DSR in most studied cases. Kosch et al. [20] support the use of AODV in inter-vehicle ad hoc networks and list the reasons for this, along with presenting an adjustment to AODV to allow for positioning information. Vehicle positioning obtained from Global Positioning System (GPS) and vehicle speed are added to the route request (RREQ) and route reply (RREP) messages of AODV. By using the position information of the vehicle, the distance between nodes that have participated in the route is calculated and added to both RREQ and RREP messages. A prototype implementation of vehicles equipped with a computer, digital maps and special Omni directional antenna was presented in this work. The performance of inter-connected MANETs was studied in the work of Seah et al. [7]. AODV and optimized link state routing (OLSR) protocols were used as representatives of reactive and proactive routing protocols, respectively. It was demonstrated that it is possible to roam between different MANETs running different routing protocols. In the work of Lee and Gerla [21], a modified AODV was introduced by adding a scheme that utilises a mesh structure and multiple alternate paths. The modified AODV was tested on 50 nodes with a maximum speed of 20 m/s (72 km/h), and the results showed an improvement in packet delivery ratio at the maximum speed. An extension to AODV by incorporating the concept of load-balancing was presented in the work of Joo-Han et al. [22], and simulation results showed that the proposed algorithm was highly efficient in throughput, packet end-to-end delay and routing overhead, especially in high-traffic loads. Another modified AODV was presented in the work of Fan et al. [23], where the hop count metric that was used in the original AODV is replaced by a new cost metric based on MAC delay, and the RREQ and RREP packets and routing table of each node are IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
modified by adding a ‘path loss’. Simulation results showed an improvement in throughput at low node speed and higher packet size. Improving the neighbour detection algorithm in AODV, based on the differentiation between good and bad neighbours, using signal-to-noise ratio (SNR) value was proposed and experimentally verified in the work of Krco and Dupcinov [24]. Other approaches that employ locationbased and cluster-based location routing algorithms for inter-vehicle and vehicle-to-vehicle communications were reported in the work of Santos et al. [25, 26], respectively, and their performances were compared to AODV and DSR protocols. It was reported that enabling Hello messages decreased AODV performance, compared to the other two protocols. AODV with path accumulation added to the routing table was introduced in the work of Gwalani et al. [5], where during RREQ and RREP, each node generates or forwards messages by adding its own address to the message as well. Each node receiving any of these messages updates its routing table with path accumulation information as well as other AODV route information. In the work of Tickoo et al. [27], the RREQ message is enhanced with three additional fields. Each node receives a RREQ from all other nodes, and measures the level of signal power of the message, then, by comparing power measurements from each node it can judge if a node is approaching or departing and retains this information for 100 s before purging it. A rectangular topology and random waypoint models are used in this work. In a motorway scenario, it is not guaranteed that two messages from the same vehicle will be received when they are travelling in opposite directions, since occasionally two vehicles are in point-to-point communication range for ,8 s. 2.3
Queue priority related work
Queuing dynamics algorithms for MANET nodes have been studied initially in the work of Chun and Baker [28], where several data packets priority algorithms were introduced and the effects of routing packet priority were investigated via simulator different data rates, node speeds and node buffer sizes. The packet priority mechanism in a MANET using DSR based on modifying the IEEE802.11 MAC layer has been investigated in the work of Pallot and Miller [29]. Static priority scheduling (SPS), defining different distributed inter-frame space times and calculating the random back-off time depending on the priority level of each packet are three procedures proposed in this work along with a model that was developed for simulating these mechanisms. Simulation results demonstrated a gain in both throughput and packet delay at high- and mediumpacket rates. In the work of Birk and Bloch [30], prioritized dispersal (PD) was introduced where data packets are received with higher priority than redundant packets and distributed via different paths from source to destination. First-in first-out (FIFO), as well as last come first served (LCFS) queuing approaches for high-priority packets were considered, analysed and compared. The proposed Q-priority method considered order changing mechanisms and simulation results demonstrated that PD-based schemes outperform non-prioritized approaches. A quality of service (QoS) architecture for real time data support was presented in the work of Lei and Heinzelman [31], where network layers and QoS requirements in each layer were studied. No-QoS and two QoS models using the IEEE802.11 MAC were tested in this work. The No-QoS model utilises regular AODV routing and FIFO-based queues. The proposed QoS1 uses QoS-aware IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
