Sei sulla pagina 1di 4

IEEE COMMUNICATIONS LETTERS, VOL. 18, NO.

10, OCTOBER 2014 1803

Fast Pairing of Device-to-Device Link Underlay


for Spectrum Sharing With Cellular Users
Li Wang, Member, IEEE and Huaqing Wu

Abstract—Device-to-device (D2D) communications has attracted This work develops a new pairing approach to achieve high
substantial research attention recently owing to its simplicity and sum performance while reducing its complexity simultane-
its potential to improve spectrum and energy efficiency within the ously. We study a general optimization problem of rate-utility
existing cellular infrastructure. This work presents a low complex-
maximization by considering the various utility functions for
ity method for matching D2D links with cellular user equipments
(CUEs) to form partners for spectrum sharing. The overall ob- CUEs and D2D links simultaneously. Thus, we focus on how
jective is to maximize well defined performance metrics of all D2D to match D2D links and CUEs into partners so as to achieve
links and CUEs after successful pairing under power and QoS con- optimized pairing for maximizing the overall performance met-
straints. We begin by identifying a set of D2D link candidates for ric such as sum data rate of all users. In particular, to reduce
each CUE uplink for a requisite SINR level to reduce the number the computation complexity, we divide our pairing selection
of potential pairing searches. More importantly, we present a sim- approach into two simple steps. We will first narrow a candidate
ple pairing algorithm to reduce the computational cost well below
the well known Hungarian (Kuhn-Munkres) algorithm used in the
D2D link set for each CUE, while assuring the required quality
literature for the pairing problem. Our new algorithm exhibits low of service (QoS) to both sides in terms of minimum SINR.
complexity and is effective in improving the sum rate of all spec- Furthermore, we will investigate how to optimize pairing selec-
trum sharing CUEs and D2D links with modest performance loss. tion among cooperative CUEs to the D2D links from the pre-
Index Terms—Device-to-Device (D2D) communications, spec-
established candidate set while, at the same time, optimizing
trum sharing, power control, sum rate, bi-partite graph. the power for each respective pair of CUE and the D2D link.
We organize rest of this paper as follows. Section II describes
I. I NTRODUCTION the D2D link and CUE pairing problem with simultaneous
power optimization in D2D underlay. We briefly depict the selec-

D EVICE-TO-DEVICE (D2D) communications has re-


cently emerged as a candidate technology in cellular
networks for improving the spectrum efficiency and off-loading
tion of D2D candidate links for each of the CUEs willing to share
its spectrum before outlining the existing pairing approach
based on the Hungarian (Kuhn-Munkres: KM) algorithm [5].
cellular traffics [1]. D2D can serve as an underlay within Section III presents our new algorithm that can substantially
the cellular network under cellular control. However, D2D lower the computational complexity. We then present our sim-
underlay must limit its interference to its partner cellular user ulation test results in Section IV before the conclusion section.
equipment (CUE). Thus, the issue of pairing D2D links and
CUEs for low-interference spectrum sharing becomes critical. II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
Well optimized pairing can improve total throughput, reduce
A. Notations and Assumptions
mutual interference, and even assure better secrecy for both
D2D links and CUEs. This work presents a low complexity We investigate spectrum sharing for cellular D2D underlay
pairing algorithm for spectrum sharing between D2D links and by focusing on a single cell, subject to their respective min-
CUEs with strong performance. imum QoS requirement, where a number of (M ) D2D links
Pervious works on pairing D2D and CUE either focus on coexist with N CUEs that occupy N orthogonal channels.
one side to make the selection, or attempt exhaustive search We limit our scope to the sharing of cellular uplink (UL)
by using classic means such as the Hungarian (Kuhn-Munkres) resources by D2D links. Here, we use C = {1, . . . , N }, and
algorithm [2]–[4]. Such searches, though optimum, are compu- D = {1, . . . , M } to denote the index sets of CUEs and D2D
tationally costly. The computational complexity grows signifi- links respectively. Note that we assume that the nodes in our
cantly if the CUE and D2D powers must be jointly optimized scenario are generally stationary or moving at moderate speed
while pairing, as shown in [3]. to allow channel state information (CSI) acquisition. Here we
apply (distance based) large scale path loss channel model for
Manuscript received May 23, 2014; revised July 23, 2014; accepted August
both CUEs and D2D users [6].
6, 2014. Date of publication August 28, 2014; date of current version October 8, Our basic assumptions are:
2014. This work is supported by National Natural Science Foundation of China • Every active UE uses one unit of resource blocks.
(No. 61201150 and 61372117), State Major Science and Technology Special • Each D2D link wants to share at most one unit of resource
Projects of China (No. 2012ZX03004001),Science Technology Innovation
Foundation for Young Teachers in BUPT (No. 2013RC0202), and Beijing block belonging to a particular CUE.
Higher Education Young Elite Teacher Project (No. YETP0442). The associate • Base Station (BS) in charge of pairing D2D link and CUE
editor coordinating the review of this paper and approving it for publication was for spectrum sharing has the following information:
I.-R. Chen. 2
The authors are with Beijing Key Laboratory of Work Safety Intelligent — Background noise level σN .
Monitoring, School of Electronic Engineering, Beijing University of Posts and — Uplink gain between CUE i and the BS gi,B and
Telecommunications, Beijing 100876, China (e-mail: liwang@bupt.edu.cn). channel gain of D2D link j, denoted as gj .
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. — Channel gains of interference links, from the transmit-
Digital Object Identifier 10.1109/LCOMM.2014.2351400 ter of D2D link j to the BS, hj,B .

