Research Article Experimental Investigation of Cooperative Schemes on a Real-Time DSP-Based Testbed

Transcription

1 Hndaw Publshng Corporaton EURASIP Journal on Wreless Communcatons and Networkng Volume 29, Artcle ID , 5 pages do:.55/29/ Research Artcle Expermental Investgaton of Cooperatve Schemes on a Real-Tme DSP-Based Testbed Per Zetterberg, Chrstos Mavrokefalds, 2 Ars S. Lalos, 2 and Emmanoul Matgaks 3 ACCESS Lnnaeus Center, Royal Insttute of Technology, Osquldasväg, 44 Stockholm, Sweden 2 Research Academc Computer Technology Insttute, Patras Unversty Campus, 2654 Patras, Greece 3 Department of Electronc and Computer Engneerng, Techncal Unversty of Crete, Kounoupdana Campus, Chana, 73 Crete, Greece Correspondence should be addressed to Per Zetterberg, per.zetterberg@ee.kth.se Receved 9 November 28; Accepted 3 March 29 Recommended by Xaver Mestre Expermental results on the well-known cooperatng relayng schemes, amplfy-and-forward (AF), detect-and-forward (DF), cooperatve maxmum rato combnng (CMRC), and dstrbuted space-tme codng (DSTC), are presented n ths paper. A novel relayng scheme named selecton relayng (SR), n whch one of two relays are selected base on path-loss, s also tested. For all schemes except AF receve antenna dversty s as an opton whch can be swtched on or off. For DF and DSTC a feature selectve where the relay only forwards frames wth a receve SNR above 6 db s ntroduced. In our measurements, all cooperatve relayng schemes above ncrease the coverage area as compared wth drect transmsson. The features antenna dversty and selectve mprove the performance. Good performance s obtaned wth CMRC, DSTC, and SR. Copyrght 29 Per Zetterberg et al. Ths s an open access artcle dstrbuted under the Creatve Commons Attrbuton Lcense, whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted.. Introducton MULTIPATH fadng s one of the major obstacles for the next generaton wreless networks, whch requre hgh bandwdth effcency servces. Tme, frequency, and spatal dversty technques are used to mtgate the fadng phenomenon []. Recently, cooperatve communcatons for wreless networks have ganed much nterest due to ts ablty to mtgate fadng n wreless networks through achevng spatal dversty, whle resolvng the dffcultes of nstallng multple antennas on small communcaton termnals. In cooperatve communcaton, a number of relay nodes are assgned to help a source n forwardng ts nformaton to ts destnaton, hence formng a vrtual antenna array. Varous cooperatve protocols have been proposed and analysed n the lterature. In [2], Laneman et al. proposed two cooperatve protocols: the amplfy-and-forward (AF) protocol and the decode-and-forward (DF) protocol, where the relays would ether purely amplfy and retransmt the nformaton to the destnaton, or decode the nformaton frst and then transmt these nformaton bts to the destnaton. In [3], Anghel and Kaveh showed that the conventonal maxmum rato combnng (MRC) was the optmum detecton scheme at the destnaton for the AF and t could acheve the full dversty order of K +,wherek s the number of relays. When t comes to the DF, the optmum maxmum lkelhood (ML) detector was proposed n [4, 5]. Furthermore, many suboptmum detecton schemes have been proposed, ncludng the λ-mrc [4, 6], the smple adaptve decode-and-forward scheme [7], the cooperatve MRC (CMRC) [8], and the lnk-adaptve regeneraton (LAR) [9]. Recently, many works have been devoted to mprove the bandwdth effcency of cooperatve networks, ncludng the dstrbuted space-tme codes [] and the relay selecton [, 2]. Among those technques, the relay selecton s very attractve. The basc dea s to let the relay wth the best channel condton relay the sgnals. Snce only one relay s workng at each tme slot, a very strct tme and carrer synchronsaton among the relays s not needed. Furthermore, because the transmsson of one nformatonbearng symbol s completed wthn two-tme slots, the relay selecton has hgher bandwdth effcency than the repetton-based cooperatve strategy. In [3] the authors mplement a cooperatve codng scheme [4]. The scheme s compared wth a tradtonal

2 2 EURASIP Journal on Wreless Communcatons and Networkng noncooperatve one whle transmttng frames of a vdeo clp. From the experments, t s observed that cooperaton ncreases the qualty of the vdeo clp. In [5], the authors perform detaled smulatons of two varatons of the decodeand-forward protocol [4, 6] usng low-densty party check (LDPC) codes, and a drect transmsson scheme. It s concluded that the cooperatve schemes outperform the drect transmsson. Most of the mplementaton work that has appeared n lterature focus on mplementng varatons of a sngle protocol. Heren, we are presentng an expermental nvestgaton of several cooperaton schemes, some of whch are sophstcated. We also put focus on presentng quanttatve results and measurements n a relevant propagaton envronment. Specfcally, n ths work we have mplemented the wellknown AF, DF, and CMRC protocols, where the sgnal receved at the destnaton s combned accordng to the MRC detecton rule. Furthermore we provde expermental results for some more technques that have been recently proposed, ncludng a DSTC scheme based on the Alamout codng and a novel relay selecton scheme. The mplementatons are made on a real-tme DSP-based testbed. Fnally, n the experments, we compare the performance of the mplemented schemes n terms of outage probablty, complexty and a novel mplementaton loss measure. The paper s organsed as follows. The mplementaton of the schemes s descrbed n Secton 2 of ths paper. The expermental results are descrbed n Secton 3. The results show that, compared wth drect transmsson, the proposed cooperatve schemes ncrease coverage. By means of mplementaton loss analyss we show that the results are farly close to the theoretcal results. A more full dscusson of the conclusons drawn are gven n Secton 4. LO Laptop LO2 Fgure : Pcture of a node. SW SW PA PA TXM RXM TXM RXM DA/AD DA2/AD2 DSK673 GPIO 2. The Implementatons The testbed conssts of four nodes, where each node has two antennas, two transmtters, and two recever chans, a DSP board for processng, and a laptop PC for control. The symbol- and sample-rate used are 96 Hz and 48 khz, respectvely. A pcture of a node s shown n Fgure and a schematc s shown n Fgure 2. As shown n Fgure 2, the base-band processng s made on the DSK673 board, whch s a DSP board provded by Texas Instruments. The A/D and D/A converters receve and transmt a sgnal wth khz carrer frequency. The up- and donwnconverson between RF (766.6 MHz) and base-band s done n the transmtter (TXM) and recever modules (RXM), respectvely. More nformaton about the hardware and software are gven n [7, 8]. The system uses sharp crystal flters n both the transmtter and the recever. Ths confnes the transmt bandwdth to 96 Hz wth lttle leakage outsde ths bandwdth. However, these flers ntroduce ntersymbol nterference. The ntersymbol nterference s 5 2 db weaker than the desred sgnal. Ths s neglgble for QPSK but degrades the performance for hgher-order constellatons. In ths paper only QPSK modulaton s used. The nodes act as source, relay, destnaton, base-staton, or moble-staton n the mplementatons heren. One of IP Fgure 2: Schematc of a node. The acronyms are RF-swtch (SW), power amplfer (PA), transmtter module (TXM), recever module (RXM), and general purpose nput/output (GPIO). the nodes s called the master. Ths node sends out a synchronsaton sgnal whch s detected by the other nodes. A snusod follows the synchronsaton sequence enablng the other nodes to adjust ther up- and downconverson frequences. These synchronsaton sequences are sent at a power level of dbm whle the actual data s sent at a power level of 2 dbm. The synchronsaton s rough and gves a remanng error of one sample. The master can be any of the nodes. In our measurements the source s usually the master. However, n a few measurements the source could not be used snce the path-loss to the destnaton was too hgh. In ths case, relay 2 was nstead used as the master. The power level used for transmttng payload data s 2 dbm. Ths results n a transmtted power spectral densty of 3 dbm/khz. Ths s comparable to what can be expected to be the case n future wreless LAN-type applcatons whch may use 2 dbm transmt power over a

