CROSS-LAYER OPTIMIZATION PERFORMANCE OF SINGLE CELL MILLIMETER WAVE OFDM WIRELESS NETWORK UNDER RAIN FADING

Transcription

1 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun 009 CROSS-LAYER OTIMIZATION ERFORMANCE OF SINGLE CELL MILLIMETER WAVE OFDM WIRELESS NETWORK UNDER RAIN FADING E. Endroyono 1, G. Hendrantoro 1, A. Matsushma 1) Jurusan Teknk Elektro, Fakultas Teknolog Industr ITS Surabaya Indonesa 60111, e-mal: endroyono@ee.ts.ac.d, gamantyo@ee.ts.ac.d ) GSST Kumamoto Unversty Kumamoto Cty Japan, e-mal: matsua@cs.kumamoto-u.ac.jp ABSTRACT Ths paper presents cross-layer optmzaton mplementaton for mllmeter wave OFDM network n the presence of ran fadng. Cross-layer optmzaton s proposed to overcome the power neffcency and complexty of conventonal fadng mtgaton technque. The network s modeled as downlnk mult-user sngle cell OFDM network. The ran fadng data s generated based on Synthetc Storm Technque usng ran rate data measured n Surabaya Area. To show the nterest of cross-layer optmzaton, a combnng optmzaton between physcal and MAC layer usng dynamc sub-carrer allocaton and adaptve power allocaton algorthm s demonstrated. The optmzaton objectves are the mprovement of utlty and farness n heavly ran affected cell. The result shows an mprovement of utlty for mult-user OFDM network n term of total rate. Keywords: OFDM, sngle cell mult-user, mllmeter wave ran fadng, SST, cross-layer desgn, JSA.. 1. INTRODUCTION OFDM has been chosen as future platform to fulfll the need of very hgh bandwdth n broadband communcaton network. Wreless OFDM has proven hs performance to combat nter-symbolnterference and operate n fadng envronment by usng mult-carrer technque. In tropcal regon, n addton to nose and lnk attenuaton, ran fadng s becomng one of mportant constrant to the capacty of mllmeter wave OFDM wreless system. That s why; there are many researches about characterstcs of tropcal ran fadng. The knowledge of dynamc ran fadng characterstc s necessary to develop fadng mtgaton technque (FMT) n the network ar nterface (Castanet, 003) (Gremont, 004) (Chu, 005). In tradtonal FMT, fadng s mtgated by a set of mnmum requred margn, flterng, power adaptaton, MIMO and advanced codng n physcal layer. Besde FMT, some technques and algorthms of resource allocaton are also developed. They try to acheve an effcent and effectve communcaton system beyond the lmt of conventonal FMT result. One of the most proposed approaches s cross-layer desgn (Corvno, 008). Ths approach s carred-out by takng beneft from exchangng nformaton of dfferent protocol layers. The operaton s held by knowng the physcal layer channel state nformaton (CSI) and upper layer nformaton, such as type of message and queung state nformaton (QSI), (Song, 005a). The desgn objectve s then extended from maxmzng user throughput to maxmze the overall channel utlzaton and farness of servce between users. Whle cross-layer s consdered as a promsng approach, the number of study on cross-layer approach n ran fadng envronment s stll lmted. Studes that try to combne the three aspects: OFDM, cross-layer, mllmeter wave and ran fadng channel are found n (Endroyono, 008a) (Endroyono, 008b). Some engneers sad that the probablty of ran fadng events can be neglected n desgn. However, we know that n severe ran fadng ntensty, a deep fadng wll lead vctm network to have a low capacty and utlty. A conventonal mtgaton n ran fadng s usually resultng n power utlzaton neffcency and ncreasng the complexty (Song, 005a). Ths paper s then presented to fll the space and s organzed as follow. Secton descrbes the mllmeter wave channel model. In secton 3, we present cross-layer model for a mllmeter wave OFDM network consderng ran fadng. In secton 4, we present an llustratonal result and dscusson related to the subject of nterest. Fnally, n secton 5 we gve a bref summary and future challenge of research.. MILLIMETER WAVE CHANNEL MODEL In tradtonal transmsson lnk budget, the receved sgnal wll depend on transmt power ( T ), system gan (G T, G R ) and free-space-loss of lnk n functon of frequency (f n MHz) and dstance of transmtter-recever (r n km). It s descrbed by the followng equaton. R = T + GT + GR 0log10( f ) (1) 0log ( r) 6.4 dbm 10 In tropcal regon, n addton to path loss of equaton (1) and shadowng, ran fadng wll exst. As stated n (Endroyono, 008a) and (Endroyono, 008b), a ran event at 4 GHz, wth 80 mm/hr rate could produce a condtonal excess loss up to 50 db/km, due to hydro-meteor effect. So, the ran fadng should be consdered n mllmeter wave lnk. G-7

