You are on page 1of 13

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 1

QoS-Aware Resource Allocation for


Device-to-Device Communications With Channel
Uncertainty
Daquan Feng, Lu Lu, Yi Yuan-Wu, Geoffrey Ye Li Fellow, IEEE, Gang Feng, Senior Member, IEEE, and
Shaoqian Li, Senior Member, IEEE

EVICE-to-device (D2D) communications enable two


Abstract—In device-to-device (D2D) communications, channel
state information (CSI) is exploited to manage the interference
between D2D users and regular cellular users (CUs) and improve
D users in proximity to transmit signal directly without go-
ing through the base-station (BS) and have attracted increasing
system performance. However, obtaining the accurate CSI is
usually difficult and causes high overhead, especially when the attentions from both industrial and academic communities [1]–
links are not connected to the base station (BS), such as the links [5]. In a recent report by the Third Generation Partnership
between regular CUs and D2D receivers (CU-D links). In this Project (3GPP), D2D communications have been considered
paper, we investigate the signaling overhead and performance as a promising technique in Long Term Evolution Advanced
tradeoff in D2D communications with channel uncertainty. To (LTE-Advanced) to improve local service flexibility when
limit interference to regular CUs, we only allow the resource of
a CU to be reused by at most one D2D pair. We also assume that users are within the network coverage and to provide emergent
only partial CSI of the CU-D links are available at BS and develop public safety services when users are outside the network
two different strategies to deal with the channel uncertainty: coverage [6].
probabilistic and partial feedback schemes. We first derive a D2D communications can significantly increase system
probability-based resource allocation scheme by utilizing channel spectral-efficiency (SE) and device energy-efficiency (EE) due
statistical characteristics to maximize the overall throughput of
the CUs and admissible D2D pairs while guaranteeing their to the potential proximity gain, reuse gain, and hop gain [7].
QoS in terms of signal-to-interference-plus-noise-ratio (SINR) and To obtain the potential benefits of D2D communications, many
outage probability, respectively. Then, we propose an efficient works have been done [8]–[26]. In particular, power control
feedback scheme to reduce the overhead of CSI feedback while [8], [9], scheduling [10]–[12], interference-avoiding multiple-
providing near optimal performance. In addition, we propose input and multiple-output (MIMO) transmission [13], [14] and
a combined scheme to take advantages of both probabilistic
and partial feedback schemes. It is shown by simulation that cross-layer optimization [15]–[19] have been investigated to
there exists an optimal threshold of the outage probability improve the performance of D2D communications. Moreover,
for probabilistic scheme while the partial feedback scheme is proper D2D mode switching has been studied in [20]–[24]
robust to the channel models. Furthermore, the combined scheme to fully exploit the new freedom of D2D communications.
outperforms the probabilistic and the partial feedback schemes In addition, statistical geometry approaches to analyze system
in terms of overall throughput.
performance of D2D communication have been introduced in
Index Terms—Device-to-Device (D2D) communications, re- [25], [26]. A comprehensive survey on D2D communications
source allocation, outage probability, signaling overhead, partial can be found in [27].
feedback, fading channel.
The exiting works have shown great potential gains of D2D
communications when complete channel state information
I. I NTRODUCTION (CSI) is exploited. However, most of them focus on the
network performance in terms of overall network throughput,
Copyright (c) 2015 IEEE. Personal use of this material is permitted. network coverage and D2D access rate. The signal overhead to
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. obtain the accurate CSI as well as the tradeoff between signal
Manuscript received February 1, 2015; revised August 9, 2015; accepted overhead and system performance have not been considered.
September 10, 2015. This paper was presented in part at the IEEE PIMRC In fact, obtaining the complete CSI is usually difficult and
2013, Lindon, UK and IEEE GLOBECOM 2013, Atlanta, GA, USA. This
work was supported in part by the National Science Foundation (NSF) under requires high overhead, especially when the links are not
Grants 1247545 and 1405116, the National Basic Research Program of China connected to BS, e.g., the interference links between regular
under Grant 2012CB316003, the National High-tech R&D Program of China cellular users (CUs) and D2D receivers (CU-D link), as shown
under Grant 2014AA01A707, and the NSFC under Grant 61471089. The
associate editor coordinating the review of this paper and approving it for in Fig.1. For the links are not connected to the BS, the
publication was Prof. Riku Jäntti. traditional training and channel estimation methods are not
D. Feng is with the State Radio Monitoring Center (SRMC), Beijing, China. applicable and feedback information is indispensable. Thus,
D. Feng is the corresponding author (e-mail: fdquan@gmail.com).
L. Lu, G. Y. are with the School of ECE, Georgia Institute of Technology, it is more reasonable to assume that the BS only has partial
Atlanta, GA, USA. CSI of the CU-D links including the distance-based pathloss
Y. Yuan-Wu is with the Orange Lab Network, Department of Wireless of the CU-D links that can be evaluated by users’ location
Technology Evolution, Paris, France.
G. Feng, S. Li are with the National Key Lab on Communications, UESTC, information, and the shadowing and fading of these links are
Chengdu, China. unknown at the BS.

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 2

D2D Rx1
of probabilistic scheme or the partial feedback scheme in terms
of the overall network throughput and outage probability of
h1, i D2D Tx1
1
g1( i1 ) D2D users.
CU i The rest of the paper is organized as follows. In Section
1
Base station (BS)
(i )
h1,B1 II, we describe the system model and formulate optimal D2D

. . .
g 1,B resource allocation problem in fading channels. Then, proba-
g N,B
CU1 Tx :transmitter bilistic resource allocation scheme, partial feedback scheme,
CUN and combined resource allocation scheme, are developed in
(i )
hM,MB
Rx :receiver Sections III, IV and V, respectively. In Section VI, we present
the simulation results. Finally, conclusions are drawn in Sec-
Desired signal
tion VII.
D2D TxM
hM , iM CU i
M

