Implementation of a Dominance Protocol for Wireless Medium Access

Transcription

1 Implementaton of a Domnance Protocol for Wreless Medum Access Nuno Perera, Björn Andersson and Eduardo Tovar IPP Hurray Research Group Polytechnc Insttute of Porto, Portugal {nperera,bandersson,emt}@de.sep.pp.pt Abstract Consder the problem of schedulng sporadc message transmsson requests wth deadlnes. For wred channels, ths has been acheved successfully usng the CAN bus. For wreless channels, researchers have recently proposed a smlar soluton; a collson-free medum access control (MAC) protocol that mplements statc-prorty schedulng. Unfortunately no mplementaton has been reported, yet. We mplement and evaluate t to fnd that the mplementaton ndeed s collson-free and prortzed. Ths allows us to develop schedulablty analyss for the mplementaton. We measure the response tmes of messages n our mplementaton and fnd that our new response-tme analyss ndeed offers an upper bound on the response tmes. Ths enables a new class of wreless real-tme systems wth tmelness guarantees for sporadc messages and t opens-up a new research area: schedulablty analyss for wreless networks. 1. Introducton The sporadc model [1] has proven to be very useful n the desgn of real-tme systems. In ths model, the exact tme of a transmsson request s unknown, but a lower bound on the tme between two consecutve transmsson requests from the same message stream s known. Ths model s supported n processor schedulng [2] (where a message stream s called a task) and n wred communcaton channels [3]. Wreless communcaton s of ncreasng nterest n the desgn of dstrbuted real-tme systems, and many schedulng algorthms and analyss technques for wreless communcatons are avalable for perodc messages. But for sporadc messages such results are less well developed. Most of the current wreless protocols cannot be analyzed to offer pre-run-tme guarantees that sporadc messages meet deadlnes, and the protocols that can offer such guarantees rely on pollng, whch s neffcent when the deadlne s short and the mnmum tme between two consecutve requests s large. In wred networks, sporadc messages can be scheduled effcently usng the CAN bus [4]. It has a medum-access control (MAC) protocol whch s collson-free and prortzed, and hence t s possble to schedule the bus such that f message characterstcs (perods, transmsson tmes, jtter, etc.) are known, then t s possble [3] to compute upper bounds on message delays. Ths MAC protocol belongs to a famly called domnance protocols or bnary countdown protocols [5], whch works as follows. Messages are assgned unque prortes and when messages contend for the channel, they perform a tournament such that the hghest-prorty message s granted access. Ths tournament s performed bt-by-bt of the prorty, startng wth the most sgnfcant bt. If a node (let us call t node ) contends wth a recessve bt but t detects that another node transmtted a domnant bt then node loses and t does no longer partcpate n the tournament. Fnally, there s only one node that wns and t transmts ts message. The domnance protocol that was mplemented n CAN uses open-collector/open dran crcuts. Clearly, ths does not easly extend to wreless channels. For ths reason, researchers n the feld recently desgned domnance protocols for wreless channels; they are based on modulatng prorty bts usng on-off keyng [6-9]. These protocols can support sporadc messages effcently and to the best of our knowledge, no other publshed protocol can do ths (see [9] for a survey of relevant lterature). Unfortunately, the recently proposed domnance protocols [6-9] for wreless channels all have n common that they were not mplemented and tested. Due to nondealtes n transcevers and the nature of the wreless medum, t s not obvous how these protocols should be mplemented. There exst prorty levels for whch the protocols need to swtch between transmt and receve modes for every prorty bt, and ths s potentally wasteful because many transcevers are not desgned for frequent swtchng and hence every swtchng takes non-neglgble tme. It s well known that wreless channels typcally have sgnfcantly hgher nose levels than wred channels and that detecton of pulses of short duraton s dffcult [10]. For ths reason, wreless communcaton systems often use long codes [11] and/or spread spectrum modulaton to ncrease the probablty of a correctly receved message. Unfortunately, these technques cannot be used to transmt prorty bts n the protocols [6-9]: () long codes operate on message-level and ths s too coarse; () there s the need to demodulate and decode an ndvdual bt so that a decson can be made whether the next prorty bt should be transmtted. Spread spectrum modulaton cannot be used on

