Interference Minimization in 5G Heterogeneous Networks
aa r X i v : . [ c s . I T ] J a n Noname manuscript No. (will be inserted by the editor)
Interference Minimization in 5G Heterogeneous Networks
Tao Han · Guoqiang Mao · Qiang Li · Lijun Wang · Jing Zhang
Accepted by Mobile Networks and Applications. The final publication is available at Springer viahttp://dx.doi.org/10.1007/s11036-014-0564-1.
Abstract
In this paper, we focus on one of the repre-sentative 5G network scenarios, namely multi-tier het-erogeneous cellular networks. User association is inves-tigated in order to reduce the down-link co-channel in-terference. Firstly, in order to analyze the multi-tierheterogeneous cellular networks where the base stationsin different tiers usually adopt different transmissionpowers, we propose a Transmission Power Normaliza-tion Model (TPNM), which is able to convert a multi-tier cellular network into a single-tier network, suchthat all base stations have the same normalized trans-mission power. Then using TPNM, the signal and in-terference received at any point in the complex multi-tier environment can be analyzed by considering thesame point in the equivalent single-tier cellular net-work model, thus significantly simplifying the analy-sis. On this basis, we propose a new user associationscheme in heterogeneous cellular networks, where thebase station that leads to the smallest interference toother co-channel mobile stations is chosen from a setof candidate base stations that satisfy the quality-of-service (QoS) constraint for an intended mobile station.Numerical results show that the proposed user associ-
Tao Han · Qiang Li (Corresponding author) · Jing ZhangSchool of Electronic Information and Communications, HuazhongUniversity of Science and Technology, ChinaE-mail: {hantao, qli_patrick, zhangjing}@hust.edu.cnGuoqiang MaoSchool of Computing and Communications, University of Tech-nology Sydney, AustraliaNational ICT Australia (NICTA), AustraliaE-mail: [email protected] WangDepartment of Information Science and Technology, Wenhua Col-lege, ChinaE-mail: [email protected] ation scheme is able to significantly reduce the down-link interference compared with existing schemes whilemaintaining a reasonably good QoS.
Keywords
Heterogeneous cellular networks · userassociation · performance analysis model · interferencemanagement A heterogeneous cellular network (HCN) usually con-sists of multiple tiers including a macrocell tier andprobably some small cell tiers, e.g., picocell tier, femto-cell tier and so on [1]. In general, there are three chan-nel allocation strategies among tiers, namely orthogo-nal deployment, co-channel deployment, and partiallyshared deployment [3]. In order to improve the spectralefficiency to match the ever growing demand for highdata rate nowadays and future, co-channel deploymentamong tiers and spatial frequency reuse are widely em-ployed in HCNs. In the small cell tier, since base sta-tions (BSs) are often deployed in an unplanned manner,it causes more serious co-channel interference in hetero-geneous networks than that in conventional single-tiercellular networks. In view of the severe co-channel in-terference under both intra-tier and inter-tier situations[12], interference management is very important in aHCN [7].User association, also called cell association or BSassociation, is one of the important approaches to per-forming interference management as well as to improv-ing the spectral efficiency and energy efficiency [13].Fooladivanda et al. proposed a unified static frameworkto study the interplay between user association and re-source allocation in HCNs [3]. Ghimire et al. formulated
Tao Han et al. a flow-based framework for the joint optimization of re-source allocation, transmission coordination, and userassociation in a heterogeneous network comprising ofa macro BS and a number of pico BSs and/or relaynodes [5], where the performance of different combina-tions of resource allocation schemes and transmissioncoordination mechanisms was characterized. Jin et al. proposed a marginal utility based user association al-gorithm to transform the combinatorial optimizationproblem into a network-wide utility maximization prob-lem [8]. Jo et al. developed a tractable framework forsignal-to-interference-plus-noise ratio (SINR) analysisin downlink HCNs with flexible cell association policies[6]. Madan et al. described