3 EURASIP Journal on Wreless Communcatons and Networkng 3 MHz bandwdth, whch also gves 3 dbm/khz power spectral densty. The hgher power used for synchronsaton can be motvated by the fact that when a wdeband system s synchronzed all the avalable power can be used for ths purpose, whle payload data would be transmtted on multple subcarrers usng only a fracton of the total avalable power used for a gven subcarrer. The resdual synchronsaton error of one sample has to be accounted for. Ths s done dfferently for dfferent schemes and ths s descrbed n more detals below. There s a delay of typcally 58 samples between the transmtter and the recever. Ths delay s due to dgtal antalasfltersnthed/aanda/dconvertersand(butto a less extent) delays n the analog hardware. Ths delay s taken nto account by lettng the transmttng frames be scheduled 2 symbols (whch correspond to 6 samples) before the correspondng receve frames. These delays lead to a nonneglgble overhead when swtchng between transmt and receve mode. The delay could be brought down to 4 symbols f the antalas flters of the D/A and A/D converters were removed. Unfortunately, we were not able to do ths. However, n the throughput fgures, we account for ths delay as 4 symbols nstead of 2, to show a result that better reflects the performance f ths small practcal ssue could be resolved. There s also a problem when a node transmts and then starts to receve drectly followng the transmsson. Ths leads to sx symbols beng nterfered by transents from the powerng down of the transmtter. Ths problem should be solvable wth a better hardware desgn. Therefore, we do not take these sx symbols nto account when calculatng the throughput. 2.. Amplfy-and-Forward (AF), Detect-and-Forward (DF), and Cooperatve Maxmum Rato Combng (C-MRC). Before transmttng the useful data a synchronsaton phase s executed to reduce the resdual synchronsaton error of one sample as descrbed above. In the synchronsaton phase the source frst sends a frame wth tranng symbols only, a frame whch s captured by the relay and destnaton and used to estmate the best samplng phase of the source sgnal. After recevng the tranng sgnal from the source, the relay sends a tranng sgnal so that the destnaton can be synchronsed. Twelve symbols are used to acheve the synchronsaton at the power level 2 dbm. After the synchronsaton phase, the frame structure used for transfer of payload data starts. The frame structure of the AF scheme s shown n Fgure 3. The notaton means that the node s transmttng abuffer of 48 symbols, whle means that the node s recevng a buffer of 48 symbols. Idle s a perod of 2 symbols where the node does not receve or transmt. However, processng of prevously receved sgnals does occur durng dle slots. The buffers whch are marked wth the number 6 are also dle buffers of length 6 symbols. Hardware consderatons made these extra dle slots necessary, see the ntroducton aforementoned. Note also that the transmt frames and the correspondng receve frames are offset 2 symbols due to the delay of 58 samples between the transmtter and recever, as mentoned prevously. The arrow ndcates where the frame structure s repeated. In the measurements, fve repeats are executed but n prncple any number of repeats s possble. Durng the fourth and ffth frames (wth reference to Fgure 3) the relay does the processng of the sgnal that was captured durng the prevous frame. In the case of AF, the processng conssts of downsamplng the sgnal to symbol rate. Ths sgnal s then scaled so that the maxmum sample has an ampltude whch equals the maxmum ampltude that the transmtter allows. Ths leads to a power back-off compared to the other schemes nvestgated heren, as they transmt all symbols at maxmum power level. The scaled sgnal s transmtted durng the ffth and sxth frames (wth reference to Fgure 3). Then, an dle perod of 8 symbols follows, so that the relay algns tself wth the next two bursts from the source. Optonally, the relay can decode the receved symbol sequence for debuggng purposes. The destnaton also remans dle for a perod of 2 symbols whle the source transmts. Durng the next two frames, the destnaton captures the sgnal from the source. Then, t remans dle for a perod of 2 symbols to compensate for the delay n the relay-to-destnaton chan. Then, durng the next two frames, t receves the sgnal transmtted by the relay. Durng the seventh and eghth frame (wth reference to Fgure 3) the destnaton combnes the sgnals receved from the source and relay. The crteron for selectng the th symbol x() from the th sample of the source-to-destnaton and relay-to-destnaton channels, that s, y SD ()andy RD (), respectvely, s gven by x() = arg mn x() A x w SD y SD () + w RD y RD () (w SD h SD + w RD h RD )x() 2, where A x s modulaton constellaton, h SD and h RD are the source-to-destnaton and relay-to-destnaton channels, and w SD and w RD are the recever weghts. The combnng s based on the maxmum rato combnng prncple, see [], whch means that the weghts are gven by w SD = h SD, w RD = h RD. Every burst of symbols carryng payload data s 48-symbol long. Every eght symbols, a tranng symbol s nserted whch s used for channel and nose estmaton at the recever. The modulaton constellaton used s QPSK. The detect-and-forward (DF) scheme s smlar to the AF scheme, wth the dfference that the relay detects the transmtted symbols and then retransmts the sequence of detected symbols. Thus, f there s no error n the detecton, the transmtted sgnal wll be perfect, whch s not the case wth AF. The so-called cooperatve maxmum rato combnng (CMRC) scheme s smlar to DF wth the dfference that the relay estmates ts receved SNR and encodes that nformaton so that the destnaton learns the receve SNR at the relay. Ths enables the destnaton to (partally) () (2)