1089-7798 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1804 IEEE COMMUNICATIONS LETTERS, VOL. 18, NO. 10, OCTOBER 2014

of N CUE spectral bands and M orthogonal D2D spectral


bands. The bottom side consists of N empty partners (one for
each CUE) and M active D2D links. Thus, if no D2D link can
share CUE−i’s spectrum while simultaneously satisfying the
constraints, then CUE−i will have only one link with the i−th
empty partner node with an edge weight of
wi,i = max fi , i = 1, . . . , N. (3a)
Pic ≤Pi,max
c ,Pjd =0

Similarly, if a D2D link cannot find any CUE to share spectrum,


Fig. 1. Pairing bi-partite graph.
it will be assigned an edge weight of
— Transmit powers of CUE i(Pic ) and D2D j(Pjd ).
— Channel gain hi,j from CUE i to D2D link j. wN +j,N +j = 0, j = 1, . . . , M (3b)
by connecting with its orthogonal band. All other edge weights
B. Pairing for Spectrum Sharing are defined by
First, we define a maximized information capacity function wi,N +j = max (fi + gj ), i = 1, . . . , N,
P c ≤P c ,P d ≤P d
by considering the achieved data rate for both CUEs and D2D i
c
i,max j
c
j,max
f ≥f , g d ≥g d
links simultaneously, after pairing the CUEs and D2D links for i i,min j j,min

spectrum sharing. We define an indicator variable j = 1, . . . , M. (3c)



1, D2D link-j shares CUE-i channel.
qi,j = (1) III. A N EW PAIRING S TRATEGY
0, otherwise.
Since it is hard to calculate the overall sum rate in one step
We assumed that at most one D2D link  can share at most one jointly, we present an optimization problem that consists of
CUE channel such that i qi,j ≤ 1 and j qi,j ≤ 1. Note that two steps and utilizes a “mutual selection” principle. Step 1
qi,j = 1 implies mutual interference. We can first use fi and gj finds, for each active CUE−i, a set of candidate D2D links
to denote QoS objective functions of CUE−i and D2D link−j, D(i) that can share the spectrum with CUE−i without severely
respectively, that depend on Pic , Pjd , and qi,j . Both CUE and impacting the rate of CUE−i. After narrowing the number
D2D link should have a minimum QoS requirement denoted by of potential pairing options, step 2 presents a low complexity
fic ≥ fi,min
c
and gjd ≥ gj,mind
. Hence our goal is to maximize the pairing algorithm.
sum objective function under power constraints Pic ≤ Pi,max c