D2D RxM II. S YSTEM MODEL AND P ROBLEM F ORMULATION


g M( iM ) Interference
signal
As in [15], we consider D2D pairs sharing the uplink
Fig. 1: D2D communications underlying uplink cellular net- resource with regular CUs. We assume that D2D pairs can
work. be accessed only by reusing the resource of the active CUs
and the BS can adjust the transmit powers of the CUs and
The partial CSI assumption for D2D underlying networks D2D pairs to satisfy the QoS requirements. Without loss of
has been considered in [28], [29]. In [28], the number of generality, we also assume that the uplink spectrum is divided
D2D pairs that can be accessed is optimized based on a into N non-overlap sub-channels equally and each sub-channel
certain outage probability of the CU at the BS to maximize is randomly assigned to one active CU. In the following, we
the system throughput. In [29], outage performance of D2D denote C = {1, ..., N } and D = {1, ..., M } as the index sets
assisted three-stage transmission is analyzed. However, in [28], of active CUs and D2D pairs, respectively, and call the sub-
the power coordination between regular CU and D2D pairs channel occupied by CU i as channel i.
has not been considered while in [29], the time-division based
D2D cooperation transmission design is mostly concerned.
A. Channel Model
In this paper, we investigate resource allocation and analyze
the tradeoff between feedback overhead and performance of We assume independent block fading channels for all the
D2D communications underlying cellular networks in fading links in the network as in [15], [29], [32]. That is, channel state
channels, where the BS only knows partial CSI of the CU- remains constant over a block of symbols and is independent
D links. We propose two different approaches to reduce CSI of the previous state [33]. Thus, within a block, the instan-
feedback overhead: probabilistic and partial feedback schemes. taneous channel power gain of the interference link between
We utilize the channel statistical characteristics to avoid the CU i and the receiver of D2D pair j can be expressed as
instantaneous CSI feedback and increase the access distance
hi,j = Cβi,j L−α
i,jRx , (1)
between CUs and D2D receivers to counteract the channel
fading. Since the exact signal-to-interference-plus-noise-ratio where C and α are the pathloss constant and exponent,
(SINR) of D2D users can not be determined, we use the respectively, βi,j is the channel fading component, and Li,jRx
maximum outage probability to guarantee QoS for D2D pairs. is the distance between CU i and the receiver of D2D pair
We will derive a closed-form expression of the distance factor j. Similarly, we can express the channel gain between CU i
for different channel models to satisfy the access conditions and the BS as gi,B , and channel gain between the transmitter
(i)
of both CUs and D2D pairs and also develop a probability- and the receiver of D2D pair j when using channel i as gj 1 ,
based resource allocation framework to maximize the overall and the interference channel gain from D2D pair j to the BS
network throughput of CUs and D2D pairs. To avoid user (i)
when using channel i as hj,B . The power of additive white
outage and reduce CSI feedback, we then propose an efficient Gaussian noise on each channel is assumed to be σN 2
.
and simple partial feedback scheme as an alternative to deal (i)
In practice, gi,B and hj,B can be obtained at the BS by the
with channel uncertainty. Part of our work has been published classical channel estimation method with the help of training
in [30], [31]. Specifically, in [30], we have proposed the sequence since these links are directly connected to the BS,
probabilistic resource allocation scheme in Rayleigh fading (i)
and gj can be estimated when potential D2D transmitters
channels while introduced partial CSI feedback scheme in send discovery beacons to the neighboring nodes [34] and the
[31]. In this paper, we will further investigate probabilistic re- BS can get the information when D2D connection request is
source allocation in lognormal and Rayleigh-lognormal fading sent. Thus, in this paper, we assume that the BS has the perfect
channels and analyze the tradeoff between signal overhead and (i) (i)
knowledge of gi,B , gj and hj,B .
system performance. In addition, to exploit benefits of both the
probabilistic and the partial feedback schemes, we develop a 1 Throughout the paper, when comparing with the algorithm in [15], we
combined scheme, which is less sensitive to the change of (i)
change the notation gj in [15] as gj to more accurately describe the channel
channel statistics and D2D cluster radius. It is shown that the gain of D2D pair j when reuses the channel of CU i; similarly, we also correct
(i)
combined scheme can significantly improve the performance the notation hj,B in [15] as hj,B .

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 3

Since the CU-D link is not connected to the BS and According to [36], when Φ = 1, the received signals cannot
obtaining its channel gain is difficult and may cause high be successfully decoded at the receiver, thus we regard the
overhead, we only exploit the distance-based pathloss for the throughput, Rjd Φ=1 = 0; when Φ = 0, the transmitted signals
channel gain, hi,j , of the CU-D link at BS to perform resource can be successfully decoded, thus the throughput Rjd Φ=0 =
allocation. The distance information can be evaluated by log2 (1 + ξjd ). Therefore, the expected effective throughput for
users’ location information. In the following context, we use D2D pair j can be expressed as,
hi,j = C ·L−α i,jRx to denote the distance-based pathloss of hi,j . { }
Besides, we assume that the channel fading component, βi,j , is E [log2 (1 + ξjd )] Φ = 0 , (5)
ergodic and stationary with probability density function (pdf), where E{·|·} denotes conditional expectation.
f (βi,j ), and cumulative distribution function (cdf), F (βi,j ). Note that in (5), when D2D pair j uses the channel of CU
i, the channel fading component, βi,j , is the only unknown
B. QoS for CUs and D2D Pairs variable for a given transmit power pair, (Pic , Pjd ). According
We assume that BS provides guaranteed QoS for each CU to [37], (5) can be calculated as,
{ }
and D2D pair. Since all the related channel power gains, gi,B E [log2 (1 + ξjd )] Φ = 0
(i)
and hj,B , are assumed to be known for all CUs at the BS, (i) (i)
{ Pjd gj Pjd gj }
)]
c
minimum SINR, ξi,min can be guaranteed2 . The SINR of CU = E [log2 (1 + ≥ ξj,min
d
2
+ Pic βi,j hi,j σN
σN 2 + P cβ h
i can be expressed as i i,j i,j
(i)
Pic gi,B { Pjd gj }
ξic , ∑ ≥ ξi,min
c
, ∀i ∈ C, (2) = E [log2 (1 + 2 )] βi,j ≤ l
2
σN + d (i)
ρi,j Pj hj,B σN + Pic βi,j hi,j
j∈D ∫ l (i)
Pjd gj f (βi,j )
where ρi,j is the resource reuse indicator, ρi,j = 1 when = [log2 (1 + 2 cβ h
)] dβi,j , (6)
0 σN + P i i,j i,j
F (l)
D2D pair j uses channel i; otherwise, ρi,j = 0, Pic and Pjd
(i)
denote the transmit power of CU i and that of D2D pair j, P dg −ξ d σ2
where l = j P jc ξd j,min N
is the ’cutoff’ value of βi,j and
respectively. i j,min hi,j

For the D2D pair, the channel fading component, βi,j , is any fading gain higher than the ’cutoff’ value will cause an
unavailable at the BS and therefore the exact SINR can not outage.
be determined. Thus, outage probability is used to provide the Therefore, the optimal resource allocation problem for max-
QoS for D2D users. When D2D pair j uses the channel i imizing the overall network throughput of regular CUs and the
(ρi,j = 1), the outage probability can be expressed as, admissible D2D pairs with guaranteed QoS in fading channels
{ } can be expressed as
Pr ξjd < ξj,min
d
≤ ψ, {
(3) ∑∑[
(i)
max log2 (1 + ξic ) +
P dg ρi,j ,Pic ,Pjd
where ξjd , σ2 +Pj j
c
d
is the SINR of D2D pair j, ξj,min i∈C j∈S
}
i hi,j
}]
N
denotes the desired SINR at D2D receivers, and ψ denotes { d

the maximum acceptable outage probability for D2D users. ρi,j E [log2 (1 + ξj )] Φ = 0 , (7)

Pic gi,B
C. Problem Formulation s.t. ξic , ∑ (i)
≥ ξi,min
c
, ∀i ∈ C, (7a)
2
σN + d
ρi,j Pj hj,B
As the BS provides guaranteed QoS for both CUs and j∈S
D2D pairs, we call a D2D pair admissible pair only when { Pjd gj
(i) }
the constraints in (2) and (3) are simultaneously satisfied. In Pr ξjd , 2 +
∑ < ξ j,min ≤ ψ, ∀j ∈ S,
d
σN ρi,j Pic hi,j
the following, we denote S (S ⊆ D) the set of admissible i∈C
D2D pairs. (7b)
As indicated in [36], without perfect channel knowledge ∑
ρi,j ≤ 1, ρi,j ∈ {0, 1}, ∀i ∈ C, (7c)
at the transmitter, outage occurs whenever the channel cannot
d j
support the target SINR, ξj,min . As in [32], for the convenience ∑
of analysis, we introduce the outage indicator function, Φ, for ρi,j ≤ 1, ρi,j ∈ {0, 1}, ∀j ∈ S, (7d)
the outage event of D2D pairs: i
 Pic ≤ Pmax
c
, ∀i ∈ C, (7e)
 1, ξjd < ξj,min
d
;
Φ= (4) Pjd ≤ d
Pmax , ∀j ∈ S, (7f)
 0, otherwise.
c d
where Pmax and Pmax denote the maximum transmit power
of CUs and D2D pairs, respectively.
2 Note that we have assumed that channels are randomly assigned to CUs, When perfect CSI of all links are available at BS, the
thus as in [35], we first do feasibility check for CUs. If the minimum SINR above optimization problem can be solved by the three-step
requirement for a CU is not achieved at the peak power, Pmax c , we adjust
the minimum SINR requirement for the CU as the SINR at the peak power algorithm in [15]. However, when the instantaneous channel
without D2D communications. power gain of CU-D link, hi,j , is unknown, it is difficult to

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 4

obtain the solution directly. In the following two sections, we all the constraints in (8) can be satisfied without outage (ψ =
will derive probabilistic and partial feedback strategies to solve 0). Fig. 2 illustrates the three possible shapes of admissible
the problem, respectively. area for this case with the lightly shaded pattern, where line
lc represents constraint (8a) with equality and line ld represents
III. P ROBABILISTIC R ESOURCE A LLOCATION S TRATEGY ξjd = ξj,min
d
in the probability of (8b).