2 prorty bts because t requres nodes that attempt to detect the prorty bts be accurately synchronzed wth the senders: there are many senders and they can all send a prorty bt at approxmately the same tme so a node tryng to receve the prorty bt cannot be synchronzed wth all of the senders. Ths makes t non-obvous whether wreless domnance protocols could work. In ths paper, we mplement one of the proposed domnance protocols [9] and we call the mplementaton WDOM. We evaluate t and fnd that n our expermental envronment, the probablty that a message s transmtted collson-free, correctly prortzed and correctly receved by all other nodes (that s nether lost nor corrupted) s at least 99.99%. We beleve ths relablty justfes the development of schedulablty analyss technques for sporadc messages n wreless networks. We do so; we adapt the response-tme formulatons from the CAN bus to WDOM and test the valdty of the analyss. We fnd that more than 99.99% of all messages that were proven to meet deadlnes also dd so n practce. The remander of ths paper s structured as follows. Secton 2 provdes the necessary background on the target platform, ntroducng the man aspects relevant for the mplementaton. Secton 3 presents our mplementaton of the protocol n TnyOS, outlnng ts man software components. Next, n Secton 4, the response-tme formulaton for our protocol s presented. The evaluaton of the mplemented protocol, n Secton 5, entals the descrpton of several experments performed n order to test the propertes of our protocol. Secton 6 dscusses the performance and the use of the protocol and compares t wth prevous work. Fnally, Secton 7 provdes conclusons and future work. 2. The Platform We mplemented the domnance protocol on an embedded computer platform known as McaZ [12]. It s a sensor network platform, offerng a low power mcrocontroller, 128 kbytes of program flash memory and an IEEE complant rado transcever CC2420 [13], capable of 250 kbts/s data rate. The McaZ platform s supported by TnyOS [14], an open-source operatng system desgned for wreless sensor networks. Ths platform was found to be an attractve alternatve for the mplementaton of our experments because of the followng relevant characterstcs: () t allowed us to replace the exstng MAC protocol n TnyOS easly; () the tmers avalable were suffcently precse for our applcaton; () the rado can be put nto a specfc test mode, where t s possble to transmt an unmodulated carrer for an arbtrary duraton; (v) the rado has bult-n RSSI (Receve Sgnal Strength Indcator)/energy detecton functonalty and Clear Channel Assessment (CCA) s avalable through a dgtal output pn; (v) the spread spectrum modulaton used makes data frames resstant to nose and dstorton. Due to (v), the man term that affects message transmsson relablty s collsons. Our protocol s (as we wll see) collson-free. The CC2420 provdes a packet level nterface for sendng and recevng, meanng that packets are sent by wrtng to the chp s memory over a bus and ssung a send command. Recepton s sgnalled when the chp trggers an nterrupt, and at that tme the communcaton stack should read the bytes out of the chp s memory. Domnance protocols n wred meda requre that a node can smultaneously transmt whle t detects the transmssons from other nodes. Unfortunately, ths s not possble n most rado transcevers, ncludng the CC2420 because the transmtted energy s much hgher than the receved energy. For ths reason, the CC2420 can only be ether n transmsson mode or n recepton mode and t can take up to 192 µs to swtch between these two modes. WDOM needs to transmt a carrer wave and the CC2420 rado can do ths, ether a modulated carrer or an unmodulated carrer n a transmtter test mode. The RSSI obtaned when a node sends an unmodulated carrer s 9dBm stronger than the RSSI obtaned when a node sends a modulated carrer (when the node transmts a data packet) [13]. Hence, for WDOM, we use the unmodulated carrer n the tournament; the modulated carrer s transmtted only as a result of transmsson of data. WDOM also needs to detect whether other nodes transmt a carrer wave. For ths, t uses the CC2420 support for CCA. The CCA functonalty of the CC2420 rado computes the average RSSI over the last 128 µs. To make a decson, ths average s compared to a confgurable threshold and then CC2420 sets the CCA dgtal output pn accordngly. Ths pn s sampled by our software communcaton stack to detect f other nodes are sendng carrer pulses. After the rado s n receve mode, t takes 128 µs to make the frst vald CCA operaton. Our protocol s heavly dependent on tmers. The McaZ s ATmega128 mcrocontroller provdes two 8-bt tmer/counter and two 16-bt tmers. For our mplementaton, we use the 8-bt Tmer/Counter2 for tmng, snce ths s the tmer used n CC2420 TnyOS communcaton stack, whch we are partally replacng. We bult a component that abstracts the tmng by provdng an nterface for settng tmeouts and to delver the clock tcks to a carrer-detecton component that uses them to drve the CCA pn pollng. Ths component confgures the tmer to gve nterrupts every µs. For ths reason, when we choose tmeouts, these wll be selected as multples of µs.

