Fountain Coded Cooperative Communications for LTE-A Connected Heterogeneous M2M Network

Transcription

1 Fountain Coe Cooperative Communication for LTE-A Connecte Heterogeneou M2M Network Ahaanun Nea, Stuent Member, IEEE, Michel Kaoch, Senior Member, IEEE, Bo Rong, Member, IEEE, Abtract Machine to machine (M2M) communication over long-term evolution avance (LTE-A) network ha emerge a a new communication paraigm to upport a variety of application of Internet of Thing. One of the mot effective technique to accommoate a large volume of evice (noe) in LTE-A i clutering where MTC noe are groupe into number of cluter an forwar their traffic to the bae tation (e.g. LTE enoeb) through ome pecial noe calle cluter hea (CH). In many application, the CH change location with time that caue variation in itance between neighboring CH. When thee itance increae, the performance of ata tranmiion may egrae. To are thi iue, we propoe to employ intermeiate non-ch noe a relay between neighboring CH. Our olution cover many apect from relay election to cooperative formation to meet the uer QoS requirement. A the number of total relay play a ignificant role in cooperative communication, we firt eign a ratele-coe-incremental-relay election (RCIRS) algorithm bae on greey technique to guarantee the require ata rate with a minimum cot. After that, we evelop both ource-feeback an non-ource-feeback bae fountain coe cooperative communication protocol to facilitate the ata tranmiion between two neighbor CH. Numerical reult are preente to emontrate the performance of thee protocol with ifferent relay election metho uner Rayleigh faing channel. It how that the propoe ource-feeback bae protocol (SFprotocol) outperform it non-ource-feeback (NSF-protocol) counterpart in term of a variety of metric. Inex Term Internet of thing, fountain coe, machine to machine communication, cooperative communication. I. INTRODUCTION Over the lat couple of ecae, future Internet eign ha attracte ignificant attention from both acaemia an inutry. The future Internet i likely to integrate heterogeneou communication technologie, both wire an wirele, contributing ubtantially to aert the concept of Internet of Thing (IoT) []. IoT i efine a a worlwie network of interconnecte object, where object are equippe with tiny ientifying evice, an intelligently tranfer ata over the network without any human intervention. Wirele enor network play an increaingly important role in the evelopment of thi emerging paraigm. It act a an important ata reource for IoT that collect the information from the urrouning environment, analye the collecte ata an tranmit to the Internet. Ahaanun Nea( ahaanun.nea.@en.etmtl.ca) an Michel Kaoch ( michel.kaoch@etmtl.ca) are with the Department of Electrical Engineering, Ecole e Technologie Superieure, Univerite u Quebec, Montreal, QC H3C K3, Canaa an Bo Rong ( bo.rong@ieee.org) i with Communication Reearch Center Canaa, Ottawa, ON K2H 8S2,Canaa. Manucript ubmitte on Augut 5, 206. With the emergence of IoT, the conventional enor noe are being replace by more technologically avance noe, namely machine noe thoe have higher information proceing feature. Thi introuce a new paraigm in communication reearch, namely machine type communication (MTC) or machine to machine (M2M) communication [2]. Smart infratructure, health care an electricity gri are a few poible application [3] [5] in which MTC may play a leaing role to improve the quality of people live. M2M network can be coniere a a generalization of traitional enor network in term of their higher proceing capabilitie an i very often intalle in remote area for ening or urveillance purpoe, where hunre or thouan of machine type evice (noe) are enely locate over mall or meium area [6]. In the pat, tremenou technology evelopment an commercial ucce have incurre in mobile cellular network (MCN). In cellular network paraigm, long-term evolution avance (LTE-A) i the latet cellular communication tanar, evelope by the Thir Generation Partnerhip Project (3GPP) to meet the requirement of International Mobile Telecommunication-Avance (IMT-Avance) ytem [7], [8]. The ongoing maive eployment of LTE-A make it poible to intall M2M network in mot urban an remote area, an by uing LTE-A backhaul network, a eamle network communication i being etablihe between MTCnoe an -application. It i expecte, in the following year more than 2 billion of MTC evice will be interconnecte in LTE-A network to upport ifferent application in IoT paraigm [9]. Therefore, when a maive number of noe will attempt to connect to an LTE-A bae tation (e.g. enoeb), it i very likely that the network will be congete, an both MTC an non-mtc traffic will be affecte. Many IoT application, uch a health care, inutry automation an robot control eman tringent QoS requirement in term of reliability, latency an calability. Hence, it i crucial yet challenging to achieve reliability in M2M communication pecially for time contraine application. Noe clutering ha been proven to be a promiing olution in congetion control an reliable ata tranmiion [0] [2] in M2M communication. In thi architecture, the noe are groupe into a number of cluter an ome noe are choen a CH which are ometime alo referre a gateway [3]. The non CH noe in each cluter en their ata to their repective CH an inter cluter traffic are forware through the CH to the LTE-A bae tation (enoeb). In many eployment cenario, ome machine are allowe to move an change their location in the eployment area with very low

2 2 Application omain Network omain M2M omain CH Non- CH HSS MME S-MME Long range interface Capillary network interface MTC- AAA S-U SMS-SC MTC-IWF SGW S-MME En Uer LTE Relay SME S-U S5 SCS PGW MTC erver Bae tation (enoeb) Fig.. An LTE-A connecte heterogeneou M2M networking architecture. mobility, uch a urveillance application, animal in habitat monitoring application, an traffic monitoring application [4], [5]. A a conequence, when the neighboring CH move apart from each other becaue of the mobility of the noe, the performance of the ata tranmiion may eteriorate below the threhol level. Therefore, CH nee to be reelecte in uch cae. However, frequent re-election of CH reult in counter effect on routing an reconfiguration of reource allocation aociate with CH-epenent protocol. We ientify thi problem a a phyical layer problem an a a olution we propoe to employ cooperative communication between neighboring cluter. Cooperative communication i an effective way of improving the throughput, power efficiency an coverage in wirele network [6] [9]. Currently, the reearcher are giving more emphai to explore cooperative trategie an protocol to achieve higher performance an reliability in communication [6], [20] [23]. In [24], [25], the author analyze cooperative communication bae on itribute pace-time block coing (STBC) for wirele enor network an howe that cooperative communication i more energy efficient than non-cooperative irect communication. However, itribute STBC require trict ynchronization among tranmitter, which i ifficult an even impoible in mot mobile cae. Previou work [24], [25] aree the fixe rate coe. The outage probability in their work cannot reach zero if the precie channel tate information (CSI) i unavailable at the tranmitter. Among many other channel coing technique, fountain coe - a channel coing technique [26] [28] that aapt it rate accoring to the variation of channel conition, ha attracte a lot