Packe-Oriened Communicaion Proocols for Smar Grid Services over Low-Speed PLC Michael Bauer, Wolfgang Plapper, Chong Wang, Klaus Doser Insiue of Indusrial Informaion Technology Universiä Karlsruhe (TH) Karlsruhe, Germany E-mail: bauer@iii.uni-karlsruhe.de Absrac Advanced Meering Infrasrucures require reliable communicaion links. This paper proposes a proocol sack for smar meering based on he parameers of a robus low-speed PLC physical layer. The proocol sack under consideraion implemens IPv6 on op of a conenion-free media access conrol sub-layer, hereby allowing for maximum flexibiliy. Simulaion resuls show ha his proocol sack allows daa ransfers a accepable laencies, despie low daa raes a he physical layer. Keywords IPv6, proocols, power line communicaion, smar meering I. INTRODUCTION ITH he increasing awareness regarding he scarciy of Wresources, he need arises for new echnologies leading o a more efficien end-use of energy [1]. In his conex, erms like Auomaic Meer Reading (AMR), Smar Meering [2], and Advanced Meering Infrasrucure (AMI) become increasingly imporan. Such erms commonly imply he need for cross-linking spaially disribued devices like elecronic meers and an infrasrucure for daa collecion and processing. For such auomaion echnological applicaions, highly reliable and available communicaion sysems are indispensable. Power line communicaion seems desirable as an access echnology for communicaion [3]. However, i is difficul o guaranee communicaion sysem reliabiliy and availabiliy under exremely difficul channel condiions as offered by he PLC ransmission channel [4],[5]. The channel properies provide numerous challenges when designing a proocol sack for low-speed PLC. Addiionally, applicaion-specific design guidelines need o be aken ino accoun, which can be challenging, if neiher sandardized daa formas are available nor he amoun of daa o be ransferred is known precisely. Our approach o proocol design focuses on economic efficiency, flexibiliy, and scalabiliy. For he upper layers of he OSI model [6], we use IPv6 a he Inerne Layer and TCP for he Transpor Layer. A he Daa Link Layer, we use polling as a conenion-free MAC sraegy, Auomaic Repea- Reques (ARQ) and CRC for error conrol. The Physical Layer is based on OFDM, aiming a maximum reliabiliy. The overall communicaion proocol sack is simulaed and is performance evaluaed by simulaion in OPNET Modeler. II. PHYSICAL LAYER Due o he unfavorable properies of he PLC ransmission channel, he physical (PHY) layer does no allow for daa raes comforable for proocol design. Modems can feaure a maximum of ens of kbi/s, if he sysem is designed for robusness and reliabiliy. This especially holds rue, if he frequency range beween 9 khz and 95 khz [6] is used. The fac ha his is only a narrow frequency band compared o broadband PLC is even worsened by he fac ha i is only useful o uilize he upper frequencies of his specrum due o he PLC channel properies. OFDM is an appropriae mehod for modulaion, as explained in [7]. The problem of robus OFDM frame synchronizaion is solved by using he zerocrossings of he mains volage, as described in [8]. Focusing on robusness, a low-speed PLC PHY can operae wih he parameers in Table I, aken from [7]. TABLE I LOW-SPEED PLC PHY PARAMETERS Parameer Type / Value Modulaion Scheme OFDM (DBPSK) Carrier Spacing 325.5 Hz Number of Carriers 48 Ne Bi Rae Approx. 11 kbi/s III. LAYER-2 PROTOCOL MAC proocols specify a resource sharing sraegy by conrolling he access of muliple users o a shared ransmission medium. Our design of a MAC proocol for PLC aims a avoiding collisions, herefore we selec polling as he preferred MAC sraegy. Due o is deerminisic behavior, polling avoids unraceable complicaions caused by he socalled hidden-node problem [1]. A. Topology The proocol is designed for a sysem where ime allocaion o cusomers wihin one low volage cell is managed by a cenral uni. We denoe his daa concenraor as he maser, whereas he meers wihin he cell are slaves assigned o he maser. The proocol suppors bidirecional communicaion beween maser and slaves. We disinguish so-called downlink (maser o slave) and uplink (slave o maser) modes. Wihin 978-1-4244-379-/9/$25. 29 IEEE 89
