Performance Evaluaion of a Proocol for Radio over Fiber Wireless LAN operaing in he -GHz band Hong Bong Kim, Adam Wolisz Telecommunicaion Neworks Group Deparmen of Elecrical and Compuer Engineering Technical Universiy of Berlin Sekr FT- Einseinufer 7 Berlin Germany Email: hbkim, wolisz @ee.u-berlin.de Absrac Wireless neworks using radio over fiber (RoF) echnology operaing in millimeer-wave bands have been suggesed as promising soluions o mee increasing user bandwidh and mobiliy demands. A sysem based on his echnology has properies quie differen from hose of convenional WLAN sysems in ha every room in a building hoss a leas one picocell having is own base saion (BS). Thus a challenging problem lies in he medium access conrol () proocol design so ha i can suppor QoS requiremens as well as a fas and easy handover. A proocol (Chess Board Proocol) based on frequency swiching (FS) codes has been proposed by he auhors considering he siuaion [7]. In his paper performance evaluaion resuls for simple six varians of i are described and discussed. I. INTRODUCTION In order o mee modern ever increasing user bandwidh and mobiliy demands, a wireless nework based on radio over fiber (RoF) echnology operaing in millimeer-wave (mmwave) bands has been suggesed as a promising soluion. In his nework mm-wave signal is ransmied over opical fiber beween he conrol saion (CS) and base saions (BSs), and he BSs serve as access poins for mobile hoss (MHs). Since he quie differen properies of mm-wave from hose in he usual wireless LAN (WLAN) bands (. or GHz) he size of he picocells is limied o a mos a room in an indoor environmen, hus leading o a very large number of BSs and frequen handovers of MHs beween picocells. Therefore simple and cos-effecive BS will be a key o he success of he sysem. Recelly much of he research in his field has been focused on such componens operaing in mm-wave bands [] []. Of mm-wave bands especially he GHz band is of much ineres since a massive amoun of license-free specrum has been allocaed wih a worldwide overlab of GHz (9 GHz) []. A proocol (Chess Board proocol) for he sysem feauring fas and easy handover and QoS suppor has been proposed by he auhors [7], which is based on frequency swiching (FS) codes. Adjacen picocells employ orhogonal FS codes o avoid possible co-channel inerference. This Fig.. CS BS BS BS f f Down-link f Up-link f f fm+ fm+ fm+ fm+ fm+ Picocell Picocell Picocell (a) BWoal 7 9 (c) 7 9 frame( f ) slo( s ) (d) BWch Opical Fiber BWg f fm fm+ fm Sysem descripion. (a) A RoF WLAN sysem operaing in he millimeer-wave band, (b) The oal sysem bandwidh is subdivided ino channels, where, and are he oal sysem bandwidh, he channel bandwidh and he guard bandwidh, respecively, (c) and (d) show frequency swiching (FS) paerns for down- and up-link when he number of channels is five. mechanism allows a MH o say uned o is frequency during handover, which is a major characerisic of he proocol. In his paper simulaion sudy for simple six varians of he proocol is described. The paper sars wih a brief descripion of he Chess Board proocol. In secion III simulaion resuls are shown and discussed. Secion IV summarizes he work. II. CHESS BOARD PROTOCOL DESCRIPTION A. Basic Operaions A brief descripion of he Chess Board proocol is given in his secion, please refer o [7] for more deail. The simple srucure of he BS and -GHz wave characerisics leads o a cenralized nework archiecure wih many picocells, where mos of he BS funcions of convenional WLANs are shifed o he CS (Fig. (a)). (b)
By subdividing he oal sysem bandwidh, (frequency)!! channels "#%$ are obained (Fig. (b)), where channels are used for down-link ransmission and he oher channels &#(')#('*"!"#%$ for up-link ransmission. A pair of frequency channels +, #('*+$ are used ogeher o + suppor #('*+ $ frequency division! "! duplex (FDD) operaion, where (-/. ) are for down- and up-link ransmission, respecively. In addiion, he ime axis is also subdivided ino ime slos of equal lengh and ime slos are grouped ino a frame. Every BS suppors all channels, each of hem, however, being suppored in each proper ime slo. Fig. (c) and (d) show FS paerns for down- and up-link, respecively, when is five. During every frame ime, each of he ime slos and channels is uilized once and only once. Adjacen picocells mus no use idenical FS paerns o avoid possible co-channel inerference. And one FS paern can be used in several picocells if hey are spaced far enough ha here is no co-channel inerference beween hem. For proper operaions