new paradigms for the designand operation of HCNs, where cell splitting, cell rangeexpansion, semi-static resource negotiation on third-party backhaul connections, and fast dynamic interfer-ence management for quality-of-service (QoS) via over-the-air signaling were investigated [10].In HCNs, an active mobile station (MS) needs toassociate itself with a particular cell, which belongs toone of the tiers in a multi-tier network. Convention-ally, a MS is associated with the nearest BS or the BSthat provides the highest received SINR. However, theseMS association schemes do not consider the possible co-channel interference caused to other active MSs. Moti-vated by this, in this paper using stochastic geometrymethods [2], we propose a MS association scheme inmulti-tier networks that is able to significantly reducethe down-link co-channel interference while guarantee-ing a predefined QoS of mobile users in HCNs withopen-access small cell. Consider the interference in up-link is not always minimized when we minimize the in-terference in down-link by a user association scheme, wewill not investigate the issue on interference in up-linkin this paper. The contributions of this paper are:1. A Transmission Power Normalization Model (TPNM)for analyzing the performance of multi-tier HCNs isproposed, which significantly simplifies the analysisof the performance of multi-tier HCNs.2. Based on TPNM, a new user association scheme isproposed to minimize the down-link co-channel in-terference, which can be used in both conventionalsingle-tier cellular networks and multi-tier HCNs.3. Extensive simulations are conducted, the results demon-strate that the proposed scheme can significantly re-duce the down-link interference under the constraintthat predefined QoS requirements are satisfied.The rest of the paper is organized as follows. Section2 describes the system model. Section 3 introduces theproposed TPNM for performance analysis in multi-tierHCNs. Based on TPNM, we proceed to propose a new user association scheme to minimize the down-link co-channel interference in section 4. Section 5 shows thenumerical results for the performance of the proposeduser association scheme. In section 6, we conclude thepaper.
We consider a heterogeneous cellular multi-tier networkthat is composed of K -tier networks where K ∈ N withonly a single BS located winthin each cell of the multi-tier networks. The transmission powers at the BSs ofthe k -th tier network are assumed to be equal and de-noted as P k . We assume that the distribution of theBSs in the k -th tier network follows a homogeneousPoisson point process Φ BS k with intensity λ BS k . Assum-ing that the multiple cells of different tiers are over-laid in the same area geographically, then the distri-bution of the BSs in multi-tier HCNs is governed by aPoisson point process Φ BS = S Kk =1 Φ BS k with intensity λ BS = P Kk =1 λ BS k .Furthermore, we assume that the distribution ofactive MSs which are associated with the BSs in the k -th tier network follows a homogeneous Poisson pro-cess Φ MS k of intensity λ MS k . Thus the distribution of allMSs in multi-tier HCNs is also governed by a Poissonpoint process Φ MS = S Kk =1 Φ MS k with intensity λ MS = P Kk =1 λ MS k .This paper focuses on the down-links in multi-tierHCNs, where all BSs reuse the same frequency that isdivided into orthogonal channels. A BS allocates differ-ent orthogonal channels to the MSs in a cell. Under suchcircumstances, there is no intra-cell interference. How-ever, due to the frequency reuse across cells, there mayexist severe inter-cell co-channel interference in multi-tier HCNs if the same sub-channel is occupied in dif-ferent cells [11]. For example, given that the BSs assignthe channels randomly and independently, at a partic-ular time instant, only a fraction of the BSs, denotedby Poisson point process Φ N_BS of intensity λ N_BS , areusing a specific channel C n simultaneously to transmitto the corresponding MSs, denoted by Poisson pointprocess Φ N_MS of intensity λ N_MS = λ N_BS , where theBSs in Φ N_BS and the MSs in Φ N_MS are communica-tion pairs.Assuming BSs assign down-link channels to the MSsassociated with them randomly, then the MSs using thesame channel C n , i.e. Φ N_MS , can be considered to fol-low a homogeneous Poisson point process [15], which isthinned from point process Φ MS . Then we define Φ N_MS as an interfering set, in which MSs are interfered by the nterference Minimization in 5G Heterogeneous Networks 3