routing with adaptive feedback and priority packet scheduling, whereas QoS2 extends the priority packet scheduling with arbitration inter-frame space (AIFS) using smaller window sizes for high-priority data packets. Simulation results showed that the QoS architecture improves real time data transmission. A fuzzy-based priority scheduler for MANETs was proposed in the work of Gomathy and Shanmugavel [32]. Packet expiry time, packet data rate and queue length of the nodes are the three input variables to be fuzzified with the packet priority index being the output. Simulation results demonstrated a gain in packet delivery ratio at higher node mobility and average packet end-to-end delay. A weighted hop priority control scheme was introduced in the work of Atoche et al. [33]. This technique gives higher priority to incoming packets that arrive from farther destinations (forwarded from larger number of hops) rather than those packets arriving from nearer destinations. With regard to outgoing packets, priority is given to those packets that are transmitted to farther destinations.. Simulation results revealed performance gains in data delivering ratio. Three different queue scheduling models were introduced and compared in the work of the Heng et al. [34]. The first model is the queue scheduling model where packets are served on an FIFO basis. The second model, is the priority queue model where multicast and unicast packets are kept in separate queues and priority is given to the multicast queue. The third model is the dynamic priority scheduling model, where multicast and unicast packets are kept in separate queues and priority is assigned according to the queue length comparison for predefined thresholds in both queues. The unicast queue is given priority only if it exceeds its threshold and the multicast queue is below its threshold. A new scheduling scheme based on multiple queuing systems and dynamic weights for each queue was proposed in the work of Brahma et al. [35] along with an architecture for the forwarding table in the MAC layer. Two weighted queue were introduced by considering the traffic load and node energy. Simulation results established reductions in the delay of high-priority packets, whereas low-priority packets experience more delay than in the original scheme. Node-disjoint multipath QoS routing protocol for supporting differentiated services (MQRD) was introduced in the work of Xuefei and Cuthbert [36]. In this work, the authors use the node-disjoint multipath routing (NDMR) protocol [37] with differentiated services (DiffServ) [38] for QoS support in MANETs. To reduce the number of high-priority packets dropping, the random early detection (RED) algorithm [39] was employed in this work. Two distinct queues were utilised for high- and low-priority packets. High-priority packets represent real time applications. The simulation comparison between NDMR and MQRD demonstrated that MQRD outperforms NDMR in both packet delivery ratio and average delay as well as routing overhead. 3
Opposite direction effect solution
In motorway scenarios, the relative speed between vehicles travelling in the same direction is minor [40]. On the other hand, relative speed between vehicles travelling in opposite directions is very high. Hence, vehicles travelling in the same direction form a network with lower mobility. In contrast vehicles travelling in the opposite direction form a high-mobility network. Conversely, permitting packet routing over inward travelling vehicles reduces network 49
partitioning compared to routing packets over same direction vehicles only [41]. Due to the vehicles mobility, electromagnetic waves that have been broadcasted by vehicles during point-to-point communication are exposed to the Doppler effect phenomenon. This Doppler effect can be interpreted as relative speed between transmitter and receiver, hence, a vehicle that receives these waves can determine the transmitter’s travelling direction (either in the same or in the opposite direction to that particular vehicle). In this paper, we propose a solution where packets can be routed over vehicles travelling in the same direction whenever this is possible, and gives priority in queue to packets leaving vehicles travelling in the opposite direction, since vehicles in the opposite direction are in point-to-point communication for only a short time period which can be ,8 s for vehicle speed of 120 km/h and 250 m transmission range.
Wait RRL sec then reply to the route with higher seq. no. and min. hop count and higher RDI
Yes Am I destination? No
RDI = RDI +1
Yes
Is the broadcaster moving in my direction? No
Re-broadcast RREQ
RDI = RDI -1
Re-broadcast RREQ
Routing protocol solution
Fig. 2 RDI nodes at receiving RREQ
to the first RREQ but wait for the route reply latency (RRL) time to elapse so that the same RREQ from different broadcasters can be received and evaluated to choose the best one, based on maximum route sequence number, minimum number of hop counts and max RDI, respectively, before replying. 3.2
Data queue priority
In motorway scenarios, vehicles travelling in the opposite direction have a very short time period to exchange data (routing or data packets). In highly dynamic networks, routes are more likely to be broken and hence, routing messages are likely to be more frequently broadcast. As a result data packets will wait longer in the interface queues until a route becomes available and all routing packets are sent out, since routing packets are given higher priority over data packets in MANETs [28, 42]. Consequently, this causes data packets to be delayed and in worse cases to be dropped. To overcome this problem, or to reduce its effect, we introduce a new queue priority mechanism based on cross-layer collaboration. Fig. 3 illustrates the cross-layer communications in the network stack. Network Layer
