A Link Layer Analyical Model for High Seed Full- Dulex Free Sace Oical Links Pi Huang and A. C. Boucouvalas Mulimedia Communicaions Research Grou School of Design, Engineering and Comuing Bournemouh Universiy, Fern Barrow, Poole, BH12 5BB, U.K. Email: {huang, boucouv}@bournemouh.ac.uk Absrac The relacemen of elecrical lines by free-sace or guided-wave oics in high-erformance digial sysems is currenly of acive research ineres. Such sysems are execed o be oeraing in full-dulex mode delivering Gbi/s daa rae in boh direcions. The FSO links mus be robus and offer high daa rae ransfer no only a he hysical layer bu also a he uer layers. This aricle lays ou a sysemaic analysis which includes roagaion delay for full-dulex oudoor FSO links using he HDLC roocol. We carry ou a deailed analysis and derive he link layer hroughu equaion. Based on he mahemaical model, we examine he link erformance a high daa raes wih secial aenion on he roagaion delay. 1. Inroducion The high available bandwidh, world wide license-free oeraion, deloymen simliciy and low cos, of free-sace oic sysems (FSO), reresen one of he mos romising aroaches for addressing he emerging broadband access marke and is las mile boleneck [1] [2]. Such robus sysems, which esablish communicaion links by ransmiing laser beams direcly hrough he amoshere, have maured o he oin ha mass-roduced roducs are now available [3]. In recen years, numerous FSO roducs have been se u oeraing a very high seeds. Available sysems offer caaciies in he range of 100Mbs o 2.5Gbs, and demonsraion sysems reor daa raes as high as 160Gbs [4], [5]. For such high seed links, i is imoran o ensure ha he 1
uer layers can deliver he informaion efficienly all he way u o he alicaion. In his aer, we focus on HDLC as he adoed link layer roocol and examine he link layer erformance under varying link bi error rae. The firs significan sudy on he link layer erformance evaluaion was carried ou by Bux [6] who is he recursor of subsequen sudies. Barker [7], Visas [8], Samaras [9] and Knuson [10] exended he work by conducing heir analysis on he secific area of IrDA (indoor infrared) shor range links. everheless, so far no sysemaic evaluaion has been carried ou for oudoor (long disance) full-dulex infrared links. Using he ARM mode of HDLC as he link layer roocol [11], we derive a mahemaical model for he link layer hroughu of FSO links using he conce of window ransmission ime [6]. The res aer is organised as follows: We begin wih he HDLC ransmission model, followed by a deailed sudy leading o a derivaion of sysem hroughu. Then, a number of resuls are given o evaluae he sysem erformance. Finally, we conclude he aer and give direcion in furher works. 2. Bidirecional HDLC Transmission Model and Parameer Definiion The mos oular layer roocol used in FSO is he High-Level Daa Link Conrol (HDLC), which has been widely imlemened ino roducs. HDLC offers hree differen modes of oeraion [11]: ormal Resonse Mode (RM), Asynchronous Balanced Mode (ABM) and Asynchronous Resonse Mode (ARM). RM is he mode in which he rimary saion iniiaes ransfers o he secondary saion, and is only used on he simlex links. For insance, he indoor infrared IrDA link roocol IrLAP is based on RM [12] [13]. ABM uses combined saions [11]. All combined saions are able o send and receive commands and resonses wihou any ermission from any oher saions on he link. ABM is no used widely oday. For ARM, he rimary saion does no iniiae ransfers o he secondary saion. The secondary saion does no have o wai o receive exlici ermission from he rimary saion o ransfer any frames. This mode can be used on boh simlex and full-dulex links. Asynchronous Resonse Mode is beer for oin o oin links, as i reduces overheads. I is widely adoed for he full-dulex links. The analysis in his aer is carried ou by using ARM mode of HDLC. Go-Back- (GB) scheme is seleced as he auomaic reea reques (ARQ) mechanism of he link [14]. 2
