A New Channel Assgnment Scheme for Interference-Aware Routng n Vehcular Networs Peppno Fazo, Florano De Rango, Cesare Sottle D.E.I.S. Department, Unversty of Calabra 87036, Rende, Italy e-mal: {pfazo, derango}@des.uncal.t sottlecesare@lbero.t Carlos Calafate D.I.S.C.A. Department, Unversdad Poltecnca de Valenca 460, Valenca, Span emal: calafate@dsca.upv.es Abstract Moble computng and vehcular communcatons are becomng a very mportant paradgm for wreless communcatons, manly because of ther ablty to adapt to dfferent moble applcatons. In ths paper, we propose a new scheme for reducng the nterference level durng moble transmssons n the VehculAr nter-networng (VANET) envronment, tang the advantage of the mult-channel nature of IEEE80.11p standard. In order to releve the effects of the cochannel nterference perceved by moble nodes, transmsson channels are swtched on a bass of a perodcal Sgnal-to- Interference Rato () evaluaton. The attenton s focused on the routng level of VANET and we propose an nterference aware routng scheme for mult-rado vehcular networs, wheren each node s equpped wth a mult-channel rado nterface. A new metrc s also proposed, based on the maxmzaton of the average level of the connecton between source and destnaton. Our soluton has been ntegrated wth the AODV routng protocol to desgn an enhanced Sgnal-to- Interference-Rato-AODV (-AODV). NS- has been used for mplementng and testng the proposed dea, and sgnfcant performance enhancements were obtaned, n terms of throughput, pacet delvery and, obvously, nterference. Keywords-Mult-channel routng, VANET, Interference Aware Routng, 80.11p, WAVE, DSRC,. I. INTRODUCTION Nowadays, dstrbuted vehcular communcatons are consdered a promsng technology when mprovng effcency and comfort of everyday road travel. Due to the hgher sgnalng burden than the one of nfrastructure systems, communcaton protocols become very complex and, sometmes, sgnalng overhead may waste bandwdth avalablty. VANETs are able to provde wreless networng capablty n stuatons where the communcaton among nodes can be ether drect or made va relayng nodes, as n classcal ad-hoc networs. The overall perceved Qualty of Servce (QoS) strongly depends on how the routng protocol overhead affects the avalable bandwdth and on how dfferent channels are selected n order to mnmze nterference levels. In ths wor, the avalablty of dfferent communcaton channels s consdered n order to mprove the system performance. QoS routng n mult-hop wreless networs s very challengng due to nterferences among dfferent transmssons, but VANETs offer the chance to reduce them snce multple smultaneous transmssons are possble. In ths paper a new nterferenceaware routng protocol for VANET envronments s proposed, tang the advantage of a dynamc allocaton of the Dedcated Short Range Communcatons (DSRC) spectrum, n order to reduce nterference level among moble nodes. In a dstrbuted mult-hop archtecture, a moble node may potentally fnd multple routes for all the destnatons. When evaluatng networ topology through ts routng table and, n the consdered case, the avalablty of dfferent avalable channels, a protocol may enhance the qualty of communcaton. So, n ths scenaro, each node should select the best route n terms of QoS, not only consderng a typcal cost metrc (bandwdth, delay, traffc load or a combnaton of them), as n the classcal mult-hop archtecture, but tang nto account the benefts that can be obtaned f dfferent nterference levels,.e. dfferent channels, are consdered. The proposed dea s manly based on the AODV [5] protocol, whch has been properly modfed to tae nto account the chance of dynamcally changng the channel used for data transmsson. In partcular, a new metrc has been defned, based on the Sgnal-to-Interference () evaluaton on the dfferent avalable channels; the proposed routng protocol ams to choose dfferent channels, one for each hop on the path, n order to obtan a global maxmzaton for the connectons between sources and destnatons. Ths paper s organzed as follows: Secton II ntroduces an n-depth overvew on the related