In this section, we derive a probability-based resource


allocation scheme for D2D underlaying cellular networks in Pi c ld ld ' lc
different fading channels by utilizing channel statistical char- '
C
c
acteristics. We first develop a probabilistic minimum distance Pmax C D
metric to find the admissible D2D pairs, and then a joint
power and channel allocation strategy to maximize the overall
A'
throughput of CUs and the accessible D2D pairs. A

A. Probabilistic Admission Control of D2D Users


To solve the optimization problem in (7), we first need
to know whether a D2D pair can be admitted or not. As a
fully loaded cellular network is considered, D2D pair j can
be admitted on channel i only when constraints (7a), (7b),
d
Pmax Pjd
(a)
(7e), and (7f) are satisfied, that is,
Pi c
Pic gi,B ld ld '
(i)
≥ ξi,min
c
, (8a) c
2 + P dh Pmax
σN j j,B E C'
| {z }
ξic
E'
{ (i)
Pjd gj } lc
Pr 2 + P ch
d
< ξj,min ≤ ψ, (8b) A'
σN i i,j
| {z } F
A
ξjd A'
Pic ≤ Pmax
c
, Pjd ≤ Pmax
d
. (8c)
Without considering the channel fading component of the
CU-D link (βi,j = 1), the admissible D2D pairs can be found d
Pmax Pjd
by the minimum distance metric derived in [15], which defines (b)
the minimum access distance between CU i and the receiver Pi c ld ld '
of D2D pair j to satisfy all access constraints. This metric '
c
C
comprehensively considers the QoS of users and the related Pmax C O
link gains as well as the power constraint of users. According
to whether the minimum SINR of CU i can be satisfied at lc
c d
the peak power, (Pmax , Pmax ), or not, the minimum access F
distance can be expressed as follows when βi,j = 13 . A
A'


1

 Cξ c
ξ d
P c
h
(i) α

 i,min j,min max j,B

 [ ]

 c g
(Pmax 2 (i)
i,B − ξi,min σN )gj − ξi,min ξj,min σN hj,B
c c d 2 (i)




c
Pmax gi,B

 if ≤ ξi,min
c
,
 2 d
σN + Pmax hj,B
(i) d
min Pmax Pjd
Li,jRx = (c)

1

 c
Cξi,min d
ξj,min (σN2
+ Pmaxd (i)
hj,B )
α

 Fig. 2: D2D admissible area. Light- and heavy- shaded pat-

 [ ]

 g (P d g (i) − ξ d σ 2 )
terns represent the admissible area without and with fading,


i,B max j j,min N


c
Pmax gi,B respectively.

 if c
> ξi,min .
 2 d
σN + Pmax hj,B
(i)

(9) Note that in the admissible area, point D or F is the most


According to the conclusion in [15], as long as the actual accessible point4 for the D2D pair j as suggested in [15]. Par-
distance between CU i and the receiver of D2D pair j, min
ticularly, at the most accessible point, when Li,jRx = Li,jRx ,
min
Li,jRx ≥ Li,jRx , (10)
4 D2D pair j cannot be accessed at other point if it is unable to be accessed
3 For the details, we refer the interested reader to [14]. at point D or F since D or F is the only point for the equality hold in (10).

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 5

the constraint in (8b) can be guaranteed with However, the R-L distribution is not mathematically tractable.
(i) In [39], [40], it is shown that the R-L distribution can be
Pjd gj d well approximated by the K-distribution. Thus, we also use
min
= ξj,min , (11)
2 + P ch
σN i i,j
K-distribution for mathematical analysis. According to [39],
[40], the cdf of βi,j then can be expressed as
min min −α min
where hi,j = C · Li,jRx . Therefore, Li,jRx = Li,jRx , is
2(Ξβi,j )
k/2 √
the required minimum distance between CU i and the receiver F R L (βi,j ) = 1 − Bk (2 Ξβi,j ), βi,j > 0, (17)
of D2D pair j for D2D pair j to be accessed on channel i when Γ(k)
the effect of fading is not considered. where k is the fading parameter related to the parameters of
Obviously, when channel fading, βi,j , is considered, the Rayleigh and lognormal distributions, Ξ , k/E [βi,j ], Γ[·]
min
minimum distance, Li,jRx , should be changed by a distance is the Gamma function, and Bk [·] is the k-th order modified
factor, γ, that is Lb i,j = γLmin
i,jRx to counteract the fading
Bessel function of the second kind.
effect and thus to satisfy the access constraints in (8). In From (12),
1
this case, the corresponding channel gain can be expressed γRL = [F R L −1 (1 − ψ)] α , (18)
as b
min
hi,j = Cβi,j (γLi,jRx )−α . The distance factor can be
determined by where F R L −1 [·] denotes the inverse function of the cdf,
F R L [·].
{ (i)
Pjd gj }
d Note that in all channel models, the minimum distance
Pr < ξj,min
σN2 + P cb factor, γmin , is monotonically decreasing with the outage
i hi,j
{ P d g (i) (i) } threshold, ψ. For example, in (14), when ψ ≤ 1e , γR ≥ 1;
j j Pjd gj
= Pr < otherwise, γR < 1. That implies that the less the desired out-
2 + P cb
σN 2 + P ch min
i hi,j σN i i,j
age probability, the bigger the minimum distance requirement.
{ } { } It also shows that the minimum distance factor shares similar
b min min min
= Pr hi,j > hi,j = Pr Cβi,j (γLi,jRx )−α > hi,j idea as the fading margin type approach. In brief, our focus
{ min min
} { }
= Pr γ −α βi,j hi,j > hi,j = Pr βi,j > γ α ≤ ψ. (12) is to increase the minimum access distance between the CUs
and the D2D pairs to counteract the unknown fading while in
Denote γmin to be the minimum { distance factor
} satisfying most cases, fading margin is used in the link budget for proper
(12). Then, it can be found by Pr βi,j > γmin α = ψ. From power control.
the above, the distance factor depends on the pdf of fading. When the minimum distance factor is obtained, we can
We will discuss this issue under different cases. express the modified minimum distance metric for D2D access
1) Rayleigh fading: For Rayleigh fading channel, βi,j is in fading channels as follows
exponentially distributed. Here, assuming it is with unite mean, b min min
L i,jRx = γmin Li,jRx . (19)
thus the cdf can be expressed as
Then, we can determine whether a D2D pair can be admitted
F R (βi,j ) = 1 − e−βi,j , βi,j > 0. (13)
or not and also find all the potential partner CUs for the D2D
Substituting it into (12), the minimum distance factor in pair if it is accessible by this metric. Specifically, let Rj denote
Rayleigh fading channels can be expressed as the set of potential partner CUs of D2D pair j, i ∈ Rj only
1/α when Li,jRx ≥ L b min , and D2D pair j is admissible (j ∈ S)
i,jRx
γR = (− ln ψ) . (14)
if and only if Rj ̸= Ø.
2) Lognormal fading: In wireless systems, the shadowing When Li,jRx > L b min , more than one accessible point will
i,jRx
fading caused by obstacles is usually modeled as lognormal be available. To find all the accessible points, let the outage
distribution. Here, we assume βi,j has zero mean and standard probability constraint in (8b) with equality, we can get
deviation, σ, in dB unit, thus its cdf can be written as, (i)
Pjd gj − ξj,min
d 2
σN
1 1 ln βi,j Pic = , (20)
F L (βi,j ) = + erf[ √ ], βi,j > 0, (15) F −1 (1 − ψ)ξj,min
d hi,j
2 2 2σl
∫x where F −1 (·) is the inverse function of F (·). Let line ld in

where erf(x) , √2π 0 e−t dt denotes the error function and


2

Fig. 2 represent Equation (20). Obviously, all the points on the


σl = ln1010 σ. ′
right of line ld will satisfy the constrain (8b) and at the same
Together with (12), the minimum distance factor under time all the points above line lc satisfy the constrain (8a).
lognormal shadowing fading channels can be written as, Thus, the modified admissible area for D2D pairs in fading
√ −1
2σl erf (1 − 2ψ) channels can be shown as the heavily shaded areas in Fig. 2.
γL = e α , (16)
where erf −1 (·) denotes the inverse function of erf(·). B. Power Control
3) Rayleigh-lognormal fading: When both multipath fading In the previous section, we have proposed the modified
and shadowing are considered, the received signal can be minimum distance for D2D access in fading channels with
modeled as Rayleigh-lognormal (R-L) distribution [38], [39]. outage requirement. In the following, we introduce the joint