Before describing he ransmission model, we look a he differen funcionaliies of he HDLC frame forma: Informaion Forma: The frame is used o ransmi end-user daa beween wo devices. Suervisory Forma: The conrol field erforms conrol funcions such as acknowledgmen of frames and requess for reransmission. I has he highes rioriy o be ransmied. Unnumbered Forma: This conrol field forma is also used for conrol uroses. I is used o erform link iniialisaion, link disconnecion and oher link conrol funcions. In order o ensure reliable ransmission, each frame has a conrol field in he header. The conrol field conains he frame sequence number and he POLL/FIAL bi (P/F). In our ransmission model, Receive ready (RR) and rejec (REJ) Suervision (S)-frame resonses are only considered. Informaion (I)-frames conain a send sequence number s, which circles hrough cerain values. A he end of each window, P bi is se for he las sen frame. When a full window is successfully received, he receiver sends an S-frame RR and ses he F bi. The resonses RR conain a receive sequence number r which acknowledges he correc receion of frames u o r-1, hus indicaing ha r is he nex frame execed. However, if an error is found during he ransmission, he receiver will send an S-frame REJ immediaely o inform he ransmier for he corrued frame reransmission. S-REJ conains a receive sequence number r which rejecs frame r and hus acknowledges ha correc receion of frames u o r-1 and indicaing ha r is he nex (reransmi) frame execed. o informaion field is conained in S- frames, which means he lengh of he S-frames is he same as he frame overhead of I-frame. The link roagaion delays are also considered and added in boh resonses. In GB scheme, all he frames will be discarded afer any frame in error is deeced, and an S-frame REJ is sen o reques reransmission wih r number se o ha erred frame. The erformance model we emloy examines bi-direcional daa ransmission. The ransmier has o send daa frames, as well as he S-frames in order o acknowledge error or successful window ransmission for he receiver side. The link ransmission model is illusraed in Fig.1 and Fig.2. Fig.1 demonsraes a 7 frames window ransmission on a dulex link wih wo channels ATx-BRx and BTx-ARx. Fig.1 (a) illusraes error free ransmission for ATx-BRx, while here is an error in frame#22 for BTx-ARx. Fig.1 (b) illusraes errors in frame#12 and 18 for ATx-BRx while here are no errors for BTx- ARx. oe ha roagaion ime is no necessarily smaller han he suervision frame ime, S. 3
SRR I w1 SREJ ATx : 10 11 S22 12 13 14 15 16P 17 18 28F 19 20 : BRx BTx : 21 22 23 22 23 24 25 26 17F 27 28P SREJ 29 : ARx w2 w 3 (a) ATx : BTx : w1 w2 SREJ 10 11 12 13 12 13 14 15 28F 16 17 18P 18 SRR 21 22 23 S12 24 25 26 27P SREJ 28 29 30 S18 31 : BRx : ARx w3 (b) Figure 1: Window ransmission of a dulex link wih wo channels of ATx-BRx and BTx-ARx. Where Tx sands for Transmier and Rx sands for Receiver. Fig.2 illusraes BTx acknowledging for an error or a successful window ransmied from ATx. In Fig.2 (a), BRx noices an error in frame#21 from ATx while BTx is ransmiing frame#13 a he same ime. BTx sends he REJ S-frame #21 afer he end of frame#13. Fig.2 (b) illusraes BRx receiving he I-frame#21 wih P bi se (final frame of he window) from ATx while BTx jus sared o ransmi frame#13. BTx sends he RR S-frame#22 afer he end of frame#13. In Fig.2 (c), BRx noices an error in frame#21 from ATx before BTx sars ransmiing frame#13. BTx hen sends he REJ S-frame#21 before ransmiing frame #13. I SREJ SRR BTx: 12 13 s21 14 :ARx BTx: 12 13 22F 14 :ARx ATx: 21 22 23 21 B (a) :BRx ATx: 21P B (b) 22 :BRx SREJ BTx: 12 s21 13 :ARx ATx: 21 22 21 :BRx B (c) Figure 2: Acknowledgemens for an error or a successful window ransmission from BTx o ATx- BRx channels. 4