of reearch interet. For it capacity approaching potentiality an ratele property, it appear a a promiing olution for ata communication in many wirele communication ytem incluing relay network [9], [29] an cognitive raio ytem [30]. A a promiing feature, it oe not require any ynchronization mechanim like virtual multiple-inputmultiple-output (MIMO). Alo it oe not require CSI at the tranmitter a fixe rate coe. In open literature, the work relate to cooperative communication [20] [25], [29] mainly focue on the performance analyi an theoretical boun for certain pecific moel. There ha been little reearch one on a full an practical olution for heterogeneou M2M network. In thi work, we evelop a et of cheme to upport cooperative communication in LTE-A connecte heterogeneou M2M network. We invetigate clutere heterogeneou M2M network, an propoe both ource-feeback an non-ourcefeeback bae cooperative communication protocol uing fountain coing. We aume that the network conit of two or more type of noe with repect to battery upply an functionality. The network i then further partitione into a number of cluter, where more powerful noe have been choen a CH. CH emulate the role of the bae tation in their repective cluter, an inter cluter traffic are forware through the CH to the LTE-A network. Then LTE-A network relay the meage to erver an en uer locate in the application omain a hown in Fig.. Our cheme ugget that intermeiate non-ch noe between two neighboring CH act a relay by forwaring the receive information from the ource CH to the etination CH. In our previou work [32], cooperative communication protocol uing LT [26] coe have been analyze for ifferent poition of relay noe over Rayleigh faing channel. In thi work we propoe ourcefeeback an non-ource-feeback protocol uing Raptor coe [27] an compare the performance of thee protocol uner ifferent relay election metho. In the ource-feeback protocol, the ource CH an relay tranmit to the etination until the etination i capable to ecoe the ource CH information correctly. After uccefully ecoing the ource CH information, the etination CH en acknowlegement (ACK) to ource an relay. In the non-ource-feeback protocol, ource CH tranmit to relay an the etination an then ceae it tranmiion after receiving ACK of ucceful reception from relay hoping that etination can uccefully receive the remaining information from relay. In both aforementione protocol, all the noe ue their own fountain coing routine an ifferent orthogonal channel to encoe an tranmit their ata imultaneouly. We tuy their performance in Rayleigh faing channel an how that by employing cooperative protocol the performance of ata tranmiion can be ramatically improve than the irect communication between two neighboring CH. Eviently the performance i improve a more relay noe get involve. However, employing more relay between the ource an the etination pair caue more ignaling overhea an higher power conumption. In thi regar, we focu on relay election between two neighboring CH aiming at achieving the target ata rate between two cluter with the involvement of minimum number of relay noe. The target ata rate between two cluter, which i aociate with the require en to en ignal quality epening on the type of

3 3 application, i provie by the network eigner. When a large number of non-ch noe are available to be a relay, fining the optimum et by uing conventional Brute-Force Search (BFS) woul require huge computational effort that limit it application in practical ytem. In orer to reuce the computational complexity of earching the optimum et, we propoe ratele coe incremental relay election (RCIRS) algorithm bae on greey technique. In our approach, ource CH collect the channel tate information of neighboring noe an inclue ome non-ch noe in the cooperative group one by one o that the cooperative ata rate i incrementally increae until it meet the target ata rate. The relay election propoe here i mainly for ource-feeback protocol. However, it coul eaily be extene to the cenario of non-ourcefeeback protocol. The main contribution of thi paper are the following. ) We ientify the CH-to-CH channel a a bottleneck in heterogeneou M2M network, an preent a full olution to upport fountain coe cooperative communication. The propoe olution conit of everal cheme from relay election to cooperative trategy, while taking into account uer QoS requirement. 2) In mot of the exiting work, the number of cooperating noe i either fixe [22] or ranom [2], [23], [24] epening on the channel an noie realization. Thi work elect the cooperative noe accoring to the ata rate requirement. The propoe RCIRS algorithm take into account the link between ource CH to relay caniate an etination CH to relay caniate. Then it juiciouly elect a minimum number of noe while achieving the require ata rate with much le computational complexity than BFS. 3) In thi work, we alo evelop two communication protocol bae on fountain coing, namely ourcefeeback an non-ource-feeback cooperative communication protocol to improve the performance of ata tranmiion between two neighboring CH. We evaluate the performance of our propoe protocol with ifferent relay election metho uing the metric of tranmiion efficiency rather than the outage probability. Thi trategy i more realitic ince in ratele coe ytem, the outage probability i alway riven to zero if there i no retriction on ecoing elay. The ret of the paper i organize a follow. We firt review the exiting work in Section II an introuce our ytem moel of cooperative communication over clutere M2M network in Section III. We then preent the relay election algorithm an fountain coe cooperative communication protocol in Section IV an Section V, repectively. Finally, Section VI emontrate the imulation reult an Section VII conclue our paper. II. RELATED WORKS In open literature, there exit everal work invetigating the performance of ratele coe over wirele relay channel. Catura et al. [29] introuce the firt ratele coing framework over relay channel, where the relay ait the ource a a econary antenna once it ecoe ource information uccefully. In thi framework, the relay noe ynchronize itelf with the ource before tarting tranmiion. The ource an relay then tranmit ata to the etination uing pace time Alamouti coe. Xi et al. [3] tuie everal ingle relay cooperative cheme an erive their achievable rate in flat Rayleigh faing channel. Yang et al. [33] tuie the performance of fountain coe in low power regime. Abouei et al. [34] tuy the performance of fountain coe in wirele boy area network with repect to reliability an energy conumption. In [7] Zhang et al. propoe a joint network-channel coing (JNCC) with Raptor coe for a 2- - ytem. They optimize the egree itribution of the Raptor coe an howe that their propoe JNCC with optimize egree itribution outperform conventional JNCC cheme in term of BER an throughput performance. In [35], Nikjah et al. propoe everal ingle parameter relay election protocol bae on information accumulation an ratele coing cheme. There, relay noe en acknowlegement to the etination after uccefully ecoing the ource information, an the etination then fin out the bet relay to tranmit information to the etination. In [36], Molich et al. tuie the performance of fountain coe in the preence of multiple relay noe uner block faing Rayleigh channel, an howe that information accumulation conume le energy an require le tranmiion time than that of energy accumulation. In