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 6

power and channel allocation scheme for multiple D2D pairs we estimate the expected D2D throughput gain, defined as the
and their potential partner CUs to maximize the overall ergodic difference between the maximum expected sum throughput
throughput. and the maximum throughput of the partner CU without D2D,
We first consider the optimal power control for a single that is,
D2D pair j and its potential partner CU i. Mathematically, ∗ P c gi,B
T i,j = f (Pic ∗ , Pjd ) − log2 (1 + max2
G
the problem can be expressed as, ). (23)
σN

{
(Pic ∗ ,Pjd ) = arg max log2 (1 + ξic )+ By now, when there is only one D2D pair j in the cell, we
Pic ,Pjd
}}
can easily find the optimal partner CU as,
{
E [log2 (1 + ξjd )] Φ = 0 , (21) G
i∗ = arg max T i,j . (24)
i∈Rj
s.t. ξic ≥ ξi,min
c
, (21a)
{ d } When multiple D2D pairs exist in the cell, the allocation
Pr ξj < ξj,min ≤ ψ,
d
(21b)
problem becomes much more complicated since different D2D
Pic ≤ Pmax c
, Pjd ≤ Pmax d
. (21c) pairs may have the same optimal partner CU. To deal with
Note that in the previous subsection, we have derived the this issue, we adopt the maximum-weight-bipartite-matching
modified admissible area for D2D pairs, where all power based method that we have introduced in [15]. Particularly, in
pairs satisfy the constraints in (21). Thus, the optimization the user match procedure, we first separate the accessible D2D
problem becomes to find the optimal power pair in the pairs and the union of all the partner CUs into two groups in
modified admissible area as shown in Fig. 2. Here, we adopt a bipartite graph, then connect D2D pair j and CU i with
G
the same method as in [15] to find the solution of (21), the weight, w i,j (w i,j = T i,j if i ∈ Rj , otherwise, wi,j = 0).
which chooses the end point in boundary of admissible area Finally, we obtain the optimal D2D pair and CU match by
as the potential {operation points. Denoting
} f (P i
c
, P d
j ) , utilizing the classic Kuhn-Munkres algorithm [41].
d
log2 (1 + ξi ) + E [log2 (1 + ξj )] Φ = 0 , from the appendix,
c So far, we have found all admissible D2D pairs and all
the power allocation in (21) can be obtained as follows, corresponding reuse candidates for each admissible D2D pair
 through the modified minimum access distance in Sec. III-A;

 arg max f (Pic , Pjd ) then, we have proposed an efficient power allocation method

 (Pic ,Pjd )∈P1

 for the admissible D2D pairs and its reuse partners based on


c
Pmax gi,B
(i) ≤ ξi,min ,
c


if
the optimal power control in [15] in Sec. III-B; finally, optimal
 σ 2 +P d h


N max j,B

 c d channel allocation is achieved by the Kuhn-Munkres algorithm
arg cmax f (Pi , Pj )
c∗ d∗ (Pi ,Pjd )∈P2 in Sec. III-C. Thus, QoS-aware D2D resource allocation in
(Pi , Pj ) =

c
Pmax gi,B
(i) > ξi,min and Pj,C ′ ≥ Pmax ,
 if c d d fading channels is derived by the probabilistic strategy and

 2 +P d


σN max hj,B can be illustrated by the algorithm in Table I.



 arg max f (P c
, P d
)

 (Pic ,Pjd )∈P3
i j


 if
 P c
g
max i,B c d d IV. PARTIAL F EEDBACK S TRATEGY
2 d (i) > ξi,min and Pj,C ′ < Pmax ,
σN +Pmax hj,B
(22) In the previous section, we have introduced a probabilistic
where resource allocation strategy to deal with the channel uncertain-
P1 = {(Pmax
c d
, Pj,C c d
′ ), (Pmax , Pj,D )},
ty of CU-D links. In this section, we will focus on efficient
partial feedback strategies, as an alternative way to address this
P2 = {(Pi,E
c d c d
′ , Pmax ), (Pi,F , Pmax )},
issue. Particularly, for partial feedback strategies, no outage
P3 = {(Pmax
c d
, Pj,C c d c d
′ ), (Pmax , Pmax ), (Pi,F , Pmax )}, will occur at D2D receivers while as with the cost of the
d
ξj,min c
[Pmax hi,j F −1 (1−ψ)+σN
2
] feedback overhead. In the following, we propose a simple
d d
and Pj,C ′ = (i) and Pj,D = user selection based partial feedback scheme, where each D2D
gj
c
Pmax gi,B −ξi,min
c 2
σN receiver only feeds back CSI of K best potential partner CUs
(i) are the power of D2D pair j at point C′
c
ξi,min hj,B to reduce feedback overhead.
(i)
c
d
Pmax d
(Pmax gj −ξj,min
d 2
σN ) For the partial feedback strategies, the key is to find the
and D, respectively, and Pi,E ′ = c
Pmax hi,j F −1 (1−ψ)
and
d (i) 2 c
best potential partner CUs for each D2D pair as the CSI of
(Pmax hj,B +σN )ξi,min
c
Pi,F = gi,B are the power of CU i at point E′ different CU-D links is not equally important. For example,
and F, respectively. the CUs that are far away from the D2D receivers are more
likely to be the reuse partners and thus the corresponding CSI
will be more important. Therefore, to choose the best potential
C. Channel Allocation partner CUs, one intuitive method is to choose the CUs farthest
As in [15], after obtaining the power allocations for a D2D away from the D2D receiver and we call this method, KFAR.
pair and its corresponding partner CUs, we can estimate the However, other factors, such as the QoS requirements of CUs
D2D throughput gain brought by the D2D pair. However, in and the channel power gain of CU-BS link, will also affect
min
this paper, we cannot obtain the exact throughput gain as in the access of the D2D pair. Thus, we consider Li,jRx /Li,jRx
[15] since the channel gain of CU-D link is unknown. Instead, as the metric to choose the best potential partner CUs. As we

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 7

TABLE I: Probabilistic Resource Allocation Algorithm


Algorithm 1 Probabilistic Resource Allocation Algorithm MDR metric scheme, and the whole procedure is illustrated
in Table II.
1: C : The set of existing cellular users TABLE II: Resource Allocation with Partial Feedback
2: D : The set of D2D pairs
3: Rj : The set of reuse candidates of D2D pair j Algorithm 2 Resource Allocation with KMDR
4: ψ : Accepted maximum outage probability for D2D pairs min
1: M DR :The maximum distance ratio (Li,jRx /Li,jRx )
5: f (Pic , Pjd ) : Ergodic throughput of D2D pair j and the 2: Cj : The selected K CU set for D2D pair j
partner CU i 3: g(Pic , Pjd ) : Throughput of D2D pair j and the partner
6: Step 1 Probabilistic Admission Control CU i
7: for all j ∈ D and i ∈ C do 4: Step 1 K CU Selection
calculate Lb min with given ψ;
8: i,jRx 5: for all j ∈ D and i ∈ C do
9: if Li,jRx ≥ L b min then min
i,jRx 6: calculate Li,jRx and M DR;
10: i ∈ Rj ; 7: sort (MDR, column, ’descend’);
11: end if 8: Cj ← the first K sorted list in jth column of M DR;
12: if Rj = ∅ then 9: end for
13: D = D − j; 10: Step 2 Optimal Power Allocation
14: end if 11: for all j ∈ D and i ∈ Cj do
15: end for calculate Lmin
12: i,jRx (the actual minimum access distance
16: Step 2 Power Allocation with fading);
17: for all j ∈ D and i ∈ Rj do if Li,jRx ≥ Lmin