wor about routng n VANETs; Secton III ntroduces the consdered scenaro and the proposed protocol. Then Secton IV offers a deep descrpton of the obtaned results. Fnally Secton V concludes the dscusson. II. STATE OF THE ART AND RELATED WORKS There are many recent wors n the lterature on VANETs, focusng mostly on nvestgatng new approaches to enhance routng operatons. The behavor of the routng protocols s manly trggered by events le tmeouts and the recepton of routng messages, and the mpact that these events have on them s dfferent. For nstance, n AODV, whch s a representatve reactve routng protocol, tmeouts have a great nfluence on the route establshment and mantenance process. The one and two hop neghbor lsts of OLSR are affected by tmeouts, whch results n neffcent floodng of topology control messages as a consequence of errors n the multpont relay set calculaton. In [1] the authors evaluated the
performance of OLSR and AODV n an urban envronment, adoptng the Vehcle Moblty Model to generate realstc moblty patterns. In [7,8], authors enhanced a tradtonal MANET routng protocol (AODV) amng at mprovng route stablty and obtanng less networ overhead, thus mang AODV sutable for VANETs. Ther study showed that more approprate routes can be found wth and wthout moblty predcton. Also, poston-based routng protocols have been proposed n VANET-related lterature. B. Karp et al. [1] proposed the GPSR poston-based protocol, that forwards data pacets by consderng geographc nformaton of the nodes whch are close to destnaton. It has been shown that GPSR does not perform optmally when large cty envronments are consdered, mostly because t uses drect communcaton among nodes that can be nterrupted by obstacles. Tradtonal ad-hoc routng protocols have also been nvestgated [] through a deep performance analyss n hghway scenaros; smulaton results showed that the consdered protocols ncrease the routng load on the networ and decrease the pacet delvery rato and the end-to-end delay. The AQOR protocol [11] also mantans neghbor nformaton to ncorporate nterference, and broadcasts route requests. By usng the neghborhood bandwdth nformaton for the new flow, feasble paths are detected; the fnal choce s made at the destnaton. More recently, Zhu and Corson [4] proposed other algorthms to determne the exact schedule of slots for a flow through the networ, guaranteeng the bandwdth by tang nterference nto account. Johansson et al. [13] used NS- to smulate the ncrease of ln breas and the decrease of relablty wth hgher node speeds. It s clear that the collected smulaton results strongly depend on the mplementaton of the protocols and ther confguratons. The man contrbutons of ths paper consst manly n the proposal of a new verson of the AODV protocol, properly modfed n order to tae the neghbors nterference level nto account. III. VEHICULAR INTERNETWOKING OVERVIEW AND PROPOSED PROTOCOL A. Vehcular communcatons through VANET The IEEE 80.11p, also called Wreless Access n Vehcular Envronments (WAVE) [9], s an extenson of the IEEE 80.11 standards famly for vehcular communcatons. It ams at provdng the standard specfcatons to ensure the nteroperablty between wreless moble nodes of a networ wth rapdly changng topology (that s to say, a set of vehcles n an urban or sub-urban envronment). The DSRC spectrum s dvded nto 7 channels, each one wth a 10 MHz bandwdth; t s allocated n the upper 5 GHz range, as shown n Fg. 1. Fgure 1. Drect Short Range Communcaton spectrum allocaton. VANET provdes wreless communcaton among vehcles and vehcle-to-road-sde equpments. The PHY layer employs 64-subcarrer OFDM. 5 out of the 64 subcarrers are used for actual transmsson consstng of 48 data subcarrers and 4 plot subcarrers. Possble modulaton schemes are BPSK, QPSK, 16-QAM and 64-QAM, wth codng rates equal to 1/, 1/3, 3/4 1/, 1/3, ¾ and an OFDM symbol duraton of 8μs. The WAVE standard reles on a mult-channel concept whch can be used for both safety-related and entertanment messages. The standard accounts for the prorty of the pacets usng dfferent Access Classes (ACs), havng