3 Fgure 1. Automaton descrbng the mplementaton of the protocol. 3. The Protocol To smplfy presentaton of the mplementaton, we present t usng a tmed-automata lke notaton. States are represented as vertces and transtons are represented as edges. An edge s descrbed by ts guard (a condton whch has to be true n order for the protocol to make the transton) and an update (an acton that occurs when the transton s made). In Fgure 1, we let / separate the guards and the updates; the guards are before / and the update s after. We let = denote test for equalty and let := denote assgnment to a varable. For those transtons wth an update havng many lnes of code, t s assumed that the lnes are executed sequentally. Our domnance protocol assumes that all nodes can hear each other. The man dea of the protocol s that a message s assgned a statc prorty and when a message contends for the channel, nodes perform a tournament such that the hghest-prorty message s granted access to the channel. Ths tournament s performed bt-by-bt, startng wth the most sgnfcant bt (States 5, 6 and 7 n Fgure 1). A bt s assgned a tme nterval. If a node contends wth a domnant bt then a carrer wave s transmtted n ths tme nterval; f the node contends wth a recessve bt, t transmts nothng but lstens. Ths makes t possble for a node wth a recessve bt to detect that another node has transmtted a domnant bt, and hence the node wth the recessve bt wthdraws. After the tournament, a node s a wnner f t requested to transmt a message, and when t lstened, t never heard a domnant bt. The wnnng node wll then transmt (States 8 and 9 n Fgure 1). A node whch loses wll receve, but f t does not receve a message t wll tmeout (transton 10 1 n Fgure 1). In order for ths scheme to work, nodes must agree on whch tme nterval to use. Ths requres a conventon, somethng that s easy to state and whch we do. Before the begnnng of a tournament, we must also establsh a common reference pont n tme (ths s ensured by States 0-5 n Fgure 1).

4 WDOM BareSendMsg ReceveMsg WDOMRado BareSendMsg ReceveMsg WDOMCarrerPulse WDOMCarrerSense WDOMClock WDOMClockTcks WDOMTme HPLCC2420FIFO HPLCC2420FIFO CC2420Control CC2420Control HPLCC2420Interrupt HPLCC2420Interrupt HPLCC2420Capture HPLCC2420 HPLCC2420 Clock HPLTmer2 Fgure 2. TnyOS component assembly. Rectangles are mplementaton modules of components. Components names are n bold, the correspondng module has the same name wth an M appended. For smplcty, only the most relevant modules and nterfaces are depcted, and confguratons wrng components are omtted. Shaded rectangles are TnyOS components reused n the mplementaton. Trangles pontng nto a rectangle are provded nterfaces. Trangles pontng out represent used nterfaces. The names of the provded nterfaces are n talcs. The protocol automaton (Fgure 1), refers to several tmeout values. These tmeout parameters were selected accordng to constrants gven n [9], usng NPRIOBITS=10: E=312 µs, F=21770 µs, G=555 µs, ETG=520 µs, H=1145 µs, L=5 µs, where L s the maxmum computatonal tme for a transton n the automaton. The tmeout parameters are based on the assumpton that a carrer pulse must have a duraton (tmeout named TFCS) of 486 µs so that the other nodes may detect t, and SWX=192 µs. The value of TFCS was expermentally obtaned; further detals are gven n Secton 6 and the relatonshp between these parameters can be found n [9]. The protocol has been mplemented n TnyOS usng nesc [15]. The man TnyOS software components of the mplementaton are presented n Fgure 2, whch provdes a smplfed overvew of the mplementaton component assembly [14, 15]. Components CC2420Control, HPLCC2420FIFO, HPLCC2420Interrupt and HPLCC2420 are part of the Hardware