For he full-dulex links, one FSO saion has boh ransmier and receiver oeraing in arallel. However, unlike simlex infrared links [12], he saion does no need o change heir ransmission/receion mode eriodically. Therefore, no urnaround ime is needed for he devices. The relevan arameers and symbols used in he analysis are shown in Table 1. S, SRR, SREJ, I, and are given by: l ' l + l' d l+ l' S = SRR = SREJ =, I =, = and = 1 (1 b ) C C c According o [11], a fixed lengh of frame overhead l =72 is aken because only high daa raes are examined in his aer. Using [12] as a guide, he maximum value of window and frame sizes are 512 and 16384bi resecively. Table 1: Parameers Used in Modelling HDLC Throughu Symbol Parameer Descriion Uni C Link daa rae bi/s d Disance beween ransmier/receiver m b Link bi error rae - Frame error rae - umber of frames in one window (window size) - l Informaion (I)-frame message daa lengh (frame size) bis l Suervision (S)-frame lengh/ I-frame overhead 72bi I Transmission ime of an I-frame sec S Transmission ime of an S-frame sec SRR Transmission ime of a Receive Ready (RR) S-frame sec SREJ Transmission ime of a Rejec (REJ) S-frame sec Proagaion ime sec A Averaged ime for sending S frames er daa frame. sec B Average ime sen on waiing for he S-frame sec w Average window ransmission ime sec full Probabiliy of a successful full window ransmission - D b1 Throughu of ATx-BRx channel bi/s D b Throughu of eiher side of he full-dulex link bi/s 3. Mahemaical Modelling In his secion, we develo a mahemaical model which allows derivaion of link hroughu for he FSO full-dulex link in he resence of bi errors. The model is derived using he conce of window ransmission ime (WTT) [6]. WTT denoes he average ime needed for a window ransmission. I incororaes ime needed for daa frame ransmissions, reransmission of error frames, acknowledgemen S-frames, waiing ime for he acknowledgemens and roagaion 5
delays. I is he average ime aken from he beginning of he window s firs frame ransmission o he beginning of he firs frame of he nex window. We assume here is always a frame ready o ransmi from boh direcions and he frame rocessing ime is small enough o be ignored [15]. Due o he comaraively small size of he suervision frame, we also assume S-frame ransmissions for acknowledgemen uroses are error free. As he assumed symmery of he full-dulex link, sudy of one link direcion is sufficien. Throughu of he channel consising of ransmier A and receiver B, (ATx-BRx), as shown in Fig.1, is considered for he analysis. For he urose of deriving he roocol hroughu, we sudy hree differen asks of he dulex ransmission searaely: sending S-frames o acknowledge he oher channel, waiing for he S-frames from he oher channel and Informaion frames ransmission. By considering hese hree roocol asks, he ransmission model is described in Fig.3 which uses he execed average ime of sending S-frame A a he beginning of any informaion frame and he execed average waiing ime for S-frame B. ATx : BTx : + w1 w 2 B A I 10 11 12 13 12 13 14 15 16 17 18P 19 A I B 21 22 23 24 25 26 27P 28 29 30 w3 B 31 : BRx : ARx Figure 3: Use he averaged ime er frame for sending S-frames A and average waiing ime for receiving S-frames B o illusrae he ransmission model. ATx-BRx channel: Reransmission frames due o error a I=12; BTx-ARx channel: Error free ransmission of a window. Firs, we consider he average ime A consumed on sending S-frames from A o B. Due o duliciy of he link, he S-frames o acknowledge B have o be sen during daa ransmission, eriod. There are wo yes of S frames: S-RR which acknowledges a received error free window and S-REJ which acknowledges a frame received in error. The robabiliy full of error free receion of he las frame (wih P bi se) of a window denoing a successful window ransmission is also he robabiliy for a ransmier o send an S- RR frame acknowledging he correc receion of a full window. By considering he robabiliies 6
