Urban Vehicular Network Performance Optimization

Transcription

1 Urban Vehicular Network Performance Optimization Xiang Yu University of Warwick Dr. M. S. Leeson University of Warwick Prof. E. L. Hines University of Warwick Abstract In urban vehicular ad hoc network (VANET) design, it is always a problem to guarantee the quality of peer node communication for moving nodes. The reason is two-fold, namely the dynamic nature of an ad hoc network, and the complexity of an urban environment. Firstly, communication links between neighboring nodes are constantly established and torn down in an ad hoc network. This momentary phenomenon results in highly unpredictable network behavior, and it degrades the signal transmission performance severely. The second factor, complexity of the urban environment also affects network transmission: a wireless signal may travel across different buildings and materials with different fading qualities; before it finally reaches its destination, the quality of transmitted data becomes unforeseeable. In this paper, we propose a novel wireless sensor network (WSN) to assist VANET transmission in urban areas. Sensors are placed along streets and junctions to assist wireless signal transmission. Then we propose a novel algorithm called the Smart Routing Algorithm (SRA) to control the sensor network. It will search for communication links which guarantee bit error rates (BER) using a genetic algorithm optimization tool. Finally we verify SRA s performance by comparing its quality to the mainstream ad hoc network routing protocol Ad hoc On-demand Distance Vector routing (AODV), using the US city of Boston as an example environment. Simulation shows that SRA can optimize signal quality from 10-2 to 10-3 within 10 epochs of optimization rounds, and can optimize links ranging from distances of 100m to 1km. It also can locate links with the minimum number of sensors, for energy conservation purposes. Compared with AODV, SRA is a more efficient and feasible option for path routing optimization. 1. Introduction Keywords: VANET, Network Routing, Optimization, Genetic Algorithm In the past decade, ad hoc networks have been studied by many researchers due to their widespread application in daily life. However, much of the work has been carried out in the free-space environment and assumed that signal propagation follows a simple line of sight (LOS) transmission model [1], [2]. There was thus a dearth of research were executed in the real urban area environment, where wireless signals need to transmit through buildings and across concrete walls before reaching their destinations. This situation lasted until the 2000s, when a transportation based system, the Vehicular Ad hoc Network (VANET) was introduced and investigated in several industrial projects [3], [4], [5]. As described in the subsequent reports and surveys, a VANET is a mobile ad hoc network (MANET) where vehicles act as nodes which are restricted to move along city streets[6], [7]. The VANET is clearly a more realistic modeling of urban ad hoc networks. Unfortunately, previous routing algorithms, such as Ad hoc On-demand Distance Vector routing (AODV) [8] and GPSR [9] were reported to have unsatisfactory performance within the VANET environment [10]. This is because signal transmission in VANETs does not follow the LOS propagation model, and signal attenuation and reflections can no longer be ignored. Thus, we need a more accurate propagation model to describe signal transmission in VANETs, and a novel routing algorithm to guarantee communication quality. In 2005, an analytical formula for path loss prediction in urban street grid area was proposed [11] and has become established in the literature [12]. In this model, the signal strength at the receiver side is decided by the distances between the junctions and the transmitter and the receiver side, as well as by the widths of the streets. As shown in Figure 1.1, signals between two ad hoc nodes with the same geographic locations but different environments travel along different paths. Signal power attenuations, expressed as a ratio between transmit power P t and receive power P r, are calculated by different equations. So, a Advances in information Sciences and Service Sciences(AISS) Volume5, Number11, June 2013 doi: /aiss.vol5.issue11.1 1