dfferent channel access settngs. Ths shall ensure that hghly relevant safety pacets can be exchanged tmely and relably even when operatng n a dense urban scenaro. Each staton contnuously alternates between the Control Channel (CCH) and one of the Servce Channels (SCHs) or the safety channels. The MAC layer n WAVE s equvalent to the IEEE 80.11e Enhanced Dstrbuted Channel Access (EDCA) Qualty of Servce (QoS) extenson. Fgure. A typcal urban VANET scenaro. Therefore, applcaton messages are categorzed nto dfferent ACs, where AC0 has the lowest and AC3 the hghest prorty. Wthn the MAC layer a pacet queue exsts for each AC. Fg. shows a typcal VANET scenaro. An mportant ssue n VANET s the choce of an approprate transmsson channel, not only consderng the type of traffc (emergency, securty, platoonng, etc.) but, manly, focusng on the reducton of the nter-node nterference. The man contrbutons of ths paper consst manly n the proposal of a new verson of the AODV protocol, properly modfed n order to tae the neghbors nterference level nto account. The evaluaton of the new metrc s based on: - Management of the mult-channel capablty of the WAVE standard at the routng level through a hgherlevel channel selecton, whch s based on a nterference-aware algorthm; - Implementaton of a -ray propagaton model n order to tae path-loss between Transmtter and Recever nodes nto account; - Perodcal Sgnal-to-Interference Rato () estmaton on the avalable transmsson channels; - Defnton of a threshold value n order to choose f a new transmsson channel must be selected; - Transmsson of synchronzaton pacets n order to advse the recevng node of a new channel selecton. B. Sgnal-to-Interference-Rato-based-AODV (-AODV) Our attenton s focused on the networ layer of a VANET, and t s assumed that the channel router of the WAVE MAC layer s able to analyze the LLC data unt n order to choose the rght prorty queue. As n the tradtonal scheme, the path dscovery process s ntated whenever a source node needs to communcate wth another node for whch t has no routng nformaton n ts table. When a new node enters nto the networ t dscovers ts neghbors through the broadcastng of
HELLO messages. The source node ntates path dscovery by broadcastng a Route REQuest (RREQ) pacet to ts neghbors. If a neghbor can satsfy the RREQ, t sends a Route REPly (RREP) bac to the source; otherwse the RREQ s forwarded agan. So, the proposed protocol called Sgnal-to-Interference- Rato-AODV (-AODV) has the bass of the AODV [5], from whch t nherts control pacets and pacet exchange procedures. HELLO messages n -AODV have the same meanng of those n the tradtonal protocol, and so they are broadcasted n the coverage area n order to now the dentty of neghbor nodes and to valdate the avalablty of lns. 1) Proposed nterference-aware metrc and assumptons The novelty of the proposal conssts n the adopted metrc for the choce of the optmal route from source to destnaton, and n the route mantenance procedure: t s not based on the mnmum hop count, as for the tradtonal AODV, but on the nterference concept, as explaned later. Also, t was necessary to modfy some control pacets. In partcular, the MAX feld s added to both RREP and RREQ pacets. The -AODV s based on the followng assumptons: - Data pacets can be delvered on sx Servce CHannels (SCH - 17,174,176,180,18 and 184), whle sgnalng ones are transmtted only on the Control CHannel (CCH - 178); - Each node can transmt/receve on one channel, so no smultaneous transmssons per node are allowed; - Each node s equpped wth a sngle nterface (wth multple channels); - Channel synchronzaton tme s related to the sgnalng pacets delvery delay, needed for channel swtchng among a couple of nodes; - The tme needed for channel swtchng s neglgble (n terms of the 80.11p MAC mplementaton, the channel router only has to forward data unts to a dfferent queue). For the -AODV, t s also supposed that a node nows exactly the level on the avalable channels for each neghbor and pacet transmsson over the fnal optmum path from a source node n S to a destnaton node n D wll