check the dest. vehicle flag
Data to be written into routing table including direction of transmitting node measured by Doppler Shift
Packets carried information including direction of receiver node
Queue
High Priority Packets to be transmitted
In motorway ad hoc network scenarios, the number of successive routers in the same direction between source and destination is an important factor for the route reliability. The higher the number of successive routers in the same direction, the more reliable it will be for the same number of hop counts. In Fig. 1, two routes between the source and the destination of a transmission are illustrated. Both arbitrary selected routes have three routers. Any route segment between any two vehicles travelling in the same direction will last more than the route segment between vehicles in the opposite direction, because the relative speed between the two vehicles in the same direction is much less than that of vehicles moving in opposite direction and, hence, it is assigned a router direction index (RDI) of 1. In contrast, any route segment between any two vehicles in the opposite direction is less reliable and, hence, has a RDI of 21. The route represented by solid lines has four segments; three of them are between vehicles travelling in the same direction (AB, BC and KM) and one in the opposite direction (CK). Thus, the RDI for this route is 1 þ 1 2 1 þ 1 ¼ 2. Furthermore, the route illustrated by the dashed lines has four segments, three of them are between vehicles in the opposite direction (AG, HD and DM), whereas GH is in the same direction. Hence, the resulting RDI is 21 þ 1 2 1 2 1 ¼ 22. In AODV protocol, two important parameters are taken into consideration when updating the routing table. These are the maximum route sequence number and the lowest number of hop counts. In this work, RDI is introduced as a third parameter to be utilised in updating the routing table. AODV employing RDI will be abbreviated as RDI –AODV for the remainder of this work. The behaviour of RDI – AODV-based vehicles during receiving route request RREQ is shown in Fig. 2. RDI is added to the AODV by extending the format of the RREQ message. Destination nodes in RDI – AODV do not reply immediately
E
D
C
B
A src
dst
M
K
H
G
F
Physical Layer
Movement direction
Data Link Layer
3.1
Receive RREQ
Dop. Measurement Doppler affected received signal Channel
Movement direction
Fig. 1 50
Idea of RDI
Fig. 3 Illustration of cross layer communications in the network stack IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
In the proposed algorithm, when a vehicle receives routing or data packets, it determines from the available Doppler information the transmitter direction and passes this information to the network layer, where it is stored and utilised at request for as long as the corresponding node is available in the routing table. The direction information is stored in the routing table as 1 if that vehicle is in the opposite direction, or 0 otherwise. When the vehicle is to send or forward a data packet, it checks its routing table to determine the next hop for that packet and the direction of the next hop with respect to the present position of the vehicle. Subsequently, the data packet is queued for transmission. The direction information should be amended to the packet header. At the link layer, before the packets are sent to the interface queue, a priority decision is considered and the next hop direction is checked. If it equals 1, then the packet is given the same priority as routing packets and is put at the head of the queue. If the queue is full, then the last packet in the queue is pushed out (drop tail) to clear a space for the packet that was given higher priority. If the next hop direction equals 0, then the packet remains in the queue and waits for its turn to be sent after all high-priority packets and the preceding low-priority packets are sent or dropped if the queue is full at that time. 4
Simulation model and setup
A 5-km dual lane motorway simulation scenario is assumed. This scenario is chosen so that routes with higher hop counts (more route segments) can be formed to study the effect of introducing RDI to AODV. The freeway mobility model is used to simulate a motorway scenario with 200 vehicles [43]. Maximum vehicles speed can be achieved during simulation, as in the traffic model fast cars cannot overtake slow vehicles but keep behind it at a safe distance, until the front vehicle changes its speed. A gateway is located adjacent to, and midway through, the motorway. Each vehicle that leaves the simulation area re-enters randomly from one of the lanes’ ends. Vehicle speeds are distributed randomly between 60 km/h, as the minimum speed in both simulations, and 90 km/h and 120 km/h as the maximum speeds, respectively. Each node changes its speed randomly with a maximum acceleration of 3 m/s2. It is assumed that each vehicle receiver is capable of measuring the phase shift caused by the Doppler effect, by which other vehicle’s direction is determined. The simulations were conducted using the network simulator ns2 [44]. Data are sent from the gateway to the vehicles, and a number of different UDP traffic sources (20, 30, 40 and 50) from the gateway to vehicles are tested. The data rate is 10 packets/s with a packet size of 512 bytes, and the node transmission range is 250 m.