13: i,jRx then
18: (Pic ∗ , Pjd ) = arg max f (Pic , Pjd ) as in (22); 14:

(Pic ∗ , Pjd ) = arg max g(Pic , Pjd ) as in (12) of
Pic ,Pjd
Pic ,Pjd
19: end for [14];
20: Step 3 Optimal Channel Match 15: i ∈ Rj ;
21: if |D|=1 then 16: end if
G ∗ P c gi,B
22: i∗ = arg maxT i,j , f (Pic ∗ , Pjd )−log2 (1+ max
σ2
); 17: if Rj = ∅ then
i∈Rj N

23: else 18: D = D − j;


24: get i∗ from Kuhn-Munkres algorithm; 19: end if
25: end if 20: end for
21: Step 3 Optimal Channel Match : The same procedure as in
∗ P c gi,B
have mentioned in Sec. III-A,
min
Li,jRx
is the required minimum
G
Algorithm 1 with Ti,j , g(Pic ∗ , Pjd )−log2 (1+ max
σ 2 )
N

distance between CU i and the receiver of D2D pair j to satisfy


all the access constraints in (7) without considering the fading
effect of the CU-D link (by setting βi,j = 1) and Li,jRx is the
actual distance between CU i and the receiver of D2D pair
j, we call this metric as the maximum distance ratio (MDR) V. C OMBINED R ESOURCE A LLOCATION S TRATEGY
metric. In the above, we have introduced the probabilistic and the
When the MDR metric is calculated for D2D pairs, the partial feedback schemes to deal with channel uncertainty. In
BS can select the K CUs with the largest MDR as the best fact, with the probabilistic scheme, the D2D user outage is
potential partner CUs for each D2D pair. We use MDR as the inevitable while the partial feedback scheme is with feedback
user selection metric is for the following two reasons: overhead. In this part, we derive a combined scheme to
min
i) As Li,jRx is required minimum distance to satisfy all reduce the outage probability and feedback overhead by taking
the access constraints without considering the fading the advantages of both the probabilistic and partial feedback
effect, thus the bigger the ratio, the more the ability to schemes.
counteract the fading; In the combined scheme, the KMDR scheme is first exe-
min cuted to find the K most potential partner CUs and to trace
ii) A lower Li,jRx implies that the CU can tolerate more
D2D interference5 and thus can get higher D2D through- and report the CSI of CU-D links. And then, the probabilistic
put gain if the CU is selected. strategy is adopted to find the potential partner CUs from
the unselected CUs in the KMDR. In this way, on one hand,
After the user selection procedure is finished, each D2D pair
KMDR scheme helps the probabilistic scheme to improve the
can trace and report the CSI of the respective K selected CUs
reliability since more accurate CSI is obtained, and on the
as the method introduced in [42]. When the CSI is available
other hand, the probabilistic scheme help the KMDR scheme
at the BS, the optimal power allocation as well as CUs and
to increase the diversity since more potential partner CUs are
D2D pairs matching algorithm in [15] can be used. Thus, the
found. For the selected partner CUs in the KMDR scheme
problem in (7) can be solved by the selected-K feedback with
and the probabilistic scheme, the power allocation in [15] and
5 In[15], it has analyzed that a CU with a better channel power gain, gi,B , in the Section III are used, respectively. The whole procedure
c
and a lower SINR requirement, ξi,min
min
will bring a lower Li,jRx and thus of the combined resource allocation strategy is illustrated in
can potentially tolerate more interference. Table III.

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 8

TABLE III: Combined Resource Allocation


Algorithm 3 Combined Resource Allocation Scheme cessed D2D pairs without outage, and the ratio of the number
of accessed D2D pairs in outage state and the number of the
1: ψ : Accepted maximum outage probability for D2D pairs
min total accessed D2D pairs, respectively. Moreover, we compare
2: M DR :The maximum distance ratio (Li,jRx /Li,jRx )
the proposed resource allocation strategies in fading channels
3: Cj : The selected K CU set for D2D pair j
with the optimal resource allocation presented in [15], where
4: Step 1 D2D Admission Control
perfect CSI of all links is known at BS.
5: for all j ∈ D and i ∈ C do
min
6: calculate Li,jRx and M DR; TABLE IV: Simulation Parameters
7: sort (MDR, column, ’descend’);
8: Cj ← the first K sorted list in jth column of M DR; Parameter Value
9: if i ∈ Cj then Cell radius (R) 500 m
10: calculate Lmini,jRx ;
Uplink bandwidth 5 MHz
2
11: if Li,jRx ≥ Lmin i,jRx then
Noise power (σN ) -114 dBm
12: i ∈ Rj ; Pathloss exponent (α) 3.5
13: end if Pathloss constant (C) 10−2
14: else Maximum CU Tx power
b min with given ψ; c 24 dBm
15: calculate L i,jRx (Pmax )
16: if Li,jRx ≥ L b min then Maximum D2D Tx power
i,jRx
d 21 dBm
17: i ∈ Rj ; (Pmax )
18: end if c Uniform distributed in [0,
Cellular user SINR (ξi,min )
19: end if 25] dB
20: end for D2D user desired SINR Uniform distributed in [0,
d
21: Step 2 Power Allocation (ξj,min ) 25] dB
22: for all j ∈ D and i ∈ Rj do D2D cluster radius (r) 20, 40, 60, ...., 160 m
23: if i ∈ Cj then Number of active CUs (N ) 20

24: (Pic ∗ , Pjd ) = arg max g(Pic , Pjd ) as in (12) of Number of D2D pairs (M ) 4
Pic ,Pjd
Outage probability thresh- 0.001, 0.01, 0.02, 0.04,
[14];
∗ P c gi,B old (ψ) 0.06, ..., 0.40
25: G
Ti,j , g(Pic ∗ , Pjd ) − log2 (1 + max2 );
σN
Fading parameter of K-
26: else 0.63
∗ distribution (k)
27: (Pic ∗ , Pjd ) = arg max f (Pic , Pjd ) as in (22);
Pic ,Pjd Exponential distribution
G ∗ c Rayleigh fading
f (Pic ∗ , Pjd ) with unit mean
Pmax gi,B
28: ,
T i,j − log2 (1 + 2
σN
);
29: end if Lognormal distribution with
Shadowing
30: end for standard deviation of 8dB
31: Step 3 Optimal Channel Match: The same procedure as
in Algorithm 1 and Algorithm 2
8
VI. N UMERICAL R ESULTS optimal outage
threshold point
7
In this section, the performance of the proposed resource
allocation strategies in fading channels is evaluated by sim-
Throughput gain (Mbps)

6
ulation. We consider a single cell cellular network, where
regular CUs are uniformly distributed in the cell with radius,
R while D2D users are uniformly distributed in a randomly 5 Optimal in [14]
located cluster with radius, r. We also assume different D2D Rayleigh
Lognormal
pairs are within different clusters, and the total bandwidth 4 Rayleigh−lognormal
are equally divided. In addition, both Rayleigh fading and
lognormal shadowing are considered in the simulation. Specif-
3
ically, for the Rayleigh fading channel and lognormal fading
channel, we assume the shadowing fading component and
Rayleigh fading component are known at BS, respectively, 2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
while for the Rayleigh-lognormal fading channel, both fading ψ
components are unknown at BS. The simulation parameters
are summarized in Table IV. Fig. 3: D2D throughput gain versus outage threshold under
Two metrics are used to evaluate the performance of the different channel models for the probabilistic algorithm when
resource allocation schemes: D2D throughput gain and outage r=100 m.
rate, defined as the increased throughput brought by the ac-

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 9

D2D receivers is increased. Besides, the figure also shows


0
10 the performance of the probabilistic scheme decreases in turn
in Rayleigh fading, lognormal fading, and Rayleigh-lognormal
−1
fading channels. That is because the fading in the three models
10
becomes more severe in this turn and also causes much more
channel uncertainty. This can also explain why the optimal
−2 outage probability threshold increases in turn for the three
10
Outage rate

channel models in the figure.


