Millimeter-Wave Multi-Hop Wireless Backhauling for 5G Cellular Networks

Transcription

1 2017 IEEE 85th Vehiclar Technology Conference (VTC-Spring) Millimeter-Wave Mlti-Hop Wireless Backhaling for 5G Celllar Networks B. P. S. Sahoo, Chn-Han Yao, and Hng-Y Wei Gradate Institte of Electrical Engineering, National Taiwan University, Taiwan Abstract The millimeter-wave (mmwave) bands, roghly referred to GHz, have been widely recognized as a promising candidate for dense deployment of small-cells backhal network. The dense small-cell deployment prodces a hge amont of backhal traffic, and directional commnication poses a significant challenge. In this paper, we propose a dynamic frame reconfigration scheme which provides greater flexibility for dynamic traffic adaptation in mlti-hop mmwave relay backhal. This allows better exploitation of the traffic dynamics in smallcell. In addition, we present a traffic load and link-qality aware mlti-hop relay backhal schedling algorithm to maximize the overall system performance. The extensive simlation reslts demonstrate the speriority of or proposed algorithm when compared with other schemes. I. INTRODUCTION Globally, the demand for higher data-rate is continally increasing. This, in trn, has led to a significant increase in capacity reqirements for the backhal network. In traditional wireless system, the mlti-hop aided transmission has been considered as an effective way to increase the coverage, throghpt and transmission reliability of networks [1][2]. Ths, the mlti-hop relaying is likely to play a significant role in mmwave celllar systems for self backhaling. The designing of schedling policies for these systems is challenging de both adaptive directional transmissions and dynamic timedivision dplexing schedles, which are key enabling featres of mmwave systems [2]. For designing schedling policies, there are two aspects of challenges in the backhal network of relay aided mmwave systems. In the first aspect, providing greater flexibility in dynamic adaptation for beam-specific transmit/receive selection to cater the hge backhal traffic. In the second, the inter-bs schedling mechanism shold serve the maximm backhal traffic demand and provide higher throghpt. The schedling mechanism shold be able to ensre fair transmission opportnity for each BS with non-niform traffic load in the relay system which is a challenging task. In this paper, we investigate the mlti-hop relaying transmission challenges for mmwave systems. We discss three soltions based on the point-to-point (P2P) link-qality and traffic-load at each BS that optimizes the overall network performance. Firstly, we present Load-iBS, a traffic-load based inter-bs schedling mechanism. This approach ses the trafficload metrics to maximize the performance withot acqiring link-state information. Secondly, we present Link-iBS, a linkqality based inter-bs schedling mechanism. This approach ses the link-qality metric to determine the schedling decision to maximize the performance. Finally, we present throghpt-optimal inter-bs, TO-iBS, a combined version of both Load-iBS and Link-iBS mechanism. This approach ses both the traffic-load and link-qality metrics, and sbseqently reallocate the power proportional to the transmission reqirements to determine the schedling decision. Henceforth, the main contribtion are smmarized as follows: We propose a frame reconfigration scheme which provides greater flexibility in dynamic traffic adaptation. We formlate the inter-bs schedling problem into a nonlinear integer programming, i.e. to maximize the nmber of flows to meet the traffic demand. We present a novel heristic schedling algorithm with adaptive power allocation method for mlti-hop relaying by extensively considering the traffic load at each BS. The proposed inter-bs schedling mechanism showcase sperior performance compared with other schemes. Related Work: There are ample of existing MAC protocols recently proposed for mmwave system [3][5][8] that are based on TDMA inclding several schedling algorithms that spports concrrent transmissions [7][9], and dynamic TDD algorithms [6][11] designed to take the advantage of beamforming. In [7], the athors propose a spatial-time division mltiple access (STDMA) algorithm that schedles both non-interfering and interfering links simltaneosly. It aims to maximize the overall system throghpt. However, it does not allow the exploitation