Full-Duplex Device-to-Device Collaboration for Low-Latency Wireless Video Distribution

Transcription

1 Fu-Dupex Devce-to-Devce Coaboraton for Low-Latency Wreess Vdeo Dstrbuton Mansour Nascheragh 1, Member, IEEE Seyed A Ghorash 1,2, Senor Member, IEEE, Mohammad Shkh-Bahae 3, Senor Member, IEEE 1. Department of Eectrca Eng., Shahd Behesht Unversty, G.C. Tehran, Iran 2. Cyberspace Research Insttute, Shahd Behesht Unversty, G.C. Tehran, Iran 3. Centre for Teecommuncatons Research, Kng s Coege London, UK m.nascheragh@ma.sbu.ac.r, a ghorash@sbu.ac.r, m.sbahae@kc.ac.uk arxv: v2 [cs.ni] 18 Apr 217 Abstract Growng demand for vdeo servces s the man drver for ncreasng traffc n wreess ceuar data networks. Wreess vdeo dstrbuton schemes have recenty been proposed to offoad data va Devce-to-Devce D2D communcatons. These offoadng schemes ncrease capacty and reduce end-toend deay n ceuar networks and hep to serve the dramatcay ncreasng demand for hgh quaty vdeo. In ths paper, we propose a new scheme for vdeo dstrbuton over ceuar networks by expotng fu-dupex FD D2D communcaton n two scenaros; scenaro one: two nodes exchange ther desred vdeo fes smutaneousy wth each other, and scenaro two: each node can concurrenty transmt to and receve from two dfferent nodes. In the atter case, an ntermedate transcever can serve one or mutpe users fe requests whst capturng ts desred fe from another devce n the vcnty. Anaytc and smuaton resuts are used to compare the proposed scheme wth ts haf-dupex HD counterpart under the same transmtter estabshment crtera to show the achevabe gan of FD-D2D scheme n vdeo content devery, n terms of sum throughput and atency. Index Terms ceuar vdeo cachng, wreess vdeo dstrbuton, D2D communcaton, fu-dupex, haf-dupex. I. INTRODUCTION Increasng demand for hgh data rate and ve vdeo streamng n ceuar networks has attracted researchers attenton to cache-enabed ceuar network archtectures [1]. These networks expot D2D communcatons as a promsng technoogy of 5G heterogeneous networks, for ceuar vdeo dstrbuton. In a ceuar content devery network asssted by D2D communcatons and smary n peer-asssted networks [2], user devces can capture ther desred contents ether va ceuar nfrastructure or va D2D nks from other devces n ther vcnty. Recenty, severa studes n both content pacement poces and devery strateges are conducted to mnmze the downoadng tme, and to maxmze the overa network throughput n terms of rate and area spectra effcency. From the content pacement perspectve of vew, contents can be paced on coaboratve nodes formery, ether accordng to a predefned cachng pocy reactve cachng [3], or more ntegenty, accordng to statstcs of user devces nterest proactve cachng [4]. The theoretca bounds for D2D cachng network proposed n [5], ndcates that cachng most popuar contents n users devces s optma n amost a system regmes. Cross-ayer resource aocaton methods are aso nvestgated for supportng vdeo over wreess n mutuser scenaros [6]. It s shown that quaty-aware resource aocaton can mprove vdeo servces n wreess networks. However, the conventona archtectures of content devery n both wreess ceuar and D2D networks, are based on hafdupex HD transmsson and to the best of our knowedge, fu-dupex FD capabty and ts advantages have not yet been nvestgated n both wreess ceuar vdeo dstrbuton and D2D cachng systems. Recent advances n FD rado desgn [7], materazed by advanced sgna processng technques that can suppress sef-nterference SI at the recever, have enabed smutaneous transmsson and recepton over the same frequency band. From theoretca pont of vew, FD communcaton can potentay doube the spectra effcency of a pont-to-pont communcaton nk, provdng SI s entrey canceed. In