Fig. 4 illustrates the actual D2D outage rates under different
10
−3 channel models and different outage thresholds. From the
figure, in all channel models, the actual D2D outage rate
increases with the outage threshold. This is because that as
10
−4
the outage threshold increase, the required minimum distance
Rayleigh
Lognormal between the CU and the D2D receiver for the D2D to access
Rayleigh−lognormal decreases. It is also seen that for a certain outage threshold,
−5
10
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
the D2D outage rate increases in turn in Rayleigh fading,
ψ lognormal fading and Rayleigh-lognormal fading channels
since more and more severe fading is presented in turn in
Fig. 4: D2D actual outage rate versus outage threshold under the three fading models.
different channel models for the probabilistic algorithm when Fig. 5 compares the performance of the proposed KMDR
r=100 m. scheme with the optimal resource allocation in [15] and also
the intuitive method, KFAR, under different channel fading
models. Since the optimal scheme has the full CSI and the
KFAR scheme only relates to the distance between users, their
performance does not change with channel models. As the
same reason for the probabilistic scheme, the performance of
8 the KMDR scheme also decreases in turn in Rayleigh fading,
lognormal fading and Rayleigh-lognormal fading channels.
7 However, compared to the probabilistic scheme in Fig. 4, the
KMDR scheme is much less sensitive to the channel models.
6
From the figure, for both the KMDR and the KFAR schemes,
Throughput gain (Mbps)

5
performance is improved with the number of the selected
CUs since more correct CSI will help the BS to make more
4 reasonable decisions. In particular, for K=6, the performance
of the KMDR scheme for all the channel models is almost the
3 same as that of the optimal scheme. Therefore, the KMDR
Optimal in [14] scheme can reduce around 70% of total feedback information
2
KFAR at the D2D receivers while still providing a near optimal
KMDR, Rayleigh performance. It is also seen that performance of the KMDR
1 KMDR, Lognormal
KMDR, Rayleigh−lognormal
scheme is much better than that of the KFAR scheme. This is
0 because, when the KMDR scheme selects the most potential
1 2 3 4 5 6 7 8 9 10
K CUs, it comprehensively considers the effect of QoS of CUs,
the channel power gain of CU-BS links, and the distance
Fig. 5: D2D throughput gain versus the number of feedback between CU and D2D receivers while the KFAR scheme only
CUs under different channel models for the partial feedback considers the last one.
schemes when r=100 m. Fig. 6 demonstrates the D2D throughput gain of the com-
Fig. 3 compares the performance of the proposed prob- bined scheme (Comb), the probabilistic scheme (Prob) and
abilistic resource allocation scheme under different channel the KMDR scheme for different D2D cluster radii. From the
fading models with the optimal scheme in [15]. From the figure, the performance of the probabilistic scheme is much
figure, D2D throughput gain for the probabilistic scheme better than that of KMDR scheme with K=1 and K=2, and
increases with outage probability threshold, ψ, at first, and worse than that of the KMDR scheme with K=4. Hence, the
then decreases after the maximum values. As a result, there probabilistic scheme outperforms the partial feedback scheme
exists an optimal threshold of the outage probability. This can when the size of feedback is low. While as the size of feedback
be easily understood. A lower outage threshold will bring a increases, the partial feedback scheme will eventually become
lower outage probability for the accessed D2D pairs, and will better than the probabilistic scheme. With the same size of
also increase the difficulty for the D2D pair to be accessed feedback, performance of the combined scheme is always bet-
since the required minimum distance between CU and the ter than that of the KMDR scheme. For any size of feedback,
performance of the combined scheme is also better than that of

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 10

link is declined and thus less interference can be tolerated. It


18 is also seen that the outage rate of the combined scheme is
Optimal in [14]
KMDR, K=1 lower than that of the probabilistic scheme, especially when
16 KMDR, K=2 the the size of feedback is large. Together the results in Fig. 6,
KMDR, K=4
14 Prob we can conclude that the combined scheme can significantly
Comb, K=1 improve the performance of the probabilistic scheme and the
Throughput gain (Mbps)

Comb, K=2
12 Comb, K=4 partial feedback scheme.

10 VII. C ONCLUSIONS
8
In this paper, we have investigated the resource alloca-
tion problem for D2D communications underlying cellular
6 network in fading channels with QoS requirements. To deal
with channel uncertainty, we have presented two different
4 schemes: a probability-based resource allocation scheme u-
tilizing channel statistics and an efficient and simple user
2
20 40 60 80 100 120 140 160 selection based partial feedback scheme. In addition, we also
r (m)
develop a combined scheme to take the advantage of both
probabilistic and partial feedback schemes. It is shown that
Fig. 6: D2D throughput gain versus D2D cluster radius under
for the probabilistic scheme, there exists an optimal outage
Rayleigh fading channels, where ψ = 0.1 for Comb and Prob.
probability threshold for D2D users with respect to the overall
network throughput. And it is also shown that the combined
scheme can significantly improve the network performance
compared to use probabilistic scheme or partial feedback
scheme alone.
−1
10 A PPENDIX A
P ROOF OF THE POWER VECTOR ALLOCATION
Define,
g(Pic , Pjd ) , log2 (1 + ξic ) + log2 (1 + ξjd ), (A.1)
then, (A.1) can be proved to be convex on the boundary of
Outage rate

−2 the maximum power constraints in [15].


10 c
Pmax gi,B
For scenario (i) 2 d (i) ≤ ξi,min
c
, the modified ad-
σN +Pmax hj,B
missible area is shown in Fig. 2a. We have proved in [15]
that the optimal power for (21) with full CSI and zero outage
Prob probability (ψ = 0) lies on line CD. And we have also proved
Comb, K=1 that (A.1) is convex on CD. Thus, in the modified admissible
area, one of the point C′ and D will be the closest point to the
Comb, K=2
Comb, K=4
10
−3
optimal point (C or D). Thus, for the optimization problem in
20 40 60 80
r (m)
100 120 140 160
(21), we also consider point C′ and D to be the two potential
operation points. Therefore, the power vector can be expressed
Fig. 7: D2D actual outage rate versus D2D cluster radius under as,

Rayleigh fading channels, where ψ = 0.1 for Comb and Prob. (Pic ∗ , Pjd ) = arg max f (Pic , Pjd ), (A.2)
(Pic ,Pjd )∈P1
the probabilistic scheme. That is because the combined scheme
can take the advantages of both the probabilistic scheme and where P1 = {(Pmax
c d
, Pj,C c d d
′ ), (Pmax , Pj,D )}, and Pj,C ′ =
partial feedback scheme as we have indicated in Section V.
d
ξj,min c
[Pmax hi,j F −1 (1−ψ)+σN
2
] d
c
Pmax gi,B −ξi,min
c 2
σN
(i) and Pj,D = c (i)
The figure also shows that for both the combined scheme gj ξi,min hj,B

and the probabilistic scheme, when the D2D cluster radius are the power of D2D pair j at point C′ and D, respectively.
is smaller, performance gap with the optimal scheme also Similarly, we can get the solutions for the scenarios in Fig.
becomes smaller. Therefore, the probabilistic scheme is more 2b and Fig. 2c. This completes the proof.
useful when D2D communication distance is short.
R EFERENCES
Fig. 7 illustrates the actual D2D outage rates of the com- [1] K. Doppler, M. Rinne, C. Wijting, C. Ribeiro, and K. Hugl, “Device-
bined scheme and the probabilistic scheme for different D2D to-device communication as an underlay to LTE-advanced networks,”
cluster radii. From the figure, the actual D2D outage rate IEEE Commun. Mag., vol. 47, no. 12, pp. 42–49, Dec. 2009.
[2] M. Corson, R. Laroia, J. Li, V. Park, T. Richardson, and G. Tsirtsis,
increases with the D2D cluster radius. This is because as the “Toward proximity-aware internetworking,” IEEE Wireless Commun.
D2D cluster radius increases, the channel power gain of D2D Mag., vol. 17, no. 6, pp. 26–33, Dec. 2010.