4 4 EURASIP Journal on Wreless Communcatons and Networkng Repeat Source Idle Idle 6 Relay Idle Idle 6 Dest Idle Idle Tme Idle 6 Tme Fgure 3: Frame structure of AF and DF schemes. compensate for erroneous decsons that may have been made at the relay, see [9]. The compensaton s made by reducng the nfluence of the relay-to-destnaton channel n the crteron () by scalng the relay-to-destnaton weght w RD as w RD = γ eq γ RD h RD, (3) where γ eq γ RD. The optmum choce of γ eq (n terms of ) s derved n [9]. The optmum γ eq s a rather complex functon of γ SR and γ RD. We chose to approxmate ths expresson wth γ eq = mn ( γ SR, γ RD ), (4) whch s an approxmaton of the optmal γ eq at hgh SNR. In our mplementaton of CMRC we used two symbols to encode the SNR. Of the four avalable bts, two are used for actually encodng the SNR and the other two consttute a redundancy check. The relay frst estmates the SNR based on the tranng sequence. The encodng s then done so that f the SNR of sgnal receved at the relay s below 3 (n lnear scale) the two bts are set as. If the SNR s n the range 3 9, 9 27, or larger than 27, the SNR two bts are set as, and, respectvely. The two redundancy bts are set as the complement of the frst two bts. At the destnaton, the SNR of the source-relay path s assumed to be zero f the redundancy check fals. Otherwse, the low-end value of the SNR range s assumed. We set γ eq to be the mnmum of the source-relay and relay-destnaton SNRs, as s defned n (4). In an attempt to mprove on DF, prmarly to prevent the forwardng of erroneously detected bts, a selectve feature s ntroduced. Thus f the source-relay SNR s below 4 (n lnear scale), the relay stays slent durng the slots allocated for forwardng. Ths s a selectable feature. In Secton 3 we wll present results for both swtched on and swtched off mode. Another selectable opton, antenna dversty, was also ntroduced. When swtched on, the receved sgnal from two antennas s combned by means of MRC at the relay and at the destnaton. However, ths approach was only mplemented for the DF and CMRC schemes and not for AF. Assumng that the frame-structure of Fgure 3 s repeated many tmes, the overhead due to the extra frames needed for synchronsaton s neglgble. Assumng further that the dle frames can be shortened, as suggested prevously, the duty cycle of AF and DF s 43%. Ths means that 43% of the symbols receved at the destnaton contans useful unque data. Ths number ncludes overhead due to the tranng sequence. The CMRC approach has a slghtly lower duty cycle of 4% due to the overhead ncurred by transmttng the source-relay SNR. We have also mplemented a drect transmsson mode, where no relayng occurs. Ths mode uses the same ar nterface, that s, 48-symbol long frames wth sx tranng symbols and QPSK modulaton. Ths scheme has a duty cycle of 87%, snce the only overhead ncurred comes from tranng symbols Dstrbuted Space-Tme Codng (DSTC). In the synchronsaton phase of the DSTC scheme the source node sends a frame wth tranng symbols that s captured by the two relays and the destnaton, and used to estmate the best samplng offset of the source sgnal. After recevng the tranng sgnal from the source, the relays take turns sendng a tranng sgnal to the destnaton. The destnaton estmates the best samplng offset for each relay from the tranng sgnal. At ths stage somethng happens whch does not occur n the other approaches. In the other approaches the samplng offset can be taken nto account at the recever. But n DSTC the two relays are transmttng smultaneously, and a sngle offset at the recever may thus not ft both relays. Therefore, n the case of DSTC the compensaton s nstead done at the transmtter. Hence, the relays adjust the tmng of ther outgong frames one sample backward or forward (or no adjustment). In order to let the relays know n whch drecton to adjust ther tmng, ths nformaton s fed back from the destnaton to the relays n a specal frame. After havng acheved synchronsaton, the sgnallng goes nto the frame structure ndcated n Fgure 4 one that s dentcal to the frame-structure of AF, DF, and CMRC except that the two relays are transmttng at the same tme. After capturng the sgnal from the source and storng t n a buffer, the relays downsample the sequence to get symbol-spaced samples. Then, the channel s estmated and the symbol sequence s detected. The next step s to create the Alamout code sequence. Each relay plays the role of one antenna n the conventonal Alamout dversty, [2],so each relay creates a dfferent sequence.

5 EURASIP Journal on Wreless Communcatons and Networkng 5 Repeat Source Idle Idle 6 Relay Idle Idle 6 Relay2 Idle Idle 6 Dest Idle Idle Idle 6 Tme Fgure 4: Frame structure of DSTC scheme. h SR R R h RD R h RD h SD S D S D S D h SR2 R2 R2 h R2D R2 (a) Phase (b) Phase 2 (c) Phase 3 Fgure 5: The three-phase transmsson of the cooperatve system. In Phase, S transmts to the other nodes. In Phase 2, the best relay s decded. Fnally, n Phase 3, the best relay (e.g., R) transmts to D. The destnaton does not use the sgnal whch comes drectly from the source. Durng the sxth and seventh frame (wthrespect to Fgure 4), the destnaton captures the sgnal from the relays. In Alamout codng every par of symbols s, s 2 s mapped onto two consecutve outgong symbols as s, s 2 at relay and s 2, s at relay 2. The sgnal receved at the destnaton n two consecutve symbols, y and y 2, then becomes y = s s 2 h + w, (5) y 2 s 2 s h 2 w 2 where h and h 2 are the channel coeffcents assocated wth relay and 2, respectvely, and w and w 2 are nose samples. Wth h and h 2 known, s and s 2 are detected based on x and x 2 whch are obtaned as x = h y + h 2 y 2 = ( h 2 + h 2 2) s + h w + h 2 w 2, (6) x 2 = h 2 y h y 2 = ( h 2 + h 2 2) s 2 + h 2 w h w 2, (7) respectvely. In order to obtan h and h 2, symbols wth number 7, 8, 5, 6, 23, 24, 3, 32, 39, 4, 47, 48 are used for channel estmaton (the frames have 48 symbols). The equatons for obtanng a channel estmate from two consecutve tranng symbols are gven n Appendx A. As n the case of DF, the two optons selectve and antenna dversty exst. When the selectve opton s swtched on the relays are slent f the SNR s less than 4. When the antenna dversty opton s swtched on the sgnals receved from both antenna branches are combned n the relays as well as n the destnaton. The combnng scheme used s maxmum rato combnng. The duty cycle of DSTC s 36% whch s somewhat lower than for DF, as more symbols are used for channel estmaton Selecton Relayng (SR). As n the DSTC case, two relays are used. The frame structure has three phases whch are llustrated n Fgures 5 and 6. In the frst phase the source sends nformaton to the two relays and the destnaton. The relays calculate the average sgnal to nose rato (ASNR,where =, 2) over all the payload frames of the frst phase. In the second phase, the relays send ther ASNR values to the destnaton n sgnallng frames. The destnaton estmates the sgnal to nose ratos of the two relay-to-destnaton lnks drectly from the sgnallng frames (ASNR,where =, 2). Usng ths nformaton, the destnaton decdes whch relay has a better overall sourcerelay-destnaton channel. The destnaton nforms the relays about whch relay s gong to be actve n the thrd phase. The format of the frames used n Phase 2 are shown n Fgures 7(a) and 7(b). In the thrd phase the selected relay retransmts the nformaton detected from the source n