be made usng a set of channels that mnmze the nter-node nterference, achevng better sgnal qualty durng the consdered sesson. The proposed metrc s based on the evaluaton of the nterference level among a couple of nodes, so an overvew on the consdered channel model should be gven. calculaton bascally conssts n the evaluaton of the receved sgnal power, and t s determned by the transmsson power and the rado propagaton condtons. Pathloss effects are domnant n VANET envronments because channel codng and frequency nterleavng mae the bt error performance of an OFDM ln n a frequency-selectve channel depend more on the average receved power than on the power of the weaest subcarrer [10]. Although real moble nodes can drectly evaluate the receved power va hardware, t s necessary to have an analytcal model, for smulaton purposes; so, for a generc recever node, we can consder the receved power P r s [3]: Pt Pr = γ d (4π ) 4πh 1 + α + α cos dλ, (1) λ where P t s the transmsson power, λ s the wavelength of the propagatng sgnal, d s the dstance between the transmtter and the recever, h s the antenna heght, α s the reflecton coeffcent of the ground surface and γ s the path-loss factor. Once a node s able to evaluate the receved sgnal power, the calculaton of for each channel can be carred out. Let us suppose that moble node n needs to evaluate the level on channel due to the presence of n neghbors n ts coverage area, then: t = n 1 P r j j = 0, () where P t s the transmsson power of n and P r j s the receved power from neghbor node j on channel. ) Dynamc channel swtchng Fg. 3 shows an example of the path dscovery mechansm n -AODV: the source node n S sends the RREQ to ts neghbors n A, n B and n C for a path towards destnaton node n D. Snce n A, n B and n C now a path to the destnaton, they wll answer wth a RREP, contanng the maxmum achevable value and the assocated channel. A B C Fgure 3. Path dscovery procedure n -AODV. When node n S receves these answers, t wll decde to store n ts routng table the next-hop n B, snce t has assocated the hghest value. The followng pseudo-code s executed perodcally by each node (every Δ seconds); T denotes the number of avalable channels for the WAVE nterface, C denotes the number of neghbor nodes and δ an nput threshold that represents the mnmum level that must be granted on each selected channel. CHANNEL ADJUST ROUTINE - For all neghbors, update the receved power level P r and store t n a vector of dmenson C; - For each avalable channel, evaluate the level through eq.() and store the values n a vector of dmenson T; - If c s the actve channel and [c]< δ: Send a Change REQuest (CREQ) pacet to a neghbor on channel c; once the CREQ has been receved, the neghbor node reples wth a Change REPly (CREP) pacet, as acnowledgement. P E
Once channels have been assgned, they need to be perodcally refreshed n order to change the assgnment f needed. Fg. 3 shows on the rght the structures of CREQ and CREP messages; CREQ s the same as a RREQ, but only the CHAN feld s used n order to mae the recever aware about the new channel; CREP contans the ACK feld to acnowledge the swtchng on the new channel. 3) Pacets formats and next-hop selecton A source node n S whch s unaware of the best path to destnaton n D can ntate the path dscovery procedure by sendng a RREQ message to ts neghbors. RREQ RREP Fgure 4. Sgnalng pacets and felds n -AODV. When a node n receves the RREQ for destnaton node n D and no entry for n D s present n ts routng table, t modfes and forwards the RREQ pacet to ts r neghbor nodes {n 1, n,, n n } nsertng the nformaton about the best value measured on the avalable channels, denoted as K MAX, so from eq. (): MAX = max { } On the other hand, f node n has nowledge of a path towards n D, t answers wth a RREP pacet, gvng to node n S the nowledge of the average along the path towards n D. Fg. 3 shows on the left the structure of the RREQ and RREP pacets: n addton to the tradtonal AODV felds, and CHAN felds have been added to them; they are used by a node when forwardng the pacet and when the