Presentaton Layer (HPL) for the CC2420, and are regular TnyOS components reused by the mplementaton. These components provde basc functonaltes for handlng the rado. They are used by WDOMRado to do operatons as startng/stoppng the rado, confgurng the rado parameters, and performng all the operatons needed for sendng/recevng packets. The HPLTmer2 s also an HPL component from TnyOS, whch drectly abstracts the 8-bt Tmer/Counter2 on the McaZ s ATmega128 mcrocontroller. Ths component s used by WDOMClock to drve the tmng of the protocol. WDOMClock confgures nterface WDOMTme { command unt32_t get(); command result_t reset(); command result_t setalarm(unt32_t when, unt8_t type); command bool salarmset(); command result_t cancelalarm(); async event vod alarm(unt8_t type); } nterface WDOMClockTcks { command result_t posttcks(bool postng); async event vod tck(); } nterface WDOMCarrerSense { command result_t start(); command result_t stop(); event result_t channelbusy(); } nterface WDOMCarrerPulse { command result_t carrertxmodestart(); command result_t carrertxmodeend(); command result_t on(); command result_t off(); } Fgure 3. WDOM mplementaton-specfc nterfaces. the tmer prescaler to delver the tmer nterrupts every µs, whch s used to drve the tmng mantaned by ths component. The WDOMClock has two functonaltes. It presents an nterface (nterface WDOMTme, n Fgure 3) to the module that runs the WDOM protocol (WDOM component) whch enables the MAC protocol to mantan ts tmng n a manner equvalent to the way x s used n the protocol automaton from Fgure 1. As seen n Fgure 3, ths nterface provdes commands to get the current tme (the same as readng the value of x n the protocol automaton) and resettng the tme ( x:=0, n the protocol automaton). Addtonally, ths nterface allows the settng of tmeouts (called alarms) when the tme reaches a certan value ( x>= condtons n the automaton). Notce that when an alarm s set, an alarm type s specfed. Ths s delvered when the alarm fres. The second functonalty offered by WDOMClock s provdng the clock tcks t receves from the hardware tmer to other components. More specfcally, because carrer sensng s made by pollng the CCA pn of the CC2420, the WDOMRado ssues a command to WDOMClock component to get the clock tcks every tme t has to perform carrer sensng. These clock tcks are then used to drve the pollng of the CCA pn. The WDOMRado component s responsble for provdng all rado functonaltes needed by the MAC protocol. It handles the nteractons wth the HPLs for sendng/recevng packets, and provdes to the WDOM component a smple send/receve nterface (nterfaces BareSendMsg and ReceveMsg are commonly used TnyOS nterfaces for these purposes). A packet ndcated for sendng to ths component wll be sent wthout any knd of arbtraton. Packets receved are ndcated to the protocol only after they have been entrely receved and fetched from the rado chp. As descrbed n the former paragraph, WDOMRado also provdes an nterface for

5 performng carrer sensng. Ths nterface, called WDOMCarrerSense, s shown n Fgure 3 and smply enables the WDOM component to turn carrer sensng on and off, whenever necessary. The WDOMRado also provdes an nterface for sendng carrer pulses. Ths nterface (WDOMCarrerPulse, detaled n Fgure 3) smply enables the WDOM protocol to turn the sendng of a carrer pulse on and off. However, because the rado must be n a transmtter test mode n order to be able to send an unmodulated carrer, the nterface also presents commands to swtch the rado nto ths transmtter test mode and to get out from ths mode nto a normal transmt mode, to be able to transmt data. Fnally, we descrbe the WDOM component tself. Ths component mplements the WDOM protocol