one cell, N slaves are conneced o one maser in a logical sar configuraion. Slaves may no exchange daa direcly beween each oher. However, every slave may communicae wih oher slaves under he adminisraion of he maser. B. Polling sraegy The maser polls he slaves in a cyclic order (i.e. 1, 2,, N, 1, 2 ). Afer compleing a visi o slave i, he maser wais for a specified ime. The period during which he maser coninuously serves slave i is called a service period of slave i, he subsequen period is called a swich-over period for slave i. The sum of service and swich-over period is defined as he maser-slave-period (M-S-period). For more deailed informaion and analysis mehods, we refer o [11]. Fig. 1 displays he separae uplink and downlink polling cycles. acknowledgemen (NAK) from he r indicaes ha he ransmied packe has been deeced as fauly, reransmission is required as a resul. Fig. 3 Sop-and-Wai ARQ D. Frame srucure Fig. 4 depics he srucure of a MAC frame used in our model. Table I shows he corresponding frame ypes. Fig. 1 Downlink and uplink cycle Wih increasing number of slaves, he ime beween wo consecuive grans of access o he communicaion medium becomes longer for each of he slaves. To mee real-ime requiremens, a gaed polling policy is preferred over exhausive or limied-1[11]. Fig. 2 depics he N-h maserslave period schemaically. The service period is limied by a imer a boh maser and slave. Fig. 2 N-h Maser-Slave period C. ARQ Error handling ARQ is preferred over FEC for error conrol. There are hree basic ypes of ARQ proocols including Sop-and-wai-, Go- Back-N- and Selecive Repea ARQ. Because he laer wo ypes of ARQ proocols require an independen reverse channel, hese ARQ ypes are no suiable for PLC. Fig. 3 shows he paern of Sop-and-Wai ARQ. A posiive acknowledgemen (ACK) from he r indicaes ha he ransmied packe has been d successfully, he ransmier may he nex packe in is queue. A negaive Fig. 4 MAC frame srucure The flag ID should only be se if he frame ype is DATA, ACK or NAK. RA, TA and URA always have o be se. DB is only se if he frame ype is DATA. The lengh of he CRC field depends on he frame ype and varies beween 9 and 14 bis. The oal lengh of he MAC frame wihou he Daa Block field is 178 bis, he maximum lengh is resriced o 813 bis including daa. In his case, he payload conribues 97.74% o he oal frame lengh. TABLE I POSSIBLE "FRAME TYPES" Frame Type Funcion POLL RTS CTS DATA ACK NAK MASTER END SLAVE END Token, Polling Reques o Clear o Daa payload Posiive acknowledgemen Negaive acknowledgemen Timeou a maser Timeou a slave E. Timing during a Maser-Slave-Period The iming diagram of one M-S-period in downlink and uplink mode is depiced in Fig. 5 and Fig. 6, respecively. Beween he ransmission of a POLL frame and he exchange of user daa, an RTS/CTS handshake is used o ensure ha he channel is free, hereby avoiding deadlocks. In downlink mode, he POLL-frame performs he same ask as an RTS. 9