of error free receion of he firs, second, o he las frame of a window being 1, 2 ( 1 ),, ( 1 ) resecively, where is he frame error rae and is he window lengh, full can be derived as: (1 ) (1 ) = = i (1 )(1 (1 ) ) (1 ) i= 1 full Since he robabiliy of a daa frame being in error is, i imlies ha he robabiliy of a user sending an S-frame REJ o inform he oosie user is also. By combining wih he robabiliy of acknowledging he correc receion of a full window full, he averaged ransmi ime er daa frame A due o sending only S-frames o BTx-ARx channel is given by: The average ime consumed on waiing for S-frames B, is given nex. Two kinds of acknowledgemen are execed o be received from users: S-frame RR, Fig.2 (b) and S-frame REJ, Fig.2 (a), (c). When an error or successful ransmission of a window is deeced, an S-frame mus be sen. Due o he randomness of such evens, he execed S-frame has o wai for a ime slo o le BTx finish ransmiing he curren daa frame. I is sufficien o assume he average ime consumed for his ime slo is half of each daa frame ransmission ime I /2. Adding bidirecional roagaion ime delay 2 and S-frame ransmission ime S, he average ime B ha is needed for waiing he acknowledgemen from BTx is given by (3). According o Fig.2 (b) and Fig.3, he window ransmission ime (WTT) of an error free window is given by: In general, when an error has occurred a he i+1 frame, where i is he correcly received frames, Fig.2 (a), (c) and 3, WTT is: i = ( i + 1)( I + A) + B (5) The robabiliy c (i) of successive i correc frame ransmissions followed by an error a he beginning of a window ransmission is given by: = ( + ) (2) A S full 1 B = I + 2 + S (3) 2 = ( + ) + (4) I A B (1) 7
i c ( i) = (1 ) ( i =1, 2,..., -1) (6) The link hroughu D b1 for ATx-BRx channel, which defined as he correcly ransmied informaion bis er second, can be derived as: i l l D (7) 1 i b1 = ( (1 ) ) + (1 ) i= 1 i The hroughu efficiency for ATx-BRx channel is: TPE = Db1 / C The average ime for ransmiing one window is given by (8) 1 i= 0 i = ((1 ) ) + (1 ) (9) w i Due o symmery of he link, he above analysis is also holds rue for ATx-BRx channel s win channel BTx-ARx. Therefore, hroughu of eiher side of he full-dulex link is Db = Db 1. 4. Link Performance a High Daa rae Transmission Based on he mahemaical analysis and using equaion (7), he link erformance of he sysem is sudied in his secion. Using BER and daa rae as arameers, wih values 10-6, 10-5 and 10-4, and 2.5Gbs resecively, hroughu comarisons are carried ou using wo differen combinaions of window and frame sizes wih values of =512, l=16kbi and =7, l=5kbi. Link hroughu efficiency is loed agains link disance in he range of 100m o 10km in Fig.4. The link when oeraed a low BER generally has, as execed, larger hroughu. The hroughu efficiency decreases as he link disance increases as i is aaren in all curves. The roagaion ime which is roorional o he link disance has considerable effec on he link hroughu for long disance links for boh combinaions of window and frame sizes esecially when he BER is low, b =10-6. However, has differen effec on hroughu when using differen and l combinaions. For insance, nearly 70% hroughu reducion occurs a he disance of 10km when =7 and l=5kbi for b 6 = 10, while has much smaller effec on he hroughu when =512 and l=16kbi in he same BER, Fig.4. Because RR S-frames have o be sen frequenly while small and l values are used, he ime for waiing for he S-frame RR becomes large when he link disance is large ( is large). Therefore, he ransmier wais a long ime for 8