(d) Normalized routing load: number of transmitted routing bytes divided by data bytes delivered successfully to their final destination. The network performance based on goodput and packet delivery ratio at maximum vehicle speeds of 90 and 120 km/h are shown in Figs. 4 and 5, respectively. The figures show that AODV –RDI outperforms AODV at medium traffic load (30– 40 traffic sources). At lower traffic load and vehicle speed, the performance of the two protocols is similar, as the network has less contention for medium access and AODV – RDI ensures that destination vehicles wait for some time (RRL) before replying to the request. This RRL plays a negative role in this situation, because, while the medium is not busy, RRL causes packet accumulation at the vehicle interface queue and, hence, reduces data delivery. Furthermore, RRL plays a negative role at higher traffic load. At high traffic source, vehicles wait for RRL as well as back-off time due to the high medium access contention. The AODV – RDI also performs well at high vehicle speed. By increasing the vehicle’s maximum speed, the route segments that were previously established between vehicles moving in the opposite direction break more often. AODV – RDI reduces these weak segments, however, despite the route establishing delay in AODV – RDI (RRL), the established routes break less frequently, and become more reliable and capable of delivering more packets than the routes which were established using the normal AODV algorithm. As the maximum vehicle speed increases the difference in network performance increases in favour of RDI – AODV, since it is capable of choosing the better route between source and destination nodes. The overall packet delivery ratio at the two maximum vehicle speeds are shown in Figs. 6 and 7. They show that packet delivery ratio decreases by increasing network traffic load, which is attributed to increasing media access contention. Increasing the vehicle’s maximum speed from 90 to 120 km/h results in a marginal reduction in packet delivery ratio. This is due to the frequent breakage of routes, especially between vehicles travelling in the opposite direction. Reducing the maximum vehicle speed increases the point-to-point communication time between vehicles travelling in the opposite direction and, hence, more data can be delivered. By comparing the curves entitled AODV and Q-priority shown in Figs. 4 – 7, a good achievement in goodput and 130
125
120
Simulation results and discussion
In order to assess network performance based on the proposed RDI and Q-priority methods, the following metrics are considered:
115
Goodput (kbps)
5
RDI & QPriority AODV RDI-AODV QPriority
110
105
100
(a) Goodput: total data packets received successfully by their destinations divided by the simulation time. (b) Packet delivery ratio: ratio of packets received successfully by their final destination to packets produced by sources. (c) Average packet end-to-end delay: average time that is taken by data packets since leaving its source to reach its final destination. IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
95
90
85 20
25
30
35
40
45
50
Traffic Sources
Fig. 4 Goodput at maximum vehicle speed of 90 km/h 51
120
24
22
Packet Delivery Ratio (%)
115
110
Goodput (kbps)
RDI & QPriority AODV RDI-AODV QPriority
RDI & QPriority AODV RDI-AODV QPriority
105
100
95
20
18
16
14
12
90
10 85 20
25
30
35
40
45
8 20
50
25
30
T raffic Sources
Fig. 5
Goodput at maximum vehicle speed of 120 km/h
50
1.4
RDI & QPriority AODV RDI-AODV QPriority
1.3
Average Packet End To End Delay (s)
24
22
Packet Delivery Ratio (%)
45
certain route and, hence, reduce the need for route repair that normally causes packets to be delayed at the node that missed the route to destination. At higher vehicle speed, the point-to-point communication time between vehicles travelling in the opposite direction becomes shorter and hence, accumulations of packets in the nodes’ queues are more likely to occur and, as a consequence, packets are delayed. At higher traffic load, the problem manifests itself more clearly since the back-off time due to higher medium contention reduces the possibility of exploiting the short time of communication between nodes travelling in the opposite direction. At higher traffic load, the network that employs RDI and Q-priority exhibits less packet delay. By choosing the more reliable route (the route that has higher successive routers travelling in the same direction), the need for route repair which normally causes delay (RRL and back-off) is reduced. In Figs. 10 and 11, the normalized routing load is shown for different traffic loads and at vehicle speeds of 90 and 120 km/h, respectively, where it clearly evident that the routing overhead is proportional to the vehicle speed and traffic load. The use of RDI is shown to reduce the effect of vehicle speed on routing overhead, especially at hightraffic load. For example, whereas it is changed from 13 to 14 overhead byte/data byte at 50 traffic sources by changing the vehicle speed from 90 to 120 km/h in the absence
26
20
18
16
14
12
10
1.2 1.1 RDI & QPriority AODV RDI-AODV QPriority
1 0.9 0.8 0.7 0.6 0.5
25
30
35
40
45
50
T raffic Sources
Fig. 6 Packet delivery ratio at maximum vehicle speed of 90 km/h 52
40
Fig. 7 Packet delivery ratio at maximum vehicle speed of 120 km/h
packet delivery ratio can be seen in favour of the new Q-priority algorithm, by exploiting the short time of point-to-point communication between vehicles driving in the opposite direction. By adding the two features of choosing the better route between sources and destinations (RDI algorithm) and exploiting the short point to point communication time (Q-priority algorithm), the network performs better, especially at higher vehicle speed where employing the two algorithms have a profound impact on the network performance. Average packet end-to-end delay performance results at different traffic load and at vehicle speeds of 90 and 120 km/h are shown in Figs. 8 and 9, respectively. Despite the improvement in packet delivery ratio using RDI, it should be noted that the RRL time causes packet delay, where the destination (final destination or node which has a route to final destination) waits for RRL before replying. When the route to destination breaks at any node on the way to a certain destination, the node holds the packet and tries to repair the route by traditional RREQ and RREP mechanism. If this occurs several times on the way of a certain packet, then it will suffer a significant delay compared to AODV algorithm as can be seen in the graphs entitled AODV and RDI – AODV. Employing the Q-priority algorithm reduces the possibility of route breakage by exploiting the availability of vehicles travelling in the opposite direction that participate in a
8 20
35
T raffic Sources
0.4 20
25
30
35
40
45
50
T raffic Sources
Fig. 8 Average packet end to end delay at maximum vehicle speed of 90 km/h IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
14
1.3
13
1.2
12
Routing Bytes / Data Bytes
Average Packet End To End Delay (s)
1.4
1.1 RDI & QPriority AODV RDI-AODV QPriority
1
0.9
0.8
0.7
11
10
9
8
7
0.6
0.5 20
RDI & QPriority AODV RDI-AODV QPriority
6
25
30
35
40
45
50
T raffic Sources
5 20
25
30
35
40
45
50
T raffic Sources
Fig. 9 Average packet end to end delay at maximum vehicle speed of 120 km/h
Fig. 11 Normalised routing load at maximum vehicle speed of 120 km/h
of RDI, it is changed only from 12 to 12.4 overhead byte/ data byte for the same parameters when using RDI. By employing both RDI and Q-priority, the network routing overhead is reduced to minimum. RDI reduces the route dependency on vehicle in the opposite direction and Q-priority exploit the short time of point-to-point communication between vehicles in opposite direction if the route building through vehicle in opposite direction cannot be avoided by using RDI mechanism.