2 routing algorithm designed for VANETs has to consider street grid distributions and node movement [13]. Several types of routing algorithms have been proposed for VANETs; one of these is Geographic Source Routing (GSR) [14], which is a basic shortest path algorithm utilizing satellite positioning technology (GPS) information to update ad hoc node positions and surroundings information. Based on this simple model, an advanced fuzzy-logic routing algorithm was proposed to predict driving directions of vehicles when they cross over street junctions [15]. Experimental results proved that while in sparse rural city areas traditional routing algorithms such as AODV and GPSR worked well, in dense urban areas knowledgebased routing algorithms performed more efficiently. Figure 1. Free space, Urban and Sensor propagation diagrams Due to recent technical advances, mass production of low-cost sensors has become feasible. Wireless Sensor Networks (WSN) [16] containing a large number of sensors have thus become popular in many research fields. In WSNs, sensors perform sensing, computation, and wireless communication with each other, and they are controlled by an external base station (BS). However, low cost sensors are severely restricted in energy consumption, and can only perform simple tasks. Routing protocols in WSNs, as reviewed in a survey [17], are categorized into 3 classes: flat-routing, hierarchical routing and location-based routing. Among the flat-routing algorithms, Sensor Protocols for Information via Negotiation (SPIN) [18] and Directed Diffusion (DC) [19] are two of the most widely applied for self-organizing WSNs and data centric WSNs. For hierarchical WSNs, in which sensors are classified into different clusters according to their conditions (in most cases, battery depletion status), Low Energy Adaptive Clustering Hierarchy (LEACH) [20] and Small minimum energy communication network (MECN) [21] are popular protocols in non-gps and GPS-assisted WSNs. In the location-based routing class, Geographic and Energy Aware Routing (GEAR) [22] and a set of protocols [23] named Most Forward within Radius (MFR), Geographic Distance Routing (GEDIR) are examples of how to take advantage of location information to assist routing. In this paper, we propose a WSN (detailed in Figure 2) whose sensors support packetforwarding and power amplification of received wireless signals. Sensors are placed along every street in a city and three sensors installed at each street junction to enable a triangle formation. These sensors in WSN perform wireless signal detection (particularly signal path loss from their neighbors in real time) so that when an ad hoc node is moving within their coverage, the information will be conveyed to their BS. This information is sent to our algorithm, the Smart Routing Algorithm (SRA), as a communication request and SRA will return optimized routing information back to WSN. The routing decision will be broadcast by the BS to sensors. The sensors so activated will turn on their power amplifiers and forward the data packets. In the second section, the urban environment model is described followed by section 3 giving details of SRA itself. In section 4, we further explore how to establish links between 10 sets of randomly selected nodes located within a real urban area, viz. Boston city. Conclusions and suggestions further work are provided in section 5. 2

3 2. Urban environment modeling To keep the bit error rate within an acceptable range and to also reduce the number of sensors used in communication, this algorithm needs to solve an optimization problem which considers: (a) urban area modeling; (b) ad hoc and sensor network topologies; (c) a network routing protocol; (d) digital transmission system requirements. An optimal solution to this problem will locate a route between two ad hoc nodes which consists of the smallest number of sensors that also satisfies the signal transmission requirement. In this section, we will describe how to setup a virtual network environment with ad hoc nodes and wireless sensors on a real city map. 2.1 City map modeling The urban network is represented as an undirected graph G ub = (V ub, E ub ), with each node pair representing a street on a city map. The set V ub represents node vectors which contain the coordinates of each node and set E ub represents link vectors which contain the node pair. The data is read in from satellite pictures of real cities (here we use the US city of Boston as an example), and street information extracted using MATLAB from this file into a matrix containing street node coordinates, street lengths and an adjacent link matrix. This street matrix will be used by algorithm as described in Table 1. Table 1. City map data elements Symbol Name Description Starting and ending nodes of each street, readable from N ub Nodes satellite map, reordered from geographic location view V ub Node coordinates Geographic information (latitude and altitude respectively), readable from satellite map E ub D ub Streets Distance matrix Streets in the city, readable from satellite map, reordered for algorithm optimization Distance matrix of street nodes, calculated with V ub and E ub, if two nodes belong to the same street, then their distances equal to this street length, otherwise Ad hoc network modeling The Ad hoc network model is simplified into a random selection of starting node and its ending peer within the city map. It is a data matrix of 10 rows and 2 columns, each row representing a node pair of Ad hoc network, and these nodes are randomly drawn from city map data V ub. Table 2. WSN data elements Symbol Name Description Sensor nodes (those placed in street junctions), associated N sn Sensor nodes with city map nodes N ub, each N ub links three N sn V sn Sensor coordinates Sensor node information, inherited from city map node information V ub, three V sn share one V ub E sn S sn D sn Sensor links Hop number matrix Sensor distance matrix Sensor node link status, if two sensors are on one street, E sn =1, otherwise 0 Distance matrix of number of hops required to jump over one street, calculated using information on matrix D ub Sensor node distance matrix inherited from D ub, if two sensor nodes belong to the same street, then their distances equal this street length plus their geographic distances, otherwise 0 3