wthn a functon that receves an nteger representng a message type. These message types can be any of the events that make the automaton evolve: every tme the WDOMClock sgnals an alarm, ths functon s nvoked, passng as message type the alarm fred; Whenever a message s queued for sendng, the protocol functon s nvoked wth an assocated message type; Smlarly, f WDOMRado sgnals the recepton of a packet, the end of a transmsson, or detecton of a busy channel. The protocol automaton mantans a state varable ndcatng the current state and, for each state there s a swtch statement, where the behavour s mplemented. When the WDOM protocol functon s nvoked, the code for the current state s executed. Wthn each state there s a swtch statement for each of the messages that may be receved n that state and make the protocol evolve. 4. Response-tme Calculatons Let us now ntroduce the response-tme calculatons for the WDOM protocol. Ths analyss s based on the analyss of non-preemptve statc-prorty schedulng used n the CAN bus, but subtletes about the synchronzaton n the protocol requres our schedulablty analyss to deal wth aspects that the CAN bus dd not have to deal wth. Consder the sporadc model [1] wth a system of n message streams: τ 1,τ 2,,τ n. Each message stream τ s characterzed by T, D and C wth the nterpretaton that () T s the mnmum tme between two consecutve transmsson requests from τ, () every tme a message from τ s requested to be transmtted t needs to fnsh the transmsson at most D tme unts after the request and () C denotes the tme requred to transmt a message from message stream τ. We assume :D T. Every message has assocated protocol overhead that should be added to C. The tme to transmt a message and performng the tournament when nodes are already synchronzed s denoted C. The tme to transmt a message and performng the tournament when nodes are not already synchronzed s denoted C. We now compute C and C from C and state the response-tme equatons. We use a message sze of 64 bytes of data (the length of data n a packet s ncluded). Addng 3 bytes used for preamble, the tme to transmt a message s gven by: 1 C = ( ) 8 = 2093μs (1) Inspectng the automaton n Fgure 1 (and applyng reasonng from [9]), yelds: ( G + H ) ( nprobts 1) C = C + 2H + G + + 2L + ETG = C μ s = 20768μs and takng nto account also the ntal dle tme (state 2 n Fgure 1) yelds: C = C + F + E + SWX = C μs = (3) = 43042μs Let us assume that the release jtter s equal to zero. We also assume that the granularty of the tme s Q bt =4/250000s=16 µs. Ths s because the rado uses Drect-Sequence Spread-Spectrum such that every 4 bts s modulated as 16 chps and the data rate s 250 kbts/s (ths s equvalent to 2Mchp/s). Usng these assumptons we obtan that the response tme can be calculated (smlar to [3]) as a sum of the watng tme w and C. where C (2) R = w + C (4) s defned as n (3). The watng tme s: = + w + F + E + SWX + Q bt w B j hp( ) T j where hp() s the set of all message streams wth a hgher prorty than τ. Observe that (5) dffers from the analyss used n the CAN bus. Wth WDOM, t s necessary to add F+E+ SWX whch s the tme before the next message s dequeued after the prevous message has been transmtted. Fgure 4 provdes the ntuton behnd ths. Fgure 4b shows two computer nodes N 1 and N 2 and how ther states change as tme progresses. We can see that a message s dequeued when nodes have made the transton to State 5. Let us assume that there s a message stream τ 1 on N 1 and a message stream τ 2 on N 2 and that τ 1 has hgher prorty than τ 2. If τ 1 and τ 2 make a transmsson request at the same tme when N 1 and N 2 are n State 1, then we wll get one response tme of τ 2 and f they make transmsson requests just before both nodes make a transton to State 5, τ 2 wll get another response tme. But nether of them maxmze the response tme ofτ 2. C j (5)