BSs that are transmitting to other MSs in the set be-cause they use the same channel C n .For ease of exposition, only path loss effect is con-sidered in the wireless channel models. Without loss ofgenerality, we consider a given BS x and a desired MS y . Then the desired signal power P xy received at y isexpressed as P xy = P x l ( x − y ) , (1)where P x denotes the transmission power of the BS and l ( · ) = k·k − α denotes the path loss in wireless channelswhere α is the path loss exponent.In this paper, we focus on the interference-limitedscenario. When a MS y is associated with a BS x , thesignal-to-interference ratio (SIR) at y is given as SIR ( x, y ) = P xy I y = P x l ( x − y ) P x i ∈ Φ N_BS \{ x } P x i l ( x i − y ) , (2)where I y denotes the interference received from the BSsin Φ N_BS except x i .In view of the severe co-channel interference, weconsider a user association scheme where the MS y ∈ Φ N_MS chooses a BS x ∈ Φ BS to associate with, andat the same time the interference from x to other MSs Φ N_MS \ { y } is minimized.In order to minimize the interference caused by thechosen BS x to other co-channel MSs, for ease of anal-ysis, we consider a MS z ∈ Φ N_MS \ { y } that receivesthe most severe interference I xz from BS x [9]. Thenthe minimization of the interference seen at z probablyimplies a minimization of the co-channel interference.On the other hand, to satisfy a reasonable QoS con-straint, it is assumed that the distance between the spe-cific MS y and the corresponding BS x ∈ Φ BS shouldbe no more than the distance between y and any BS ∀ x i ∈ Φ N_BS transmitting in the same channel. In otherwords, we intend to choose a suitable BS for y such thatthe co-channel interference caused to other MSs is min-imized, under the constraint the QoS of the MS y is sat-isfied. If in Φ BS there is no BS satisfies this constraint,then MS y will try to search another channel. (cid:37)(cid:54)(cid:3)(cid:76)(cid:81)(cid:3)(cid:55)(cid:76)(cid:72)(cid:85)(cid:3)(cid:20)(cid:37)(cid:54)(cid:3)(cid:76)(cid:81)(cid:3)(cid:55)(cid:76)(cid:72)(cid:85)(cid:3)(cid:21) (cid:37)(cid:54)(cid:3)(cid:76)(cid:81)(cid:3)(cid:55)(cid:76)(cid:72)(cid:85)(cid:3)(cid:22)(cid:48)(cid:54) '11 BS x BS x BS x BS x BS x MS y '21 BS x MS y '22 BS x '31 BS x Fig. 1
By TPNM, each tier is scaled by using the location of aspecific MS y as the scaling center. In order to analyze the multi-tier HCN, we pro-pose a TPNM in this paper, which is able to convert amulti-tier HCN to a virtual single-tier cellular networkby first scaling each tier according to its correspond-ing transmission power and path-loss effect, and thencombining the different tiers into a single-tier cellularnetwork, such that all BSs have the same normalizedtransmission power and the signal power and interfer-ence received at a specific MS from the BSs in the vir-tual single-tier cellular network are exactly the same asthose received from the BSs in the original multi-tiercellular network.As an example, consider a -tier HCN shown in Fig.1. There are BSs in various tiers including BS x in tier , BS x and x in tier , and BS x in tier , withdifferent transmission powers. MS y receives the desiredsignal from the associated BS and interference from theother BSs. For ease of analysis, we set the location ofthe MS y as the origin and scale each tier by usingdifferent factors such that virtual BSs x ′ , x ′ , x ′ and x ′ with the same normalized power are obtained,and the received signal/interference powers at y fromthese virtual BSs are exactly the same as those receivedfrom the original BSs, e.g., the power received at MS y from BS x before scaling is exactly the same as thatfrom virtual BS x ′ after scaling.In a K -tier HCN, the BSs in tier k, k ∈ { , , . . . , K } ,have transmission power P k and follow a Poisson pointprocess Φ BS k of intensity λ BS k . Without loss of general-ity, we consider a MS y located at origin o , then thereceived signal power P xy at MS y from a BS x ∈ Φ k isgiven as P xy = P k l ( x − o ) = P k l ( x ) = 1 · (cid:16) P − α k k x k (cid:17) − α = 1 · (cid:13)(cid:13)(cid:13) P − α k · x (cid:13)(cid:13)(cid:13) − α = 1 · l (cid:16) P − α k · x (cid:17) , (3)where is the normalized transmission power and l ( x − o ) is the path loss function from x to y . Tao Han et al.