2 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun Wreless Lnk Model n Ran Fadng Fgure 1, gves an example of the CDF of ran attenuaton statstc n rany season (condtonal) and full year statstc. It shows the reason that some desgners exclude the ran fadng from the calculaton to smplfy the desgn. It s clear that there s a low percentage of ran fadng to exceed certan lmt. However, n a very hgh bt rate broadband system, an outage of 1% of tme wll cause a very mportant mpact. A small outage wll result n very hgh number of packet loss or decreasng the utlty. two other parameters of ran fadng (fade duraton and fade slope) are also mportant n the cross-layer desgn of a mllmeter wave. arameters of power control, adaptve modulaton and advanced codng should be desgned to adapt the value of fade slope and fade duraton. Fgure 1. Example of Ran Fade CDF To smulate ran fadng effect, a sngle cell wreless lnk consstng of Base Transcever Staton (BTS) and mult-user termnals s proposed. User termnals are dstrbuted randomly around the BTS. The ran fadng s generated based on a model called Synthetc Storm Technque, SST (Kanellopoulos, 1986) (Castanet, 003) (Fontan, 005). In SST model, the ran attenuaton A (k) of a lnk segment Δ s not measured drectly from ste. The Ln attenuaton seres s generated based on the samplng ran ntensty (ran rate data, R ), the assumpton of ran velocty, ran drecton (relatve angle to the ran drecton for segment number, = 1,,3... n ), relatve dstance, and ITU s coeffcents a, b n frequency 30 GHz; as followng equaton: N A ( k) = ar * ΔL () n= 0 b ( k n) n The relaton between ran cell, ran drecton and lnk drecton s shown n Fgure. In our case, ran rate data was gathered from multple measurement stes n ITS Surabaya that the detal s presented by (Mahmudah, 008). As llustrated n Fgure 3, each lnk n cell wll experence dfferent fade ntensty due to ther poston and respectve dstance. The graphs show smlar statstc trends, but they have dfferent probablty n hgher attenuaton due to ran fadng correlaton factor. Correlaton factor and dfferental gan wll be mportant to be consdered n the crosslayer approach. In general, SST attenuaton model s enough to present an effect of movng ran n a cell. However, Fgure. Cell Structure n SST Model Fgure 3. Lnk Attenuaton Statstc Based on SST 3. CROSS-LAYER MODEL FOR OFDM NETWORK IN RAIN FADING In ths secton, a smple cross-layer optmzaton for OFDM wreless network s presented. 3.1 Cross-layer Model Fgure 4 gves a generc model of cross-layer approach for a down-lnk sngle cell mult-user wreless OFDM. The optmzaton dea s smlar to the proposton of (Song, 005a) (Song, 005b). The optmzaton s performed by Resource Allocaton Algorthm block; n combnng the nformaton from physcal layer and lower part of MAC layer. The contrbuton of study s to show that cross-layer approach s an effectve method to optmze the network n a ran fadng envronment. The operaton of resource allocaton s depends on channel condton of users (the Channel Status Informaton, CSI) and farness nformaton. CSI s necessary to maxmze the avalable capacty (throughput). The farness nformaton (or prorty) G-8