cooperative communication, a relay noe i locate between the ource an the etination an ait the communication in a two hop manner to achieve higher ata rate than the ata rate of irect link. In the literature, comprehenive tuie have been given on relay election in the cooperative communication with fixe coe rate [37] [40]. In [37], Zhao et al. howe for a ingle ource-etination pair, the election of the bet relay noe among multiple relay noe can offer ame level of performance a employing many orinary relay noe. In [38], Bleta et al. propoe a itribute cheme for the election of the ingle relay noe in the preence of multiple relay bae on intantaneou channel conition. In [39], the author propoe threhol-bae elective relaying cheme in orer to minimize the en-to en bit error rate (BER) in cooperative ecoe an forwar ytem. In [40], Bali et al. propoe a itribute cheme for multiple relay election bae on ource an relay channel an threhol optimization. They aume that all the relay have ame average SNR to the ource an optimize the threhol value on thi aumption. However, in practical, the relay noe are locate ranomly between the ource an the etination an experience SNR at relay noe may vary accoring to the itance an faing of the wirele channel. Mot exiting work on fixe coe rate ytem are energy combination of orthogonal tranmiion from ifferent path at the receiver. In ratele coe ytem, however, receiver collect mutual information from ifferent inepenent path to ecoe information. Therefore, the relay election cheme for fixe rate coe cannot provie the optimal throughput when applie to ratele coe cooperation. In fact, the invetigation on the relay election ha not receive much

4 4 TABLE I NOTATION r 3 r Broacat phae r 2 Cooperation Phae CH Non- CH Fig. 2. Cooperative communication framework over clutere M2M network. attention yet regaring the cooperative communication uing ratele coe. In [36], a oon a relay noe ecoe the meage, they tart tranmitting to the etination with the poibility of having a poor link. The exiting work in [35], [38], [4] mainly focu on how to elect the bet relay in the preence of multiple relay for a ingle ource-etination pair o that the highet tranmiion rate can be achieve [35], [4]. In [35], the etination elect the bet relay from a et of noe after ecoing the ource information. The achievable ata rate with the cooperation of a ingle relay may not be able to attain the target level. Moreover, in wirele M2M network, ingle relay election may not be robut enough to provie reliability ue to the link intability an reource limitation of the MTC noe. Multiple relay election utilizing the avantage of fountain coe become a critical challenge in uch a reearch context, that we olve in our work. III. SYSTEM MODEL In thi paper, we conier a clutere heterogeneou M2M network etting where the MTC noe are groupe into a number of overlapping an ijoint cluter an ome powerful noe among them are electe a cluter hea (CH) a illutrate in Fig.. In our etup, each cluter ha a cluter hea which coorinate tranmiion within the cluter, collect ata, communicate with neighboring cluter an forwar ata to the LTE-A network. We alo aume that noe are equippe with both a long range an a hort range interface. Noe ue hort range raio interface to create an a-hoc capillary network among the noe in local M2M omain an long range interface to connect to LTE-A network. The reaon behin clutering i to avoi congetion an increae the efficiency of ata tranmiion by allowing limite number of noe to acce LTE-A network irectly. Eviently the level of congetion control epen on the number of CH. On the other han, accommoating a large number of noe in a cluter may alo caue congetion inie a cluter. In the literature, comprehenive tuie have been given on cluter eign. Communication capability, communication link quality, torage tatu an battery tatu of each noe are mainly coniere uring CH election [0] [2]. However, cluter eign an CH election are out of the cope of thi paper. In thi paper, we only focu on eigning cooperative trategie with relay election between two neighboring cluter. Let aume, the CH in a ource cluter want to en a meage to the CH of a neighboring cluter. Unlike the conventional irect communication between CH, we employ Symbol H C target C C ri C ri C cr γ ab τ ab P 0 C A x A min N R ζ Ω(x) Z co G a,b Definition Number of bit in relay probe meage Target en to en ata rate of - pair Achievable ata rate for - pair Achievable ata rate for -r i pair Achievable ata rate for of r i - pair Achievable cooperative ata rate of the network Channel gain of a-b pair Time taken by b to ecoe relay probe meage tranmitte by a Tranmiion power of each noe Set of initial relay caniate Set of ucceful relay caniate Set of electe relay noe Number of initial caniate in et C Number of relay in et A min The tranmiion efficiency Degree itribution of Raptor coe in noiy channel Channel log likelihoo ratio Path lo between noe a an noe b cooperative communication to improve the reliability of ata tranmiion between two neighboring CH. In Fig. 2, we how an example ytem moel that conit of two overlapping cluter with many ege noe that can ene ignal from both CH. Thi figure illutrate a imple cenario of ata tranmiion between two overlapping cluter where ource CH, tranmit ata targeting etination CH,. Exploiting the intrinic nature of wirele meium, the neighbor noe r, r 2, r 3 overhear the tranmiion between two CH an are able to facilitate the tranmiion by forwaring the receive information to the neighboring CH. Correponingly, r, r 2, r 3 act a relay between an, an thu achieve iverity by receiving ignal from ifferent faing path. We aume, the tranmit power of each tranmit noe i retricte to P an the noie at each receive noe j i aume to be white gauian with a variance of σj 2. Denoting the tranmitte ymbol of a X, the receive ymbol at r i an can be expree a Y j = H j X + W j, () where j = {r i, }; W j i the gauian noie at j with variance σ 2 j ; H j i moele a G j h j where h j CN (0, ) which i a circularly ymmetric complex Gauian ranom variable with zero mean an unit variance, an G j enote the path lo between noe an noe j. Then the receive SNR at j for a pair -j can be expree a γ j = H 2 jp /σ 2 j. (2) In orer to avoi interference, we aume that orthogonal channel [42], [43] are allocate to ifferent terminal i.e., inter-uer orthogonality. The ource tranmit to the etination an relay through one channel an the relay tranmit

