Reint Efficient Voice Communications ove IEEE80.11 WLANs Using Imoved PCF Pocedues E. Ziouva and T. Antonakooulos The Thid Intenational Netwok Confeence - INC 00 JULY 00 Coyight Notice: This mateial is esented to ensue timely dissemination of scholaly and technical wok. Coyight and all ights theein ae etained by authos o by othe coyight holdes. All esons coying this infomation ae exected to adhee to the tems and constaints invoked by each autho's coyight. In most cases, these woks may not be eosted o mass eoduced without the exlicit emission of the coyight holde.
Efficient Voice Communications ove IEEE80.11 WLANs Using Imoved PCF Pocedues Eustathia Ziouva 1 and Theodoe Antonakooulos 1 Comutes Technology Institute, Riga Feaiou 61, 61 Patas, Geece Deatment of Electical Engineeing and Comutes Technology Univesity of Patas, 6500 Rio - Patas, Geece Tel: +30-61-997 346, Fax: +30-61-997 34, e-mail: antonako@ee.uatas.g Abstact This ae esents a new dynamically adatable olling scheme fo efficient suot of voice communications ove diffeent IEEE80.11 netwoks. The oosed olling scheme is simle to imlement and does not equie any modification on the existing access otocol. The analytical aoach that models the oosed olling scheme, uses a discete-time Makov chain and oves that the secific olling scheme, when silence detection is used at the wieless teminals, imoves the caability of IEEE80.11wieless LANs fo handling voice taffic efficiently. Keywods Wieless Netwoks, Centalized access contol, Voice communication 1. Intoduction The IEEE80.11 standad (IEEE, 1997) fo wieless local aea netwoks (WLANs) coves the Medium Access Contol (MAC) sub-laye and the hysical laye of the Oen System Inteconnection (OSI) efeence model. A gou of wieless teminals unde the contol of a Distibuted Coodination Function (DCF) o a Point Coodination Function (PCF) foms a Basic Sevice Set (BSS). A BSS can eithe be an indeendent netwok o an infastuctue netwok, in which an Access Point (AP) links the wieless teminals to a backbone netwok (like Asynchonous Tansfe Mode, ATM), theefoe extending thei ange to othe BSSs via othe APs. In a BSS, the wieless teminals and the AP can use eithe DCF fo asynchonous data tansmissions, o PCF fo contention-fee acket tansmissions. In DCF mode, the netwok is in a Contention Peiod (CP) and the stations contend in ode to gain access using the Caie Sense Multile Access / Collision Avoidance (CSMA/CA) access method. In PCF mode, the AP coodinates the medium usage and the netwok is in a Contention-Fee Peiod (CFP). The medium can be altenated between CP and CFP accoding to the Contention-Fee Peiod Reetition (CFPR) inteval that is the eciocal of the ate at which the AP initiates the CFP. As the seed and caacity of WLANs incease, so does the demand fo imoving the quality of sevice (QoS) of eal-time alications. (Zahedi and Pahlavan, 000) used the DCF mode in ode to ovide eal-time alications and showed that the efomance is oo by consideing the delay equiements. (Romans and Touilhes, 1998), (Deng and Chang, 1999)
and (Liu and Wu, 000) oosed modifications to the DCF mode in ode to cay acket telehony taffic. (Sheu and Sheu, 001) oosed a novel modified otocol to ovide asynchonous and multimedia taffic ove IEEE80.11 WLANs that is comliant with the IEEE80.11 standad. Howeve, this otocol was not designed fo eal-time taffic exchange between stations belonging to diffeent BSSs. Fo eal-time taffic exchange between stations fom diffeent BSSs, all eal-time ackets have to be tansfeed to the AP of the BSS and thus a centalized access method like PCF is equied. Although PCF intoduces high ovehead in each tansaction, it can satisfy time-bounded equiements if it is oely adjusted. (Cow and et al, 1997) showed with simulations that olling is inefficient fo handling on/off seech taffic. In (Ziouva and Antonakooulos, 001-1) we examined how PCF can efficiently be used to cay Constant Bit Rate (CBR) voice taffic between stations of the same BSS, while the DCF mode suots asynchonous data taffic. The numbe of voice uses suoted by PCF can be futhe inceased by using silence detection at each voice use and a new management scheme on the AP s olling list, the so called Cyclic Shift olling ocess, that does not equie any modifications on the wieless teminals access mechanism (Ziouva and Antonakooulos, 001-). In this wok, we extend the Cyclic Shift olling ocess by adding a temoay emoval ocedue fo stations that ae in silent state. The AP s olling list does not emain constant duing the lifetime of a convesation, but stations that ente silent state duing thei call ae emoved fom the AP s olling list temoaily and ae einstalled on the olling list afte a secific duation elated to the hangove used by the silence detecto. This dynamically adatable management scheme imlemented on the AP s olling list is called Cyclic Shift and Station Removal Polling ocess (CSSR). We conside