6 6 EURASIP Journal on Wreless Communcatons and Networkng Source Phase Phase 2 t t t t t Relay r r r r r t r Relay 2 r r r r r t r Destnaton r r r r r r r t Phase 3 2 t t t t t r r r r r Symbols Fgure 6: Frame defnton. the frst phase. Note that whle Fgure 6 shows fve payload frames beng transmtted n the frst and thrd phase. Ths number s actually ncreased to ten durng the measurements presented n Secton 3. Durng the second phase, the ntegrty of the frames used for sgnallng s checked by estmatng the SNR of the frames based on ther tranng sequences. If the SNR s lower than 4 (n lnear scale), then the frame s assumed to be n error. The correspondng relay wll then not be elgble for transmsson n the thrd phase. Lkewse, the relays wll not transmt f the frame sent from the destnaton to the relays durng the second phase has an SNR of less than 4. The destnaton wll not use ether of the two relays f the frames receved from both relays n the second phase are n error. If both frames are receved correctly, then the followng crteron s used for relay selecton best = arg max {mn{asnr,snr }}. (8) ={,2} TheASNRandSNRvaluesusednthecrteron(8) for selecton of the best relay are estmated dfferently from all other SNR values used n the cooperatve schemes. The dfference les n the way the nose s estmated. In the case of the ASNR and SNR values n (8) the nose s estmated n an ntal frame whch s sent before the executon of Phase, Phase 2, and Phase 3, and where there s no other transmsson. In the other cases, the nose s estmated as the dfference between the receved sgnal samples and the sgnal obtaned by multplyng the estmated channel wth the tranng symbols. A detaled descrpton of the procedure used for estmatng and sendng the SNR and ASNR values of (8) sgvennappendx B. 8 symbols Tranng symbols 2 symbols (a) Frame structure 9 symbols Tranng symbols 2 symbols (b) Frame structure 2 4 symbols Quan. ASNR 3 symbols Index Fgure 7: The transmt frame structures used n phase 2. The relay usage s reduced by 5% compared wth DSTC as only one relay out of two s chosen. The dea behnd the scheme s that channel varatons are composed of short-term varatons, due to Doppler fadng, and longterm varatons, due to obstacles between the nodes and obstructons, for example, walls. Wth the proposed scheme we should be able to select the best relay when the dfference n channel condtons between the two relays s large because of the long-term propertes, even though tme delays may somewhat alter the propagaton condtons between the moment of selecton and use. The careful reader may have notced that we have not started wth a synchronsaton phase as n the other approaches descrbed above. Instead, synchronsaton s done by embeddng known tranng symbols n the frst frame of Phase, n all the frames sent durng Phase 2, and n the frst frame sent durng Phase 3 (n the last case

7 EURASIP Journal on Wreless Communcatons and Networkng 7 ndrectly snce t relays the data sent from the source). Regardng the frst frame n Phase and Phase 3, wetreatt as known data when we synchronse, whle we assume the data to be unknown durng the detecton (the data s not used for channel estmaton though), and therefore we can calculate the also based on ths data. When we calculate the duty cycle we assume that these symbols were actually carryng payload data. The results should be the same as n a case where synchronsaton had occurred n a dedcated synchronsaton phase. The ar nterface employed for payload data s the same as for AF and DF, that s, 48 symbols, where every eght symbol s tranng. The duty cycle s 4% where the overhead of Phase 2 s ncluded, but where we have assumed that the delay from the transmtter to the recever s reduced from the actual value of 2 symbols down to 4 symbols. There s room for reducng the overhead of phase 2 by shortenng the control frames and by slght modfcatons of the scheme. Snce there s a possblty for the destnaton to select nether of the two relays, t would be possble to skp phase 3 f ths nformaton can be relayed to the source. Ths was however never mplemented. As n all the other approaches (except AF) there s an antenna dversty opton where the sgnals from the two antenna branches are combned by MRC at the relay and the destnaton. 3. Results A measurement campagn was conducted n an ndoor offce envronment (see Fgures and ). In the campagn a source (S), two relays (R, R2), and a destnaton (D) were used, although relay R2 s only n DSTC and SR. Some of the postons of these nodes durng the measurements are llustrated n Fgure 2. Inordertobeabletocompareallfveschemeswth dfferent optons, a measurement procedure consstng of measurement runs was developed. Wthn each measurement run twenty-four dfferent confguratons were run n sequence. In Table below we lst the sequence of confguratons n one measurement run. The reader may note that some confguratons are dentcal. Each measurement run was conducted under statonary condtons, that s, there were no people movng on the floor plan and the source, the relays and the destnaton were all standng stll. Ths s not a requrement for the schemes to work but t makes t more lkely that the schemes see the same propagaton channels. The fact that some confguratons n one measurement run are dentcal can be used to verfy the smlarty of the channel condtons under whch the dfferent confguratons are tested. A total of 47 measurement runs were conducted. The postons of the two relays and the destnaton were changed before every run. Each scheme transmtted ten payload frames of 48 symbols. The channel estmates obtaned durng these frames were saved and made avalable for postprocessng. We also calculate the bt error rate () and the number of clockcycles used by the DSPs. In addton to these metrcs, some scheme specfc results are also measured. The nose level Table : Lst of confguratons n one measurement run. Confguraton Scheme Antenna dversty Selectve opton Drect No No 2 AF No No 3 DF No No 4 CMRC No No 5 DSTC No No 6 SR No No 7 Drect Yes No 8 AF No No 9 DF Yes No CMRC Yes No DSTC Yes No 2 SR Yes No 3 Drect No No 4 AF No No 5 DF No Yes 6 CMRC No No 7 DSTC No Yes 8 SR No No 9 Drect Yes No 2 AF No No 2 DF Yes Yes 22 CMRC Yes No 23 DSTC Yes Yes 24 SR Yes No was measured and found to be very smlar on all antenna branches of all the nodes. In Fgures 8 and 9 the cumulatve dstrbuton of the SNR of all propagaton paths that are nvolved n the schemes s shown (the SNR s calculated by dvdng the channel estmate level wth the nose level of the recever n queston). The curves show that the relay 2 generally has a better channel to the source whle relay has better channel to the destnaton. The worst channel s that between the source and the destnaton. It can also be noted that the SNRs are very low whch represents challengng condtons. In Secton 3. we do a straghtforward analyss of the measurement results at hand whle n Secton 3.2 we do an analyss whch provdes more nsght and s less dependent on the scenaro chosen. 3.. Straghtforward Comparson. The most straghtforward way of comparng the dfferent schemes s to look at the bt error rate statstcs over the 47-measurement runs. In Table 2 we show the outage probablty. We defne ths probablty as the fracton of frames whch have at least one bt n error. In order to make a far comparson of the drect scheme, whch has a duty cycle of about two tmes that of the other schemes, we assume that the drect scheme repeats every frame two tmes and that the recever s able to determne whch of the two copes of the same frame has the least number of bt errors (ths reduces outage probablty from 74% to 7%).