and Pjd ≤ Pj,max


d
. As a result, we aim to optimize the following A. D2D Link Candidacy
general maximization problem:
⎛ ⎞ Our method first establishes an admissible D2D link set for
  each CUE, and the CUE to be shared resource as reuse partner.
max ⎝fi + qi,j gj ⎠ (2a) We use D(i) to indicate the admissible candidate D2D set for
{qi,j ,Pic ,Pjd } i j CUE i, representing all the D2D links that have edge connec-
fic ≥ fi,min
c
, gjd ≥ gj,min
d
(2b) tions to CUE−i resource, whereas C (j) denotes the available
Pi ≤ Pi,max , Pj ≤ Pj,max
c c d d
. (2c) CUE resource set for D2D j, representing all CUEs that have
connecting edges to D2D link−j.
The objective for the coordinating BS is to find the best pairing Let ξic and ξjd denote the SINR of CUE i and that of D2D
variable qi,j and the respective power Pic and Pjd to maximize link-j. Instead of presenting a highly complex and nearly
the sum metric for the cell. Here, we can take the sum of CUE intractable combinatorial problem with power optimization,
rate and D2D link rate as the objective. Such problem is non- our approach is to identify a set of D2D links that can coexist
convex combinatorial. The solution requires the optimization of with the active CUE uplinks by simultaneously satisfy a
two constrained variables Pic and Pjd for each possible pairing minimum SINR (rate) requirement ξjd of the D2D link without
of CUE and D2D. It naturally extends to other performance causing strong interference to any CUE uplinks in terms of
metrics such as secrecy rate [7]. SINR, ξic , by guaranteeing their minimum SINR ξi,minc
. Note
that we can obtain ξic ,
C. QoS Constrained Bi-Partite Graph ⎛ ⎞−1

Recall that each D2D user has one of the two modes: D2D ξic = Pic gi,B ⎝ 2 ⎠
qi,j Pjd hj,B + σN , ∀i ∈ C. (4)
underlay (by sharing CUE band) and dedicated mode (with or- j∈D
thogonal channel). Because of the QoS requirement and power
constraints, pairing is not always feasible and some CUEs can Similarly, the j-th D2D link SINR can be expressed as
not share their spectra whereas some D2D links have to be
−1

assigned orthogonal bands (with 0 spectral gain), by using bi- ξjd = Pjd gj qi,j Pic hi,j + σN
2
.
partite graph as shown in Fig. 1. This means that there does i∈C
not need to be an edge linking one CUE spectrum with one
D2D link. Hence, different from [8], we construct a bi-partite We now need to determine which of the D2D links can
graph with N + M nodes on both sides. The top side consists take part in sharing the uplink spectrum resource blocks of the
WANG AND WU: DEVICE-TO-DEVICE LINK UNDERLAY FOR SPECTRUM SHARING WITH CELLULAR USERS 1805

CUEs. Thus, one objective is for the BS to find qi,j . For a large pairing search by leveraging the KM algorithm. However, KM
number of CUEs and D2D links, the optimized pairing problem algorithm incurs high complexity, particularly for large number
becomes a large combinatorial problem. of potential partners. Thus, to reduce complexity, we also
Different from the scheme illustrated in [3], our goal is to propose a new simple suboptimum alternative: IPPO (Inverse
select admissible candidate D2D set, D(i) , for each CUE i first. Popularity Pairing Order) algorithm aimed at speeding up the
However, we can borrow the idea of picking candidate CUE pairing process without significant performance loss, as sum-
set for each D2D presented in [3]. Since the SINR is reflective marized in Algorithm 1.
of the Shannon rate, here we will take it as the basic mutual
constraint to CUE and D2D to narrow the candidate set. Thus,
we define this candidate set as

D(i) = j ∈ D : ξjd Pic , Pjd ≥ ξj,min
d
, (5)

subject to

ξic ≥ ξi,min
c
, Pjd ≤ Pj,max
d
, Pic ≤ Pi,max
c
, ∀i ∈ C, ∀j ∈ D,
c d
where Pi,max and Pj,max denote the power limit of CUE and
d
D2D link respectively, and ξj,min denotes the minimum re-
quired SINR of D2D link-j. Finding this candidate set requires
the BS to check each of the potential D2D links interested in
spectrum sharing.
Once the selection of the candidate set D(i) is identified for
an active CUE−i, we have a set of candidate D2D links that
can coexist with their cellular uplinks (in terms of SINR) while
meeting their own rate requirement.