5 5 m m m m m (a) m (b) m 2 m 2 m 2 m 2 m 2 m 2 (c) () m 4 m 4 m 4 (e) m 5 m 5 m 5 m5 m m5 5 (f) Relay caniate Selecte relay Fig. 3. Relay election proceure: a) Network topology conit of ource CH, etination CH in re circle an a number of neighbor noe in grey circle, b) broacat relay probe meage m to it neighborhoo, c) en relay probe meage m 2 a an acknowlegement, ) Neighbor noe that uccefully ecoe the relay probe meage of both CH en relay repone meage, e) en relay requet m 4 to the electe relay noe, f) Selecte relay noe confirm the requet an en relay confirmation meage m 5 to an. to the etination through other ifferent orthogonal channel. Thi aumption can be eaily implemente in practice uing orthogonal frequency iviion multiplexing (OFDM) moulate WiFi technology, where ifferent ub-carrier are aigne to ifferent uer. After relay election, we aume that all the noe ue their own fountain coing routine to encoe the ata. Table I lit the notation ue in thi paper. In the ubequent ection, we will ue ource noe (etination noe) an ource CH (etination CH), interchangeably. IV. RATELESS CODED RELAY SELECTION It ha been proven that cooperative trategie are energy efficient than non-cooperative irect tranmiion [24], [36]. However, involving more relay noe between ource an etination impoe more ignalling overhea an power conumption. On the other han, the election of the relay noe play a critical role to improve the performance of the cooperative ytem [3], [38], [4]. Coniering the cot an performance iue in M2M network, we propoe a relay election algorithm calle ratele coe incremental relay election (RCIRS), aiming at achieving the given target ata rate by employing minimum number of relay noe. We conier a ecoe an forwar relaying ytem where ource CH tranmit it information to the etination an relay uing fountain coe. After uccefully ecoing the ource CH information, relay noe encoe the receive information uing fountain coe. The relay noe an the ource noe then continue their tranmiion to the etination until the ource information i uccefully ecoe at the etination. A. Relay Selection Proceure We illutrate our propoe relay election proceure in Fig. 3. Fig. 3(a) how an example network topology with a ource CH an a etination CH by re colore circle, an a number of neighbor noe by grey colore circle. In the beginning of relay election, ource CH, broacat relay probe meage enote by m a hown in Fig. 3(b) uing preefine ratele coe to it neighbor noe an continue it broacating until it receive an acknowlegement from etination CH,. Let H bit of ata to be broacate a relay probe meage. We aume H i a long enough number. The aree of the ource an etination CH are inclue in the relay probe meage. After uccefully ecoing of m, en a relay probe meage enote by m 2 a an acknowlegment to uing ratele coe which i hown in Fig. 3(c). Then etimate the en to en ata rate for - pair bae on the time it take for the ucceful ecoing of relay probe meage tranmitte by. Thi relay probe meage tranmitte by i reliably ecoe at when τ 0 C = τ 0 log 2 ( + γ ) H, (3) where τ 0 i the uration of tranmiion an γ i the receive SNR at for - pair. In ieal fountain coe, the receiver i capable to recover the ource information a oon a the accumulate mutual information equal the entropy of the coewor. However it i impoible to generate univeral fountain coe that i imultaneouly perfect at all poible rate. In practical fountain coe, the receiver i capable to recover the ource information with overhea that i not too

6 6 large [28]. An overhea like thi will only lea to a lo of rate an oe not impact much on the analyi here. In orer to implify the analyi, ieal fountain coe an ecoer are aume at the receiver in thi ection. Therefore, the achievable ata rate per unit banwith for - pair i given by C = log 2 ( + γ ) = H/(τ 0 ). (4) At the ame time, neighbor noe proce the receive relay probe meage from ifferent CH an meaure the achievable ata rate. Let aume neighboring noe r i take τ ri an τ ri time to ecoe relay probe meage of an, repectively. Hence, the achievable ata rate for -r i pair i C ri = log 2 ( + γ ri ) = H/τ ri, (5) where γ ri the receive SNR at r i. Similarly, the achievable ata rate for r i - pair i given by C ri = log 2 ( + γ ri) = H/τ ri, (6) where γ ri i the receive SNR at r i for r i - pair. Among the neighbor noe, only thoe who receive relay probe meage from both ource an etination CH repon to the relay probe with relay repone, which i enote by a hown in Fig. 3(). The relay repone meage inclue the ata rate C ri an C ri a well a the i of both ource an etination CH. Source CH, then analyze each relay repone meage an take only thoe noe whoe ource to relay ata rate, C ri are greater than C. We refer thee noe a initial relay caniate an the et of caniate relay a initial caniate et C. The next tep i to earch for a ubet A min from the initial caniate et C. Here, A min enote our eire et of relay noe that conit of minimum number of relay noe an contribute to meet the target CH-to-CH ata rate. In regar to thi, we propoe a earch criterion an a greey algorithm, namely RCIRS algorithm. In the next two ubection, we explain the propoe earch criterion an RCIRS algorithm. After the election of et A min, the ource en the relay requet meage m 4 to the electe noe a hown in Fig. 3(e). Then the electe noe confirm thi requet an en a relay confirmation meage, which i enote a m 5, to both an a hown in Fig. 3(f). Thu, uring the eign phae of the network, we juiciouly elect the minimum number of relay noe from a large et of neighbor noe. When the network tart to actively operate, the ource noe en ata only targeting the electe relay noe along with the etination noe. Then the electe noe forwar the ource ata only to the etination an thu the overall cot i reuce while the target ata rate i maintaine. B. Relay Selection Problem In thi ubection, the relay election criterion an RCIRS algorithm are preente. The main goal of the relay election problem i to fin a et of relay noe A min from C, that ha minimum number of relay noe an alo atifie the target ata rate contraint, C target. To olve thi problem, we etablih a earch criterion. In the earch criterion, we introuce a parameter, namely achievable cooperative ata rate, C cr that repreent the ata rate between ource CH an etination CH attaine by the co-operative participation of ource, relay an etination noe. The earch criterion i tate a below: where C cr C target, (7) C + C cr = + r i A x C ri r i A x C ri C ri, (8) where A x C i a ucceful caniate et that atifie the criterion in Eq. 7. The relationhip of the parameter C cr with previouly efine ata rate of -, -r i an r i - pair hown in Eq. 8, i erive by uing Eq. 4, 5, 6 an the following equation, τ.c + r i A x (τ τ ri )C ri = H, (9) where τ i the minimum time taken by the etination to uccefully ecoe the ource meage. It i worth noting that C cr repreent the average of cooperative ata rate, which ha nothing to o with H. In practice, H, the number of tranmitte bit, influence the tart time of relay tranmiion. For thi reaon, the coe cooperative communication icue in thi paper have a weakne of incontant ata rate, i.e., the etination achieve higher ata rate a more relay join the tranmiion later. Incontant ata rate may negatively affect realtime application, uch a vieo an auio. It i, neverthele, not a ignificant problem for ata collection tak in M2M omain. Let conier, N i the total number of relay in the initial caniate et C. So, the number of all poible ucceful caniate et A x coul vary from 0 to 2 N. In the next paragraph, we have hown an example problem formulation of earching A min from et C. Let α i be an inicator variable which i equal to if r i i electe a relay an 0 otherwie, the minimal election problem can be formally efine a follow: min N i= α i.t. C cr C target, α i {0, }, r i C. (0) The above problem i a non linear integer optimization problem. Now if the ize of the network i high i.e. if N i a large number, fining A min from all poible A x C et by uing conventional Brute-Force earch woul require huge computational effort. The computational complexity of BFS increae exponentially with the number of relay caniate N, that limit it application in practical ytem. With N initial relay caniate in the election proce, an exponential complexity of 2 N woul be require to fin out A min et. In orer to reuce the computational complexity of earching A min, we propoe a greey algorithm RCIRS in the next ubection.