8 8 EURASIP Journal on Wreless Communcatons and Networkng Probablty SNR<x Fgure : Node nsde offce (db) S > R, antenna S > R, antenna 2 S > R2, antenna S > R2, antenna 2 Fgure 8: Cumulatve dstrbuton of the SNR of the channel between the source and the relays..9.8 Probablty SNR<x R > D, antenna R > D, antenna 2 R2 > D, antenna (db) 2 4 R2 > D, antenna 2 S > D antenna S > D antenna 2 6 Fgure : Node n corrdor. feature s swtched on. Lkewse, the performances of DSTC and SR are very smlar, agan assumng the selectve feature s swtched on. Table 3 shows the probablty of a hgher than 5%, that s, we allow a few bt errors n each frame. Under ths crteron, the performance of AF s better than the performance of DF and CMRC. The comparson n ths secton can be crtcsed for beng hghly dependent on the selecton of postons for the source, relays, and destnaton. Therefore, we analyse the performance n terms of mplementaton loss n the next secton. Fgure 9: Cumulatve dstrbuton of the SNR of the channels to the destnaton. As may be notced, some of the confguratons are actually dentcal. For nstance, the second row of Table 2 shows the results for AF repeated four tmes. However, they correspond to dfferent measurement tme slots n the sequence of Table.The dfference between multple values for the same confguraton s n the range 3%. Ths shows that the relatve comparsons between the dfferent confguratons based on Table 2 are meanngful. We may mmedately conclude that the features selectve and antenna dversty consstently mprove the performance. The performance of CMRC s better than that of AF. The performance of DF and CMRC s smlar f the selectve 3.2. Implementaton Loss Analyss. As has been mentoned, we use QPSK modulaton n our measurements. The bterror rate () versus SNR (γ) n an addtve whte Gaussan nose (AWGN) channel for ths scheme s gven by where Q(x)sdefnedby Q(x) = = Q ( γ ), (9) t=x exp ( t 2 /σ ). () 2πσ Ths s a theoretcal expresson whch assumes no mperfectons such as frequencyoffset, synchronsaton errors, and so forth. When a Raylegh fadng model s used, γ s assumed to

9 EURASIP Journal on Wreless Communcatons and Networkng 9 D R D D R2 R2 R2 R S 33 m Fgure 2: Some of the postons of the nodes used durng the measurements. S = source, R = relay, R2 = relay 2, D = destnaton. Table 2: Outage probablty: the percentage of frames wth one bt error or more. The notaton (A) ndcates that antenna dversty s swtched on, whle (S) ndcates that the selectve feature s used. Drect 7 62 (A) 7 64 (A) AF DF 6 52 (A) 54 (S) 49 (A,S) CMRC (A) (A) DSTC (A) 38 (S) 26 (A,S) SR (A) (A) Table 3: Outage probablty: the percentage of frames wth 5% bt errors or more. The notaton (A) ndcates that antenna dversty s swtched on, whle (S) ndcates that the selectve feature s used. Drect 64 5 (A) (A) AF DF (A) 4 (S) 3 (A,S) CMRC (A) (A) DSTC (A) 26 (S) 5 (A,S) SR 25 4 (A) 28 5 (A) be exponentally dstrbuted wth mean γ. The dstrbuton functon of γ s then gven by f γ = γ exp( x/γ ). () The mean average over fadng can then be calculated as = E γ { Q ( γ )}. (2) Ths equaton can be used as the bass for obtanng the mean under any propagaton model by generatng a lot of snapshots of the SNR (.e., γ) from the propagaton model and then calculate the for each snapshot usng the Q( γ) formula, and fnally calculatng the average. In the case of two-branch receve dversty n Raylegh fadng, wth maxmum rato combnng (MRC), the SNR of the combned channel can be smulated as γ = γ + γ 2, (3) where γ and γ 2 are the SNR of the two branches. If the two branches are ndependent Raylegh fadng the SNR of combned channel, γ, wll be χ 2 (4) dstrbuted. The combned channel wll have a hgher mean SNR and a lower varance than the two ndvdual branches. Ths wll concentrate the dstrbuton of the resultng. Ths s often a desrable effect and s known as channel hardenng. The concept of channel hardenng s also what s used n cooperatve relayng. In cooperatve relayng the hardenng comes from gatherng the energy from several dstrbuton paths for the transmtted sgnal. The queston from an mplementaton pont of vew s whether n practse we are able to combne all the dfferent channels so that (2) stll apples. A straghtforward ad hoc modfcaton of (2)s ( )} γ = E γ {Q, (4) γ loss where γ loss s the mplementaton loss. If we can characterse the mplementaton loss, the performance n any gven envronment can be obtaned once the propagaton scenaro and user dstrbuton s known. In our reference scheme, drect transmsson, the SNR s that of the source-destnaton channel, and wth dversty we add the SNRs of the two dversty branches, just as we dd above. For AF, DF, and CMRC we combne the source to destnaton channel wth the channel that passes through the relay. It may be argued that the relay n ths case acts as two concatenated AWGN channels and therefore the channel through the relay can be seen as one AWGN by addng the nose of the source-to-relay and relay-to-destnaton lnks. Thus the SNR of the resultng channel s gven by γ AF = γ DF = γ CMRC = γ SD +(γ SR + γ RD). (5) When dversty s appled n DF or CMRC each SNR n the equaton above should be the sum of the SNR of the two dversty branches. In the DSTC scheme there s no drect path but an attempt to combne the energy of both relays and therefore the resultng SNR s gven by γ DSTC = (γ SR + γ RD) +(γ SR2 + γ R2D). (6) In the SR scheme fnally, we select the best of two relay paths and therefore (5) above generalses to γ SR = γ SD +max ( (γ SR + γ RD),(γ SR2 + γ R2D) ). (7) In Fgures 3 to 25 we have marked the measured bt error rate () and the combned SNR (as defned for each scheme by the equatons above), for every receved frame wth an x. We have also plotted the as defned by (4) usng dfferent values for the mplementaton loss γ loss.the dea s to subjectvely select a value of γ loss that seems to ft well wth the measurement ponts. When we do ths, t seems