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 11

[3] M. J. Yang, S. Y. Lim, H. J. Park, and N. H. Park, “Solving the data in cellular networks,” in Proc. IEEE Conf. Commun. (ICC’ 13), June
overload: Device-to-device bearer control architecture for cellular data 2013, pp. 101–105.
offloading,” IEEE Veh. Technol. Mag., vol. 8, no. 1, pp. 31–39, Mar. [24] G.-D. Yu, L.-K. Xu, D.-Q. Feng, R. Yin, G. Y. Li, and Y.-H. Jiang,
2013. “Joint mode selection and resource allocation for device-to-device
[4] P. Phunchongharn, E. Hossain, and D. Kim, “Resource allocation for communications,” IEEE Trans. Commun., vol. 62, no. 11, pp. 3814–
device-to-device communications underlaying LTE-advanced networks,” 3824, Nov. 2014.
IEEE Wireless Commun. Mag., vol. 20, no. 4, pp. 91–100, Aug. 2013. [25] Q. Ye, M. Al-Shalash, C. Caramanis, and J. G. Andrews, “Device-to-
[5] D.-Q. Feng, L. Lu, Y. Yuan-Wu, G. Y. Li, S.-Q. Li, and G. Feng, device modeling and analysis with a modified matern hardcore bs loca-
“Device-to-device communications in cellular networks,” IEEE Com- tion model,” in Proc. IEEE Global Telecommun. Conf. (GLOBECOM’
mun. Mag., vol. 52, no. 4, pp. 49–55, Apr. 2014. 13), Dec. 2013, pp. 1–6.
[6] 3GPP, “3rd generation partnership project; technical specification [26] X. Lin and J. G. Andrews, “Optimal spectrum partition and mode
group services and system aspects; study on architecture selection in device-to-device overlaid cellular networks,” in Proc. IEEE
enhancements to support proximity-based services (ProSe),” Global Telecommun. Conf. (GLOBECOM’ 13), Dec. 2013, pp. 1–6.
TR23.703 V0.4.1., Release 12, Dec. 2013. [Online]. Available: [27] A. Asadi, Q. Wang, and V. Mancuso, “A survey on device-to-device
http://www.3gpp.org/DynaReport/23703.htm communication in cellular networks,” IEEE Commun. Surveys Tuts.,
[7] G. Fodor and N. Reider, “A distributed power control scheme for vol. 16, no. 4, pp. 1801 – 1819, Nov. 2014.
cellular network assisted D2D communications,” in Proc. IEEE Global [28] S. Shalmashi, G. Miao, and S. Ben Slimane, “Interference management
Telecommun. Conf. (GLOBECOM’ 11), Dec. 2011, pp. 1–6. for multiple device-to-device communications underlaying cellular net-
[8] C.-H. Yu, K. Doppler, C. Ribeiro, and O. Tirkkonen, “Resource sharing works,” in Proc. IEEE 24th Int. Symp. on Personal, Indoor and Mobile
optimization for device-to-device communication underlaying cellular Radio Commun. (PIMRC’ 13), Sept. 2013, pp. 223–227.
networks,” IEEE Trans. Wireless Commun., vol. 10, no. 8, pp. 2752 [29] A. Abu Al Haija and M. Vu, “Spectral efficiency and outage performance
–2763, Aug. 2011. for device-to-device cooperation in uplink cellular communication,”
[9] B. Kaufman and B. Aazhang, “Cellular networks with an overlaid device IEEE Trans. Wireless Commun., vol. 14, no. 3, pp. 1183 – 1198, Mar.
to device network,” in Proc. IEEE 42nd Asilomar Conf. on Signals, Syst. 2015.
and Comput., Oct. 2008, pp. 1537–1541. [30] D.-Q. Feng, L. Lu, Y. Yuan-Wu, G. Y. Li, G. Feng, and S.-Q. Li, “Op-
[10] T. Chen, G. Charbit, and S. Hakola, “Time hopping for device-to- timal resource allocation for device-to-device communications in fading
device communication in LTE cellular system,” in Proc. IEEE Wireless channels,” in Proc. IEEE Global Telecommun. Conf. (GLOBECOM’ 13),
Commun. and Networking Conf. (WCNC’ 10), Apr. 2010, pp. 1 –6. Dec. 2013, pp. 1–5.
[11] H. Min, W. Seo, J. Lee, S. Park, and D. Hong, “Reliability improvement [31] ——, “User selection based on limited feedback in device-to-device
using receive mode selection in the device-to-device uplink period communications,” in Proc. IEEE 24th Int. Symp. on Personal, Indoor
underlaying cellular networks,” IEEE Trans. Wireless Commun., vol. 10, and Mobile Radio Commun. (PIMRC’ 13), Sept. 2013, pp. 1–5.
no. 2, pp. 413–418, Feb. 2011. [32] X. Kang, R. Zhang, Y.-C. Liang, and H. K. Garg, “Optimal power
[12] M. Zulhasnine, C. Huang, and A. Srinivasan, “Efficient resource al- allocation strategies for fading cognitive radio channels with primary
location for device-to-device communication underlaying lte network,” user outage constraint,” IEEE J. Sel. Areas Commun., vol. 29, no. 2, pp.
in Proc. IEEE 6th Int. Conf. on Wireless and Mobile Computing, 374–383, Feb. 2011.
Networking and Commun. (WiMob’ 10), Oct. 2010, pp. 368–375. [33] R. McEliece and W. Stark, “Channels with block interference,” IEEE
[13] P. Jänis, V. Koivunen, C. Ribeiro, K. Doppler, and K. Hugl, Trans. Inform. Theory, vol. 30, no. 1, pp. 44–53, Jan. 1984.
“Interference-avoiding MIMO schemes for device-to-device radio un- [34] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklos,
derlaying cellular networks,” in Proc. IEEE 20th Int. Symp. on Personal, and Z. Turanyi, “Design aspects of network assisted device-to-device
Indoor and Mobile Radio Commun. (PIMRC’ 09), Sept. 2009, pp. 2385– communications,” IEEE Commun. Mag., vol. 50, no. 3, pp. 170–177,
2389. Mar. 2012.
[14] D. Zhu, W. Xu, H. Zhang, C. Zhao, J. C. Li, and M. Lei, “Rate- [35] L. P. Qian, Y. J. Zhang, and J. Huang, “MAPEL: Achieving global
maximized transceiver optimization for multi-antenna device-to-device optimality for a non-convex wireless power control problem,” IEEE
communications,” in Proc. IEEE Wireless Commun. and Networking Trans. Wireless Commun., vol. 8, no. 3, pp. 1553–1563, Mar. 2009.
Conf. (WCNC’ 13), Apr. 2013, pp. 4152–4157. [36] D. Tse and P. Viswanath, Fundamentals of wireless communication.
[15] D.-Q. Feng, L. Lu, Y. Yuan-Wu, G. Y. Li, G. Feng, and S.-Q. Cambridge university press, 2005.
Li, “Device-to-device communications underlaying cellular networks,” [37] A. Papoulis and S. Pillai, Probability, random variables, and stochastic
IEEE Trans. Commun., vol. 61, no. 8, pp. 3541–3551, Aug. 2013. processes, 4th ed. NY: McGraw-Hill, 2002.
[16] F. Wang, C. Xu, L. Song, Q. Zhao, X. Wang, and Z. Han, “Energy-aware [38] G. L. Stüber, Principles of Mobile Communication, 3rd ed. New York,
resource allocation for device-to-device underlay communication,” in NY: Springer US, 2012.
Proc. IEEE Conf. Commun. (ICC’ 13), June 2013, pp. 6076–6080. [39] A. Abdi and M. Kaveh, “K distribution: an appropriate substitute for
[17] C. Xu, L. Song, Z. Han, Q. Zhao, X. Wang, X. Cheng, and B. Jiao, “Effi- rayleigh-lognormal distribution in fading-shadowing wireless channels,”
ciency resource allocation for device-to-device underlay communication Electron. Lett., vol. 34, no. 9, pp. 851–852, Apr. 1998.
systems: A reverse iterative combinatorial auction based approach,” [40] P. Theofilakos, A. Kanatas, and G. Efthymoglou, “Performance of
IEEE J. Sel. Areas Commun., vol. 31, no. 9, pp. 348–358, Sept. 2013. generalized selection combining receivers in K fading channels,” IEEE
[18] R. Yin, G. Yu, H. Zhang, Z. Zhang, and G. Y. Li, “Pricing-based in- Commun. Lett., vol. 12, no. 11, pp. 816–818, Nov. 2008.
terference coordination for D2D communications in cellular networks,” [41] D. West et al., Introduction to Graph Theory. Upper Saddle River, NJ:
IEEE Trans. Wireless Commun., vol. 14, no. 3, pp. 1519–1532, Mar. Prentice Hall, 2001.
2015. [42] P. Jänis, V. Koivunen, C. Ribeiro, J. Korhonen, K. Doppler, and
[19] R. Yin, C. Zhong, G. Yu, Z. Zhang, K.-K. Wong, and X. Chen, “Joint K. Hugl, “Interference-aware resource allocation for device-to-device
spectrum and power allocation for D2D communications underlaying radio underlaying cellular networks,” in Proc. IEEE 69th Veh. Technol.
cellular networks,” IEEE Trans. Veh. Technol., vol. PP, no. 99, pp. 1–1, Conf. (VTC Spring’ 09), Apr. 2009, pp. 1–5.
2015.
[20] D.-Q. Feng, G.-D. Yu, Y. Yuan-Wu, G. Y. Li, G. Feng, and S.-Q.
Li, “Mode switching for device-to-device communications in cellular
networks,” in Proc. IEEE Global Conf. on Signal and Inform. Process.
(GlobalSIP’14), Dec. 2014.
[21] D.-Q. Feng, G.-D. Yu, C. Xiong, Y. Yuan-Wu, G. Y. Li, G. Feng,
and S.-Q. Li, “Mode switching for energy-efficient device-to-device
communications in cellular networks,” IEEE Trans. Wireless Commun.,
vol. PP, no. 99, pp. 1–1, 2015.
[22] K. Doppler, C.-H. Yu, C. Ribeiro, and P. Jänis, “Mode selection
for device-to-device communication underlaying an LTE-advanced net-
work,” in Proc. IEEE Wireless Commun. and Networking Conf. (WCNC’
10), Apr. 2010, pp. 1–6.
[23] S. Wen, X. Zhu, X. Zhang, and D. Yang, “Qos-aware mode selection and
resource allocation scheme for device-to-device (D2D) communication