community to invest in such a technology may come when these devices become smaller and cheaper as they are predicted to do over the next 5 years and indeed eventually become pervasive in nature as is envisaged by the ‘smartdust concept’. Research into current ‘Smartdust’ technology called ‘Motes’ (wireless devices) and their configuration into MANETs using on-road trials in the ASTRA project [46] is funded by the UK’s DfT as part of the Horizons Research Programme. The project has a number of aims, including a future scoping of the technology, how the technology could be applied to transport and a proof of concept trial to investigate whether these devices (and their networking protocols) could be suitable for development in the ‘moving’ mobile environment, that is, vehicle mounted devices interacting with other vehicle mounted devices or infrastructure mounted devices. The project has already shown that MANETS represents a flexible new technology that can offer dynamic solutions to meet complex traffic scenarios and innovative demand management strategies. Prototypes have been evaluated using frequency bands around the 800– 900 MHz, 2.45 GHz and 5.8– 5.9 GHz – future ‘in-house’ prototypes will migrate to a suitable frequency for widespread, rather than experimental use as shown in Fig. 12. The pervasive nature of the technology enables cars to be ‘always connected’ to the infrastructure in the same way
6 Developing a future ITS environment with ad hoc networks This paper has provided an insight into the development of a new technique to achieve a robust exchange of data in a wireless ad hoc network established between vehicles carrying data and passing each other at speed. This is probably the most challenging of the environments for the mobile ad hoc networks. However, the incorporation of a part infrastructure-based network with some fixed points (possibly wireless devices fitted to streetside infrastructure) will broaden the capabilities of such future wireless systems (where investment in the intelligent infrastructure seems appropriate) [45]. The business model in the ITS 13
12
RDI & QPriority AODV RDI-AODV QPriority
Routing Bytes / Data Bytes
11
10
9
8
7
6
5
4 20
25
30
35
40
45
50
T raffic Sources
Fig. 10 Normalised routing load at maximum vehicle speed of 90 km/h IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
Fig. 12 Prototype Zigbee Mote 53
that home broadband users enjoy ‘always-on’ Internet access thus opening up the scope for an intelligent, configurable ITS infrastructure that will be available for a range of services to support travel and travellers. Thus, road users will perceive direct benefits from the introduction of the technology thereby easing user acceptance. Research to deliver this leads to the opportunity to consider for the first time realistically a fully connected Intelligent Transport System for the future. In 2006, the Office of Science and Technology published the findings Foresight Intelligent Infrastructure Study (IIS) which investigated how technology may evolve over the next 50 years to deliver a robust, sustainable and safe transport infrastructure in the future [47]. Among the many recommendations and predictions on how the technology may deliver more intelligence into infrastructural systems was the view that pervasive wireless systems will have a significant future role in transport. Recently, the use of RFID (radio frequency identification) for transport applications has begun to emerge as a key technology, particularly for use in the freight and logistics sector for tracking containers, pallets and individual products, for car-parking, ticketing and possibly for future road user charging. However, Foresight recognised that RFID is just the starting point for a raft of more exciting possibilities with future wireless mobile ad hoc networks, with much more capability and ‘intelligence’ than current RFID and thus moving towards a more ‘all-seeing, allknowing’ network. Research is currently focused on filling in the knowledge and technology gaps in pervasive, mobile ad hoc wireless systems for a range of transport applications. Mobile wireless systems are beginning to be proven as a future tool that will enable the joining up of vehicles, individuals and infrastructure into a single ‘connected’ intelligent infrastructure system. Embedding this technology in infrastructure (such as environmental sensors in lampposts or vehicles or goods, and connecting to individuals through their PDAs, mobile phones or even bespoke wearable wireless interfaces) offer the potential for a more all-seeing, allknowing ITS infrastructure. If for example, vehicles are continually in wireless communication with the infrastructure (through small wireless sensors embedded in the infrastructure), new paradigms for traffic monitoring and control could be considered, road space allocated more efficiently and incidents dealt with in an optimum way. If vulnerable users have such wireless devices, the infrastructure could warn vehicles to slow down and the drivers be more vigilant – indeed wireless devices attached to children could, for example, warn drivers that children are playing out on the street, just around the corner so reduce speed now. Such devices could help with security and safety