3 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun 009 s necessary to prevent bad users from zero utlty condton. Fgure 4. Cross-layer Model n OFDM System 3. Cross-layer Optmzaton arameters The optmzaton keys of optmzaton are the number of user nvolved n optmzaton, number of orthogonal sub- Carrer of OFDM, bandwdth per sub-carrer, CSI and performance. arameters used n the paper are lsted n Table 1. Tabel 1. System Desgn arameters Notaton arameter concerned M, Total number of user, user ndex Total number of Sub-carrer, K, k sub-carrer ndex Total bandwdth and bandwdth W, Δf of each sub-carrer, Δ f = W / K Channel gan of user for subcarrer k or for each sub-carrer H[, H[ Δ f ] bandwdth. Nose power of user for subcarrer k or for each sub-carrer N[, N[ Δ f ], bandwdth. [ Channel condton of user for ρ sub-carrer k, related to SNR ran Total power and transmt power, p[, p( Δ f ) C, c [ U ( r ) on sub-carrer k Capacty, achevable transmsson rate (bps/hz) per of user sub-carrer k r, r Total rate and data rate of user Set of allocated sub-carrer to D user Utlty functon related to achevable rate of user r To focus the optmzaton problem, we assume that each user always have nformaton n ther buffer to be transmtted va an optmzed channel, havng the followng Shannon s capacty formula. C = W log (1 + SNR) bps (3) In our case, the formulaton wll be normalzed to achevable rate of user wth power, c C W, bps/hz (4) / In regard of mult-carrer system, the equaton (4) could be rewrtten as c ( Δ f ) related to the transmsson power p( Δ f ), the channel condton H ( Δ f ), and SNR-gap β for requred BER, wth β = 1/ 5/( ln(5ber). p f H f bps c f N f β. ( Δ ) ( Δ ) ( Δ ) = log 1 + (5) ( Δ ) Hz To start wth cross-layer approach, the sgnal power, margn and SNR n the lnk budget are assumed to be enough to fulfll BER and throughput requrement n clear sky condton. So, we just consder the ran fadng mpact to the system. The SNR under ran ( SNR ) s therefore the SNR n ran clear sky (n decbel) mnus the attenuaton by ran captured from channel gan n ran. SNRran = SNRclear Aran (db) (6) In Fgure 5 we can see the example of temporal varaton of SNR n ran wth SNR clear = 30dB. Fgure 6 shows that condtonally probablty of havng SNR ran 0dB s more than 4%. It s clear that an adaptaton s requred. Fgure 5. CSI n Ran Fade, wth SNR clear =30 db Fgure 6. CDF of SNR rany, wth SNR clear =30 db So, equaton (5) can be wrtten by knowng the SNR target SNR and the condton of ran clear fadng A ran. β. SNRclear bps c f A ( Δ ) = log 1 + (7) ran Hz bps c ( Δ f ) = log ( 1+ β. CSI ) (8) Hz G-9

4 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun 009 The CSI n equaton (8) s assumed always avalable n BTS. In dscrete form, the CSI s the sample of lnk status ρ [ at a known power p [ and nose densty N [. H [ ρ [ = (9) N[ Therefore, c [ = log ( 1+ β. p[. ρ[ ) bps / Hz (10) From equaton (10), the optmzaton wll then depend on requred BER (β), power allocated p [ and number of allocated sub-carrers based on ρ [k ]. In our case, we have two cross-layer allocaton adaptaton: dynamc Sub-carrer Allocaton (DSA) and Adaptve ower Allocaton (AA). The two adaptaton are based on a selecton of an optmum set of carrer D and power p [ to optmze the avalable rate r and maxmze the utlty U(.) of each user, M. r = c [. Δf (11) k D max D, p[ k ] M U ( r ) (1) In utlty based optmzaton, the value of utlty should always be bgger than zero. The possblty to maxmze number of sub-carrer n regard of CSI for multple users M > 0 should follow the farness crtera n equaton (1) used by (Song, 005a) (Song, 005b). U () r = ln( r 0.3) (13) The maxmum utlty wll be acheved n condton that U ( r) = r = throughput n physcal layer maxmzng utlty U ( r ) means mprovng the Farness n cross-layer approach. 3.3 Dynamc Sub-carrer Allocaton Dynamc Sub-carrer Allocaton (DSA) s carred-out by controllng the allocaton of subcarrers k D based on CSI n a fxed transmtted power. The achevable capacty s the sum of each ndvdual sub-carrer havng bandwdth Δ f, r = c [ Δf (14) k D To show the advantage of DSA n ran fadng, sortng search algorthm proposed by (Song, 005b) s compared to fxed sub-carrer allocaton (FSA). By attrbuton of sub-carrer ndex x k to user wth best channel condton, the optmzaton formulaton s then denoted by: M M 1 1 ( ) = Δ U r U f. c [. xk (15) M = 1 M = 1 k K 3.4 Adaptve ower Allocaton Adaptve ower Allocaton (AA) s performed by controllng transmtted power lke n the conventonal power control, but n