using FS paerns, we assume ha he sysem is synchronous. Fig. shows down- and up-link slo formas and essenially, hey have he same forma. The down-link slo begins wih a address indicaing he desinaion of he slo. This address is followed by a permission field, which auhorizes ransmission of he MH specified in he Address in he following up-link slo. The nex field is for down-link payload, desined o he MH specified by he address. The las field consiss of anoher address and reservaion resul which indicae if he reques for bandwidh from he addressed MH is successfully confirmed or no. The up-link slo consiss of wo pars. The firs one is used for uplink daa ransmission and i consiss of MH s address, piggyback field, and payload field. The second par is for reservaion and i mus no be used by he MH currenly using he he firs par bu may be used by any oher MHs having daa o send, and no being assigned bandwidh in his channel so far. If he piggyback field is se i means ha MH sill has more daa o send and requess assignmen of one more slo, and non-seing his field indicaes ha is ransmi buffer is empy. Each payload field (boh up-link and downlink) allows packing several small packes or fragmenaion for a large size packe. To reques a permi for up-link ransmission on some channel, he MH sends a reques o he CS, using he reservaion field in any slo in his channel (Fig. ). B. Mobiliy Suppor As long as a MH remains uned o a cerain channel (frequency pair) and ime slo pair, i.e., lisening on he downlink channel, he MH knows precisely ha a slo from he CS should appear every frame ime $ so ha i can expec he ime insances a which down-link slos arrive. However, when i moves ino an adjacen picocell using a differen FS paern, he MH will receive he down-link slo in an unexpeced ime insance. Thus he MH can easily realize wihin a mos a single frame ime ( ) ha i moved ino DOWN UP Fig.. Nex ch Permi P Daa Reserv. addr. ime Piggyback P addr. Packe Daa Packe Reserv. guard ime Slo formas for he down- and up-link daa ransmission. anoher picocell. Furhermore, using he following slo for he uplink ransmission he MH can and should make a reques for ransmi permission/reservaion in he new pair of ime slos. Noe ha he CS knows ha he MH wih his address has been ransmiing previously in a differen picocell, so i is possible a) o rea such reques wih higher prioriy han requess for new reservaions and b) release any possible reservaion in he old picocell. C. Six Varians of Chess Board Proocol Six varians of he Chess Board proocol, classified ino wo groups, are considered in he paper as shown in Table I. In he firs group (group A), MHs are assumed no o have a capabiliy o change channels during operaion, whereas MHs in group B are assumed o be able o change channels. A is he simples in which he CS allows a single channel o be used by only one MH, and i has no queue for requess. When he channel is being used, furher requess are blocked. In paricular, if a MH moves ino anoher picocell in which is channel is already used, here is no way o coninue ransmission. A is similar wih he only difference ha he CS does enqueue requess for busy channels raher han rejecing he reques, and when he channel is released he CS assigns i o he MH a he head of he queue. In A inerleaved usage of slos in a single channel by several MHs is performed. Slos are assigned using a simple round-robin fashion hereby assuring equal porion of he capaciy o each candidae MH. Essenially group B proocols are similar o hose in group A, wih he difference, ha in he case of group B proocols he CS aemps o find capaciy in oher channels, no only in he channel on which he reques has been issued. For insance in B when a reques from a MH arrives a he BS, he CS invesigaes each channel if i is reserved or no. If a channel is free i assigns he channel o he MH by sending he channel number. When all channels are reserved he reques is rejeced. A queue for requesing MHs is mainained in B, and as soon as a channel is released i is allocaed o he MH a he head of he queue. B is similar o B wih an excepion ha when he number of requesing MHs is greaer han he number of channels, all he channels are shared by he MHs in a round-robin fashion. In his case whenever he CS ransmis a permi o a MH i mus also inform he MH of he nex channel number in he permi