voice activities between stations fom diffeent BSSs and we estimate an ue bound on the numbe of voice stations that a BSS can suot, while keeing low voice acket delay and guaanteeing edetemined minimum bandwidth fo data taffic. The CSSR olling ocess is evaluated by a discete time Makov chain and is also comaed with the Cyclic Shift olling ocess. Ou studies wee efomed using 3 kbs Adative Diffeential Pulse Code Modulation (ADPCM) voice coding. Fo focusing on diffeent asects of PCF efomance, we assumed an eo-fee channel. Section biefly descibes the PCF oeation fo handling voice taffic while Section 3 esents the CSSR olling scheme and its Makov model. In Section 4, we evaluate the efomance of the oosed scheme deiving the maximum numbe of voice stations handled by PCF. Finally, Section 5 discusses some numeical esults.. Voice Taffic Management We conside a numbe of BSSs inteconnected via a backbone netwok. The AP initiates the CFP with the tansmission of a Beacon fame and teminates it with the tansmission of a CF- END fame. The CFP has a maximum duation that defines the maximum numbe of voice stations that can be suoted. Due to the DCF taffic, the CFP may be stetched. Since the actual duation of CFP and CP may vay, the IEEE80.11 standad defines a maximum value of time that the CFP is stetched and a minimum duation fo CP. All timing aametes that detemine the coexistence of PCF and DCF ae contained in each Beacon fame. Each voice station uses an ADPCM code at 3 kbs and a silence detecto. We exloit the silent eiods of a voice souce in ode to incease the numbe of calls but at the exense of voice ackets doing ate. The used voice model is the well-known model oosed in (Bady, 1969). The
duations of voice talk suts and silence eiods follow the exonential distibution with aveage d t equal to 400 ms and d s equal to 600 ms esectively (Siam and et al, 1999), (Onvual, 1994). The silence detectos use a technique to avoid sudden end-cliing of seech and to bidge shot seech gas such as those due to sto consonants, the so called hangove. This technique esults to silent gas lage than the hangove duation. A voice acket is geneated evey CFPR inteval when the CFP is scheduled to begin. A voice acket is tansmitted ove the netwok each time the station is being olled by the AP. If a new acket is geneated befoe the evious acket has been tansmitted, the olde acket is discaded. All stations on the olling list ae olled once duing each CFP. The voice taffic management guaantees that the voice acket delay (Ziouva and Antonakooulos, 001-1) is less than the CFPR inteval used by the AP. A station in talk sut geneates a voice acket (called talk acket) that is tansmitted when the AP olls the station. A station in silent eiod does not geneate any voice ackets and tansmits a NULL acket (called silent acket) when it is olled. When the AP takes contol of the medium, it stats olling accoding to Figue 1. Contention-Fee Peiod Reetition (CFPR) Inteval Contention-Fee Peiod (CFP) Contention Peiod (CP) PIFS Beacon Voice Packet+Poll Voice Packet Voice Packet+Poll NULL CF- END Figue 1. Voice tansmissions in PCF mode PIFS: The time inteval the AP has to sense the medium idle to gain access : The time inteval between the tansmitted fames 3. The CSSR olling ocess The basic functionality of the dynamically adatable olling scheme is the following: At the beginning of each CFP ound, the AP cyclically shifts the stations on its olling list, so the fist station at the evious ound becomes the last station at the cuent ound and all othe stations advance one osition towads the stat of olling. At the beginning of each CFP ound, the olling stats fom the beginning of the shifted olling list. When the stat of a silence eiod is detected fo a station, the AP does not oll this station fo a few PCF ounds. The PC maintains two olling lists, a main olling list that contains all stations having established connections, iesective of thei state and an active olling list (a subset of the main olling list) which contains only the stations that have to be olled at the next PCF ound. Both olling lists ae cyclically shifted in each PCF ound. Wheneve the AP cannot comlete its active olling list duing a CFP ound, the AP stats its olling sequence with the fist station on its olling list at the next CFP ound. The advantages of the CSSR olling scheme ae the following: The cyclic shift of the olling list seads unifomly the doed voice acket to all active stations in the netwok and theefoe it inceases the numbe of voice stations handled by PCF, comaed to the case of using only silence detection (Ziouva and Antonakooulos, 001-). The temoay station emoval ocedue ovides moe bandwidth fo actual voice tansmissions, thus the netwok efomance is futhe imoved. Finally, the AP s olling list management scheme does not equie any modification on the wieless stations MAC otocol and is athe simle to imlement.