From Eq. (3), it is observed that the signal powerreceived at y from x is equal to that received from avirtual BS x ′ = P − α k · x with transmission power andlocated at P − α k · x . Following this, the Poisson pointprocess Φ BS k can be scaled to a point process Φ BS ′ k = P − α k · Φ BS k of intensity λ BS ′ k = (cid:18) p − αk (cid:19) λ BS k = P α k λ BS k ,in which all the virtual BSs have the same transmissionpower and produce the same received signal power(or interference power) at the MS y as the original BSsin Φ BS k .From the above analysis, we scale the Poisson pointprocesses in all K tiers to normalize the BSs’ transmis-sion powers to , and then combine them into a singlePoisson point process Φ BS ′ = K [ k =1 P − α k · Φ BS k , (4)which is of intensity λ BS ′ = P Kk =1 P α k λ BS k .3.2 Received signal power based on TPNMIn this section, we give an example to demonstrate theadvantage of using TPNM by considering analysis ofcell association where a MS always associate with theBS delivering the highest received signal strength. Wefirst analyze the case without TPNM, then we presentthe analysis by using TPNM. Consider a specific receiving MS y , without loss of gen-erality, we place it at the origin o ∈ R . Then the BS x = arg max x i ∈ Φ BS P x i k x i k − α , (5)which can produce the highest received signal power at y , is selected to associate with y .Assuming there are K tiers in the network with cor-responding transmission power P k , k ∈ { , , . . . , K } ,we find the nearest BS x k from each tier Φ BS k to y , i.e., x k = arg min x kj ∈ Φ BS k k x kj k . (6)According to Slivnyak theorem [14], Φ BS k ∪ { o } has thesame properties as the Poisson point process Φ BS k , sothe distance R k , k x k − y k = k x k − o k = k x k k be-tween BS x k and MS y satisfies the following probabil-ity density function (PDF) f R k ( r k ) = 2 πλ BS k r k · e − λ BS k πr k . (7) Then the signal power received at MS y , i.e., P x k y = P k R − αk , has the following PDF f P xky ( p xky ) = f R k ( r k ) · (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:18) P k p x k y (cid:19) α ! ′ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 2 πλ BS k αp x k y (cid:18) P k P x k y (cid:19) α e − πλ BS k (cid:16) Pkpxky (cid:17) α . (8)Defining D x k y , P − α x k y = P − α k R k , we obtain thePDF of D x k y as follow: f D xky ( d x k y ) = f P xky ( P x k y ) · (cid:12)(cid:12)(cid:12)(cid:0) d − αx k y (cid:1) ′ (cid:12)(cid:12)(cid:12) = 2 πλ BS k P α k d x k y · e − πλ BS k P αk d xky , (9)and the corresponding cumulative distribution function(CDF) is derived as F D xky ( d x k y ) = ˆ d xky −∞ f D xky ( d x k y ) d d x k y = 1 − e − πλ BS k P αk d xky . (10)The BS that MS y is associated with should havethe largest P x k y , so it should have the smallest D x k y aswell. Denote the smallest D x k y by D xy , and the largest P x k y by P xy , we obtain P xy = max k ∈{ , ,...,K } P x k y , (11)and D xy = min k ∈{ , ,...,K } D x k y . (12)Since random variables D x k y , k ∈ { , , . . . , K } aremutually independent, then the CDF of D xy can bederived as F D xy ( d xy ) = 1 − K Y k =1 (cid:18) − (cid:18) − e − πλ BS k P αk d xky (cid:19)(cid:19) = 1 − e − πd xky P Kk =1 λ BS k P αk , (13)and the PDF of D xy can be derived as f D xy ( d xy ) = (2 π ) d xy K X k =1 λ BS k P α k · e − πd xky P Kk =1 λ BS k P αk . (14)Because D xy = P − α xy , we have F P xy ( P xy ) = F D xy (cid:16) P − α xy (cid:17) = 1 − e − πP − αxy P Kk =1 λ BS k P αk . (15) nterference Minimization in 5G Heterogeneous Networks 5 By using TPNM, we scale each tier with factor P − α k ,and then combine them to a virtual Poisson point pro-cess Φ BS ′ of intensity λ BS ′ = P Kk =1 P α k λ BS k . Denote thedistance between MS y and the nearest BS x ∈ Φ BS ′ by R , which has the following CDF F R ( r ) = 1 − e − λ BS ′ πr = 1 − e − P Kk =1 P αk λ BS k πr . (16)Because P xy = 1 · R − α and thus R = P − α xy , we have F P xy ( p xy ) = 1 − e − P Kk =1 P αk λ BS k π (cid:18) p − αxy (cid:19) = 1 − e − πp − αxy P Kk =1 λ BS k P αk . (17)Since Eq. (17) is of the same form as Eq. (15),TPNM can be used to analyze the received signal andinterference powers at an arbitrary MS with path losseffect, which significantly simplifies the derivations. y , by using TPNM, themulti-tier HCN can be transformed to a virtual single-tier cellular network Φ BS ′ = S Kk =1 P − α k · Φ BS k of inten-sity λ BS ′ = P Kk =1 P α k λ BS k . Then the average fraction ofusers that are served by tier k in open access is givenas ¯ N k = λ BS ′ k λ BS ′ = λ BS k P α k P Kk =1 λ BS k P α k . (18)Then in tier k , the nearest BS x k is selected asso-ciate with MS y . To evaluate the interference causedby x k to other co-channel MSs, we consider the nearestMS to x k other than y , i.e., z k = arg min z i ∈ Φ N_MS \{ y } k x k − z i k , (19)then the received interference at z k from x k is given as I x k z k = P k l ( R x k z k ) = P k k x k − z k k − α , (20)where the distance R x k z k = k x k − z k k between x k and z k follows the following CDF and PDF respectively: F R xkzk ( r x k z k ) = 1 − e − λ N_MS πr xkzk , (21) f R xkzk ( r x k z k ) = 2 πλ N_MS r x k z k · e − πλ N_MS r xkzk . (22)According to (18), we consider the probability that theBS x , which is serving MS y , belongs to the k -th tieralso as ¯ N k . Then the expectation of the interferencereceived at MS z from BS x can be derived as E ( I xz ) = K X k =1 ¯ N k ˆ ∞ f R xkzk ( r x k z k ) · P k l ( r x k z k ) d r x k z k = P Kk =1 λ BS k P α +1 k P Kk =1 λ BS k P α k · ˆ ∞ l ( r x k z k ) f R xkzk ( r x k z k ) d r x k z k . (23)4.2 Interference minimized user association schemeIn the proposed user association scheme, consider a spe-cific MS, the BS x opt that generates the largest receivedSIR at this MS is selected under the constraint on thepredefined QoS.Consider an arbitrary MS y , we first transform themulti-tier HCN to a virtual single-tier cellular network Φ BS ′ = S Kk =1 P − α k · Φ BS k of intensity λ BS ′ = P Kk =1 P α k λ BS k by TPNM. Then we have the interfering set that trans-mit simultaneously in channel C n as Φ N_BS ′ ⊂ Φ BS ′ ofintensity λ N_BS ′ , which is transformed from Φ N_BS byTPNM as well.To satisfy the QoS constraint, not all BSs in Φ BS ′ can be chosen to communicate with the MS y , we denotethe subset of BSs that are allowed to communicate with y by T y ⊂ Φ BS ′ . Then the distance between y and eachBS in T y is no more than the distance between y andany other transmitting-in-the-same-channel BS, i.e., T y = (cid:8) x ′ i : k x ′ i − y k ≤ (cid:13)(cid:13) x ′ j − y (cid:13)(cid:13) , x ′ i ∈ Φ BS ′ , ∀ x ′ j ∈ Φ N_BS ′ (cid:9) . (24)We denote the number of BSs in subset T y by a ran-dom variable N T y . N T y is the number of points from Φ BS ′ in the void ball V = b (cid:0) y, R N_BS ′ (cid:1) of Φ N_BS ′ ,where R N_BS ′ is the void distance of Φ N_BS ′ , whosePDF is [14] f R N_BS ′ (cid:0) r N_BS ′ (cid:1) = 2 πλ N_BS ′ r N_BS ′ · e − πλ N_BS ′ ( r N_BS ′ ) . (25)Then an estimated value ¯ N T y of N T y is given as ¯ N T y , E (cid:0) N T y (cid:1) = λ BS ′ · A ( V )= λ BS ′ ˆ ∞ π (cid:0) r N_BS ′ (cid:1) · f R N_BS ′ (cid:0) r N_BS ′ (cid:1) d r N_BS ′ = λ BS ′ λ N_BS ′ , (26) Tao Han et al. where A ( V ) denotes the area of V .We assume that the proportion of the transmittingBSs in each tier is the same. Then according to (26),we obtain ¯ N T y = λ BS ′ λ N_BS ′ = λ BS ′ λ N_BS · λ BS ′ λ BS = λ BS λ N_BS . (27)According to (18), the average fraction of users servedby tier k in open access can thus be derived as ¯ N k = λ BS k P α k P Kk =1 λ BS k P α k . (28)Then the BS that satisfies x opt = arg min x ′ i ∈ T y max z ′ j ∈ Φ N_MS ′ \{ y } · (cid:13)(cid:13) x ′ i − z ′ j (cid:13)(cid:13) − α (29)is selected to associate with y . For the co-channel MSthat receives the largest interference from x opt , i.e., z opt ,we have z opt = arg min z ′ i ∈ Φ N_MS ′ \{ y } k x opt − z ′ i k . (30)Denote the distance between x opt and z opt by R opt = k x opt − z opt k , then the CDF and PDF of R opt are de-rived as F R opt ( r opt ) = (cid:16) − e − λ N_MS πr (cid:17) ¯ N Ty , (31) f R opt ( r opt ) = 2 π ¯ N T y λ N_MS r opt · e − λ N_MS πr · (cid:16) − e − λ N_MS πr (cid:17) ¯ N Ty − , (32)respectively.And then the expectation of the interference re-ceived at MS z opt from BS x opt can be similarly derivedas E (cid:0) I xz opt (cid:1) = ˆ ∞ f R opt ( r opt ) · K X k =1 ¯ N k P k · l ( r opt ) d r opt = P Kk =1 λ k P α +1 k P Kk =1 λ k P α k · ˆ ∞ f R opt ( r opt ) · l ( r opt ) d r opt . (33) N o r m a li z ed i n t e r f e r en c e conventional, α =2.5conventional, α =3conventional, α =4proposed, α =2.5proposed, α =3proposed, α =4 Fig. 2
Interference in a 3-tier cellular network where P k = { , , . } , λ BS k = { . , . , } and λ BS /λ N_BS = 3 . N Ty Normalized intensity of receiving mobile stations N o r m a li z ed i n t e r f e r en c e proposed, N Ty =1proposed, N Ty =2proposed, N Ty =3proposed, N Ty =5 Fig. 3
Interference in a 3-tier cellular network where P k = { , , . } , λ BS k = { . , . . } and α = 4 . In this section, we present the analytical results of theproposed BS association scheme and compare it to theconventional scheme that is subject to severe co-channelinterference. To avoid the singularity of path loss func-tion l ( · ) in (23) and (33), we use l ( r ) = (1 + r α ) − [4]instead of l ( r ) = r − α in deriving the analytical results.In Fig. 2, the interference in a 3-tier cellular networkis demonstrated and compared between the proposedscheme and the conventional scheme. The BS transmis-sion powers in tier 1, 2 and 3 are , and . respec-tively, and the intensities of BSs in tier 1, 2 and 3 are . , . and respectively. The result indicates thatthe proposed scheme is effective to reduce the interfer-ence in multi-tier cellular networks. Fig. 2 also showsthat for both the proposed interference minimized userassociation scheme and the conventional scheme, withmore severe path loss effect, the interference caused toother co-channel MSs is reduced. nterference Minimization in 5G Heterogeneous Networks 7 In Fig. 3, we show how interference is affected byvarious values of N T y . A greater N T y means that thereare more candidate BSs to chose from such that it ismore probably to select a BS which leads to less in-terference to other co-channel receiving MSs. Whereaswhen N T y → , the proposed scheme degenerates to theconventional scheme. In this paper, we first propose a transmission power nor-malization analysis model, which significantly simplifiesthe analysis of the received signal and interference, thusSIR, in multi-tier HCNs. Then we propose an interfer-ence minimized user association scheme, which can beapplied in both single-tier and multi-tier HCNs. Usingthe proposed TPNM, we proceed to analyze the inter-ference in multi-tier HCNs. Results demonstrate thatthe proposed scheme significantly reduces the down-linkinterference in multi-tier HCNs, meanwhile the con-straint on the QoS of users is satisfied.
Acknowledgements
The authors would like to acknowledgethe support from the International Science & Technology Co-operation Program of China (Grant No. 2014DFA11640, 0903and 2012DFG12250), the National Natural Science Foundationof China (NSFC) (Grant No. 61471180, 61271224 and 61301128),NSFC Major International Joint Research Project (Grant No.61210002), the Hubei Provincial Science and Technology Depart-ment (Grant No. 2013CFB188 and 2013BHE005), the Funda-mental Research Funds for the Central Universities (Grant No.2013ZZGH009, 2013QN136, 2014TS100 and 2014QN155), the Spe-cial Research Fund for the Doctoral Program of Higher Educa-tion (Grant No. 20130142120044), and EU FP7-PEOPLE-IRSES(Contract/Grant No. 247083, 318992 and 610524). This researchis also supported by Australian Research Council Discovery projectsDP110100538 and DP120102030.