ths paper, we propose an FD-based scheme for D2D wreess vdeo dstrbuton. Detas aong wth the man contrbutons are as foows: The proposed scheme has been nvestgated n two dfferent scenaros: a user devces operate n bdrectona FD mode n whch two users can exchange data smutaneousy at the same frequency and b user devces can concurrenty transmt to and receve data from two dfferent nodes at the same frequency..e., an ntermedate node can receve ts desred content from one node and smutaneousy serve for other user s demand at the same frequency. We have anayzed throughput and deay n both scenaros and compared them aganst conventona HD systems. In contrast wth the works n the terature [3], where ony one actve node per custer s consdered, we consder D2D communcaton among mutpe nodes n our proposed scheme. We have derved cosed form expressons for FD/HD- D2D coaboraton probabtes whch prevousy obtaned by numerca evauatons n [8]. The remander of paper s structured as foows. In Secton II system mode s ntroduced. In secton III, throughput anayss for the proposed FD-enabed ceuar system s provded. In secton IV smuaton resuts are expaned and concusons

2 are presented n secton V. II. SYSTEM MODEL We consder a ceuar network wth a snge ce, one base staton BS and n randomy dstrbuted users Fg.1 a accordng to unform dstrbuton. Assumng that nterce nterference s neggbe or canceed out, anayss can be extended to mut-ce scenaros. We dvde the whoe ce area nto ogca equay szed square custers Fg. 1a and negect co-channe nterference and neghborng ce users nfuence, for the sake of smpcty. We consder an n-band overay spectrum access strategy for D2D communcatons [9]. Thus, there s no nterference between ceuar and D2D communcatons. A D2D communcatons are under fu contro of the BS. We aso assume that SI canceaton aows the FD rados to transmt and receve smutaneousy over the same frequency band. However, snce a D2D pars n a custers share the same resource bocks, nter- and ntra-custer nterference s taken nto account. Denote the set of popuar vdeo fes as V {v 1, v 2,..., v m } wth sze m. We use Zpf dstrbuton for modeng the popuarty of vdeo fes and thus, the popuarty of the cached vdeo fe v s n user u ω, denoted by f ωs, s nversey proportona to ts rank,.e., f ωs s 1 γr m g1, 1 s m. The Zpf exponent g γr γr characterzes the dstrbuton by controng the popuarty of fes for a gven brary sze m. Contents are paced n users caches n advance accordng to a cachng pocy n whch each user wth a consderabe storage capacty can cache a subset of fes F V from the brary,.e., F {f 1, f 2,..., f h }, h m. We assume that there s no overap between users caches,.e., F p F q φ. Each user randomy requests a p q vdeo fe from the brary accordng to Zpf dstrbuton. Techncay, to schedue and estabsh a D2D connecton, necessary sgnang messages are needed to be exchanged between D2D pars and the BS [1]. However, the sgnang mechansms do not affect our anayss n ths work. Hence, we adopt the protoco mode of [11] to setup D2D communcatons, whch s based on a dstance threshod; A par of users/devces u, u j can potentay ntate a D2D communcaton for vdeo fe transfer provdng that the dstance between u and u j s ess than a threshod n Fg. 1a and one of them fnds ts desred vdeo fe n the other devce. Fg. 2b ustrates the schematc of typca D2D communcaton graphs nsde a custer. Each user generates a random request accordng to the Zpf dstrbuton. BS s assumed to be aware of a contents n the users caches. We defne a drected edge from u pontng to u j f the user u j requests a fe that has been prevousy cached by u. Snce we assume that each user can make ony one request as shown n Fg. 1b, there w be at most one ncomng nk to the