4 2.3 Sensor network modeling Figure 2. Sensor network paradigm A sample WSN paradigm is illustrated in Figure 2, where two cars (ad hoc nodes) are moving in an urban area consisting of 17 streets. The WSN consists of 24 wireless sensors that are deployed in advance; 11 of them are currently activated to maintain live communication. Each sensor is equipped with sensing and amplifying blocks, enabling it to forward data packets and amplify received signals. The sensor network model is also represented as an undirected graph G sn = (V sn, E sn ), which is similar to an urban street map G ub. The initial placement of sensors is shown by red dots for the Boston example in Figure 3(a). In this paper, only those sensors placed in street junctions will be involved in the routing optimization process. Thus we only include these sensors in our modeling. Those sensors placed along streets will be represented in S sn as an attribute of city streets - the number of hops it will use to send signals over the street. Detailed description of WSN data elements can be found in Table Smart routing algorithm (SRA) In this section, we introduce an efficient calculation method termed the Smart Routing Algorithm (SRA), which is based on a genetic algorithm (GA). The idea is to divide the urban area into mesh grids, each of which contains one triangle formation of sensors placed in a street junction. 3.1 Triangle sensor formation Figure 3. (a) Sensor placements on Boston city map; (b) sensor triangle formation 4

5 These grids are coded into variable binary strings and different strings from different areas are combined together into one string, which represents a complete link between two ad hoc nodes. A randomly generated string will be used as an input of SRA, and then it will be manipulated using genetic techniques to reach an optimal string which represents the route between two ad hoc nodes with the lowest BER. Sensor triangle formation is a path routing strategy applied in SRA algorithm. In each grid, one sensor triangle formation is defined to allow multiple path generations. As shown in Figure 3(b), one sensor triangle formation provides three additional paths. Sensor triangle formations can only be found in street junction areas, where two adjacent sensors are close enough to send signals through buildings. Outside street junction areas, signals are not strong enough to penetrate concrete walls and reach the other side (we will not take indoor signal amplifier in consideration). Sensor triangle formations make path routing optimization possible with multiple candidate links available, SRA can calculate their fitness values and choose the optimal one from the set. 3.3 Path optimization In order to make this evolution efficient and successful, a fitness function is defined by: n 2 Fitnessi ( BERk, i BER0 ) N sn, i [1,4] (1) k 1 In equation (1), BER 0 is a threshold value for the digital communication system in question; for an audio system it is 10-3, and for a video system BER k,i is the cumulative BER value of link i traversing a number k of sensor grids. The fitness value for optimization describes the absolute distance from the desired BER per sensor exhibited by the current link. An optimal link shall be the one with the least number of sensors and also having good signal quality. The operation of the algorithm is illustrated in detail in Figure 4. Figure 4. Flowchart of smart routing algorithm 5

6 4. Simulation results As stated previously, in this paper a city map of the US city of Boston was used as a benchmark. This map information was read from mapping toolbox of MATLAB and is used to create three urban areas shown in Figure 5: Figure 5. Boston city map and area definitions 4.1 Adhoc On-Demand Vector Routing algorithm (AODV) Our simulation result with the AODV algorithm is shown in Figure 6, from which one can observe that AODV routing is not efficient in optimizing radio link signal quality; half of the links were not optimized at all. Also, as more streets are covered in communication links there is less opportunity for these links to be optimized. AODV tends to keep the original link as long as it is working, and is reluctant to evaluate new links; it thus performs badly in the urban environment. 4.2 Smart routing algorithm (SRA) Figure 6. Simulation outputs using the AODV algorithm Optimization on distance Figure 7 shows the results of optimization using SRA on a static scenario setting on Boston city central area. The algorithm defines 10 groups with evenly distributed spatial distances and randomly chooses 10 node pairs for each group. The x-axis in figure 7 is the maximum spatial distance of each group (first group distance (0, 72), then (72,144) ). The y-axis is the average signal bit error rate (BER) over 10 node pairs in each group, indicating received signal quality strength on receiver node side. From Figure 7 two findings can be gleaned: (a) both initial signal quality and optimized signal quality are not dependent on spatial distance between transmitters and receivers; (b) optimization efficiency is also not dependent on spatial distance but related to the initial conditions. 6