F. Collision avoidance As shown in Fig. 7, collisions may happen a he insan he slave s imer expires. If he slave immediaely s a SLAVE END frame, he incoming ACK or DATA frame will clash. Fig. 7 shows how hese collisions are avoided by he slave delaying he ransmission of he SLAVE END frame. Maser Am Slave UPLINK Timeou (Slave) Am Collision Maser Am-1 Slave Am-1 DOWNLINK Timeou (Slave) Collision D DATA A ACK SLAVE END Fig. 7 Possible collisions and avoidance sraegy Fig. 5 Timing Diagram for uplink mode A imer clock is sared a he insan he maser s a POLL frame. Anoher imer is sared a he slave node a he insan i s he POLL frame. The slave imeou is defined for a shorer period han he maser imeou in order o avoid collisions. If he imer a he maser (slave) expires, i s a MASTER END (SLAVE END) frame. The MASTER END (SLAVE END) may also be sen before he imer expires. This happens in he case when he maser (slave) has no packes queued for ransmission in downlink (uplink) mode. Fig. 6 Timing Diagram for downlink mode IV. HIGHER LAYERS A. Nework Layer On he nework layer, he Inerne Proocol in version 6 (IPv6) [13] is employed due o is larger address space compared o IPv4 and is convenien hierarchical addressing scheme. In [12], he minimum hos requiremens of IPv6 for Low Cos Nework Appliances (LCNA) are discussed. As he inended applicaion relies on embedded sysems wih limied resources, i.e. elecronic meers, he rules for such end nodes apply. Address resoluion from IPv6- o MAC addresses is provided by he Neighbor Discovery Proocol ha srongly relies on he broadcas abiliy of he MAC layer [14]. To reduce he communicaion effor from N² messages necessary o resolve MAC- and IPv6 addresses of all N slaves, a sar-up phase has been implemened in which every slave s an Unsolicied Neighbor Adverisemen o he maser s MAC address. Addiionally, he maser acs as a public Neighbor Cache ha answers o Neighbor Soliciaions wih auhoriaive, valid Neighbor Adverisemens. This is possible as any slave-o-slave communicaion has o be handled by he maser, and helps reducing Neighbor Discovery-relaed broadcass o a minimum. B. Transpor Layer On he ranspor layer, he Transmission Conrol Proocol is employed, providing a communicaion service wih congesion conrol, fragmenaion, checksum calculaion and connecion racking. The correc order of packes is ensured wih sequence numbers and deadlocks are avoided by means of 91
imers. Due o selecive reransmission mechanisms, Fragmenaion on TCP layer is more efficien han on he nework layer: if a fragmen on ranspor layer is los, only he missing daagram needs o be reransmied. C. Applicaion Layer A very simple file ransfer proocol is employed for he ransmission of user daa. I is considered TFTP-based [15], relying on TCP. A small reques daa packe is sen from he maser o every slave. Every slave hen answers wih is meering daa packes. Separae acknowledgemen packes for every daagram are no necessary a applicaion layer, as he correc recepion is already confirmed by TCP ACKs. TABLE II ND PARAMETER TTINGS Consan specified value used value Rouer Lifeime -9 s 9 s Rouer Adverisemen Reransm. 3 3 Minimum Delay Beween Rouer > 3 s, Adverisemens < 18 s 18 s Maximum Rouer Adverisemen Delay Time -.5 s -.5 s Rouer Reachable Time < 36 432 s Maximum Neighbor Soliciaions 3 3 Maximum Neighbor Adverisemens 3 3 Node Reachable Time 3 ms 432 s Neighbor Soliciaion Reransmission Timer 1 ms 3-9 s V. SIMULATION A. Sysem Capaciy Esimaion The oal number of slaves and he amoun of user daa ha can be ransmied by each slave can be derived numerically from an empirical formula which approximaes waiing and handshaking imes by aking ino accoun he average bus uilizaion bus_uil. Wih N slaves and n requess per hour, he amoun of daa on MAC layer (mac_amoun bis consising of MAC-, IPv6-, TCP header and user daa) and a daa rae of bus_speed in bi/s of he bus, his yields equaion (1). N mac _ amoun n daaraio = (1) bus _ speed bus _ uil 36 The value of daaraio=1 represens a sysem