of the traffic dynamics. Recently, athors in [9] proposed a QoS-aware schedling algorithm that partially considers the traffic dynamics, however, it does not exploit the flexibility to adapt flctating capacity reqirements in the backhal. A fndamental problem is ths to exploit the traffic dynamics and flctating capacity reqirements to determine the backhal schedling decision. In this work, a dynamic dplexing scheme is considered, where each BS determine its transmit-receive dplex pattern. A. Network Model II. SYSTEM OVERVIEW We consider a mlti-hop celllar network consisting of a set of mmwave-enabled small-cell enodeb (SeNB), where a macrocell enodeb (MeNB) provides a root from which connections goes to the SeNB at the edge via relay SeNBs that are self-backhaled on mmwave band as shown in Fig. 1. Copyright 2017 held by B. P. S. Sahoo, Chn-Han Yao, and Hng-Y Wei. Pblication Rights Licensed to IEEE

2 The backhal network, N (B, L), operates in slotted time t {0, 1, 2,... }, where B {1,..., B} denotes the set of SeNBs and L {1,..., L} denotes the set of P2P links, indexed by b and l respectively. We denote the cardinalities of these sets as B and L respectively. Fig. 1: Illstration of 5G mmwave self-backhal network. We assme, all the nodes have adaptive beamforming capabilities, and are able to transmit to or receive from mltiple neighbors concrrently. The transmission over each link is defined by the general half-dplex constrained. The channel condition remains the same throghot a frame, bt may vary dynamically from frame to frame. The system rns a bootstrapping program by which each node is capable of detecting and identifying its set of neighbors. B. Physical Model For mmwave commnications based on the abstraction sed in the prior stdy [2], the received power at receiver v B from a transmitter B with transmitting power P t (, v), can be calclated as: P r (, v) = P t (, v)ψ(, v)p L(, v) 1 (1) where ψ is the combined antenna gain of the receiver and transmitter, and P L(, v) 1 is the associated path-loss in db. Withot loss of generality, we consider the Line-of-Sight (LOS) model, expressed in (2), since Non-Line-of-Sight (NLOS) transmissions sffers from higher attenation than LOS transmission, hence, not sitable for backhal transmission [9]. It is noted that an extension to inclde both LOS and NLOS model is straightforward. P L(, v)[db] = α + 10β log 10 v + ξ (2) where ξ N(0, σ 2 ), and α and β are provided in Table I. Let G 2 max be the maximm gain on the desired backhal link v. We assme, SINR v is the instantaneos SINR vale of link v, which is compted based on the active links in the crrent network schedle at t. This ratio is defined in the expression (3) as below SINR v = P t(, v)g 2 maxp L(, v) 1 I 0 + σ 2 N (3) where P t (, v) is the transmitting power from node to node v, assming I 0 is the interference de to concrrent transmission, and σn 2 is the thermal noise power. Given the SINR and allocated bandwidth, we assme that the maximm data rate on link v, i.e., the achievable capacity, expressed in (4), can be estimated as R v = ηw log 2 (1 + SINR v ) (4) where W is the mmwave system bandwidth allocated to the link, η is the bandwidth overhead. C. Backhal Traffic Flows Let data that is destined for node v B be labeled as. Every node keeps internal qees that store data according to its destination. Let (t) be the crrent amont of data for node v in node, where v N(), the set of neighbor nodes of. The nits of (t) can be integer nits of packets or ( real valed ) nits of bits. Specifically, we define Q (t) = (t) as the matrix of crrent qee backlogs at. We Q (N()) assme that Q () (t) = 0 for all slots t, as no qee stores data destined for itself. The qee backlogs from one slot to the next satisfy the following, for all, v B, sch that v: (t + 1) max [ Q (t) v=1 (t), 0 ] + A (v) (t) where A (v) (t) is the amont of new data that arrives at node from v N() on slot t. (t) be the matrix exogenos arrival on t, for the node B, it is calclated by adding the traffic from access network (UE BS) and the traffic from all other backhal nodes (BS BS) as Let A (v) (5) A (v) (t) = λ 0 + λ (v) p v (6) v=1 where λ 0 is the traffic from access network to node, and p v is the probability that the traffic is transmitted from node v to, where v N(). We assme that λ () = 0 for all B, as no data arrives that is destined for itself. Ths, A (v) (t) is a matrix of non-negative real