7 Figure 7. Optimization on link signal quality single objective optimization The first finding is explained by the fact that the signal quality at receiver side is heavily affected by the urban environment, so it cannot be predicted using a free-space transmission model. In simulation, the signal quality is a proportional to two factors: the number of streets and the length of the streets. The initial bit error rate, which is an average of 10 receiving node signal qualities, is a random value because the links between starting and receiving node pairs are randomly chosen with randomly generated bit strings. The second finding further strengthens the opinion that the optimization target is not dependent on spatial distance, but relies on initial link selection. Figure 7 shows optimization curves after 4 rounds, 8 rounds and 20 rounds, respectively, and these retain a similar profile to the initial curve but shifted down in BER. This finding indicates that choosing a good initial point will accelerate the optimization process Optimization on signal quality and sensor numbers The search space defined in algorithm contains two factors: signal quality indicator (BER) at the receiver side and the number of sensors used on the link. These may be plotted to form an optimization factor space as shown in Figure 8. Figure 8. Optimization factor space single objective optimization 7

8 In Figure 8, the fitness function is single objective, so only link signal quality is optimized. As optimization advances, the average link signal quality decreases. However, the x-axis: number of sensors is not optimized. Figure 9. Optimization factor space multiple objective optimizations In Figure 9, the fitness function has multiple objectives: 1 Fitnessn log10( FitnessSensor, n ) (2) log10( FitnessBER, n) The optimization progress evolves as a clustering of factor space, which means both the link signal quality and the number of sensors are optimized. From observing the results in Figure 9, we can conclude that SRA is optimizing its routing decision to clustering thus eliminating both the worst and the best routing solutions. It leaves those solutions with tolerable signal quality and sensor number usage. This simulation proves that SRA is able to manage multi-objective optimization tasks by modifying its fitness function. 5. Conclusions and future work In this paper, we have proposed a novel wireless sensor network to assist vehicular network point-to-point transmission. A novel routing algorithm, SRA, was defined on this network infrastructure. By forming sensor triangles, SRA will locate multiple candidate routes satisfying transmission requirements and then optimize signal quality or hop number for power efficiency. Our simulation results show that compared with the mainstream ad hoc routing protocol AODV, SRA is superior in both route searching and link signal quality optimization. SRA is also able to manipulate optimization to multiple-objective tasks, for example to find an optimal link with guaranteed signal quality using the minimum number of hops to save the energy. A potential urban wireless communication optimization extension is stochastic geometry [24], capturing the urban streets using one Poisson point process model and the survival of inter-node communications by another. SRA could use this survival rate within its optimization objective thus capturing both mathematical expectation and realistic routing information. 6. References [1] D. Johnson, and D. Maltz, Dynamic Source Routing in Ad hoc Wireless Networks, Mobile Computing, pp : Springer US [2] M.K. Marina and S.R. Das, Ad-hoc on-demand multi-path distance vector routing, Wireless Communication, Mobile Computing, vol. 6, no. 7, pp , Dec