B. UL Resource Matching and Power Allocation


For each CUE i, we have established a candidate D2D link
set D(i) in the previous section. At most one D2D link in Aiming to match more spectrum sharing partners of CUE
set D(i) may share CUE−i spectrum that can meet the rate and D2D link, IPPO starts with the D2D link with the fewest
requirements for itself without hurting CUE simultaneously, edges (least popular) and finds its best match with the largest
while optimizing the transmit powers for both. sum rate through D2D and CUE power optimization. We then
Let Ric and Rjd denote the Shannon rate of CUE−i and move down the list with the second D2D link with the next
D2D link−j, respectively. Without loss of generality, we can fewest number of edges and so on. Because we always go to the
consider the special case of fi = Ric , and gj = Rjd . Note that remaining least popular D2D link, this algorithm provides bet-
the achievable data rate of CUE i can be expressed as ter opportunity for overall matching success without iteration
⎡ ⎤
(or increasing complexity). This also provides certain fairness
⎢ Pic gi,B ⎥ assurance for D2D nodes with fewer connections. Note that, we
Ric = log2 ⎣1 +  2 ⎦
. (6)
qi,j Pjd hj,B + σN can define different edge weights for different QoS objective
j∈D(i) functions as shown in Section II.
Without substantial performance loss, our focus is on the
Similarly, define a CUE subset for the j−th D2D link reduction of computation complexity by IPPO over the tradi-
  tional KM. Specifically, we examine two different parts of the
C (j) = i : i ∈ C, such that j ∈ D(i) . computation time: pairing only (P O) and the total time from
both pairing and power optimization (P P O). As shown in [9],
Sharing CUE resources, the j−th D2D link achieves rate the P O complexity of KM is O((N + M )3 ), whereas that of
⎛ ⎞ our proposed IPPO is only O(N + M ). On the other hand,
⎜ Pjd gj ⎟ the complexity of power optimization for KM and IPPO are
Rjd = log2 ⎝1 +  2 ⎠. (7) O((N + M )2 ), and O((N + M )) respectively.
qi,j Pic hi,j + σN
i∈C (j)

IV. N UMERICAL R ESULTS AND A NALYSIS


C. Sum Rate Maximization and A Simplified Solution
In our simulation test, we consider the CUEs and D2D links
Consider maximization of sum rate of all CUE-resources:
 within a cell of radius R = 300 m and σN 2
= −40 dBm. The

max Ci,j , Ci,j = Ric + qi,j Rjd . (8) distribution of D2D user equipments (DUE) follows Poisson
qi,j ,Pic ,Pjd
i∈C (j) ,j∈D(i)
Point Process with density μ = 1 nodes/m2 . Select 100 nodes
randomly as the DUEs, from which we pick the closest M
Traditionally, as shown in [8] and [9], one optimum solution to pairs of DUEs to form D2D links. Simultaneously, we assume
the maximum total throughput problem is through exhaustive that a variable number of N conventional CUEs are uniformly
1806 IEEE COMMUNICATIONS LETTERS, VOL. 18, NO. 10, OCTOBER 2014

Fig. 2. Sum rate with different number of CUEs. Fig. 4. Computation time with different number of CUEs.

c d
Fixing Pi,max = 24 dBm and Pj,max = 21 dBm, Fig. 4
shows that, as CUE number N grows, the complexity of IPPO
remains nearly unchanged whereas the KM computation time
grows to approximately 3 orders of magnitude higher when
N reaches 200. Since LTE scheduling is for each (1-ms)
subframes, computation time of several seconds leads to very
large scheduling latency. Our results clearly demonstrate the
computation advantage of our newly proposed algorithm with-
out substantial performance loss.