nmbers, with zeros on the diagonal. D. TDD Frame Reconfigrations for Backhaling To better exploit the traffic variations, enhanced Interference Mitigation and Traffic Adaptation (eimta) was introdced [4], which notably allows dynamic adaptation of the TDD pattern. Each TD-LTE frame consists of 10 sbframes, which is preconfigred for downlink (D) or for plink (U) resorces. LTE offers 7 different patterns to statically configre the sbframes as D, or L, as shown in Fig. 2. On the downside, it sffers from nbalanced allocation of UL and DL time-slots in neighboring nodes. On the other hand, with an increased focs on small-cell deployments, the basis for the backhal resorce allocation design shold be dynamic TDD

3 where a sbframe carrying data (or the whole frame) shold be assigned either for transmission or for reception as part of the dynamic schedling decisions. We proposed a novel dynamic frame reconfigration mechanism, where each sbframe cold be allowed to transmit to or receive from mltiple nodes concrrently, a maximm-size data frame according to the channel capacity. This allows better exploitation of the traffic dynamics in small-cell deployments. Fig. 2: The LTE TDD configrations. Specifically, or proposed scheme incldes two steps: firstly, we assme a mechanism is in place to poll the transmission reqests from neighboring nodes, so that each node cold be aware of the traffic demand of its neighbor nodes; secondly, the sbframes are reconfigred dynamically as part of the dynamic schedling algorithm. E. Problem Formlation The schedling of sbframes is the TDD configration that is sed for inter-bs data commnication. We consider a qeed traffic network where an end-to-end flow with a minimm throghpt reqirement is transmitted throgh mltiple hops. We calclate the traffic load at each BS and establish a schedle to carry inter-bs data by exploiting concrrent transmissions. For this prpose, we present a schedling mechanism to find the set of transmitting and receiving nodes, T and R, respectively. Assming the crrent amont of traffic that is destined to node v from is (t) on slot t. Let max (t) be µ v the maximm amont of traffic that can be delivered from v with the crrent P2P link condition µ v on t. It means choosing a higher µ v is amonts to maximizing the amont of traffic than can be transmitted. Ths, we want to choose an improved µ v for each P2P link so as to maximize the amont of traffic that can be delivered on slot t over the whole network. We assme the traffic demand is Q N() (D) on each link N(), we aim to maximize the total amont of traffic demand served over the whole network by allowing concrrent transmission nder the crrent channel conditions, and halfdplex transmission constraints. Given the hge traffic load at each BS in the backhal network, and the limited nmber of timeslots; the optimal schedle, while providing a fair transmission opportnity for each BS, shold also served as mch traffic demands as possible. The optimal backhal traffic schedling problem can be formlated as follows. Q = arg max µ v sbject to =1 =1 B L b=1 l=1 µ v (7) P t () P max, B (8) W W max, B (9) l t + l t v 1, if node and v are adjacent;, v (10) This is a non-linear integer programming problem, and is NP-hard. Constraint (8) indicates the total power is shared dring concrrent transmission, (9) indicates the total bandwidth is shared among mltiple links, and (10) indicates the halfdplex assmption, where adjacent links cannot be schedled concrrently in the same schedle. Since it is difficlt to solve the problem (7) in polynomial time, we propose a distribted heristic algorithm instead to solve this in Section III. III. THE SCHEDULING POLICY The proposed heristic algorithm yields the T and R set for each sbframes of Flexible Frame, while enabling the spatial rese to maximize the total throghpt of the system. A. Set Membership Metric To achieve high transmission data rate for each traffic flow, links with higher channel qality are sally preferred in link activation selection. Therefore, nodes involved in each active flow can transmit or receive maximm traffic. In addition, considering a qeed network, when traffic aggregates at some nodes, congestion may occr and these nodes become bottleneck of the network. Therefore, selection of appropriate nodes for transmission or reception is important to improve the network throghpt, considering both the link qality and the traffic