recever must be aware about MAX. If P(n,n D )= {l 1,l,,l m } s the best path, n terms of a lst of lns from n K to n D, values on lns l,, =1 m, are nown snce they have been evaluated through eq. 3. Thus, each ntermedate node n the path dscovery procedure nows the average, denoted wth AVG_RREP, whch s evaluated as follows: m l MAX l l=, (4) AVG RREP = 1 _ m where l MAX s the evaluated on the l-th ln l (belongng to node n Kl ). In ths way, each node has the nowledge about the average towards a destnaton f a partcular next-hop s chosen durng forwardng operatons. CREQ CREP (3) IV. PERFORMANCE EVALUATION The protocol proposed n Secton III.B has been mplemented n the NS smulator; frst of all, the QoS MAC of IEEE80.11e has been ntroduced and then t has been extended n order to nclude all the functonaltes of the multchannel IEEE80.11p standard. Dfferent classes (.h and.cc) have been created or modfed (mac_80_11e, aodv, aodv_pacet, aodv_rtable, pacet, ntermod) and a practcal OTcl scrpt has been mplemented n order to have the opportunty of smulatng dfferent scenaros. The CtyMob generator [6] has been used to create Manhattan patterns, wth the followng parameters: map dmensons 1000m x 1000m, maxmum vehcle speed 15 m/s, number of damaged vehcles 3, downtown area 400m x 400m. The path-loss has been consdered through eq.1, wth γ=4, α=0.1 and h=1.5 m [3]. Transmsson rate has been fxed at 3Mbps, and the number of moble nodes vares from 0 to 80. Many smulaton runs have been carred out n order to determne the optmal value of some smulaton parameters. The number of concurrent connectons vares from to 10 but, due to space constrants, only results for 4 and 10 are shown (as n the captons of fgures); Δ has been fxed to 60ms and δ has been fxed to 10 9 : the chosen values of Δ and δ led to the best results for the consdered parameters. The -AODV protocol has been compared to the tradtonal AODV (AODV SINGLE n the captons) and the tradtonal AODV wth a random channel selecton (AODV MULTI RANDOM n the captons). Fg. 5 shows the average aggregated throughput of the networ (the total amount of bts receved by all nodes durng smulaton tme): t can be seen how t decreases for hgher number of moble nodes. Ths s mostly due to the hgher overhead burden, although the -AODV outperforms the classcal schemes AODV SINGLE/MULTI and a consderable gan (about Mbps) has been obtaned. If the number of connectons s low, e.g. 4, the system s under-utlzed, whle for a hgher number of actve connectons the throughput s near the maxmum achevable one. 3 1 19 17 15 13 11 Throughput (Mbps) 9 4 --AODV 4 -AODV MULTI 4 -AODV SINGLE 10 - AODV SINGLE Fgure 5. The average throughput (Mbps) for the smulated networ. PDR (%) 100 95 90 85 80 75 70 65 60 4-AODV MULTI RANDOM 4-AODV SINGLE 10 - - AODV 10-AODV MULTI RANDOM 10-AODV SINGLE Fgure 6. The average Pacet Delvery Rato (PDR) for the smulated networ.
Fg. 6 shows how the protocols perform n terms of Pacet Delvery Rato (PDR): when the VANET accommodates a hgher number of vehcles, as well as a hgher number of concurrent connectons, PDR decreases ndependently of the adopted routng scheme but, also n ths case, the -AODV has better performance and an enhancement of about 8% s reached. Fg. 5 and Fg. 6 demonstrate how an nterferencebased metrc can ncrease the performance of the system: collsons and nterference errors are heavly reduced. 9 8 7 6 5 4 3 1 0 Fgure 7. The average overhead of the -AODV. The ntroducton of the perodcal channel refresh leads to the exchange of CREQ and CREP messages as ntroduced n secton III. Fg. 7 shows how the ncreasng of the -AODV overhead (evaluated as the rato between the number of sgnalng pacets and the number of total pacets) s neglgble (near to %), when compared wth tradtonal schemes. It ncreases for hgher number of connectons number and moble nodes. AODV SINGLE and MULTI have the same overhead performance because no new messages are ntroduced n the MULTI case, but only a random selecton of a transmsson channel. 