7 7 Algorithm RCIRS Input: C, C target Output: A min, C cr : C cr C, Sum C r 0, Sum C r /C r 0 2: A min 3: while C cr < C target o 4: for each i( C\A min o the following ) o C 5: ri = +Sum C r +C ri +Sum C r /C r+c ri /C ri 6: en for 7: r i arg max ri, ri = max( ri ) i C\A min 8: A min A min r i 9: C cr ri 0: Sum C r Sum C r + C ri : Sum C r /C r Sum C r /C r + C r i C ri 2: en while 3: return A min, C cr C. Ratele Coe Incremental Relay Selection (RCIRS) algorithm In Algorithm, ifferent tep of RCIRS algorithm i epicte. The baic iea of RCIRS algorithm i that noe in the et C are inclue in cooperative group one by one o that the cooperative ata rate i incrementally increae until the conition in Eq. 7 i atifie. In orer to minimize the number of relay, each time the caniate that make the maximum increment to C cr i ae in the cooperative group. At the beginning, electe A min i an empty et, ince no relay ha been electe, an C cr i initialize with C by coniering the ignal receive from only (line -2). After the initialization, C cr i compare with C target (line 3). If C cr i maller than C target, the achievable cooperative ata rate ri i compute by uing each caniate relay r i (line 4-6). Then the caniate relay that maximize ri i electe an inclue in the electe relay et A min (line 7-8). C cr, Sum C r an Sum C r /C r are upate accoringly (line 9-). Thi proce i repeate for remaining element of C until Eq. 7 i atifie (line 2). At the en, the algorithm return the eire relay et A min an cooperative ata rate of the ytem C cr (line 3). In thi way, the RCIRS algorithm fin a near minimal A min by limiting the caniate earch pace within N + (N ) +... = N(N + )/2 et, which i much lower than the earch pace of BFS (2 N ). V. FOUNTAIN CODED COOPERATIVE PROTOCOLS We propoe two fountain coe cooperative communication protocol, namely ource-feeback an non-ource-feeback where ource CH an relay noe ue their own fountain routine to encoe information an then tranmit to the etination through ifferent orthogonal channel. We aume that the ecoer know the encoing egree itribution of receive encoe ymbol via an extra robut irect equence prea pectrum (DSSS) channel with long equence length. Thi i ifferent from uing ientical ranom number generator at both encoer an ecoer a in [3], which nee trict ynchronization an increae the complexity in a multiple relay network. A. Source-feeback Bae Protocol (SF-protocol) In thi protocol, a ource CH encoe it information uing fountain coe an tranmit to the etination CH an relay. Relay noe attempt to ecoe ource ata an forwar the ource information to the etination CH uing their own fountain coe a oon a the information i ecoe uccefully at the relay noe. Source CH elect thi relay noe by appropriate relay election algorithm. Both ource CH an relay then continue their tranmiion until they receive an acknowlegment from the etination CH inicating that the reception ha been ucceful. The protocol work a follow: ) The ource tranmit it encoe ata targeting the etination an relay. 2) Both relay an etination accumulate information from ource tranmiion. 3) Since the ource to relay link i uperior than ource to etination link, relay ecoe the information fater than the etination. After uccefully ecoing the ource information, relay noe co-operate with the ource to tranmit to the etination. Each relay ue it own fountain encoing routine to encoe information an then tranmit to the etination through preallocate orthogonal channel. 4) The etination collect information from both ource an relay, an attempt to ecoe information once the accumulate mutual information i lightly greater than the ource information. If the ecoing i ucceful, then it en an ACK targeting the ource an relay. Otherwie it continue to collect more information. Thi proceure continue until ucceful reception at the etination. B. Non-ource-feeback Bae Protocol (NSF-protocol) In thi protocol, ource CH tranmit the ata tream encoe by a fountain coe targeting the etination an relay noe. The relay noe liten to the ource ata; a oon a a relay acquire ufficient information to ecoe ource ata, it tranmit an ACK to the ource that it reception wa ucceful. Intantly the relay noe witche from reception to tranmiion moe an co-operate with the ource to tranmit to the etination. Once the ource ha receive acknowlegment from all of the relay noe, it ceae it tranmiion hoping that the etination can uccefully receive the remaining information from relay. The protocol work a follow: ) Source generate a large number of encoe ymbol an tranmit it encoe ata targeting the etination an relay. 2) Relay an the etination conitently receive ignal from ource an accumulate the mutual information to ecoe the ource information. 3) A oon a a relay noe ha ufficient information to ecie on a coewor, it en acknowlegement

8 8 Information bit LDPC coing LDP coewor bit LT coing x x 2 3 v x xk... v2 v v v 5 Raptoe coewor bit c c c c 2 3 c 4 c 5 c.. 6 c7 c c m Channel output bit... y y 2 y 6 y7 y8 y 9 y m y3 y y 4 5 Fig. 4. The factor graph of Raptor coe. The graph i truncate to length m. to the ource an witche from reception moe to tranmiion moe. At thi point, it encoe information uing it own fountain encoing routine an tranmit on the pre-allocate orthogonal channel. 