of individuals, be used on airline boarding cards and other tickets, and even be used to verify HOV (high occupancy vehicles) or blue badge entitlement. When such a system is also connected to a vehicle’s CAN-bus, information on driving style or unusual driving behaviour (e.g. where there is a badly maintained stretch of road or object in the road) could be detected from the CAN data. Following detection and processing, mitigating and maintenance actions could be automatically triggered. Many of these devices can carry payloads such as sensors and the idea of monitoring pollution, with these devices in a pervasive way is beginning to be researched in the university in the MESSAGE project (with pervasive wireless environmental sensors being attached to lampposts) [48]. However if these devices become small and cheap enough (as is the future vision for Smartdust), then 54
one could imagine that we each carry our personal exposure meter. Moreover with ‘extreme’ sensor design, wireless pollution sensors could be fitted in engine manifolds and exhaust pipes to measure the actual pollution generated by a vehicle and maybe adjustments to driving style or engine management systems are advised or made to mitigate some of the pollution effects (early prototypes are being developed in the UK). If future ‘carbon allowances’ are to be considered the connected car and the pollution a driver generates will need to be measured and monitored – as proposed in the Smart Market Protocols project where auction and trading-based carbon allowances have been considered. Wireless interfaces on individuals whether through PDAs or mobile phones, or dedicated devices (such as motes integrated into jewelry – a research project currently in its early stages at Newcastle University) enable individuals to be connected and interact with the infrastructure. This thrust of research will finally provide the missing link in delivering the vision of future pervasive information delivery, whereby context specific and bespoke traveller information can be delivered to the individual on the move, through embedded screens in infrastructure, on mobile devices and for example on ‘terminator’ glasses where one is able to display traveller information on the lens of specially adapted spectacles (seen as particularly beneficial for mobility impaired users who are unable to interact readily with mobile phones and PDAs or other ICT systems). A key element of the connected environment is future pervasive traveller information whereby pervasive, bespoke information delivery may have a role in influencing travel behaviour and travel choices and hopefully could help affect a modal shift towards public transport, particularly if the cost and carbon costs of the alternatives can be readily compared. Moreover, such pervasive information may also be a driver in making more effective use of demand responsive transport and flexible transport services. Significant research is required to fully realise the potential of such wireless systems, not just on the transport application side, but challenges to reduce the size of these devices from ‘smart-lumps’ to ‘smartdust’ are critical as size and cost of these devices will dictate whether the devices will become pervasive in the transport domain. This requires detailed work on antennae design, investigation as to which is the most appropriate communications frequency, and development of communication standards including 802.11xx, WiFi and CALM [49]. Probably the most important challenge is that of battery power and the development of power scavenging or other techniques. Another key area of research, which is still embryonic, is in low-cost and robust sensor design – much work is on-going but it is uncoordinated in the transport domain. Additionally, work is required on the robustness and dependability of mobile sensor devices and suitable communication protocols and techniques to deal with the data. Finally, attention has to be paid to the issues of privacy and data protection in a potentially all-seeing, all-knowing connected world. 7
Conclusions
This paper presents a solution to the opposite road traffic direction effect on the data delivery between vehicles equipped with inter-vehicle ad hoc networks. By using the Doppler effect phenomenon, vehicles using this algorithm can determine another vehicle’s direction (approaching or departing) and, hence, using this information can choose the better route between source and destination as well as IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
using this information to give priority to packets departing to nodes travelling in the opposite direction. A modification of AODV is proposed that employs an RDI and queue priority algorithm based on the cross-layer collaboration technique. RDI and the Q-priority algorithms save memory by limiting the modification to routing table of the ordinary AODV to one parameter in RREQ message and representing the next hop towards destination by either 0 or 1 in the routing table. Simulation results were presented showing gains in network goodput, average packet end-to-end delay and a total routing overhead. The improvement in the network performance using the proposed algorithms is shown to improve, as the maximum vehicle speed increases.