operaing under full load. In order o ensure sysem sabiliy, equaion (1) should reurn values less hen.85, which has been verified by simulaion. B. Seings All simulaions were performed in OPNe Modeler (Educaional Version 14.5, Build 7545 32-bi). The scenario was chosen o be on he edge of sabiliy according o equaion (1) wih he following parameers: 1 slaves are conneced via a bus link wih a defined daa rae of 11 bi/s, according o he parameers of he physical layer described in Secion II. To ensure represenaive sabiliy analysis, all simulaions are performed over a period of 24 hours ha sars immediaely wih he sar-up phase. To obain comparable runime delay values of IP packes, he maser s an ICMPv6- Echo Reques packe conaining 64 byes of daa o all slaves afer 9 seconds. Afer one hour, he acual meering process sars, which gives he sysem enough ime o sabilize afer he Neighbor Discovery and delay deerminaion phase. All slaves are polled every 15 minues by he maser wih a 256 byes reques packe on he applicaion layer ha is answered by 55 byes of meering daa by each slave. The user daa ransmission process has o be compleed in less han 9 seconds (15 minues), afer which he nex reques is sen. The maximum packe size on MAC layer is se o 99 byes. This is a compromise beween hroughpu (which is affeced by MAC overhead) and he ime lef for one slave o occupy he medium wihou any possibiliy for oher ineracion. Anoher imporan poin is he opimum packe size as a base for CRC algorihms on he MAC layer. The 55 byes of user daa lead o 6 fragmens per meering daa se per slave which have o be ransmied and acknowledged on he MAC layer. The iming parameers on he MAC layer allow he ransmission of 2 such packes per maser-slaveperiod, he maximum number of unacknowledged packes is one ( sop and wai -algorihm). Timing seings on TCP layer have o be adaped o enable connecion esablishmen despie he large delays beween consecuive packe inerchanges. Therefore, he iniial reransmission imeou is se o 6 seconds. Timing seings for he Neighbor Discovery Proocol (see Table II) are se o limi broadcas raffic and reransmissions o a minimum, as permanen opology changes in he observed nework segmen are considered o be rare evens. Nodes being unavailable is assumed o be a emporary sae. The expiraion ime of rouer enries in he rouer cache of slaves violaes he defined values, bu as he whole sysem breaks down if he maser is unavailable, his can be acceped wih he goal o reduce higher layer proocol raffic. For he Neighbor Soliciaion Reransmission Timer, he same delays have o be aken ino accoun as for TCP reransmission imeous. Therefore, he value is dynamically chosen beween 6 and 9 seconds. VI. RESULTS A. Bus Uilizaion The bus uilizaion is he insananeous uilizaion of he bandwidh provided by he PHY in percen. I indicaes how efficienly he proocol uses he available bandwidh. I is, for example, decreased by waiing imes on MAC layer. On he conrary, a value of 1% represens full uilizaion of he medium. A he beginning of he simulaion in Fig. 8, a value of abou 8% is shown. 92
Bus Uilizaion (%) 1 5 2 4 6 8 1 x 1 4 Fig. 8 Bus Uilizaion This represens he sar-up phase where all slaves and he maser exchange heir address informaion wih Unsolicied Neighbor Adverisemens. In he subsequen ime span, only MAC proocol daa unis are exchanged he available daa rae of 11 bi/s is fully uilized. Afer 9 seconds, he iniial IPv6 delay measuremen wih ICMPv6- Echo Reques packes is shown. Afer 36 seconds, he maser applicaion ransmis he firs reques packe which leads o a TCP handshake for connecion esablishmen. The following consan exchange of daagrams leads o an average Bus Uilizaion of around 72%. B. Uplink Cycle Lengh The uplink cycle