4) Source ceae it tranmiion after receiving acknowlegement from all relay noe. The upenion of ource noe may help to ecreae the energy conumption. By thi time, if the etination can not accumulate enough partial information from ource tranmiion, it epen on the relay noe to collect the remaining information. After uccefully ecoing of ource information, the etination en ACK to relay. C. Encoing an Decoing of Raptor Coe In our propoe cooperative protocol, the ource an relay ue a pecial cla of fountain coe, namely Raptor coe to encoe their information. Raptor coe i univerally capacity achieving over the binary eraure channel (BEC) [27] an nearly capacity-achieving over other channel moel uch a binary ymmetric channel (BSC), AWGN an faing channel [28], [29]. It i a concatenation of two coe, a pre-coe or outer coe an an inner coe. Uually LDPC an LT coe are ue a outer an inner coe, repectively. However, in mot practical etting Raptor coe outperform LT-coe in every apect. Ω(x) = 0.006x x x x x x x x v k ' () We aume ource CH want to tranmit k-ymbol information block to etination CH. In Raptor encoing, the k-ymbol information {x = x, x 2,...x k } i firt encoe by the pre-coe which i typically high-rate LDPC coe to prouce a k -input ymbol {v = v, v 2,...v k }. Then LT-coe that follow the egree itribution a mentione in Eq. [27] i applie to v to create Raptor coewor {c = c, c 2,...c m...}. In our ytem, A rate-0.98 LDPC coe i implemente a the outer coe of Raptor coe. In Fig. 4, the factor graph of Raptor coe i hown that i referre a G m an truncate at block length m where channel output {y = y, y 2,...y m } correpon to tranmitte Raptor coewor {c = c, c 2,...c m }. A mentione before, the egree of the encoe information i tranmitte to the etination by an extra robut CDMA channel with long equence length. The ecoer collect information progreively an make firt ecoing attempt when the receive accumulate information i little bit larger than the ource information. In noiy channel, the ecoing of raptor coe i accomplihe uing the tanar Belief Propagation (BP) algorithm. At the l th ecoing attempt, it perform BP ecoing on factor graph G m by iteratively paing the LLR (log-likelihoo ratio) meage from input bit{v...v k } to output bit {c...c m } an then from output bit back to input bit. Let µ j,l c o,v i an νv j,l i,c o enote the meage pae from the output bit c o to the input bit v i an input bit v i to the output bit c o, repectively at the j th iteration of l th ecoing attempt. In every iteration, the following meage upate rule are applie in parallel to all input an output noe in the factor graph [27] tanh µ(j,l) c o,v i 2 = tanh (Z c o ) 2 ν (j+,l) v i,c o = o o i i tanh νj,l v i,c o 2, (2) µ (j,l) c o,v i, (3) where Z co i log-likelihoo ratio (LLR) of the tranmitte bit c o. Since we ue binary phae hift keying (BPSK) a moulation cheme, the tranmitte coewor c o 0, i of equal probability. In Rayleigh faing channel, the channel intrinic log likelihoo information correponing to the output noe c o while channel tate information i available at the receiver, i formulate a [44] Z co = log pr(y o c o = 0) pr(y o c o = ) = 2 σ 2 y o.a, (4) where σ 2 = P0, y σ0 2 o i the noiy obervation of the channel an a i the normalize Rayleigh faing factor with E[a 2 ] = an enity function f(a) = 2a exp( a 2 ). In the en of l th ecoing attempt, if the tranmitte coewor i ecoe correctly, the receiver en an ACK through a noiele feeback channel to the ource to terminate the tranmiion of the current coe wor. Otherwie it collect more output ymbol an attempt again to ecoe the coewor. D. Tranmiion Efficiency In ratele coe ytem, the ource noe generate a large number of ymbol an tranmit to the etination until it receive any ACK from the etination. Hence, the probability of outage i alway riven to zero unle any contraint i impoe on ecoing elay. In thi paper, tranmiion efficiency rather than outage probability i coniere a a primary metric to evaluate the performance of the propoe protocol. In the following, we evaluate the performance of our propoe protocol bae on the require time to ecoe the ource information correctly at etination. Let n enote the number of time lot require for the etination to collect enough information from all ource an relay before it uccefully ecoe a k-bit ource meage. In thi work, we ue BPSK moulation. BPSK allow tranmiion of one bit per channel that mean a time lot i what a

9 9 tranmitter take to tranmit bit. Therefore, the tranmiion efficiency i given by ζ = k/n where the number of require time lot are normalize by the bit rate. Since the ata rate of ource-relay link are greater than the ata rate of ourceetination link, relay noe can ecoe the ource information fater than the etination. The ource information i ecoe at relay r j once the accumulate information atifie (-,0) r 3 r (.5,0) r 2 (,0) n r j C rj k, (5) r R (-.5,0) where n r j i the require time lot for r j to ecoe ource information correctly. Since BPSK i ue a moulation, the ata rate, C ab of a link, a-b can be calculate by [7]: BP SK Cab = log 2 ( + e x )e x 2γ ab 2γ ab x, 2 2πγ ab (6) where γ ab = H ab 2 P a /σb 2. In both protocol, a oon a the relay ecoe ource information, it encoe the receive information uing fountain coe an tranmit through the pre-aigne orthogonal channel to the etination. In ourcefeeback protocol, the ource noe continue it tranmiion until the receiver uccefully ecoe the ource information. Uing BPSK a moulation cheme, in thi protocol the etination correctly ecoe the ource information when the accumulate information atifie BP SK nc + R (n n r BP SK j )Cr j k, (7) j= where, R i the number of participating relay noe, C i the ata rate of - link an C rj i the ata rate of r i - link. Retricting the moulation cheme to BPSK, the maximum achievable tranmiion efficiency ζ of thi protocol can be efine a BP SK C + R ( nr j n j= )CBP SK r j = ζ. (8) In non-ource-feeback protocol, the ource noe ceae it tranmiion after all relay noe uccefully ecoe ource information. Relay noe tranmit towar the etination until the ource information i uccefully ecoe at the etination. Therefore, in non-ource-feeback protocol, the etination may reliably ecoe the ource information when the accumulate information atifie max n r BP SK C + R (n n r BP SK j )Cr j k, (9) j= where max n r = max{n r, n r 2,...n r R } i the require time for the relay that ha minimum ource to relay ata rate, min C r = min{c r, C r2,...c rr }. The maximum tranmiion