9 [3] D.A. Magder, P. Bosch, T. Klein, P. Polakos, L. Samuel, H.Viswanathan, 911-NOW: a network on wheels for emergency response and recovery operations, Bell Labs Technical Journal, vol. 11, no. 4, 2007: pp [4] W. Enkelmann, FleetNet-applications for inter-vehicle communication, Proceedings of the IEEE Intelligent Vehicles Symposium, 2003: pp [5] D. Reichardt, M. Miglietta, L. Moretti, P. Morsink, W. Schulz, CarTALK2000: safe and comfortable driving based upon inter-vehicle communication. Intelligent Vehicles Symposium, Proceedings. IEEE, 2002: pp [6] F. Li and W. Yu, Routing in vehicular ad hoc networks: A survey. IEEE Vehicular Technology Magazine, vol. 2, no. 2, 2007: pp [7] Y. Toor, P. Muhlethaler, and A. Laouiti, Vehicle Ad Hoc networks: applications and related technical issues, IEEE Communications Surveys and Tutorials, vol. 10, no. 3, 2008: pp [8] P. Lalwani, S. Silakari, P. K. Shukla, Optimized and Executive Survey on Mobile Ad-hoc Network, Proceedings of the International Symposium on Cloud and Services Computing (ISCOS), 2012: pp [9] W. Chen, R. K. Guha, T. J. Kwon, J. Lee, Y.-Y. Hsu, A survey and challenges in routing and data dissemination in vehicular ad hoc networks, Wireless Communications and Mobile Computing, vol. 11, no. 7, 2011: pp [10] M. Torrent-Moreno, F. Schmidt-Eisenlohr, H. Füßler, H. Hartenstein., Packet forwarding in VANETs, the complete set of results. Technical report , ISSN , Germany. [11] Q. Sun, S.Y. Tan and K.C. Teh, Analytical formulae for path loss prediction in urban street grid microcellular environments, IEEE Transactions on Vehicular Technology, vol. 54, no. 4, 2005: pp [12] P. Ngor, S. Y. Tan, P. H. J. Chong, Routing performance of mobile ad hoc network in urban streetgrid environment using non-line-of-sight propagation model, Proceedings of 8th International Conference on Information, Communications and Signal Processing (ICICS), 2011: pp [13] L. Zhang, A. Lakas, H. El-Sayed, E. Barka, Mobility analysis in vehicular ad hoc network (VANET), Journal of Network and Computer Applications, vol. 36, no. 3, 2013: pp [14] J.Y. Boudec, S. Giordano, A location-based routing method for mobile ad hoc networks, IEEE Transactions on Mobile Computing, vol. 4, no. 2, 2005: pp [15] I. Skog, P. Handel, In-Car Positioning and Navigation Technologies A Survey, IEEE Transactions on Intelligent Transportation Systems, vol. 10, no. 1, 2009: pp [16] I.F. Akyildiz, A survey on sensor networks, IEEE Communications Magazine, vol. 40, no. 8, 2002: pp [17] M. Radi, B. Dezfouli, K. Abu Bakar, M. Lee, Multipath Routing in Wireless Sensor Networks: Survey Research Challenges, Sensors, vol. 12, no. 1, 2012: pp [18] J. Kulik, W.R. Heinzelman, and H. Balakrishnan, Negotiation-based protocols for disseminating information in wireless sensor networks, Wireless Networks, vol. 8, no. 2/3, 2002: pp [19] J. A. Stankovic, T. He, Energy management in sensor networks, Philosophical Transactions of the Royal Society A, vol. 370, no. 1958, 2012: pp [20] P. Jindal and V. Gupta, Study of Energy Efficient Routing Protocols of Wireless Sensor Networks and their further researches: a Survey, International Journal of Computer Science and Communication Engineering, vol. 2, no. 2, 2013: pp [21] S. K. Singh, M. P. Singh, D. K. Singh, Routing Protocols in Wireless Sensor Networks A Survey, International Journal of Computer Science & Engineering Survey, vol.1, no.2, 2010: pp [22] K. Zeng, K. Ren, W. Lou, P. J. Moran,Energy aware efficient geographic routing in lossy wireless sensor networks with environmental energy supply, Wireless Networks, vol. 15, no. 1, 2009: pp [23] E. Alotaibi, B. Mukherjee, A survey on routing algorithms for wireless Ad-Hoc and mesh networks, Computer Networks, vol. 56, no. 2, 2012: pp [24] W. Guo, T.O. Farrell, Relay Deployment in Cellular Networks: Planning and Optimization, IEEE Journal on Selected Areas in Communications, vol. 30, no. 8, 2012: pp