6 N 1 N 2 F+E+SWX+H N N 2 1 a) Arrval pattern of messages F Tme b) Protocol automaton state changes 2 Nodes check the queue at arrval to State Tme Fgure 4.Worst-case arrval pattern and protocol automata state changes. Instead, t s the arrval pattern llustrated n 4a that maxmzes the response tme of τ 2. B can be computed as follows: 0 f lp( ) = B = (6) max{ C j : j lp( ) } f lp( ) where lp() s the set of all message streams wth a lower prorty than τ. Note that the analyss consders the ntal dle tme between states 1-5 (Fgure 1) to be part of the message when we compute nterference. Ths ntal dle perod should not be ncluded when computng the blockng n (6). Let us apply the response tme analyss to calculate the values accordng to (1)-(6) n the followng example (the response tmes wll be tested emprcally n Secton 5). Example 1. Consder m=10 computer nodes wth one message stream on each node. Message streams are gven values, as shown n Table 1 (all values are gven n µs). Table 1. Message streams for Example 1. =1 =2 =3 =4 =5 T C C C B R =6 =7 =8 =9 =10 T C C C B R We assume that deadlne monotonc s used, and assume D =T. It can be seen that the perods are harmonc. We apply (1)-(6). Observe (Table 1) that the schedulng theory predcts that all deadlnes wll be met because we obtan :R T. 5. Expermental Evaluaton Havng seen the mplementaton, we now turn our attenton to evaluatng ts performance. We have the followng hypotheses: 1. The mplementaton of the protocol offers collson-free medum access for data messages. 2. The mplementaton of the protocol offers prortzed medum access. 3. The response-tme analyss equatons n (1)-(6) can be used to analyze the response-tmes of the mplementaton of the protocol. Hypothess 1 and Hypothess 2. In order to test Hypothess 1 and Hypothess 2, four experments were set up. We let d denote the maxmum dstance between any two nodes. We postoned m nodes n a crcle such that for every node, the dstance to ts neghbors wth the mnmum dstance s maxmzed. The experment runs as follows. A specal node (whch s not ncluded n the m nodes) transmts a carrer wave and all other nodes boot. All nodes request to transmt a message and they enter state 1 (from Fgure 1). These nodes stay n state 1 (Fgure 1) untl the specal node stops transmttng the carrer. We made the experment wth m=2 nodes and wth m=10 nodes. For the case m=2 nodes, all nodes make a new request to transmt a message random(0, 255) ms (Ths means generate a unformly dstrbuted random number wth a mnmum value 0 and maxmum value 255) after the prevous request. For the case m=10 nodes, all nodes make a new request to transmt a message random(0,1023) ms tme unts after the prevous request. We also vared the dameter, d=1m and d=10m. Nodes are gven numbers from 1 to 10 and ther prorty s equal d. We send messages wth a length of 64 bytes. Every node has a sequence counter, ntalzed to 1. The sequence counter s transmtted n every message and then the sequence counter on the node s ncremented. Whenever a node receved a message t compares the receved sequence counter to the prevously sequence number receved from the same node. If the new sequence number s one greater than the prevous sequence number then the recever concludes that the transmsson was collson-free; otherwse the recever takes the dfference between the sequence counters, subtracts one; ths s the number of lost messages. Snce a collson causes a lost message, ths gves us an upper bound on the number of lost messages due to collsons. We also tested whether prortzaton s functonng.