loads at the node. When a node receives a transmission traffic reqest, it calclates the traffic load at each neighboring node sing the local one-hop information, inclding the P2P channel statistics. Based on this metric, we categorise or proposed schedling algorithm into three parts and describe it in next sbsection. B. The Proposed Schedling Policy 1) Load-based inter-bs Schedler (Load-iBS): As the name sggests, in Load-iBS we consider the traffic load at each BS to determine the T and R set. BS with maximm traffic to transmit is selected as a member of T in the crrent sbframe. Therefore, the network can transmit maximm aggregate traffic at any sbframe. Note that the sccessfl reception is proportional to the transmission reqirement and

4 channel qality. The selection metric (hereinafter weight, ω) is calclated as below ω v =, v N() (11) 2) Link Qality-based inter-bs Schedler (Link-iBS): In the Link-iBS, to achieve the higher transmission data rate in each P2P link, the links with higher channel qality is activated for crrent sbframe. Therefore, the total traffic transmitted over each higher-qality P2P link is amonts to maximizing the overall network throghpt. However, this mechanism may lead to wastage transmission slot de to no traffic demand from nodes associated with that link. ω v = µ v (12) 3) Throghpt-Optimal inter-bs Schedler (TO-iBS): Considering either the load of each node or the P2P link qality may not lead to near-optimal throghpt performance in the network. Hence, differently from above two scheme, the TOiBS implements Load-iBS mechanism to obtain the T and R sets, and sbseqently implements a power reallocation mechanism to improve the active P2P link qality. We can calclate the weight as ω v = min [ Q v ], µ v (13) We tilize a proportional power allocation scheme with fractional path loss compensation for improving the link qality. We will describe it briefly. Let P max represents the maximm transmission power of the BS; P L represents the path loss between BS; SINR target represents the targeted received power level and P noise represents the noise power level. Then, the BS s transmitting power P t is set to SINR target = P t P noise P L (14) P t = ϕ P L, P min P t P max (15) where ϕ is the compensation factor and is set to ϕ [0, 1] in LTE standard with SINR target between 0 and 30dB. The psedo-code of the schedling algorithm is presented in Algorithm 1. IV. PERFORMANCE EVALUATION We considered a tree-strctred backhal network with m- hop i.e., the depth of the tree. We assme each node completes the traffic demand polling in the Anchor Frame. The basic system parameters sed for simlation are listed in Table I. The related parametres are based on [8]. In the simlation, we set two kinds of traffic model sch as: Poisson Process (PP), where each flow arrive follows a PP with arrival rate (6), and Interrpted Poisson Process (IPP), where each flow arrive follows an IPP represented by an ON-OFF process. In the ON state, a packet arrives in each time slot according to a Bernolli distribtion. In the OFF state, packets do not arrive. IPP traffic is typically brsty traffic. In order to show the advantages of or proposed scheme, we implemented the TDMA, and the Exhastive Search method for performance comparison. Algorithm 1: Millimeter-Wave Inter-BS Schedler Inpt : N (B, L, Q) Otpt: Set(T, R) 1 Implements at each SCBS b B. 2 Initialize: T = R = φ, and U = B 3 while U φ do 4 Calclate W = max v N() ω v ω v ; 5 foreach U do 6 v arg max W(v); v N() 7 if W() > W(v) then 8 if ω v ω v > 0 then v v 9 T = T {}; U = U\{}; Q () v = ( 0 v ) N(); 10 v maxq ( max ) ; v N() U 11 R = R {v maxq }; U = U\{v maxq }; Q () v maxq = 0, N(v max Q); 12 else 13 R = R {}; U = U\{}; 14 v maxq ( max v N() U = 0, ( v ) N(); ) Q v () ; 15 T = T {v maxq }; U = U\{v maxq }; Q (v maxq) = 0, N(v max Q); 16 end 17 end 18 end 19 end 20 retrn Set(T, R); Parameter TABLE I: Simlation Parameters Carrier freqency Dplex Mode System Bandwidth BS Transmit Power BS Antenna Gain Vale 28 GHz TDD 2 GHz 30 dbm 10 dbi Pathloss Parameter (P L), α = 72.0 P L = α + 10β log 10 (d) [db], d in meters β = 2.92 Beam-width We plot the network throghpt of five schemes nder the increasing nmber of flows in Fig. 3. It indicates the total throghpt of the backhal network i.e., the average sm of the throghpt of all sccessfl flows. We observe the performance trend of or proposed algorithm; more the demanding flows, more the aggregate transmission of data, and ths higher the system throghpt. In Fig. 4, we plot the nmber of sccessfl flows nder 45