user node and one or mutpe outgong nks from the user node. In ths system, no data s reayed over mutpe hops, whch means any transmssons from one node to another nodes corresponds to deverng a dfferent vdeo content. It s aso possbe that some users demand for the same vdeo content whch s prevousy cached by one user. For nstance the users n set Z demand for the a User Custer a Square ce wth equa-szed custers wthn the ce BS a u1 u2 u3 u4 u5 Haf-Dupex Transmtter Haf-Dupex Recever u6 Fu-Dupex Node Z... Data Drecton... Set of Recevers b Schematc of D2D communcatons Fg. 1. System mode and D2D communcatons graph same vdeo content from user u 6 Fg. 1b. The number of users n set Z depends on the popuarty of the vdeo content whch s desred by these users. As can be seen n Fg. 1b, there are two dfferent possbe confguratons for FD coaboraton; b-drectona fu-dupex BFD mode, n whch two users exchange ther desred vdeo content and three node fu-dupex TNFD mode, n whch an ntermedate node can receve ts desred vdeo content from one node and smutaneousy serve for another users demand see u 6 n Fg. 1b. III. ANALYSIS Both anaog and dgta SI canceaton methods can be used to partay cance the SI. However, n practce, t s dffcut or even mpossbe to cance the SI perfecty. We assume that a users transmt wth power P t. The SI n FD nodes s assumed to be canceed mperfecty wth resdua sef-nterference-topower rato β and hence, the resdua SI s βp t. The parameter β denotes the amount of SI canceaton, and 1og 1 β s the SI canceaton n db. When β, there s perfect SI canceaton, whe for β 1, there s no SI canceaton. Thus, the sgna-to-nterference-pus-nose rato SINR at recever u j due to transmtted sgna from u can be wrtten as SINR j P t h j d α j σ 2 + z Φ\{} P th zj d α zj + χβp t, 1 where z Φ\{} P th zj d zj s tota nter- and ntra-custer nterference due to the nodes n set Φ, whch s the set of concurrent transmttng nodes. Backsash n eq. 1 mpes that the node u s excuded from transmtters. h j and h zj are the fadng power coeffcents wth exponenta dstrbuton of mean one, correspondng to the channe between transmtter u and recever u j, and nterferer u z, respectvey. d j denotes the Eucdean dstance between transmtter u and recever u j nsde the custer. α s the path oss exponent. A whte Gaussan nose wth power σ 2 s added to the receved sgna. χ denotes coaboraton mode; χ, when user u operates n HD mode, and χ 1, when t operates n FD mode. A. Coaboraton Probabty For gven k users whch randomy fa nsde a custer and gven h number of cached contents for each user nsde the

3 random custer c, we defne popuarty of cached contents wthn the custer as ρ c k 1 ρ u, where ρ u f s F f s s the popuarty of cached contents by user u. For the th user, u, we defne two parameters P a and P b as foows; P a : the probabty that u cannot fnd ts desred content wthn custer. P b : the probabty that user u can serve for other users requests. Snce a requests are dentcay dstrbuted and ndependent..d at each user, gven k users nsde the custer, the probabty that u operates n HD mode s P HD u k P ap b. 2 Smary, the probabty that u operates n FD mode s P FD u k 1 P ap b. 3 However, the probabty of makng HD-D2D and FD-D2D connectons depends on parameter k. The probabty that u can coaborate n HD or FD mode s P δ u k Pu δ k Pr[K k], 4 where δ {HD, FD} s the operaton mode, Pr[K k] s the probabty that there are k users n the custer. Snce the dstrbuton of users s assumed to be unform wthn the ce area, the number of users n the custer s a bnoma random varabe wth parameters n and 2 2a,.e., K Bn, 2 2 2a, 2 where 2 2a s the rato of the custer area to the ce area. 2 Hence, the probabty that k users fa nsde the custer s n 2 k n k Pr[K k] 1 k 2a 2 2 2a 2. 