(1- ) SP (1- ) 1- (1- ) 1-1- TP NP (1- ) SSP 1-1 R-1 R- 1 1... 1 R-K Figue. The CSRR olling ocess: Finite-state Makov chain fo a voice station Fo analyzing the CSSR olling scheme, a voice station is modeled as a discete-time Makov chain that is shown in Figue. All state tansitions occu at the end of a PCF ound. Duing a CFP eiod, a voice station geneates ackets with obability and may be olled with obability. A voice station is in Silent-Polled (SP) state when it is olled and has no voice acket to tansmit. The station leaves the SP state and entes the Not-Polled (NP) state when olling cannot be comleted and the station is among the stations that ae not olled. The station leaves the SP state and entes the Talk-Polled (TP) state when a voice acket is geneated and the AP olls the station. A voice station emains in the TP state if it is in talk sut and is olled continuously. If the station is in talk sut and the AP does not oll it, the station moves to the NP state. If the station s talk sut ends and the AP olls the station, the station tansmits a NULL acket and moves to the Stat of Silence-Polled (SSP) state and emains in this state until the next PCF ound. Enteing the SSP state, the station is emoved fom the active olling list fo 1 K ounds and asses though the Removed-i (R-i) states. Afte K ounds, the station is eositioned on the olling list and, the station etuns eithe to the NP, TP o SP state, deending on its osition in the olling list and the tansmission of a voice o a NULL acket. If Π Χ is the steady state obability of a station being at state X, we can calculate the stationay distibution of the discete Makov chain if the obabilities and ae know. Π =, + TP 1 Kt 1 t SP ( 1 )( 1 t) t Π =, ( 1 ) Π SSP =, 1 + K 1 t t 1+ Kt 1 t Π = 1 NP + ( 1 ) Π = K R 1 + K 1 t t 1 Kt 1 t
Π R is the steady state obability of a station being emoved fom olling list fo K PCF ounds. If K = 0 (a olling scheme without station emoval) the CSRR olling ocess eesents the Cyclic Shift olling ocess. 4. Pefomance Evaluation In this section, an analytical aoach is esented to deive an ue bound to the numbe of voice stations that can be handled by PCF, when silence detection and the CSSR olling scheme ae emloyed. 4.1. The voice acket doing obability Since the CFPR inteval can be selected so that the voice delay equiements ae always satisfied, the aamete that defines the PCF efomance is the obability of doed ackets P do. Accoding to (Siam and et al, 1999) and (Jayant and Chistensen, 1981), 1 % of doed voice ackets loss can be toleated. In this case, we can estimate an ue bound to the numbe of voice stations accommodated by PCF fo vaious values of K, while the condition P do < 0.01 is satisfied. In ode to decease the numbe of ejected ackets when a station is emoved fom the olling list, the duation of the station emoval is limited to the hangove duation. Futhemoe, fo two connected stations A and B, when e.g. station A is in silent station B is usually in talk sut, thus the AP of station A has to send the ackets eceived by the AP of station B, to station A when this station is emoved fom its olling list. Unde these cicumstances, ackets fom station A ae doed eithe when it is in talk sut and its AP does not oll it, o when the station is in talk sut and is olled by its AP but the AP of station B does not oll station B, theefoe the ackets of station A ae not deliveed to station B. So, the obability of doed voice ackets is given by: P =Π +Π Π B ( A) ( A) ( A) ( ) do NP t TP NP Since, the steady state obabilities Π X ae functions of obabilities and and these two obabilities ae consideed equal fo the two communication stations, the obability of doed voice ackets can be calculated if and ae known. 