50 00 150 100 50 Overhead (%) 4 - AODV MULTI RANDOM 4 - AODV SINGLE 10 - AODV SINGLE Normalzed 0 4 - AODV MULTI 4 - AODV SINGLE Fgure 8. The average perceved by moble nodes. Fg. 8 llustrates the enhancement ntroduced n the average perceved (evaluated as n eq. and normalzed to the value of 10 10 ) wth the adopton of -AODV. When tradtonal routng schemes are employed, the routng protocol acts by gnorng levels and nterference problems, so the values of normalzed (near to 0) llustrated n the fgure are obtaned. Clearly, values ncrease for lower numbers of concurrent connectons. The dfferences wth the -AODV are evdent, although the trend s decreasng when the number of vehcles ncreases. When the number of nodes s too hgh the number of avalable channels s lmted (sx n the consdered case) so the nterference cannot be heavly reduced. V. CONCLUSIONS A new routng protocol for VANET envronments, - AODV, has been proposed. It s based on the tradtonal sgnalng scheme of AODV, but taes advantage of a dynamc allocaton of the DSRC spectrum, n order to reduce nterference levels among nearby moble nodes. A new metrc based on the recurrent evaluaton of the level on the dfferent lns from sources towards destnatons has been proposed; t gves the opportunty to choose the next-hop n routng operatons dependng on the best perceved value on the ln. An mplementaton for the NS smulator has been developed, and vehcular moblty has also been taen nto account. Despte of a neglgble ncrease n terms of protocol overhead, smulaton results have shown that there are good enhancements n terms of throughput, pacet delvery rato and normalzed. VI. REFERENCES [1] Jerome Haerr, Feth Flal, Chrstan Bonnet, Performance Comparson of AODV and OLSR n VANETs Urban Envronments under Realstc Moblty Patterns, Proc. of Med-Hoc-Net 006, 5th IFIP Medterranean Ad-Hoc Networng Worshop, June 14-17, 006, Lpar, Italy. [] Xong We, L Qng-Quan, Performance evaluaton of data dssemnatons for vehcular ad hoc networs n hghway scenaros, The Internatonal Archves of the Photogrammetry, Remote Sensng and Spatal Informaton Scences. Vol. XXXVII. Part B1, Bejng 008. [3] Y. Zang, L. Stbor, G. Orfanos, S. Guo, H.J. Reumerman, An Error Model for Inter-Vehcle Communcatons n Hghway Scenaros at 5.9GHz, WASUN 05, October 005, Montreal, Quebec, Canada. [4] C. Zhu and M. S. Corson, QoS Routng for Moble Ad Hoc Networs, Proceedngs INFOCOM 00, New Yor. [5] Charles E. Perns, Elzabeth M. Bldng-Royer, Samr Das Ad Hoc On-demand Dstance Vector (AODV) Routng, IETF Internet Draft, November 001. [6] Martnez, F.J. Cano, J.-C. Calafate, C.T. Manzon, P., CtyMob: A Moblty Model Pattern Generator for VANETs, Communcatons Worshops, 008. ICC Worshops '08, Bejng, 19-3 May 008. [7] O. Abed, R. Berang, and M. A. Azgom, Improvng Route Stablty and Overhead on AODV Routng Protocol and Mae t Usable for VANET, Dstrbuted Computng Systems Worshops, 009. ICDCS Worshops '09. 9th IEEE Internatonal Conference on pp. 464 467, -6 June 009. [8] M. Fathy, and J. Taghloo, Enhancng AODV routng protocol usng moblty parameters n VANET, Computer Systems and Applcatons, 008. AICCSA 008. IEEE/ACS Internatonal Conference on pp. 9-35, March 31 008-Aprl 4 008. [9] Tas Group p, IEEE P80.11p: Wreless Access n Vehcular Envronments (WAVE), draft standard ed., IEEE Computer Socety, 006. [10] Sbecas, S., Corral, C. A., Emam, S., Strats. G. and Rasor, G. Pseudoplot OFDM scheme for 80.11a and R/A n DSRC applcatons. Vehcular Technology Conference, 003. VTC 003-Fall. 003 IEEE 58 th. [11] Q. Xue and A. Ganz, Ad hoc QoS on-demand routng (AQOR) n moble ad hoc networs, Journal of Parallel Dstrbuted Computng vol. 63, pp. 154-165, 003. [1] Karp, B. and Kung, H. T., GPSR: greedy permeter stateless routng for wreless networs, In Proceedngs of the 6th Annual nternatonal Conference on Moble Computng and Networng (Boston, Massachusetts, Unted States, August 06-11, 000). MobCom '00. ACM, New Yor, NY, pp. 43-54. [13] P. Johansson, T. Larsson, N. Hedman, B. Melczare, and M. Degermar, Scenaro-based Performance Analyss of Routng Protocols for Moble Ad-Hoc Networs, n Proceedngs of 5th ACM SIGMOBILE, pp. 195-06,Seattle, USA, August 1999.