5 The probabty that u can fnd ts desred fe nsde the custer and cannot fnd on ts own cache.e., we excude sef-request 1 from user u, can be wrtten as P a ρ c ρ u. 6 We defne Q u x whch determnes the probabty that u can serve x number of users requests nsde the custer. The number of users demandng for a content whch s cached by u s a bnoma random varabe wth parameters k 1 and ρ u,.e., Q u x k 1 x ρ u x 1 ρ u x. k 2 7 It s cear that for k < 2, Q u x, whch mpes that there s no user s demand for cached content by u. And fnay, P b can be wrtten as P b Q u x. 8 By substtutng eqs. 6, 8 n eqs. 2, 3, and eqs. 2, 3 n eq. 4, respectvey, we get the fna mathematca expressons for HD and FD coaboraton probabtes. P HD u k 1 ρ c ρ u Q u x Pr[K k]. 9 1 sef-request takes pace when the user fnds ts desred fe n ts own cache. P FD u ρ c ρ u Q u x Pr[K k]. 1 k Denotng Pu sef as the probabty that user u fnds ts desred content on ts own cache, By substtutng eqs. 9 and 1 n Pu FD + Pu HD + Pu sef 1, the probabty that node u demands for a fe whch s cached by tsef s P sef u 1 k B. Throughput Anayss P serve u Pr[K k]. 11 We focus on a typca random custer c representatve custer and derve system sum throughput for ths custer. We obtan the ergodc capacty of the nk assocated wth D2D par u,u j, whch s defned by C j W E[og SINR j ], where, W s the bandwdth for D2D nk. For the wreess D2D network descrbed n secton II, the expected vaue of the throughput of the system due to estabshng node u n δ mode can be wrtten as T δ u P δ u C δ u, 12 where Pu δ s the coaboraton probabtes for δ mode, whch s derved n equatons 9 and 1. Cu δ s achevabe capacty by estabshng node u n δ mode and can be cacuated as W E[og SINR j ], 13 C HD u u j A Cu FD W E[og SINR o ] + W E[og SINR j ], 14 u j B where A and B are the set of users whch are connected to u n HD and FD modes respectvey. Frst term n eq. 14,.e., W E[og SINR o ], determnes the ergodc capacty for the nk through whch u receves ts desred fe n FD mode from u o TNFD mode. Showng the set of estabshed nodes nsde the random custer c s by Ψ {u 1, u 2,..., u τ }, the sum throughput of the respectve custer can be wrtten as C. Downoad Tme η δ c u Ψ T δ u. 15 As we descrbed n secton II, there are two fu-dupex coaboraton modes: TNFD and BFD. For better understandng the concept of downoad tme n HD and FD modes, we use the D2D communcaton graphs shown n Fg. 1b. 1 TNFD Mode: consder u, u j and Z {u 1, u 2,...u k } n whch u / Z, u j / Z. For a typca nk between u and u j, and assumng that u s transmttng vdeo fe v j to u j, the experenced average downoad tme θ j at u j can be defned as θ j bv j C j, where b vj s the number of bts for vdeo fe v j and C j s the achevabe ergodc capacty for transmttng nk from u to u j. Smary, for the set Z, we have: Θ {θ j 1, θ j 2,..., θ j k } where, θ j k bv k C j k. Due to random dstrbuton of the users ocatons, the ergodc

4 capacty for a nks assocated wth a users n set Z s not necessary the same, hence θ j p θ j q for p q. Snce a users n set Z are demandng the same vdeo content from u j, the tota average downoad tme due to one transmsson of user u j can be defned as ϖ max 1 λ k θ j λ, θ j λ Θ. 16 Denotng Du HD j and Du F D j as the tota experenced average downoad tmes by estabshng u j n HD and FD modes, respectvey, we have: D HD u j θ j + ϖ, D F D u j maxθ j, ϖ BFD Mode: In ths mode, both users.e., u 3 and u 4 n Fg. 1b exchange data smutaneousy. Denotng θ j and θ j as the experenced downoad tme for u and u j, respectvey, the tota average downoad tme can be cacuated as D HD θ j + θ j, D F D maxθ j, θ j. 18 In practce, the receved and transmtted packets may have dfferent engths. Therefore, the