4.. The basic system obabilities (1) Accoding to (Fine and Tobagi, 1986) and (Fiedman and Ziegle, 1989), the obability that a voice station is in talk sut is given by = d t /(d t + d s ). If n is the obability that a station is not olled, then will be deived by: = 1 n. The obability n deends on the numbe N of voice stations that fom the main olling list (N stations geneate N ackets, ulink and downlink), the numbe N of stations that do not aticiate in the active olling list duing a PCF ound (emaining N N ackets to be exchanged) and the numbe N of stations that ae olled duing a PCF ound. Let N tmax denotes the maximum numbe of talk ackets that can be handled by PCF. If N N N tmax, evey station is olled duing a PCF ound, but if N N > N tmax, only N stations can be olled, the stations holding the fist N ositions on the active olling list, while the est (N N N ) stations ae not olled. So the conditional obability n N,Nhat a station is not olled duing a PCF ound when thee ae N and N stations is given by: ( ) ( ) tmax = N N N N N if 0 N N N 1 n N, N
Fo a given numbe N, the numbe of stations that can be olled has a maximum value N max deending on the CFP length and the combination of the numbe of talk ackets and the numbe of silent ackets. The AP olls eithe N max stations o (N N ) stations, whateve of the two events haens fist. Finding the obability P N N that the AP olls N stations given that it has emoved N stations and the obability P N that the AP has emoved N stations, and using the total obability theoem and the above equation, we have that n = ( Nmax N N) N N N N N min, t max 1 P P N N N N = 0 N= ( Ntmax N) N N Let assume that a station is emoved fom the olling list with obability. Then the N stations duing a PCF ound have a binomial mass function and so N N N N PN = ( 1 ) N The obability can be calculated by Π R, whee is equal to N /(N N ), since duing a PCF ound a station can be at one of the N fist ositions of (N N ) stations on the active olling list. In ode to find the obability P N N, we must calculate the aametes N tmax and N max. The maximum numbe N tmax of talk ackets that the AP can handle deends on the CFP maximum duation and the talk acket tansmission time T t, which is equal to T v +, whee T v is the tansmission time of a voice acket including heades. Theefoe: ( ) Nt max = TCFPR Tmax FS PIFS TBeacon TCF END TCP Tt whee T CFPR, T maxfs, T BEACON, T CF-END and T CP ae the CFPR inteval duation, the maximum time the CFP can be stetched, the Beacon and CF-END fame tansmission time and the CP duation esectively. The diffeent BSSs use the same CFPR inteval duation, theefoe the same tansmission time is used fo both ulink and downlink talk ackets. The maximum numbe N max of stations that can be olled duing a PCF ound can be found when all exchanged ackets ae silent ackets. Deending on T t /T s (T s is the silent acket tansmission time which equals the tansmission time of a NULL acket lus the time), a talk acket can be elaced by one o moe silent ackets. The AP defines the emaining CFP duation using time infomation fom the eviously tansmitted ackets, since it does not know if the next station in the list is in talk o silent state. The AP consides that the downlink and ulink ackets of the last station that can be olled ae talk ackets and thus it guaantees that the maximum duation of the CFP is not exceeded. This exlains why we subtact two talk ackets fom N tmax. Finally, we have to subtact the N downlink ackets that have to be sent to the emoved silent stations. So is given by: N max (( ) ) Nmax = Ntmax N Tt Ts + 1 Finally, the obability P N N is detemined by the numbe N t of talk ackets and the numbe (N + N N t ) of silent ackets duing a CFP. Thus: min ( Ntmax,N+ N) N + N N + N P = (,,, N max ) N g N N Nt Nt Nt = 0 Nt Nt ( max ) N+ N Nt ( 1 ) ( ) Nt t t t s t t t f N + N N T T + N N