m m m TABLE I SIX VARIANTS OF THE CHESS BOARD PROTOCOL FOR ROF WLAN OPERATING IN GHZ BAND m Fig.. BS MH Indoor environmen for simulaion. field of downlink slo (Fig. ). I can be expeced ha group B proocols will ouperform group A proocols a he expense of increased complexiy. III. NUMERICAL RESULTS A. Simulaion Scenario and Assumpions In he simulaion sudy emphasis is placed on he performance comparison of six varians of he Chess Board proocol and he effecs of wo parameer (he number of channels and slo lengh) on sysem performance. Because of is cenralized naure of he sysem downlink ransmission is so simple ha only uplink performance is considered. The indoor environmens for simulaion includes four picocells and MHs (Fig. ). MHs are assumed o be freely moving across boundaries beween picocells according o random waypoin mobiliy model. Some assumpions for simulaions are as follows: Afer having sen a reques a MH receives he reservaion resul in one frame ime from he down-link channel. The capaciy of a BS (7*9;: ) is. The channel daa rae (7<*=?> ) is equal o 7<@9A:CBDFEHG, where is he number of channels and G is fixed size overhead including address, permi field, reservaion field and guard ime. Message raffic consiss of hree packes: byes (%), 7 byes (%), and byes (%) []. Inerarrival ime beween messages is exponenially disribued. Channel is perfec (error-free). JI -persisen algorihm is used for sending a reques packe. MHs use harmonic backoff algorihm (aemping probabiliies, /, /, "! ) in compuing heir probabiliies o ransmi a reques. Iniial channel disribuion by MHs is uniform. Oher assumpions and parameers for simulaion are summarized in Table II. Each simulaion was run for sec (simulaed ime) including warm-up phase of sec. 9 % confidence inervals were calculaed for he mean packe delay and normalized hroughpu whose variaions from he sample mean were less han % for all resuls. Channel change during operaion Reques when he channel is used Group A blocked A NO queued A A shared B blocked if no B YES queued if no B B shared if no TABLE II SUMMARY OF THE SIMULATION PARAMETERS Picocells Guard ime byes MHs FCS byes BS Capaciy ( LK ;M? ) Buffer Lengh Mbyes adrs byes Mobiliy Model Random Waypoin Permi field bye Speed of MH. m/s Reservaion slo byes Simulaed Time sec Flag byes B. Delay Performance Saisics Collecion afer sec The mean packe delay is he average ime spen by a packe from he insan i is generaed ill is ransmission is complee. Fig. and show he mean packe delay of group A and B proocols when he number of channels is five and he slo size is byes, respecively. In Fig. as he raffic load increases, A and A perform beer han A. Similarly in Fig. B and B are beer han B as he raffic load grows. We can also see ha group B proocols highly ourperform group A proocols because of he group B s runking abiliy. Since he similar rends are observed wih differen number of channels and slo size, from now on we will only consider A and B proocols. The impac of he slo size in A is shown in Fig. indicaing smaller slo size is beer han larger one. However due o he fixed size overhead oo small size may cause larger delay under heavy raffic load. In order o invesigae he effecs of he number of channels on delay performance, simulaions were run wih a fixed-size slo and various number of channels. Fig. 7 shows ha smaller number of channels is beneficial in erms of he mean packe delay, which is because he frame ime is direcly proporional o he number of channels N SR$.OQP. An ineresing fac is observed when he raffic load is Mbps. When is 9, each of he 7 channels is shared by wo MHs and he oher wo channels are shared by hree MHs
A A A Mean packe delay of A...... A A Fig.. The mean packe delay of group A s when is five and slo size is byes. A 9 7 Mbps Mean packe delay of A Mbps Slo size (byes) Fig.. The mean packe delay of A when he number of channels are five and he slo size is byes...... B B B Mean packe delay of B B B B Mbps Mean packe delay of A Mbps....... 7 7. Fig.. The mean packe delay of group B s when is five and slo size is byes. 7 9 7 9 Number of channels Fig. 7. The mean packe delay of A when he slo size is byes and he number of channels is. respecively since channel disribuion of MHs is assumed uniform and oal number of MHs is. If he hree MHs using he same channel come ogeher in he same picocell heir oal raffic load (9 Mbps) becomes greaer han he channel dae rae (. T /9), hus resuling in high mean packe delay. Whereas when is each channel is shared by only wo MHs. Even if wo MHs using he same channel say in he same picocell heir oal raffic load ( Mbps) is smaller han he channel daa rae (7.7 T /). Tha is he reason why he mean packe delay is lower when is han when is 9. Therefore, we can see ha he mean packe