8
References
1 Perkins, C.E., and Bhagwat, P.: ‘Highly dynamic destinationsequenced distance-vector routing (DSDV) for mobile computers’. Proc. ACM SIGCOMM’94 Conf. on Commun. Archit. Protocol. Appl., 1994, pp. 234– 244 2 Perkins, C.E., and Royer, E.M.: ‘Ad-hoc on-demand distance vector routing’. 2nd IEEE Workshop on Mobile Computing Systems and Applications (WMCSA), 1999, pp. 90– 100 3 Johnson, D.B.: ‘Routing in ad hoc networks of mobile hosts’. IEEE Workshop on Mobile Computing Systems and Applications, 1994, pp. 158–163 4 Haas, Z.J.: ‘A new routing protocol for the reconfigurable wireless networks’. IEEE Proc. 6th Int. Conf. on Univers. Pers. Commun., 1997, pp. 562–566 5 Gwalani, S., Belding-Royer, E.M., and Perkins, C.E.: ‘AODV-PA: AODV with path accumulation’. IEEE Int. Conf. on Commun (ICC), 2003, vol. 1, pp. 527– 531 6 Charles, E.M.R., Perkins, E., Das, S.R., and Marina, M.K.: ‘Performance comparison of two on-demand routing protocols for ad hoc networks’, IEEE Pers. Commun., 2001, 8, pp. 16– 28 7 Seah, W.K.-G., Yeo, L.-Y., Eu, Z.-A., Tan, H.-X., and Tan, K.-S.: ‘Performance modeling of MANET interconnectivity’, Int. J. Wireless Inform. Netw., 2006, 13, (2), pp. 115–126 8 Nadeem, T., Dashtinezhad, S., Chunyuan, L., and Iftode, L.: ‘TrafficView: a scalable traffic monitoring system’. IEEE Int. Conf. on Mobile Data Manag, 2004, pp. 13–26 9 Franz, W.J., Eberhardt, R., and Luckenbach, T.: ‘FleetNet – Internet on the road’. 8th World Congress on Intell. Trans. Syst., 2001, p. 73 10 The-Fleetnet-Project. Available at: http://www.et2.tu-harburg.de/ fleetnet/ 11 Virtual-Immersive-Communications (VICOM). Available at: http:// www.vicom-project.it/index.html 12 Mobile-Metropolitan-Ad-hoc-Network (MobileMAN). Available at: http://cnd.iit.cnr.it/mobileMAN/’ 13 ‘Scientific Report’. ESF/PESC Exploratory Workshop, 2002 14 Dashtinezhad, S., Nadeem, T., Dorohonceanu, B., Borcea, C., Kang, P., and Iftode, L.: ‘TrafficView: a driver assistant device for traffic monitoring based on car-to-car communication’. IEEE 59th Veh. Technol. Conf. (VTC), 2004, vol. 5, pp. 2946–2950 15 Singh, J.P., Bambos, N., Srinivasan, B., and Clawin, D.: ‘Wireless LAN performance under varied stress conditions in vehicular traffic scenarios’. IEEE 56th Veh. Technol. Conf. (VTC), 2002, vol. 2, pp. 743–747 16 Wischoff, L., Ebner, A., Rohling, H., Lott, M., and Halfmann, R.: ‘SOTIS – a self-organizing traffic information system’. 57th IEEE Semiannual Veh. Technol. Conf. (VTC), 2003, vol. 4, pp. 2442– 2446 17 Wischhof, L., Ebner, A., Rohling, H., Lott, M., and Halfmann, R.: ‘Adaptive broadcast for travel and traffic information distribution based on inter-vehicle communication’, IEEE Intell. Veh. Symp., 2003, pp. 6– 11 18 Yenliang, L., Huier, L., Yajuan, G., and Helmy, A.: ‘Towards mobility-rich analysis in ad hoc networks: using contraction, expansion and hybrid models’, IEEE Int. Conf. on Commun., 2004, 7, pp. 4346– 4351 19 Perkins, C.E., Royer, E.M., Das, S.R., and Marina, M.K.: ‘Performance comparison of two on-demand routing protocols for ad hoc networks’, IEEE Pers. Commun., 2001, 8, pp. 16– 28 20 Kosch, T., Schwingenschlogl, C., and Li, A.: ‘Information dissemination in multihop inter-vehicle networks’, IEEE 5th Int. Conf. on Intell. Trans. Syst., 2002, pp. 685–690 IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
21 Lee, S.J., and Gerla, M.: ‘AODV-BR: backup routing in ad hoc networks’. IEEE Wireless Commun. Netw. Conf., 2000, vol. 3, pp. 1311–1316 22 Joo-Han, S., Wong, V.W.S., and Leung, V.C.M.: ‘Efficient on-demand routing for mobile ad hoc wireless access networks’, IEEE J. Selected Areas Commun., 2004, 22, pp. 1374–1383 23 Fan, Z.: ‘High throughput reactive routing in multi-rate ad hoc networks’, Electron. Lett., 2004, 40, pp. 1591– 1592 24 Krco, S., and Dupcinov, M.: ‘Improved neighbor detection algorithm for AODV routing protocol’, IEEE Commun. Lett., 2003, 7, pp. 584– 586 25 Santos, R.A., Edwards, R.M., Seed, L.N., and A, Edwards.: ‘A location-based routing algorithm for vehicle to vehicle communication’. 13th Int. Conf. on Comput. Commun. Netw. (ICCCN), 2004, pp. 221 –226 26 Santos, R.A., Edwards, R.M., Edwards, A.E., and Belis, D.: ‘A novel cluster-based location routing algorithm for inter-vehicular communication’. 