V. C ONCLUSION
We consider the resource sharing problem of D2D underlay
within cellular networks under transmit power and minimum
QoS constraint. We present an augmented bi-partite graph to
Fig. 3. Sum rate with different number of D2D links (N = 30).
capture the constrained sum rate maximization problem for
sharing CUE resource with D2D links. We propose a novel
distributed within the cell. Based on this scenario, we compare algorithm to reduce the computation complexity without sac-
IPPO with the well known KM, IPPO+KM, as well as the rificing much performance over the traditional KM algorithm.
“random” pairing result in terms of achieved sum data rate Our proposed formulation and algorithm also apply to other
and computation time. Notice that, we also consider a new relevant performance metrics such as secrecy rate.
method that uses the KM algorithm to refine the results ob-
tained from IPPO. In this method labelled as IPPO+KM, after
R EFERENCES
applying IPPO to generate a subset of bipartite graph edges, we
add another KM refinement step to achieve performance gain [1] C. Xu et al., “Efficiency resource allocation for device-to-device underlay
communication systems: A reverse iterative combinatiorial auction based
over IPPO. However, since no additional power optimization approach,” IEEE J. Sel. Areas Commun., vol. 31, no. 9, pp. 348–358,
is needed to compute new edges, this IPPO+KM refinement Sep. 2013.
requires very little additional complexity, unlike the full-fledged [2] C. Yu, K. Doppler, C. B. Ribeiro, and O. Tirkkonen, “Resource sharing
optimization for device-to-device communication underlaying cellular net-
KM which must compute all edges through power optimization. works,” IEEE Trans. Wireless Commun., vol. 10, no. 8, pp. 2752–2763,
c d
Figs. 2 and 3 illustrate the effects of Pi,max , Pj,max , as well Aug. 2011.
as the number of CUEs and D2D links on the achieved sum rate. [3] D. Feng et al., “Device-to-device communications underlaying cellu-
Fig. 2 shows that the sum rate grows with increasing number of lar networks,” IEEE Trans. Commun., vol. 61, no. 8, pp. 3541–3551,
Aug. 2013.
CUEs because D2D links would have more choices of potential [4] Y. Cheng, Y. Gu, and X. Lin, “Power and channel allocation for device-
spectrum resources to share. Similarly, when the number of to-device enabled cellular networks,” J. Comput. Inf. Syst., vol. 10, no. 2,
CUEs is fixed, more D2D links make it possible for better pp. 463–472, 2014.
[5] J. Edmonds and R. M. Karp, “Theoretical improvements in algorithmic
pairing with CUE resources, thereby leading to larger sum rate, efficiency for network flow problems,” J. Assoc. Comput. Mach., vol. 19,
as Fig. 3 shows. Figs. 2 and 3 also show that larger transmit no. 2, pp. 32–38, Mar. 1972.
power limits of CUEs and DUEs lead to higher sum rates. [6] G. L. Stüber, Principles of Mobile Communications, 3rd ed. New York,
c
Larger Pi,max d
and Pj,max allow each pair to reach higher rate NY, USA: Springer-Verlag, 2012.
[7] A. D. Wyner, “The wire-tap channel,” Bell Syst. Tech. J., vol. 54, no. 8,
and hence larger sum data rates. Our IPPO algorithm only ex- pp. 1355–1387, Oct. 1975.
hibits modest performance loss against the high complexity KM [8] T. Han, R. Yin, Y. Xu, and G. Yu, “Uplink channel reusing selection
algorithm. IPPO+KM ranks second on the performance when optimization for device-to-device communication underlaying cellular net-
works,” in Proc. IEEE 23rd Int. Symp. PIMRC, Sep. 2012, pp. 559–564.
compared with KM and IPPO, which demonstrates more on [9] J. Han, Q. Cui, C. Yang, and X. Tao, “Bipartite matching approach to opti-
the tradeoff of computation complexity and performance. All of mal resource allocation in device to device underlaying cellular network,”
them achieve significantly higher rates than “Random” pairing. Electron. Lett., vol. 50, no. 3, pp. 212–214, Jan. 30, 2014.

Potrebbero piacerti anche