delay in A depends no only on he number of channels and slo size bu also on he channel disribuion of MHs. The impac of slo size of B is shown in Fig.. Jus as in A proocol smaller slo size is advanageous in delay performance bu because of fixed-size overhead small slo is no always beneficial. Fig. 9 shows how he number of channels influences he delay performance. Wih ligh raffic delay grows along wih he number of channels since frame ime is proporional o i. As raffic load increases he delay has a minimum when is around en. Noe ha in our simulaion scenario he average number of MHs in a picocell is en. So when is each MH is assigned is own channel resuling in rare collision and minimum reservaion delay. If is over en he sysem has more channels han MHs on he average, hereby increasing frame ime and wasing of bandwidh. C. Throughpu Performance Throughpu (normalized) is defined in he paper as a fracion of ime during which user daa is ransmied. Fig. and show hroughpus of A and B, respecively. Since he channel daa rae is deermined by he number of channels smaller number of channels performs beer han larger number of channels. Furhermore, larger slo size is favorable in erms of he hroughpu performance. From he poin he line is no longer linear packe loss begins o occur. I can be seen ha B highly ouperforms A. IV. CONCLUSION The wireless LAN environmen using radio over fiber echnology operaing in he millimeer-wave band imposes quie differen requiremens on he sysem design as compared o he convenional WLANs. Since he high peneraion loss of millimeer-wave signal many BSs should be employed o cover indoor areas. In such nework wih high number of small cells, he issue of mobiliy managemen has a very special significance. A proocol, called Chess Board proocol,
9 7 Mean packe delay of B Slo size (byes) Fig.. Mean packe delay of B when he number of channels are he slo size is byes. Mean packe delay of B 7 9 7 9 Number of channels Fig. 9. Mean packe delay of B when he slo size is byes and he number of channels is. Normalized hroughpu.9..7...... Throughpu of A s = s = s = ch = 7 9 Fig.. ch = Throughpu of A proocol. Normalized hroughpu.9..7...... Throughpu of B s = s = s = ch = 7 9 Fig.. ch = Throughpu of B proocol. feauring fas and easy handover and QoS suppor has been proposed by he auhors [7]. In his paper six varians of he proocol were considered and heir performance has been evaluaed by simulaion sudy. Simulaion resuls have shown ha group B proocols which are assumed o have a capaciy o change channels during operaion highly ouperform group A proocols for which a fixed channel is assumed for each MH. Delay performance of boh of hem depends on he slo size and he number of channels; moreover, in group A proocols i relies also on channel disribuion of MHs. Smaller slo size and smaller number of channels are beneficial in delay performance. On he oher hand larger slo size and smaller number of channels are in favor of hroughpu performance. To exploi bandwidh resources effecively, MH s capabiliy o change channels during operaion is required in a highly densed mobile environmen. REFERENCES [] P. Smulders, Exploiing he GHz Band for Local Wireless Mulimedia Access: Prospecs and Fuure Direcions, IEEE Commun. Mag. pp.- 7, Jan.. [] K. Kiayama, e. al, An Approach o Single Opical Componen Anenna Base Saions for Broad-Band Millimeer-Wave Fiber-Radio Access Sysems, IEEE Trans. Microwave Theory Tech., vol., pp.-9, Dec.. [] T. Kuri, e. al, -GHz-Band Full-Duplex Radio-On-Fiber Sysem Using Two-RF-Por Elecroabsorpion Transceiver, IEEE Phoon. Technol. Le., vol., no., pp.9-, Apr.. [] K. Kiayama, K. Ikeda, T. Kuri, A. Söhr, and Y. Takahashi, Full-duplex demonsraion of single elecroabsorpion ransceiver basesaion for mmwave fiber-radio sysems, Microwave Phoon. MWP, pp. 7-7,. [] G. Grosskopf, D. Rohde, and R. Eggemann, Mbi/s Daa Transmission a GHz Using an Opically Seered Anenna, ECOC, Muenchen, Germany, Sep., vol., pp. -. [] R.-P. Braun, G. Grosskopf, D. Rohde, and F. Schmid, Low-Phase-Noise Millimeer-Wave Generaion a GHz and Daa Transmission Using Opical Sideband Injecion Locking, IEEE Phoon. Technol. Le., vol., no., pp. 7-7, May 99. [7] H. B. Kim, H. Woesner and A. Wolisz, A Medium Access Conrol Proocol for Radio over Fiber Wireless LAN operaing in he -GHz Band, in Proc. h European Personal Mobile Commun. Conf., Apr.. hp://www-kn.ee.u-berlin.de/publicaions/proc.hml. [] K. Thompson, G. J. Miller and R. Wilder, Wide-Area Inerne Traffic Paerns and Characerisics, IEEE Nework, pp.-, Nov. 997.