10 EURASIP Journal on Wreless Communcatons and Networkng.7 Drect.7 Drect wth antenna dversty db mplementaton loss Fgure 3: Implementaton loss plot for drect transmsson. The x are measurement ponts and the curves are theoretcal curves for dfferent mplementaton loss values. db mplementaton loss Fgure 4: Implementaton loss plot for drect transmsson wth antenna dversty. The x are measurement ponts and the curves are theoretcal curves for dfferent mplementaton loss values. approprate to put most focus on a range of SNRs where starts to approach zero. There s a problem wth ths analyss when t comes to AF. The symbols used for channel estmaton are affected by the nose at the relay, and of the back-off. Thuswecan not estmate the relay-to-destnaton propagaton channel at the destnaton. For ths reason we have used the SNRs estmated for DF nstead of those actually estmated for AF. Ths ntroduces an error snce the channel s not entrely constant Drect Transmsson. For the drect transmsson the mplementaton loss s approxmately db n the range of SNRs from 5 to db, both wth and wthout dversty Amplfy-and-Forward (AF). Amplfy and forward has a loss of approxmately 2.5 db n the range of SNRs from 5 to db Detect-and-Forward (DF). Wthout the selectve feature, DF gves mplementaton losses of up to 2 db. Wth the feature swtched on, the loss s about 4 db wthout antenna dversty and 5 db wth antenna dversty Cooperatve Maxmum Rato Combnng (CMRC). Cooperatve maxmum rato combnng gves an mplementaton loss of about 2.5 db, both wth and wthout antenna dversty. The results of the drect comparson n Secton 3. showed a slght advantage for CMRC when amng for zero bt error rate. Ths advantage s hard to fnd when comparng Fgures 5 and 8. However, for SNRs above db the performances of both schemes are very smlar Dstrbuted Space-Tme Codng (DSTC). Wthout the selectve feature, the performance s very poor wth mple AF 2 db mplementaton loss Fgure 5: Implementaton loss plot for amplfy-and-forward, the curves are theoretcal curves for dfferent mplementaton loss values. mentaton losses of up to 2 db. Wth the selectve feature, the loss s db wth some sort of typcal value around 5 db. Ths s true both wth and wthout antenna dversty Selecton Relayng (SR). In selecton relayng (wthout antenna dversty) the maxmum mplementaton loss s db. However, f we dsregard data wth SNR less than 8 db, we see an mplementaton loss of about 2 db except for one outler (SNR =.3dB, = 3%). When antenna dversty s swtched on, the mplementaton loss s about 3 4

11 EURASIP Journal on Wreless Communcatons and Networkng.7 DF.7 CMRC db mplementaton loss Fgure 6: Implementaton loss plot for detect-and-forward (DF), the curves are theoretcal curves for dfferent mplementaton loss values. db mplementaton loss Fgure 8: Implementaton loss plot for cooperatve maxmum rato combnng (CARA), the curves are theoretcal curves for dfferent mplementaton loss values DF wth selectve 2 db mplementaton loss Fgure 7: Implementaton loss plot for detect-and-forward (DF) wth the selectve feature, the curves are theoretcal curves for dfferent mplementaton loss values CMRC wth dversty 2 db mplementaton loss Fgure 9: Implementaton loss plot for cooperatve maxmum rato combnng (CMRC) wth antenna dversty, the curves are theoretcal curves for dfferent mplementaton loss values db for SNRs above 8 db, except for one outler (SNR = 2.5, = 2.5%) Complexty. All the processng was done on 673 floatng pont processor from Texas Instruments whch runs at a 225 MHz clock. The numbers of clock-cycles consumed per frame for the dfferent confguratons are lsted n Tables 4 and 5 (usng the same orderng as n Table 2). Table 4 s about the number of clock-cycles n the destnaton whle Table 5 s about the number of clock-cycles n the relay. The code was wrtten n C and compled usng the compler provded by Texas Instruments wth all optmsatons swtched on, and set to mnmse the number clock-cycles needed. The code was wrtten so that all mportant loops are ppelned. We tred to keep the memory usage low to mnmse the number of cache msses. All programs and data were located n the nternal memory. The number of clock cycles shown below does not nclude up- and downconverson and channel flterng and pulse-shapng snce these operatons are mplemented n FPGA or ASIC

12 2 EURASIP Journal on Wreless Communcatons and Networkng DSTC DSTC wth antenna dversty db mplementaton loss Fgure 2: Implementaton loss plot for dstrbuted space-tme codng (STC), the curves are theoretcal curves for dfferent mplementaton loss values. db mplementaton loss Fgure 22: Implementaton loss plot for dstrbuted space-tme codng (DSTC) wth antenna dversty, the curves are theoretcal curves for dfferent mplementaton loss values DSTC wth selectve DSTC wth antenna dversty and selectve db mplementaton loss Fgure 2: Implementaton loss plot for dstrbuted space-tme codng wth the selectve feature (DSTC), the curves are theoretcal curves for dfferent mplementaton loss values. db mplementaton loss Fgure 23: Implementaton loss plot for dstrbuted space-tme codng (DSTC) wth antenna dversty and the selectve feature, the curves are theoretcal curves for dfferent mplementaton loss values. n a commercal mplementaton. The overhead for storng the bt error rate and SNR measurements s not ncluded. The results for the complexty of the destnaton n AF may be surprsngly hgh. The reason s that ths scheme was not as effcently mplemented as the other schemes. (The code of the AF mplementaton used some unnecessary buffers storng ntermedate results n the destnaton whch could be avoded. These buffers ncrease the number of cache-stalls and thereby the cycle-count.) So the actual value should be the same as for DF snce the same processng s done n the destnaton. The tme avalable for dong the processng s 48 symbols. Wth the symbol rate of 96 Hz the number of clockcycles avalable per frame s.25e6. Thus, we are usng less than.6% of the resources avalable n the DSP. There s a fxed-pont verson of the processor, called 646, whch has a clock frequency of.2 GHz. None of the processng done requres a large dynamc range and therefore a fxed