efficiency, ζ of thi protocol can be efine a max n r n BP SK C + R ( nr j n j= )CBP SK r j = ζ. (20) Fig. 5. Simulation moel. Number of Active Relay (R) Tranmit SNR:P 0 /σ 2 0 (B) SRCRS RCIRS BFS Fig. 6. Number of active relay v tranmit SNR(B) (N = 25, C target = 4 bit/ec). VI. SIMULATION RESULTS The imulation cenario in our tuy conit of a pair of ource CH an etination CH with a number of relay caniate where noe are place ranomly within the area x,.5 y.5. The ource i place at (-,0) an the etination i at (,0) a hown in Fig. 5. For implicity, the gauian noie at each noe i aume to be of the ame variance σ0, 2 an each noe ha imilar tranmit power P 0. Without lo of generality, we aume a unit path lo between ource an etination, i.e., G =. We then have ( ( ri ) α where, ri an ) α G ri = ri an Gri = ri are the itance between ource an etination, ource an i th relay noe r i, an etination an r i, repectively. We aume free-pace path lo, o we have α = 2. A. Performance of Relay Selection Algorithm In thi ubection, we conuct imulation to evaluate the performance of the propoe relay election algorithm. The channel between all noe are aume to be frequencyflat block-faing Rayleigh channel. We compare our relay election metho with ource relay channel bae relay election (SRCRS) a in [36] an [40] where relay noe are electe bae on the channel quality between ource an relay without coniering the relay an etination link. Fig. 6 how the number of active relay a a function of tranmit SNR, P 0 /σ0 2 for three cheme incluing RCIRS,

10 RCIRS BFS SRCRS 0.9 Cooperative Data Rate (bit/ec) Tranmit SNR:P 0 /σ 2 0 (B) Tranmiion efficiency(ζ) ζ * (SF protocol with RCIRS) 0.6 ζ(sf protocol with RCIRS) ζ * (SF protocol with SRCRS) ζ(sf protocol with SRCRS) ζ * (NSF protocol with RCIRS) 0.5 ζ(nsf protocol with RCIRS) ζ * (NSF protocol with SRCRS) ζ(nsf protocol with SRCRS) DT Tranmit SNR:P 0 /σ 2 0 (B) Fig. 7. Cooperative ata rate v tranmit SNR(B) (N = 25, R = 4). Fig. 8. Tranmiion efficiency v tranmit SNR in Rayleigh channel. TABLE II COMPARISON AMONG DIFFERENT RELAY SELECTION METHODS (N = 25). P 0 /σ0 2(B) Metho C cr (bit/ec) R BFS RCIRS SRCRS BFS RCIRS SRCRS SRCRS an BFS having a ata rate contraint, C target = 4 bit/ec. We aume N = 25. Noe that participate in the cooperative group until the etination i able to ecoe the relay probe meage are efine a active relay. From Fig. 6, it i oberve that SRCRS employ more relay noe than RCIRS to meet the target ata rate. Thi i becaue in SRCRS, noe i inclue in the cooperative group a oon a it ecoe the ource information uccefully with the poibility of having poor link towar etination. It i alo oberve that RCIRS employ equal number of relay a BFS to meet the ata rate requirement with le computation complexity. Moreover, it i hown that at lower tranmit SNR region, a large number of relay are require to meet the ata rate requirement, which ecreae graually at higher SNR region. Fig. 7 how the cooperative ata rate v tranmit SNR, P 0 /σ 2 0 for ifferent cheme with parameter, R = 4 an N = 25. It i oberve that RCIRS almot attain the performance of BFS a expecte an outperform the SRCRS. Thi i becaue in RCIRS, noe i ae in the cooperative group bae on the ata rate of ource to relay an relay to etination link while in SRCRS, only the ata rate of ource to relay i coniere. Table II lit the maximum achievable cooperative ata rate, C cr an correponing number of active relay noe, R for ifferent value of P 0 /σ0 2 an number of initial relay caniate, N = 25. It i note that SRCRS achieve the ame C cr a BFS by employing more number of relay noe than that of the BFS. BFS maximize C cr by employing minimum number of relay noe with the cot of huge computation Tranmiion efficiency(ζ) ζ, SF protocol with RCIRS ζ, SF protocol with SRCRS ζ, NSF protocol with RCIRS ζ, NSF protocol with SRCRS Number of relay (R) Fig. 9. Tranmiion efficiency v number of relay in Rayleigh channel. complexity (e.g., 2 25 ). It i worth noting that our propoe RCIRS almot attain the C cr of BFS by employing the ame number of relay noe a of BFS with le computational complexity. At P 0 /σ0 2 = 8 B, the achievable cooperative ata rate of RCIRS i bit/ec, which i only 0.5% le than that of BFS. B. Performance of Fountain Coe Cooperative Protocol In thi ubection, we compare the performance of the propoe fountain coe cooperative protocol uner ifferent relay election metho with that of noncooperative irect communication cheme. Fig. 8 preent the performance of the noncooperative irect tranmiion (DT) an propoe protocol in term of tranmiion efficiency, ζ a a function of tranmit SNR, P 0 /σ 2 0 uner ifferent relay election metho for parameter R = 2. It i oberve that cooperative protocol ramatically improve the tranmiion efficiency than DT between two neighboring CH. It i oberve that at fixe relay election metho the ource-feeback bae protocol (SF-protocol) outperform the non-ource-feeback bae protocol (NSF-protocol). The reaon of the better performance of SF-protocol, i that in thi protocol relay an ource continue their tranmiion until the etination en back an ACK, wherea in NSF-protocol,