7 Prob. of correct recepton and protzaton 100,000% 99,999% 99,998% 99,997% 99,996% 99,995% 99,994% 99,993% 99,992% 99,991% 99,990% 100% 100% 100% m=2 m=10 99,998% dstance = 1m dstance = 4m Table 2. Response tmes obtaned expermentally for settngs of Example 1. =1 =2 =3 =4 =5 r =6 =7 =8 =9 =10 r We observe that all measured response tmes are less than the calculated upper bounds on response tmes n Example 1. Ths corroborates Hypothess 3. Fgure 5. Prortzaton and collson free test results. We dd t as follows. When a node sends a message t sends ts prorty n the data packet. All nodes receve ths packet (f they dd not receve, t would be consdered as a collson, see Hypothess 1) and f the prorty of the wnner was less than the prorty of ths node then t s consdered as a prortzaton error. The experments were run untl a total of messages were sent, and the data that we obtaned s presented n Fgure 5. We can see that more than 99.99% of all messages were collson-free and prortzed (observe that messages mght get lost due to nose). Ths corroborates Hypotheses 1 and 2. Hypothess 3. In order to test Hypothess 3 we set up the followng experment. One specal node (whch s not ncluded n the m nodes) sends a carrer for approxmately 1 mnute. Durng ths mnute, the other nodes boot and enter state 1 (from Fgure 1). Then the specal node stops transmttng the carrer. Ths causes () all m nodes to reset ther tmers and () all m nodes to request to transmt a message. Then messages are requested to be sent sporadcally such that a message stream τ made a new transmsson request T + random(0,5*t ) ms after the prevous request. Every node had exactly one message stream, thus n=m. A node whch receves a message reads the current tme and calculates the response tme. Perodcally (every 3 mnutes), the specal node wll send a carrer for a long duraton n order to synchronze clocks on the m nodes agan. Ths s performed mmedately after recevng a message n order to avod dsturbng a tournament. When the node fnshes sendng the carrer, all m nodes reset ther tmers and the queung tme of prevously queued messages. Ths mantans the clocks of all m nodes more tghtly synchronzed durng the length of the experment and ths s necessary n order to measure the response tmes of messages because the recepton and enqueung tmes of a message are measured on dfferent nodes. The experment was run for the task set n Example 1, untl messages were transmtted. The results obtaned are gven n Table 2 (all values are gven n µs). Let r denote the maxmum measured response tme. 6. Dscusson Recall that the am of ths paper s to show that domnance protocols for wreless channels can be mplemented and ths s mportant for real-tme communcaton. The ssue of obtanng long range communcaton s a subject pertanng to telecommuncaton. It may requre other technques for modulatng the prortzaton bts or other technques to confgure/nterface wth the transcever. However, dong so s beyond the scope of ths paper. In order to show the potental range, we have made experments on detectng pulses of carrers when the recever s always n receve mode. These experments suggest that a communcaton range of 15 m may be possble when WDOM s used. Ths s promsng, consderng that the manufacturer states the range of the CC2420 transcever s meters ndoors (all experments were done ndoors). We set up an experment to test the ablty of our target platform to detect pulses. We set up two nodes: one sender and one recever, wth non-obstructed lne-of-sght and they were separated 15 meters apart. The experment was conducted n an ndoor offce envronment. The nodes were put at the ends of a corrdor. The default values as descrbed n the manufacturer s manual for the rado parameters were used and the experment was conducted wth new batteres n the nodes. We used dfferent duratons of the pulses sent and we selected the duratons as a multple of µs. For every duraton, we transmtted pulses, counted the number of detected pulse and computed the estmated probablty of an undetected pulse. The result s shown n Fgure 6. Observe (n Fgure 6) that wth duraton of 277 µs or less t happens (and t happens often) that the recever does not detect the pulse. Wth 486 µs all pulses are detected. Hence, we mplemented our protocol based on that assumpton. Detecton of pulses s well known to be a dffcult problem [10] and t s known that false postves can occur. For ths reason we run the experment agan but the sender does not transmt any pulses. We wat for the duraton of the experments and detect the number of detected pulses. We fnd that no pulses are detected and