5 Fig. 3: System throghpt nder different traffic model static frame configration so it cannot accommodate flctating capacity reqirements. We can observe, the difference between TO-iBS and Exhastive Search is minor, both scheme almost accommodate similar nmber of demanding flows. The Exhastive Search iteratively select the best node from the network and add it to the T and R set. The speriority of or proposed scheme comes from two facts. First, it ses local information to acqire the knowledge of traffic information at each node and sbseqently makes the schedling decision. Secondly, the power reallocation proportional to the throghpt reqirements contribtes to the overall performance. different traffic types and loads with varying nmber of flows from 10 to 100. The end-to-end flow that sccessflly reached its destination node and also achieved its throghpt reqirement is conted as sccessfl flow. Under light load, the nmber of sccessfl flows keep increases with the traffic load. Under Poisson traffic, the Exhastive Search and TOiBS improves the sccess rate abot 10% compared with IPP traffic, nder light load. Fig. 4: No. of sccessfl flows nder different traffic model Fig. 5 demonstrates the brsty traffic and Poisson traffic in general experiences very similar average delay de to the node having traffic to transmit is more likely to have the transmission opportnity at the crrent sbframe. Therefore, the flow that arrives non-niformly, in IPP arrival case, easily misses the channel to transmit. Fig. 5: Transmission delay nder different traffic model In smmary, compared with TDMA and Exhastive Search, the proposed scheme, TO-iBS, has clear edge. TDMA has a V. CONCLUSION In this paper, we consider the problem of providing greater transmission/reception flexibility in the mmwave backhal network and design a dynamic frame reconfigration mechanism to overcome the static dplexing. We have formlated the selection of dplex schedles as an optimization problem and proposed an efficient algorithm to solve this by extensively considering the traffic load at each BS. Extensive simlation shows the speriority of or algorithm in achieving similar network throghpt and nmber of sccessfl flows comparing with optimal algorithm, nder fairly niform traffic. ACKNOWLEDGEMENT This work was financially spported by the Ministry of Science and Technology of Taiwan nder Grants MOST E , and sponsored by MediaTek Inc., Hsinch, Taiwan. REFERENCES [1] Z. Pi, F. Khan, An Introdction to Millimeter-Wave Mobile Broadband Systems, IEEE Commn. Mag., vol. 49, no. 6, pp , Jn [2] S. Singh, et al. Tractable Model for Rate in Self-Backhaled Millimeter Wave Celllar Networks, IEEE J. Sel. Areas Commn., vol. 33, no. 10, pp , Oct [3] S. Singh, R. Mdmbai, U. Madhow, Distribted Coordination with Deaf Neighbors: Efficient Medim Access for 60 GHz Mesh Networks, Proceedings IEEE INFOCOM, pp. 1 9, Mar [4] Volker Pali, Yi Li, Eiko Seidel, Dynamic TDD for LTE-A and 5G, Nomor Research GmbH, Sept [5] Chan-Y Tng, Chn-Yen Chen, and Hng-Y Wei, Next-Generation Directional mmwave MAC Time-Spatial Resorce Allocation, 11th Int l Conf. on Heterogeneos Networking for Qality, Reliability, Secrity and Robstness (Qshine 2015), Taipei, Taiwan, Ag [6] R. Ford, et al. Dynamic Time-domain Dplexing for Self-backhaled Millimeter Wave Celllar Networks, IEEE International Conference on Commnication Workshop (ICC), pp , Jne [7] J. Qiao, LX Cai, X. Shen, JW Mark, STDMA-based Schedling Algorithm for Concrrent Transmissions in Directional Millimeter Wave Networks, IEEE Int l Conf. on Comms., pp , Jne [8] C.Y. Chen, Y.Y. Chen, and H.Y. Wei, Mlti-Cell Interference Coordinated Schedling in mmwave 5G Celllar Systems, 8th Int l Conf. on Ubiqitos and Ftre Networks (ICUFN 2016), pp , Jl [9] Y. Zh, Y. Ni, J. Li, D. O. W, Y. Li, D. Jin, QoS-aware Schedling for Small Cell Millimeter Wave Mesh Backhal, IEEE International Conference on Commnications (ICC), pp. 1 6, May [10] Hng-Y Wei, Samrat Gangly, Raf Izmailov, and Zygmnt Haas, Interference-Aware IEEE WiMax Mesh Networks, The 61st IEEE Vehiclar Technology Conference (VTC Spring 05), May 2005 [11] J. Garcia-Rois, et al. On the Analysis of Schedling in Dynamic Dplex mlti-hop mmwave Celllar Systems, IEEE Transactions on Wireless Commnications, vol. 14, no. 11, pp , Nov