whee f(x) and g(x) ae two functions that define the emitted combinations of N t talk ackets and (N + N N t ) silent ackets duing CFP and ae given by the following two equations: (,, t, tmax ) g N N N N = f ( x) 1 if x = 0 = 0 if x 0 0 if N + N = Ntmax Nt = 0 Nt + N + N Ntmax < Ntmax = + + = + + + 1 if Nt + N + N Ntmax > N tmax ( N ) ( ) ( t + N + N Nt = N ) max t N max t + Tt + N + N Nt Ts > Nt T max t (( Nt N N Ntmax Ntmax ) (( Nt ) Tt ( N N Nt ) Ts NtmaxTt) ) ( ( )) 5. Numeical Results Accoding to ou analytical aoach, the effect of CSSR olling scheme on the PCF efomance is deicted in Figue 3. The suoted numbe of voice stations inceases when the oosed scheme is used along with silence detection in contast to the case of CBR voice taffic. Futhemoe, the thoughut imoves when the CFPR inteval and the aamete K incease, but K must not exceed the hangove duation. The esults of ou analysis ae deived consideing the attibute values used by the high data ate extension of the IEEE80.11 standad at 5.5 and 11 Mbs. We also use the otional shot hysical heade defined in the standad fo eaching maximum thoughut. Maximum numbe of voice stations 80 70 60 50 40 30 0 CBR taffic Cyclic shift (K=0) CSSR (K=1) CSSR (K=) 5.5 Mbs 11 Mbs 10 15 16 17 18 19 0 1 3 4 5 CFPR Inteval (ms) Figue 3. The effect of CSSR on the maximum numbe of suoted voice stations
6. Conclusions In this ae, we esented the CSSR olling scheme that can be used along with silence detection fo inceasing the numbe of suoted voice communications ove diffeent IEEE80.11 WLANs. The oosed scheme imoves the PCF efomance by cyclically shifting the stations on the AP s olling list in each PCF ound and dynamically emoving stations that ente silence mode. The efficiency of the CSSR olling scheme is a function of the CFPR inteval used by the netwok and the hangove duation of the silence detectos, which also detemines the maximum duation of temoay station emoval. Finally, the CSSR scheme can be easily imlemented on the AP of an IEEE80.11 netwok, without equiing any modification on the wieless teminals access mechanism. Refeences Bady, P.T. (1969), A model fo geneating ON-OFF seech attens in two-way convesation, Bell System Technical Jounal, Vol. 48, No. 7,. 445-47. Cow, B., Widjaja, I. Kim, J.G. and Sakai, P. (1997), IEEE 80.11 Wieless Local Aea Netwoks, IEEE Communications Magazine, Vol. 35, No. 9,. 116-16. Deng, J. and Chang, R.S. (1999), A Pioity Scheme fo IEEE 80.11 DCF Access Method, IEICE Tansactions Communications, Vol. E8-B, No. 1,. 96-10. Fine, M., and Tobagi, F.A. (1986), Packet Voice on a Local Aea Netwok with Round Robin Sevice, IEEE Tansactions on Communications, Vol. COM-34, No. 9,. 906-915. Fiedman, E. and Ziegle, C. (1989), Packet Voice Communications Ove PC-Based Local Aea Netwoks, IEEE Jounal on Selected Aeas in Communications, Vol. 7, No.,. 11-18. IEEE, (1997), Daft Standad fo wieless LAN medium access contol (MAC) and hysical laye (PHY) secification, IEEE P80.11. Jayant, N.S. and Chistensen, S.W. (1981), Effects of Packet Losses in Wavefom Coded Seech and Imovements Due to an Odd-Even Samle-Inteolation Pocedue, IEEE Tansactions on Communications, Vol. COM-9, No.,. 101-109. Liu, H-H. and Wu, J-L. C. (000), Packet Telehony Suot fo the IEEE 80.11 Wieless LAN, IEEE Communications Lettes, Vol. 4, No. 9,. 86-88. Onvual, R.O. (1994), Asynchonous Tansfe Mode Netwoks: Pefomance Issues, Atech House, Boston. Romans, C. and Touilhes, J. (1998), A Medium Access Potocol fo Wieless LANs Which Suots Isochonous and Asynchonous Taffic, Tech. Re. HPL-98-35, Hewlett-Packad Com., Bistol. Sheu, S.-T. and Sheu, T.-F. (001), A Bandwidth Allocation/Shaing/Extension Potocol fo Multimedia Ove IEEE 80.11 Ad Hoc Wieless LANs, IEEE Jounal on Selected Aeas in Communications, Vol. 19, No. 10,. 065-080. Siam, K., Lyons, T.G. and Wang, Y.T. (1999), Anomalies Due to Delay and Loss in AAL Packet Voice Systems: Pefomance Models and Methods of Mitigation, IEEE Jounal on Selected Aeas in Communications, Vol. 17, No. 1,. 4-17. Zahedi, A. and Pahlavan, K. (000), Caacity of a Wieless LAN with Voice and Data Sevices, IEEE Tansactions on Communications, Vol. 48, No. 7,. 1160-1170. Ziouva, E. and Antonakooulos, T. (001-1), CBR Packetized Voice Tansmission in IEEE80.11 Netwoks, in Poc 6 th IEEE Symosium on Comutes and Communications,. 39-398. Ziouva, E. and Antonakooulos, T. (001-), Voice Communications ove IEEE80.11 Wieless LANs Inteconnected Using ATM Links, in Poc. 6 th Confeence on Local Comute Netwoks,. 60-69.