8 Prob. of Undetected Carrer 60% 50% 40% 30% 20% 10% 0% 54% 46% 0% Duraton of Carrer Pulse (μs) Fgure 6. Probablty of undetected carrer as a functon of the carrer pulse duraton. use that to obtan an estmate of the probablty of a false postve detecton. It s zero. 7. Conclusons and Future Work We have shown that a wreless domnance protocol can be mplemented. It s collson-free and prortzed and hence t allows us to compute the response tmes. The protocol s ntended for short-range communcaton n small geographc areas and for ths purpose, our protocol works relably. We observed that the tournament functons correctly n more than 99.99% of the cases. The overhead s to a large extent due to the transton tme between transmsson and recepton. Ths s a technologcal parameter that can be mproved wth better hardware as wtnessed by the fact that the Hperlan standard [16] requred a swtchng tme of 2µs. If such a transcever was avalable and offered the flexblty to desgn the MAC protocol n software, then the overhead of the protocol could be reduced by two orders of magntude. We are currently lookng for such hardware. Fnally, we pont out that besdes from beng naturally useful for schedulng sporadc real-tme traffc, WDOM also supports certan types of group communcaton [17]. 8. Acknowledgements We are grateful to the revewers for suggested mprovements of the paper. Ths work was partally funded by the Portuguese Scence and Technology Foundaton (Fundação para Cênca e Tecnologa - FCT) and the ARTIST2 Network of Excellence on Embedded Systems Desgn. 9. References [1] Mok, A. "Fundamental Desgn Problems of Dstrbuted Systems for the Hard Real-Tme Envronment Electrcal Engneerng and Computer Scence, In Electrcal Engneerng and Computer Scence, Cambrdge, Mass., 1983, Massachusetts Insttute of Technology, Cambrdge, Mass., [2] Baruah, S.K., Mok, A.K. and Roser, A.K., "Preemptvely Schedulng Hard- Real-Tme Sporadc Tasks on One Processor". In IEEE Real-Tme Systems Symposum, 1990, [3] Tndell, K., Hansson, H. and Wellngs, A., "Analysng real-tme communcatons: controller area network (CAN)". In 15th Real-Tme Systems Symposum (RTSS'94), 1994, [4] Bosch, "CAN Specfcaton, ver. 2.0, Robert Bosch GmbH, Stuttgart," Onlne: [5] Mok, A.K. and Ward, S. "Dstrbuted Broadcast Channel Access". Computer Networks, , [6] You, T., Yeh, C.-H. and Hassanen, H.S., "CSMA/IC: A New Class of Collson-free MAC Protocols for Ad Hoc Wreless Networks". In 8th IEEE Internatonal Symposum on Computers and Communcaton, 2003, [7] You, T., Yeh, C.-H. and Hassanen, H.S., "A New Class of Collson - Preventon MAC Protocols for Ad Hoc Wreless Networks". In IEEE Internatonal Conference on Communcatons, [8] You, T., Yeh, C.-H. and Hassanen, H.S., "BROADEN: An effcent collsonfree MAC protocol for ad hoc wreless networks". In IEEE Internatonal Conference on Local Computer Networks, [9] Andersson, B. and Tovar, E., "Statc-Prorty Schedulng of Sporadc Messages on a Wreless Channel". In Internatonal Conference on Prncples of Dstrbuted Systems (OPODIS 05), Psa, Italy, [10] Tobag, F.A. and Klenrock, L. "Packet Swtchng n Rado Channels: Part II - The Hdden Termnal Problem n Carrer Sense Multple-Access and the Busy-Tone Soluton". IEEE Trans. on Communcaton, 23 (12) , [11] Shannon, C.E. "A mathematcal theory of communcaton". Bell System Techncal Journal, and , [12] XBow Inc. [13] In. [14] Hll, J. "System Archtecture for Wreless Sensor Networks Computer Scence Department, In Computer Scence Department, 2003, Unversty of Calforna, Berkeley, [15] Gay, D., Welsh, M., Levs, P., Brewer, E., Von Behren, R. and Culler, D., "The nesc language: A holstc approach to networked embedded systems". In ACM SIGPLAN Conference on Programmng Language Desgn and Implementaton (PLDI'03), 2003, [16] "Broadband Rado Access Networks(BRAN);HIPERACCESS; PHY protocol specfcaton", [17] Andersson, B., Perera, N. and Tovar, E., "Usng a Prortzed MAC Protocol to Effcently Compute Aggregated Quanttes". To appear n 5th Intl Workshop on Real Tme Networks (RTN'06), Dresden, Germany, 2006.