lengh specifies he amoun of ime he maser needs o poll all 1 slaves once and o collec user daa. This cycle lengh direcly influences delay imes beween consecuive accesses o he communicaion medium by a slave. If no slave has any packes waiing in is queue, one cycle akes abou 8 seconds. Depending on he amoun of packes waiing in a slave s queue, his process can ake up o 2 seconds (see Fig. 9). The exchange of address informaion as well as he delivery of ICMPv6 packes leads o a cycle lengh of 4 o 45 seconds. Afer he sar of he applicaion, here sill do exis cycles wih he minimum lengh of 8 seconds. This suggess ha he sysem is sill sable: no every polling cycle has o be used for exchanging daa unis. Cycle Lengh (seconds) 2 15 1 5 2 4 6 8 1 x 1 4 Fig. 9 Uplink Cycle Lengh C. Downlink Cycle Lengh The Downlink Cycle measures he amoun of ime he maser needs for delivering daa once o each of he 1 slaves. If here is no daa waiing in a paricular slave s queue in he maser, he nex slave is conaced. In he beginning, in Fig. 1 one can see he cycle in which he Neighbor Adverisemens and, afer 9 seconds, he ICMPv6 Echo Requess are delivered. Afer 1 hour, he peak cycle lengh of nearly 9 seconds is caused by he esablishmen of TCP connecions o all 1 slaves and he subsequen delivery of he reques packe on he applicaion layer. As he amoun of daa delivered o he slave mainly consiss of he reques packes and acknowledgemens on TCP layer, he mean duraion is lower han he one of he uplink cycle lengh. 1 Cycle Lengh (seconds) 5 2 4 6 8 1 x 1 4 Fig. 1 Downlink Cycle Lengh D. IP End-o-End Delay On IP layer, he end-o-end delay beween wo nodes is measured beween he ime of creaion of an IPv6 packe and he ime of recepion wihin he IP module of he r. I is influenced by he waiing ime in he MAC queue of he ransmiing node and he ransmission duraion. The iniial delay deerminaion only depends on he MAC algorihm and he number of slaves. I resuls in abou 5 seconds, which is direcly relaed o he ransmission ime of he packes on he bus and he waiing ime of he ICMPv6 Echo Reply in he slaves queues, influenced by he Uplink Cycle Lengh. The ime delay varies beween 5 and 35 seconds (see Fig. 11) wih an average of abou 15 seconds. The ime span is direcly relaed o he number of packes already waiing in he queue of slave 1 and o he lengh of boh uplink and downlink polling cycles during he waiing ime in he queue. Delay (seconds) 3 2 1 2 4 6 8 1 x 1 4 Fig. 11 End-o-End Delay on IP layer beween Maser and Slave 1 E. Applicaion Delay The applicaion delay is measured beween he ransmission of he reques packe and he complee recepion of all fragmens belonging o a meering daa record. Afer having sen he reques packe o slave 1, he maser consanly has o wai abou 6 seconds unil all 6 fragmens of he 55 bye meering daa se are d and reassembled (see Fig. 12). 93
This is less han he period of 9 seconds beween wo consecuive meer daa requess. The sysem is sable. Delay (seconds) 6 4 2 2 4 6 8 1 x 1 4 Fig. 12 Applicaion Delay beween Maser and Slave 1 F. Parameer Sudies To simulae differen precondiions, several parameers have been varied. On he one hand, he number of slaves paricipaing in communicaion has been alered. As his direcly influences he lengh of a polling cycle, boh delay and hroughpu are affeced. Wih more slaves, he delay beween wo consecuive accesses o he medium increases, as well as he oal amoun of daa ha has o be delivered in ime. The amoun of meering daa mus be reduced wih increasing number of slaves o keep he sysem sable. Table III shows he resuls. TABLE III IP END-TO-END DELAY Number of IP End-o-End Configuraion slaves Delay 5 iniial delay measuremen < 4 seconds 5 daa ransfer (12 byes per slave per meering