13 EURASIP Journal on Wreless Communcatons and Networkng SR 2 db mplementaton loss Fgure 24: Implementaton loss plot for selecton relayng (SR), the curves are theoretcal curves for dfferent mplementaton loss values SR wth antenna dversty 2 db mplementaton loss Fgure 25: Implementaton loss plot for selecton relayng (SR) wth antenna dversty, the curves are theoretcal curves for dfferent mplementaton loss values. pont mplementaton could be made wthout ncreasng the number of clock-cycles. It may seem that the amplfy-and-forward technque would requre much less computatonal power n the relay than the other schemes at the relay. However, note that n a TDD mplementaton the relay must stll do synchronsaton and subsample the sgnal (one sample per symbol nstead of fve samples per symbol). Moreover, we scale every burst to make optmum use of the avalable dynamc range of the D/A converter Table 4: Number of clock-cycles used per frame at the destnaton. Drect (A) (A) AF DF (A) 2496 (S) 3996 (A,S) CMRC (A) (A) DSTC (A) 2788 (S) 486 (A,S) SR (A) (A) Table 5: Number of clock-cycles used per frame at the relay. AF DF (A) 3432 (S) 526 (A,S) CMRC (A) (A) DSTC (A) 3276 (S) 576 (A,S) SR (A) (A) What should not be forgotten regardng the complexty of relayng schemes s the memory requred for storng the sgnal to be relayed n the relays. In the DF, CMRC, and DSTC schemes the requred amount of memory are two frames of 96 bts each. In the SR schemes, ten such frames are stored. In the AF technque we need to store the samples of the receved sgnal, for example, usng 6 bts for real and magnary parts, respectvely (n our mplementaton we have stored them as floats). Thus, these lead to a memory requrement n the relay of 24 bytes for DF, CMRC, and DSTC, 2 bytes for SR and 384 bytes for AF. These number wll scale wth the bandwdth f multple subcarrers are ntroduced. Other complextes that should be consdered are the synchronsaton requrements. Here, we have assumed that the transmsson wll go on for long enough for the overheads durng the synchronsaton phase to be neglected. Ths s also a queston of the functonalty of the upper layers, that s, how the source, relay, and destnaton are set up and how spectrum resources allocated. The DSTC scheme requres the relays to adjust the tmng of the transmtted sgnal so that the sgnals from both relays to arrve algned at the destnaton. Ths s probably not a very problematc ssue n a commercal mplementaton as the destnaton wll need to acknowledge packets, and therefore there wll be sgnallng from the destnaton to the relays n any case. 4. Concluson We have mplemented four well-known cooperatve relayng schemes: amplfy-and-forward (AF), detect-and-forward (DF), cooperatve maxmum-rato combnng (CMRC), and dstrbuted space-tme codng (DSTC), and one novel scheme selecton relayng (SR), see Secton 2. In the novel scheme, SR, we select one out of two relays, on a slow bass, that s, we only am to select the relay whch has the best channel n average (takng nto account both the source to relay and the relay to destnaton path), that s, we do not am to track the fast fadng but only the path loss. For the DF and DSTC we ntroduced a feature selectve where the relay only forwards a frame f ts receve SNR s

14 4 EURASIP Journal on Wreless Communcatons and Networkng better than a threshold (4 db), and otherwse stays slent. For all schemes except AF, we also ntroduced antenna dversty by means of maxmum rato combnng. We measured the performance of all fve schemes plus drect transmsson (whch s a reference case), wth and wthout antenna dversty and the selectve optons. The measurements were done n an ndoor offce envronment under challengng condtons, that s, all lnks experenced low sgnal to nose ratos. As shown n Secton 3., all schemes mproved the coverage area over drect transmsson. The feature selectve helped mprove the performance of DF and DSTC sgnfcantly. Usng antenna dversty was also an effectve means for mprovng performance. The greatest performance mprovements were acheved usng DSTC and SR whch utlse two relays. We also analysed the mplementaton loss of our mplementatons. Ths was obtaned by calculatng the theoretcal based on measured channels from the source to the relays, from the relays to destnaton, and the drect path from the source to the destnaton. The number obtaned was compared wth the actual. By dong so t was evdent that DF and DSTC need the selectve opton to functon properly. Dong so DF and DSTC have an mplementaton loss of around 5 db, whle CMRC, AF, and SR has an mplementaton loss of around 2.5 db. Drect transmsson has the smallest mplementaton loss of approxmately db. It was noted that CMRC performs better than AF, when countng the number of frames wth bt errors whle AF performs better than CMRC when a few errors are allowed. We mplemented our system on a floatng pont DSP and used.6% of ts resources for channel estmaton and detecton. When ncreasng the bandwdth of the system the load on the DSP should ncrease proportonally to the bandwdth expansons. Surprsngly, the mplementaton of the amplfy and forward technque s not less computatonally expensve than the other approaches. Ths s related to the fact that we are usng a TDD system and therefore the relay needs to synchronse, store, and forward the receved sgnal. Fnally, the AF soluton needs more memory than, for example, DF snce t does not store decoded bts but rather sgnal samples. Appendces A. Channel Estmaton n DSTC Every 8 symbols we put two consecutve tranng symbols that the relay and destnaton can use to estmate the channels. If s and s + are tranng symbols, then we get y = s s + h + w. (A.) y + s + s h 2 w + If we stack all equatons related wth tranng data, we get the expresson y = Sh + w. (A.2) The least-squares estmate of the channel h s gven by the expresson ĥ = ( S H S ) S H y. (A.3) It can be shown that ths tranng scheme s not only convenent but also optmal, n terms of mean-square channel estmaton error, snce the columns of S are orthogonal. B. Estmaton and Encodng of the ASNR Values n SR If y(n), n =,,..., N, are the samples correspondng to the N transmtted tranng symbols s t (n), the channel s estmated by h = N N n= y(n) s t (n). (B.) The relay nodes also need to know the nose varance σ 2 n order to calculate the ASNR value that are sent to the destnaton n Phase 2. Ths s done by measurng the nput level n the frst 48-symbol long frame, see Fgure 6. The SNR s estmated by the destnaton as SNR = h 2 σ 2, (B.2) where s the 2 norms of the channel. The ASNR value whch are estmated by the relays s an average over the frame n the frst phase, that s, ASNR = N SNR(n). (B.3) N The ASNR value s normalsed n the form abs exp,where abs, and exp {,, 2,...,5}. Theabssrounded up to the nearest nteger and hence the values that t can eventually acqure are {,, 2,...,}. For example, f the ASNR value s equal to 4325, t s normalsed n the form and, so, abs = and exp = 3andafterthe roundng abs = 4. As another example, f the value s 5678, then the fnal values are abs = 6andexp= 4. The values are then transformed nto symbols (two for each). Ths s smply done by consderng the 4-dgt bnary representaton of each number. Hence, f abs = 4, the bnary form s, and so the symbols correspondng to the ndexes and are send to the destnaton. As another example, f exp = 6, the bnary form s, and, n ths case, the ndexes are and 2 (there are four ndex s,, 2, 3 correspondng to the four QPSK symbols). The symbols correspondng to these ndexes are sent to the destnaton usng the frames of Fgure 7(a). The destnaton receves the frames, does the opposte steps to get the abs and the exp and from them acqures the ASNR s. Also, t has already calculated the SNR values and, hence, t uses (8) to decde whch relay s the best. If the best relay s the relay R, the destnaton sends the symbol that corresponds to ndex. If t s nconclusve, t sends the n=