mult-carrer bass. Optmzaton of avalable capacty and utlty s carred out by compensatng channel attenuaton to acheve a sutable SNR n each sub-carrer. In ths case, the BTS try to normalze sub-carrer power for requred SNR value. ower optmzaton mechansm s based on the balancng of (margnal) utlty target and SNR target as n conventonal water fllng. Optmzaton s carred-out by controllng the subcarrer power p[ based on captured CSI. In ths case, p[ n equaton (9) wll be vared and denoted by * [ k * p ], wth p [ 0 and normalzaton power constantλ > 0 and k D. ( r ) ' * teraton U 1 p [ (16) λ β. ρ[ One of super constran s that the total power must not exceed the total power of total bandwdth W. Fgure 7. Water-fllng used n AA As proposed n (Song, 005b), a combnaton of sequental lnear approxmaton water fllng algorthm and greedy power allocaton algorthm s appled. The greedy approach s appled by allocatng power n form of adaptve modulaton p [ = f ( b), to make the transmsson of b (bps/hz) possble by knowng the CSI and SNR-gapβ. b 1 f ( b) = (17) β. ρ[ 3.5 Jont Sub-carrer and ower Allocaton JSA (jont DSA and AA algorthm) s a sequental combnaton between sub carrer allocaton and power adaptaton. The process s to fnd an optmal sub-carrer allocaton of all potental users and to maxmze data rate by approprate power allocaton. The result wll depend on the affnty of sub carrer assgnment process and stepsze of power adaptaton. FSA can be regarded as a statc DSA. + G-10

5 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun RESULTS AND DISCUSSIONS In ths secton, we wll frstly present the nterest of usng cross-layer based on Shannon s formula and then present the result of cross-layer on sngle cell OFDM mult-user system wth a 4 km cell dameter usng SST generated ran fadng. To show the advantage of JSA, an evaluaton between FSA and DSA, ther jont wth A wll be presented. 4.1 Optmzaton Interest Fgure 8 gves us an llustraton of sub-carrer number effect to the rate total n ran fadng. The graphs show smlar capacty trend, but the bgger number of sub-carrer wll result n the bgger probablty of havng hgher total avalable. It s approprate wth (Song, 005a) (Song, 005b) that the global optmal of system wll normally be acheved f the number of user and the number of sub-carrer are large enough. aspect n cross-layer approach to mprove the utlty of certan farness requrement. Fgure 10. Avalable Capacty n Dfferent BER Target 4. FSA and DSA n Ran Fadng In order to evaluate the nterest of applyng cross-layer approach usng DSA algorthm, we use the CDF dynamc sub-carrer allocaton usng DSA n Fgure 11. It gves a superor performance compared to conventonal Fxed Sub-carrer Allocaton (FSA) n term of channel capacty. An addtonal gan n FSA s less sgnfcant compared to the mpact of addtonal gan n DSA. Fgure 8. Rate Total wth Dfferent SC Number Fgure 9 shows us the effect of SNR value n a same ran fadng statstc. Wth hgher the SNR target, hgher avalable capacty wll be obtaned. Fgure 11. Comparson of FSA and DSA 4.3 AA and JSA Fgure 1 shows the result of AA whle operatng n the same statstc of ran fadng. Ths result s lke the result of normal power control applcaton. Fgure 9. Avalable Capacty n dfferent SNR The avalable capacty of system n a fxed SNR wth dfferent BER requrement (SNR gab) s shown n Fgure 10. When BER requrement ncrease, we need more margn to acheve the same capacty. So the avalable capacty wll depend on the value of SNR gap. From Fgure 8, 9 and Fgure 10, we conclude that the choce number of sub-carrer allocated to user, power, SNR and requred BER s mportant clear Fgure 1. Comparson AA n Dfferent SNR G-11