daa se) < 6 seconds 1 iniial delay measuremen < 7 seconds 1 daa ransfer (55 byes per slave per meering daa se) < 4 seconds 2 iniial delay measuremen < 15 seconds 2 daa ransfer (27 byes per slave per meering daa se) < 7 seconds The higher IP end-o-end delay regarding daa ransfer wih 5 slaves is caused by he larger amoun of daa ha can be ransmied wihou yielding an unsable sysem (compare equaion (1)). As 2 packes are ransmied per maser-slavecycle, he larger amoun of daa leads o more packes remaining in a slave s queue a a cycle s end, which hen leads o longer delays beween creaion of an IPv6 packe an is delivery. The effecive hroughpu of user daa on applicaion layer, calculaed on a 9 seconds basis, is shown in Table IV. TABLE IV DATA THROUGHPUT Number of slaves Transferred Daa Amoun per Reques Average Daa Rae 5 12 bye + 256 bye 5447 bi/s 1 55 bye + 256 bye 5116 bi/s 2 27 bye + 256 bye 5255 bi/s VII. CONCLUSION Despie he considerable overhead caused by he large IPv6 header and also TCP, he proocol suie is applicable even a low PHY Layer daa raes. In combinaion wih he robus MAC algorihm, accepable delay imes and hroughpu can be achieved. The effecive hroughpu on he applicaion layer shows ha he sysem behaves linearly. The overhead a MAC layer says consan and only slighly decreases he usable daa rae. Less han 5% of he available daa rae of 11 bi/s are used for bus arbiraion and packe acknowledgemens. ACKNOWLEDGMENT The auhors kindly hank OPNET Technologies, Inc. for heir suppor in erms of he OPNET Universiy Program [16]. REFERENCES [1] European Parliamen, European Council, Direcive 26/32/EC on energy end-use efficiency and energy services, April 26. [2] A. Moreno-Munoz, J.J.G. De La Rosa, Inegraing power qualiy o auomaed meer reading, IEEE Indusrial Elecronics Magazine, vol. 2, issue 2, pp. 1-18, June 28 [3] G. Deconinck, An evaluaion of wo-way communicaion means for advanced meering in Flanders (Belgium), IEEE Insrumenaion and Measuremen Technology Conference Proceedings, 28 [4] M. Goez, K. Doser, A Universal High Speed Powerline Channel Emulaion Sysem, Inernaional Zurich Seminar on Broadband Communicaions, 22 [5] J. Bausch, T. Kisner, M. Babic, K. Doser, Characerisics of Indoor Power Line Channels in he Frequency Range 5-5 klz, IEEE Inernaional Symposium on Power Line Communicaions and is Applicaions, 26 [6] ISO/IEC 7498-1, Open Sysems Inerconnecion Basic Reference Model: The Basic Model, 1994 [7] CENELEC EN 565-1, Signaling on low-volage elecrical insallaions in he frequency range 3 khz o 148.5 khz, 1991 [8] T.Kisner, Ein neuariges mehrrägerbasieres PLC-Sysem mi sörresisener Synchronisaion, Universiäsverlag Karlsruhe, 28 [9] T. Kisner, M. Bauer, A. Hezer, and K. Doser, Analysis of Zero Crossing Synchronizaion for OFDM-Based AMR Sysems, IEEE Inernaional Symposium on Power Line Communicaions and is Applicaions, 28. [1] S. Mangold e al., IEEE 82.11e Wireless LAN for Qualiy of Service, Proceedings European Wireless, vol. 18, 22 [11] H. Levy and M. Sidi, Polling Sysems: Applicaions, Modeling, and Opimizaion, IEEE ransacion on communicaions, vol. 38, pp. 175-176, Ocober 199. [12] Nobuo Okabe e al., Hos Requiremens of IPv6 for Low Cos Nework Appliances, hp://ools.ief.org/hml/draf-okabe-ipv6-lcna-minreq-2, IETF, December 22. [13] Deering, S. R. Hinden: Inerne Proocol, Version 6 (IPv6) Specificaion. RFC 246 (Draf Sandard), 1997. Updaed by RFC 595. [14] NARTEN, T., E. NORDMARK, W. SIMPSON H. SOLIMAN: Neighbor Discovery for IP version 6 (IPv6). RFC 4861 (Draf Sandard), 27. [15] SOLLINS, K.: The TFTP Proocol (Revision 2). RFC 135 (Sandard), 1992. Updaed by RFCs 1782, 1783, 1784, 1785, 2347, 2348, 2349. [16] OPNET Technologies, Inc., hp://www.opne.com/ 94