15 EURASIP Journal on Wreless Communcatons and Networkng 5 ndex. The decson s sent to the relays usng the frame of Fgure 7(b). As observed from the fgure, three symbols are used for ths nformaton because the destnaton repeats the ndex three tmes. Ths s done because the relays acqure the ndex by employng a majorty procedure. That s, n order to decde n favour of R, the ndex must appear at least two tmes after the detecton. If not, the relays decde that the best relay s nconclusve. Acknowledgment Ths work was performed n part wthn the framework of the EU funded IST COOPCOM Project. References [] J. G. Proaks, Dgtal Communcatons, McGrawHll, Sngapore, 4th edton, 2. [2] J.N.Laneman,D.N.C.Tse,andG.W.Wornell, Cooperatve dversty n wreless networks: effcent protocols and outage behavor, IEEE Transactons on Informaton Theory, vol. 5, no. 2, pp , 24. [3] P. A. Anghel and M. Kaveh, Exact symbol error probablty of a cooperatve network n a Raylegh-fadng envronment, IEEE Transactons on Wreless Communcatons, vol.3,no.5, pp , 24. [4] A. Sendonars, E. Erkp, and B. Aazhang, User cooperaton dversty part I: system descrpton, IEEE Transactons on Communcatons, vol. 5, no., pp , 23. [5] D. Chen and J. N. Laneman, Modulaton and demodulaton for cooperatve dversty n wreless systems, IEEE Transactons on Wreless Communcatons, vol. 5, no. 7, pp , 26. [6] A. Sendonars, E. Erkp, and B. Aazhang, User cooperaton dversty part II: mplementaton aspects and performance analyss, IEEE Transactons on Communcatons, vol. 5, no., pp , 23. [7] P. Herhold, E. Zmmermann, and G. Fettwes, A smple cooperatve extenson to wreless relayng, n Proceedngs of the Internatonal Zurch Semnar on Communcatons (IZS 4), pp , Zurch, Swtcherland, February 24. [8] T. Wang, A. Cano, and G. B. Gannaks, Effcent demodulaton n cooperatve schemes usng decode-and-forward relays, n Proceedngs of the 39th Aslomar Conference on Sgnals, Systems and Computers, pp. 5 55, Pacfc Grove, Calf, USA, October-November 25. [9] T. Wang, R. Wang, and G. B. Gannaks, Smart regeneratve relays for lnk-adaptve cooperatve communcatons, n Proceedngs of the 4th Annual Conference on Informaton Scences and Systems (CISS 7), pp , Prnceton, NJ, USA, March 26. [] J. N. Laneman and G. W. Wornell, Dstrbuted space-tmecoded protocols for explotng cooperatve dversty n wreless networks, IEEE Transactons on Informaton Theory, vol. 49, no., pp , 23. [] Y. Zhao, R. Adve, and T. J. Lm, Symbol error rate of selecton amplfy-and-forward relay systems, IEEE Communcatons Letters, vol., no., pp , 26. [2] A. Bletsas, A. Khst, D. P. Reed, and A. Lppman, A smple cooperatve dversty method based on network path selecton, IEEE Journal on Selected Areas n Communcatons, vol. 24, no. 3, pp , 26. [3] S. Sngh, E. Sddqu, T. Koraks, P. Lu, and S. Panwar, A demonstraton of vdeo over a cooperatve PHY layer protocol, n Proceedngs of the 4th Annual Internatonal Conference on Moble Computng and Networkng (MobCom 8), San Francsco, Calf, USA, September 28. [4] A. Stefanov and E. Erkp, Cooperatve codng for wreless networks, IEEE Transactons on Communcatons, vol. 52, no. 9, pp , 24. [5] M. Karkoot and J. R. Cavallaro, Cooperatve communcatons usng scalable, medum block-length LDPC codes, n Proceedngs of the IEEE Wreless Communcatons and Networkng Conference (WCNC 8), pp , Las Vegas, Nev, USA, March-Aprl 28. [6] H. van Khuong and H. Y. Kong, Performance analyss of cooperatve communcatons protocol usng sum-product algorthm for wreless relay networks, n Proceedngs of the of the 8th Internatonal Conference on Advanced Communcaton Technology (ICACT 6), vol. 3, p. 273, Phoenx Park, Korea, February 26. [7] P. Zetterberg, Wreless development laboratory (WIDELAB) equpment base, Tech. Rep. IR-S3-SB-36, Royal Insttute of Technology, Stockholm, Sweden, March 23. [8] P. Zetterberg and A. Lavas, D.3.: software and hardware support functonalty for frst selecton of schemes to be mplemented, Tech. Rep. FP , CoopCom, Chana, Greece, November 27, home.php. [9] T. Wang, A. Cano, G. B. Gannaks, and J. N. Laneman, Hgh-performance cooperatve demodulaton wth decodeand-forward relays, IEEE Transactons on Communcatons, vol. 55, no. 7, pp , 27. [2] S. M. Alamout, A smple transmt dversty technque for wreless communcatons, IEEE Journal on Selected Areas n Communcatons, vol. 6, no. 8, pp , 998.