6 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun 009 Fgure 13 shows the advantage of usng jont cross-layer approach, when a combnaton of AA and other technque s carred-out. Combnaton of smple sub-carrer allocaton n lower MAC layer and power allocaton algorthm n HY layer results n a global mprovement n term of total utlty. By exchangng nformaton from two layers, the average utlty of dynamc sub-carrer assgnment and adaptve power allocaton s extremely superor to fxed sub-carrer allocaton n conventonal system. We know that the probablty to have maxmum throughput may be dmnshed, but the probablty of servce for each user s hgher. Ths result gves a better utlty as predcted. It means that cross-layer system gve us an opportuntes to mprove the network performance wthout sacrfcng total power and more bandwdth. Fgure 13. Utlty Dfference n Ran Fadng Ths study shows another example about the convergence of the approach, as shown n Fgure 14. As the statstc of ran fadng channel and the varaton of user dstance are random, each approach wll gve dfferent temporal results. However, we can easly conclude that cross-layer (jont subcarrer and power allocaton) gves superor result to network utlty. All graphs of JFA and JSA have hgh probablty of the utlty value. We (than) see another opportunty to mprove the OFDM network capacty by mplementng cross-layer desgn usng user dversty, based on user ndependency related to spatal- temporal characterstc of ran channel. Fgure 14. Trend of Utlty n Ran Fadng 5. CONCLUSION In ths paper, we consder the cross-layer optmzaton applcaton for mllmeter wave wreless network consderng the presence of ran fadng. To show the nterest of cross-layer optmzaton, a combnaton of dynamc sub-carrer allocaton algorthm and adaptve power allocaton algorthm s demonstrated. The result of smulaton shows that a very smple cross layer technque usng Dynamc Sub-carrer Allocaton (DSA), Adaptve ower Allocaton (AA) n the form of JSA can mprove the performance of mult-user OFDM network n term of the utlty based on farness constrant. It shows that total throughput (avalable rate) has been mproved whle keepng an effectve bandwdth and power. In the future, there wll be possblty to take more beneft from spataltemporal characterstc of ran fadng n the form of space-tme mult-user dversty n mult-cell multuser OFDM network. REFERENCES B.C. Gremont and M. Flp, (004), Spato-temporal ran attenuaton model for applcaton to fade mtgaton technques, IEEE Transacton on Antennas and ropagaton, Vol. 5, No. 5, May 004. C.Y. Chu and K.S. Chen, (005), Effect of Ran Fadng on the Effcency of the Ka-Band LMDS System n the Tawan Area, IEEE Transactons on Vehcular technology, Vol. 54, No. 1, January 005. Endroyono and Hendrantoro G., (008a), Mtgaton of Mllmeter Wave Channels Attenuaton on OFDM Networks n Tropcal Regon usng Cross-Layer Optmzaton Approach, ICAST008, Kumamoto Unversty, Japan March 008. Endroyono and Hendrantoro G., (008b), Cross- Layer Optmzaton erformance Evaluaton of OFDM Broadband Network on Mllmeter Wave Channels, IEEE WOCN 008, 5th IFI Internatonal Conference on Wreless and Optcal Communcatons Networks, Surabaya, 5-7 May 008, age(s):1 5 F.. Fontan, A.Nunez, A. Valcarce and U. C. Febg, (005), Convertng Smulated Ran-rate Seres nto Attenuaton Seres Usng the Synthetc Storm Technque, COST Acton 80, 3rd Internatonal Workshop, M9104, June 005. G. Song and Y.G L, (005a), Cross-layer optmzaton for OFDM wreless network - art I: Theoretcal framework, IEEE Trans. Wreless Communcaton, vol. 4 no. pp , Mar 005. G. Song and Y.G L, (005b), Cross-layer optmzaton for OFDM wreless network - art II: Algorthm Development, IEEE Trans. Wreless Communcaton., vol. 4 no. pp , Mar 005. Kanellopoulos, J. D. and. Kafetzs, (1986), Comparson of the Synthetc Storm Technque G-1

7 Semnar Nasonal Aplkas Teknolog Informas 009 (SNATI 009) Yogyakarta, 0 Jun 009 wth a Conventonal Ran Attenuaton redcton Model, IEEE Transactons on Antennas and ropagaton, Vol. A-34, No. 5, May Laurent Castanet, Ana Bolea-Alamañac, Mchel Bousquet, (003), Interference and Fade Mtgaton Technques for Ka and Q/V Band Satellte Communcaton Systems, roc. COST- 80, 003 Mahmudah, H., (008), redks Redaman Hujan Menggunakan Synthetc Storm Technque, Thess Jurusan Teknk Elektro FTI-ITS, 008. V Corvno, et.al. (008), Cross-layer rado resources allocaton for mult-carrer ar nterfaces, n mult-cell mult-user envronments, manuscrpt accepted by IEEE Transacton on vehcular technology, ISSN: , 008. G-13