transmsson of a nodes w not end up at the same tme. Therefore, due to asymmetrc data packets at the transmtter and recever, ths stuaton s referred to as the resdua hdden node probem. However, the node that fnshes data transmsson earer can resove ths ssue by transmttng busy tone sgnas unt the other node competes ts transmsson [12]. IV. NUMERICAL EVALUATIONS In ths secton, we provde Monte-Caro smuaton to evauate the performance of our proposed FD-D2D cachng system. We assume a snge square ce as shown n Fg. 1a. Smuaton parameters are shown n Tabe 1. The proposed FD-scheme smuated based on the foowng scenaros: Cachng procedure: Each user caches mutpe fes from the brary, accordng to the descrbed cachng pocy n secton II. Ths procedure can be aunched n the off-peak hours of the ceuar network to avod traffc oad. Devery procedure: Users make and send ther request to the BS randomy accordng to Zpf dstrbuton and consequenty the BS recognzes users nterests. Moreover, users ocatons are known n advance for BS va channe state nformaton CSI procedure. Hence, BS can predct potenta D2D communcatons graphs as such n Fg. 1b for a custers by havng knowedge of users caches, nterests and ocatons. In each custer, BS determnes and estabshes τ number of nodes assocated wth most popuar cached contents. Snce a D2D communcatons n a custers use the same tme-frequency resources, nter- and ntra-custer nterferences are taken nto account. Fg. 2 shows the probabty that a node nsde a custer s n FD, HD or sef-request mode. By ncreasng D2D coaboraton threshod, the expected number of users nsde a custer ncreases and, consequenty, the expected number of nodes that coaborate n FD mode ncreases. As can be seen n Fg. 2, the probabty that users can fnd ther desred content on ther own caches, decreases as ncreases, because for ower vaues of, there are few users nsde a custer and these users prevousy stored hgh popuar fes. Hence, when the users nsde a custer make request accordng to Zpf dstrbuton, there s a hgh probabty that they make a request for a fe that they have prevousy stored on ther own caches. In contrast, as the densty of users nsde a custer ncreases.e., hgher vaues of, the number of sef-request users decreases. Fg. 3 shows the mpact of the number of users n wthn the ce on the aforementoned probabtes. As can be seen from Fg. 3, the hgher densty of users wthn a ce, the hgher s FD coaboraton probabty. Fg. 4 and Fg. 5, show the tota average rate for FD-D2D and HD-D2D systems. Athough the number of custers ncreases at ower ranges of we expect that the frequency reuse ncreases as we, nevertheess, the probabty that custers are of ow densty or that no D2D canddates are found theren, aso ncreases. Ths can be nterpreted as the fact that the probabty of fndng a user s desred fe nsde the custer decreases when the node densty decreases. As the number of custers n the ce decreases, the frequency reuse decreases too. However, the probabty that a user can fnd ts desred fe nsde the custer ncreases and hence the probabty of makng D2D communcaton ncreases. The mpact of parameter τ τ n Fg. 5 demonstrates that ncorporatng FD-enabed nodes, wth mutpe nodes estabshments nsde a custer can mprove the average gan n sum throughput by ncreasng the number of actve D2D nks. Aongsde the consderabe mprovements n system sum throughput, the gan n frequency reuse n FDenabed system s more accentuated for hgher ranges of. Fg. 6 ustrates the tota average downoad tme versus. As we dscussed n secton II, there are three possbe ways for the users to access ther desred fe; through conventona ceuar nfrastructure, va D2D coaboraton, and by sef-request. We defne the downoad tme as the deay ncurred n