11 the ource top it tranmiion after the relay noe ecoe the meage uccefully. It i alo oberve that the relay election metho ignificantly influence the performance of the protocol. The propoe protocol (SF/NSF-protocol) uing RCIRS outperform it SRCRS counterpart. Furthermore, the impact of relay election metho i more noticeable on NSF-protocol than SF-protocol. It i oberve that at P 0 /σ0 2 = B, the performance gap between SF-protocol uing RCIRS an SF-protocol uing SRCRS i 2.22% while the performance gap between NSF-protocol uing RCIRS an NSF-protocol uing SRCRS i 5.45%. Thi i becaue in the SRCRS metho, relay noe are electe bae on the ource-relay link without coniering the relay-etination link. Therefore, the poibility of having poor relay-etination link of the electe relay noe an the upenion of the ource tranmiion in NSF-protocol, ignificantly egrae the performance of the NSF-protocol uing SRCRS. The maximum achievable tranmiion efficiency, ζ of both protocol uner ifferent relay election metho are alo preente in Fig. 8. It i oberve a tranmit SNR increae, the ζ curve reache the aymptote of, ue to the BPSK moulation cheme. Comparing the achieve tranmiion efficiency, ζ with the maximum achievable tranmiion efficiency, ζ for either protocol, we notice that there i an obviou rate lo, which i ue to the uboptimality of the Raptor coe. A hown in the figure, at P 0 /σ0 2 = 0 B uner RCIRS metho the rate lo of SF-protocol an NSFprotocol are 5.7% an 9%, repectively. In Fig. 9, we illutrate the tranmiion efficiency of the propoe protocol a a function of R for fixe tranmit SNR, P 0 /σ0 2 = 6 B. Here alo relay election metho influence the performance of the protocol in term of tranmiion efficiency ζ. We oberve that SF-protocol uing RCIRS metho outperform the SF-protocol uing SRCRS metho. We alo oberve that the performance of the protocol alway improve with the involvement of more relay. However, at fixe SNR, the performance ifference of propoe protocol at a fixe relay election metho graually ecreae with the increae of relay noe a hown in Fig. 9. Thi i becaue, more relay rener more mutual information to etination an reuce the impact of irect link at etination. VII. CONCLUSION Reliability of M2M network i getting more attention a an important IoT performance inicator. With repect to thi fact, thi paper evelop fountain coe cooperative cheme an invetigate their performance in clutere M2M network. Particularly, we firt propoe an RCIRS algorithm to juiciouly elect relay noe o that the require ata rate i achieve with the minimum number of participating relay. Our propoe relay election algorithm bet uit for aaptive ytem in which the ata rate between two cluter can be ajute to attain the require en to en ata rate to enure quality of ervice of the ytem. It how that our propoe RCIRS employ le relay noe than SRCRS an nearly attain the cooperative ata rate obtaine by BFS. We then evelop ource-feeback an non-ource-feeback bae protocol to improve the ata tranmiion between two neighboring CH. Regarle of the protocol ue, tranmiion efficiency can alway be improve by involving more relay noe for cooperation at the cot of aitional complexity. Simulation reult emontrate that our eign can ignificantly improve the en-to-en performance between two CH than that of non cooperative irect tranmiion. It i alo oberve that at fixe relay election metho, the ource-feeback bae protocol (SF-protocol) alway outperform the non-ource-feeback (NSF-protocol) one in term of tranmiion efficiency. REFERENCES [] G. Kortuem, F. Kawar, D. Fitton an V. Sunramoorthy, Smart object a builing block for the Internet of thing, IEEE Internet Computing, vol. 4, no., pp. 44-5, Jan.-Feb [2] G.N.Shirazi, Optimization in Wirele Senor an Machine-Type Communication Network, Ph.D. iertation, Britih Columbia Univerity, [3] J. Chen, X. Cao, P. Cheng, Y. Xiao an Y. Sun, Ditribute collaborative control for inutrial automationwithwirele enor an actuator network, IEEE Tranaction on Inutrial Electronic, vol.57, no.2, pp , Dec [4] S.Y. Lien, T.-H. Liau, C.-Y. Kao an K.-C. Chen, Cooperative acce cla barring for machine-to-machine communication, IEEE Tranaction on Wirele Communication, vol., no., pp , Jan [5] Y. Zhang, R. Yu, S. Xie, W. Yao, Y. Xiao an M. Guizani, Home M2M network: Architecture, tanar an QoS improvement, IEEE Communication Magazine, vol. 49, no. 4, pp , Apr. 20. [6] Y. Zhang, R. Yu, M. Nekovee, Y. Liu, S. Xie an S. Gjeing, Cognitive machine-to-machine communication: Viion an potential for the mart gri, IEEE Network, vol. 26, no. 3, pp. 63, May/Jun [7] ITU-R-M.234, Requirement relate to technical ytem performance for imt-avance raio interface() [imt.tech], ITU Technical Report, [8] P. Bhat, S. Nagata, L. Campoy, I. Berberana, T. Derham, G. Liu, X. Shen, P.Zong an J. Yang, LTE-avance: an operator perpective,. IEEE Communication Magazine, vol.50, no.2, pp. 04-4, Feb [9] Machina Reearch - Machine-to-Machine connection to hit 8 billion in 2022, generating USD.3 trillion revenue, [0] Exalte Project, Firt report on LTE-M algorithm an proceure, Aug. 20. [] K.R. Jung, A. Park an S. Lee. Machine-typecommunication (mtc) evice grouping algorithm for congetion avoiance of mtc oriente lte network,. Security-Enriche Urban Computing an Smart Gri, vol.78, pp , 200. [2] R.M.Huq, K.P. Moreno, Z.Hui hu, J. Zhang,O. Ohlon an M.I. Hoain, On the Benefit of Clutere Capillary Network for Congetion Control in Machine Type Communication over LTE, Proceeing of the 24th International Conference on Computer Communication an Network (ICCCN), Aug [3] A.Scaglione, D. L. Geockel an J. N. Laneman, Cooperative communication in mobile a hoc network: Rethinking the link abtraction, IEEE Signal Proceing Magazine: Special Iue on Signal Proceing for Wirele A Hoc Communication Network, vol. 23, no. 5, pp. 829, Jun [4] Z. Zhu, L. Zhang an R. Wakikawa, Supporting mobility for internet car, IEEE Communication Magazine, vol. 49, no. 5, pp , May 20. [5] E. Ekici, Yaoyao Gu an D. Bozag, Mobility-bae communication in wirele enor network, IEEE Communication Magazine, vol. 44, no. 7, pp , July [6] J. N. Laneman, D. N. C. Te an G. W. Wornell, Cooperative Diverity in Wirele Nework: Efficient Protocol an Outage Behavior, IEEE Tranaction on Information Theory, vol. 50, no. 2, pp , Dec [7] Y. Zhang an Z. Zhang, Joint Network-Channel Coing with Ratele Coe over Multiple Acce Relay Sytem, IEEE Tranaction on Wirele Communication, vol. 2, no., pp , Jan [8] A. Nea, M. Kaoch, an B. Rong, Joint network channel fountain cheme for reliable communication in wirele network. Proceeing of the IEEE ICNC, pp , 204.