15th IEEE Int. Symp. on Pers., Indoor Mobile Radio Commun. (PIMRC), 2004, vol. 2, pp. 1032–1036 27 Tickoo, O., Raghunath, S., and Kalyanaraman, S.: ‘Route fragility: a novel metric for route selection in mobile ad hoc networks’. 11th IEEE Int. Conf. on Netw. (ICON), 2003, pp. 537– 542 28 Chun, B.-G., and Baker, M.: ‘Evaluation of packet scheduling algorithms in mobile ad hoc networks’, ACM Mob. Comput. Commun. Rev., 2002, 6, pp. 36–49 29 Pallot, X., and Miller, L.E.: ‘Implementing message priority policies over an 802.11 based mobile ad hoc network’. IEEE Military Commun. Conf., 2001, vol. 2, pp. 860–864 30 Birk, Y., and Bloch, N.: ‘Improving network performance with prioritized dispersal’. 19th Annu. Joint Conf. IEEE Comput. Commun. Soc., 2000, vol. 3, pp. 1817–1826 31 Lei, C., and Heinzelman, W.: ‘Network architecture to support QoS in mobile ad hoc networks’. IEEE Int. Conf. on Multimedia and Expo (ICME), 2004, vol. 3, pp. 1715–1718 32 Gomathy, C., and Shanmugavel, S.: ‘An efficient fuzzy based priority scheduler for mobile ad hoc networks and performance analysis for various mobility models’. IEEE Wireless Commun. Netw. Conf. (WCNC), 2004, vol. 2, pp. 1087–1092 33 Atoche, G., Hayasaka, M., Tomitsuka, S., Manodham, T., and Miki, T.: ‘Weighted hop priority control scheme for multihop wireless ad hoc networks’. Asia-Pacific Conf. on Commun, 2005, pp. 48–52 34 Heng, X., Xin, W., and Toh, C.K.: ‘Comparative analysis of scheduling algorithms in ad hoc mobile networking’. 6th Int. Conf. on Parallel Distributed Comput., Appl. Technol (PDCAT), 2005, pp. 639– 643 35 Brahma, M., Kim, K.W., Hachimi, M.E., Abouaissa, A., and Lorenz, P.: ‘A buffer and energy based scheduling in mobile ad hoc networks over link layer’. Adv. Int. Conf. on Telecommun., Int. Conf. on Internet Web Appl. Services, AICT-ICIW, 2006, pp. 69–69 36 Xuefei, L., and L, Cuthbert.: ‘Multipath QoS routing of supporting DiffServ in mobile ad hoc networks’. 6th Int. Conf. on Software Eng., Artif. Intell., Netw. Parallel/Distributed Comput., and 1st ACIS Int. Workshop on Self-Assembling Wireless Netw, 2005, pp. 308– 313 37 Xuefei, L., and Cuthbert, L.: ‘Stable node-disjoint multipath routing with low overhead in mobile ad hoc networks’. IEEE Comput. Soc.’s’. 12th Annu. Int. Symp. on Model. Anal. Simul. Comput. Telecommun. Syst. (MASCOTS), 2004, pp. 184–191 38 Blake, S., and Black, D.: ‘An architecture for differentiated services, IETF RFC2475’, 1998 39 Floyd, S., and Jacobson, V.: ‘Random early detection gateways for congestion avoidance’, Trans. Netw., IEEE/ACM, 1993, 1, pp. 397– 413 40 Durresi, M., Durresi, A., and Barolli, L.: ‘Adaptive inter vehicle communications’, Int. J. Wireless Inform. Netw., 2006, 13, (2), pp. 151– 160 41 Fu¨ßler, H., Mauve, M., Hartenstein, H., Ka¨semann, M., and Vollmer, D. Department for Mathematics and Computer Science, University of Mannheim ‘A comparison of routing strategies for vehicular ad hoc networks’, Technical Report TR-02-003 Report, 2002) 42 Das, S.R., Perkins, C.E., and Royer, E.M.: ‘Performance comparison of two on-demand routing protocols for ad hoc networks’, IEEE INFOCOM., 2000, 1, pp. 3 –12 43 Bai, F., Sadagopan, N., and Helmy, A.: ‘IMPORTANT: a framework to systematically analyze the impact of mobility on performance of routing protocols for ad hoc networks’, IEEE INFOCOM, 2003 44 http://www.isi.edu/nsnam/ns/ 45 Blythe, P.T.: ‘Intelligent infrastructure: a smart future with smart dust and smart markets’, Proc. IFAC Conf. on Transport Automation and Control Delft, August 2006 55
46 Blythe, P.T., Tully, A., and Martin, G.: ‘Investigating next-generation wireless technology to deliver pervasive road user charging’, Proc. 12th World Congress on Intelligent Transport Systems and Services, San Francisco, November 1995 47 http://www.foresight.gov.uk/Previous_Projects/Intelligent_Infrastruct ure_Systems/Index.htm
56
48 Blythe, P.T., Bell, M.C., Sharif, B., and Watson, P.: ‘Pervasive Environmental Monitoring using Smartdust: The MESSAGE Project’, Prof IET Seminar on RFID and Electronic Vehicle Identification, Newcastle, November 2006 49 Pickford, A.W., and Blythe, P.T.: ‘Road user charging and electronic tolling’ (Artech House, 2006), ISBN 1-58053-858-4
IET Intell. Transp. Syst., Vol. 1, No. 1, March 2007