downoadng a fe,.e. the tme between sendng requests by the user t capturng the fe. Downoad tme for sef-request case s zero and we excude ths case from cacuatons of downoad tme. In the proposed FD-D2D system, each D2D recever can downoad ts desred fe wth zero watng tme. Fg. 6 shows the major mpact of FD coaboraton on decreasng the atency n downoadng vdeo fes. Parameter TABLE I SIMULATION PARAMETERS Vaues V Vdeo Content Lbrary v th vdeo content n brary Number of users n [1 1] Cached contents per node h 1, 3, 5 Sze of brary m 1 Zpf exponent γ r 1, 1.6 SI canceaton factor 1og 1 β -7 db Number of estabshed nodes τ 1, 2, 3 D2D nk bandwdth W 1.2 MHz Background nose σ dbm/hz Path oss exponent α 2.6 Sze of fes [5 5] MB User transmt power P t 23 dbm Ce sze a 1 km Log-norma shadow fadng 4 db standard devaton Monte-Caro teratons 1

5 1 7 Coaboraton Probabty , Sm 1, Sm Sef, γ r Sef, γ r 1, Sm 1.6, Sm 1.6, Sm Sef, γ r Sef, γ r 1.6, Sm Average Rate Mbps FD, τ2 HD, τ2 FD, τ3 HD, τ Fg. 2. Coaboraton Probabty versus for n 5 and h 1. Fg. 5. Average rate versus for h 3, γ r 1 and n 1. Coaboraton Probabty FD, Theory FD, Sm HD, Theory HD, Sm Sef, Theory Sef, Sm n Fg. 3. Coaboraton Probabty versus n for γ r 1.6, h 5 and.2. Average Downoad Tme s , n5 1, n5 1.6, n5 1.6, n5 1, n1 1, n1 1.6, n1 1.6, n Fg. 6. Tota average downoad tme versus for h 1 and τ 1. Average Rate Mbps , n5 1, n5 1.6, n5 1.6, n5 1, n1 1, n1 1.6, n1 1.6, n Fg. 4. Average rate versus for n 5 and h 1 and τ 1. V. CONCLUSION In ths paper, we used fu dupex rados on user devces to ncrease the throughput of vdeo cachng n ceuar systems wth D2D coaboraton. We nvestgated FD-enabed networks by enabng FD rados ony for D2D communcatons. Smuaton resuts show that achevabe throughput gan can ncreases n hgh ntra- and nter-custer nterference condtons. We aso showed that aowng fu dupex coaboraton can have a major effect on the quaty of vdeo content dstrbuton by reducng downoad tme compared to HD-ony coaboraton. REFERENCES [1] N. Gorezae, A. F. Mosch, A. G. Dmaks, and G. Care, Femtocachng and devce-to-devce coaboraton: A new archtecture for wreess vdeo dstrbuton, IEEE Commun. Mag., vo. 51, pp , 213. [2] N. Anjum, D. Karamshuk, M. Shkh-Bahae, N. Sastry, Survey on Peerasssted Content Devery Networks, Computer Networks, Esever. [3] N. Gorezae, P. Mansourfard, A. F. Mosch, and A. G. Dmaks, Base- Staton Asssted Devce-to-Devce Communcatons for Hgh-Throughput Wreess Vdeo Networks, IEEE Trans. Wreess Commun., vo. 13, pp , 214. [4] E. Bastug, M. Benns, and M. Debbah, Lvng on the edge: The roe of proactve cachng n 5G wreess networks, IEEE Commun. Mag., vo. 52, pp , 214. [5] M. J, G. Care and A. F. Mosch, Fundamenta Lmts of Cachng n Wreess D2D Networks, IEEE Trans. Inform. Theory, vo. 62, no. 2, pp , Feb [6] H. Sak and M. Shkh-Bahae, Cross-Layer Resource Aocaton for Vdeo Streamng Over OFDMA Cogntve Rado Networks, IEEE Trans. Mutmeda, vo. 17, pp , 215. [7] Knox, Mchae E. Snge antenna fu dupex communcatons usng a common carrer. IEEE WAMICON, 212. [8] M. Nascheragh, S. A. Ghorash, M. Shkh-Bahae, Fu-Dupex Devceto-Devce Communcaton for Wreess VdeoDstrbuton, IET Communcatons, 217. [9] A. Asad, Q. Wang and V. Mancuso, A Survey on Devce-to-Devce Communcaton n Ceuar Networks, IEEE Communcatons Surveys & Tutoras, vo. 16, no. 4, pp , 214. [1] M. Nascheragh, L. Marand, and S. A. Ghorash, A Nove Devceto-Devce Dscovery Scheme for Underay Ceuar Networks. arxv preprnt arxv: [11] P. Gupta and P.R. Kumar, The capacty of wreess networks. IEEE Trans. Inform. Theory, 46.2, pp , 2. [12] Y. Zhang, L. Lazos, K. Chen, B. Hu, and S. Shvaramaah, FD-MMAC: Combatng mut-channe hdden and exposed termnas usng a snge transcever, n IEEE INFOCOM 214, pp