he acknowledgemen from he receiver before saring a new window ransmission. Large and l values herefore have beer hroughus for he long disance links. However, errors are more likely o occur when frames are large (l=16kbi), and longer ime eriod is required on waiing for he REJ S-frames. Small and l herefore have beer hroughu while he link disance is shor ( is small). For he same BER, he crossing oins of he wo curves ha reresen differen and l combinaions imly a beer hroughu may achieve by aroriae adaaion of window and frame size. 2.5 Gbs, on-oimum and l 1 0.9 Throughu efficiency 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1.0E+02 1.0E+03 1.0E+04 Disance (m) BER=1e-6, l=5k, =7 BER=1e-6, l=16k, =512 BER=1e-5, l=5k, =7 BER=1e-5, l=16k, =512 BER=1e-4, l=5k, =7 BER=1e-4, l=16k, =512 Figure 4: Link hroughu in varies BERs for he disance range from100m o 10km. 5. Conclusion and Fuure Work Moivaed by he fas develoing free-sace oical echnology, and in order o examine and imrove he erformance of he oular dulex HDLC link layer roocol over FSO links, a sysemaic mahemaical model for high seed full-dulex FSO links is resened in his aer using he ARM mode of HDLC as he link layer roocol. A simle hroughu formula is derived from he model in he resence of bi error rae. Based on he analyical model, we hen sudied he link erformance and he effec of roagaion ime on link hroughu. The resul shows ha he 9
roagaion ime has a significan effec for long disance links, and he effec is likely o be even more severe if higher daa raes are alied in fuure sysems. As a air of inaroriae window and frame size will have a significan derimen on he sysem hroughu, an area worh furher sudy is o invesigae he oimum value of window and frame size o acquire he maximum hroughu for he sysem. REFERECES [1] H.A. Willebrand and B.S. Ghuman, Fiber oics wihou fiber, IEEE Secrum, Aug 01,.40 [2] Demer, R. "A ray of ligh" IEE Review, Volume: 47 Issue: 2, March 2001 Page(s): 32-33 [3] Davis, C.C.; Smolyaninov, I.I.; Milner, S.D. Flexible oical wireless links and neworks, Communicaions Magazine, IEEE, Volume: 41 Issue: 3, March 2003 Page(s): 51-57 [4] Acamora, A.S. and Krishnamurhy, S.V. A broadband wireless access nework based on mesh-conneced free-sace oical links IEEE Personal Communicaions [see also IEEE Wireless Communicaions], Volume: 6 Issue: 5, Oc 1999 Page(s): 62 65 [5] Chinlon Lin, Kung-Li Deng and Chun-Ki Chan Broadband oical access neworks, Lasers and Elecro-Oics, 2001. CLEO/Pacific Rim 2001. The 4h Pacific Rim Conference on, Volume: 2, 2001 Page(s): II-576 -II-577 vol.2 [6] Bux W., Kummerle K. & Truong H.L., Balanced HDLC rocedures: A erformance Analysis, IEEE Trans. Comm, 1980, com-28,.1889-1898. [7] P. Barker, A.C. Boucouvalas, Performance modelling of he IrDA roocol for Infrared Wireless Communicaions IEEE Communicaions magazine, Vol.36, o.12,.113-117, December 1998. [8] V. Visas and A. C. Boucouvalas, Window and frame size adaiviy for maximum hroughu in IrDA links. Proceedings of The 3rd Elecronic Circuis and Sysems Conference, Braislava, Slovakia on Seember 5-7, 2001,.147-152 [9] Mehod for evaluaing error raes in infrared wireless links Samaras, K.; Sree, A.M.; O'Brien, D.C.; Edwards, D.J.; IEEE, Elecronics Leers, Volume: 33 Issue: 20, 25 Se 97 Page(s): 1720 [10] Michael G. Roberson, Sco V. Hansen, Franklin E. Sorenson and Charles D. Knuson. "Modeling IrDA Performance: The Effec of IrLAP egoiaion Parameers on Throughu." Proceedings of he Tenh IEEE Inernaional Conference on Comuer Communicaions and eworks (ICCC '01), Phoenix, Arizona, Ocober 15-17, 2001. [11] ISO 3309 High Level Daa Link Conrol (HDLC) Procedures, Frame Srucure 1991-06-01. [12] IrDA, Serial Infrared Link Access Proocol (IrLAP), Version 1.1 1996. [13] IrDA, Serial Infrared Physical Layer Secificaion for 16Mb/s Addiion (VFIR) Erraa o version 1.3, 1999 [14] V. Visas and A.C. Boucouvalas Auomaic Reea Reques Schemes for Infrared Wireless Communicaions, IEE Elecronics Leers, 28h Feb 2002, Vol. 38, o. 5.254-246. [15] P. Chazimisios and A. C. Boucouvalas ''IrLAP IrDA Proocol Throughu Deendence on Processor Seed'' Inernaional Symosium on CSDSP 2002, July 15-17,. 272-275. 10