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 12

Daquan Feng received his B.S.E degree from Henan Geoffrey Ye Li (S’93-M’95-SM’97-F’06) received
University, Kaifeng, China, in 2008 and his Ph.D. his B.S.E. and M.S.E. degrees in 1983 and 1986,
degree from the National Key Laboratory of Sci- respectively, from the Department of Wireless En-
ence and Technology on Communications, UESTC, gineering, Nanjing Institute of Technology, Nan-
Chengdu, China in 2015. He had been a visiting jing, China, and his Ph.D. degree in 1994 from
student in the School of Electrical and Computer the Department of Electrical Engineering, Auburn
Engineering, Georgia Institute of Technology, USA, University, Alabama, USA.
from 2011 to 2014. After graduation, he joined He was a Teaching Assistant and then a Lecturer
the State Radio Monitoring Center (SRMC), Bei- with Southeast University, Nanjing, China, from
jing, China. His research interests include device- 1986 to 1991, a Research and Teaching Assistant
to-device communications, full-duplex communica- with Auburn University, Alabama, from 1991 to
tions, energy-efficient wireless network design, and heterogeneous network. 1994, and a Post-Doctoral Research Associate with the University of Maryland
at College Park, Maryland, from 1994 to 1996. He was with AT&T Labs -
Research at Red Bank, New Jersey, as a Senior and then a Principal Technical
Staff Member from 1996 to 2000. Since 2000, he has been with the School of
Electrical and Computer Engineering at the Georgia Institute of Technology
as an Associate Professor and then a Full Professor. He is also holding
the Cheung Kong Scholar title at the University of Electronic Science and
Technology of China since March 2006.
His general research interests include statistical signal processing and
communications, with emphasis on cross-layer optimization for spectral- and
energy-efficient networks, cognitive radios and opportunistic spectrum access,
and practical issues in LTE systems. In these areas, he has published over 300
refereed journal and conference papers in addition to 26 granted patents. His
publications have been cited over 21,000 times and he has been recognized
as the World’s Most Influential Scientific Mind, also known as a Highly-
Cited Researcher, by Thomson Reuters. He has been involved in editorial
activities for about 20 technical journals for the IEEE Communications and
Lu Lu received her B.S.E degree and M.S.E de-
Signal Processing Societies. He organized and chaired many international
gree from the University of Electronic Science and
conferences, including technical program vice-chair of IEEE ICC’03, technical
Technology of China (UESTC), Chengdu, China,
program co-chair of IEEE SPAWC’11, general chair of IEEE GlobalSIP’14
in 2007 and 2010, respectively. She then got her
and technical program co-chair of IEEE VTC’16 (Spring). He has been
licentiate degree from Royal Institute of Technology
awarded IEEE Fellow for his contributions to signal processing for wireless
(KTH) in 2011. She is currently working toward the
communications since 2006. He won 2010 Stephen O. Rice Prize Paper Award
Ph.D. degree with the School of Electricaland Com-
and 2013 WTC Wireless Recognition Award from the IEEE Communications
puter Engineering, Georgia Institute of Technology,
Society. He also received 2013 James Evans Avant Garde Award and 2014
Atlanta, GA, USA. Her research interests include
Jack Neubauer Memorial Award from the IEEE Vehicular Technology Society.
MIMO, cooperative communications, and cognitive
Recently, he won 2015 Distinguished Faculty Achievement Award from the
radio networks.
School of Electrical and Computer Engineering, Georgia Tech.

Yi Yuan-Wu received the Engineer degree in elec-


tronic from Huazhong University, Wuhan, China, in
1982, the Master of Sciences degree in 1983 and
Ph.D. in 1987 in signal processing and telecommu-
nication from Rennes University, Rennes, France.
Between 1983 and 1987, she done the Ph.D. research Gang Feng (M’01, SM’06) received his BEng. and
on the systems of digital diffusion to mobile (DAB) MEng degrees in Electronic Engineering from the
at CCETT of Rennes. Between 1989 and 1991, University of Electronic Science and Technology of
she worked in Thomson-LGT on video diffusion China (UESTC), in 1986 and 1989, respectively,
networks. Since 1992, she joined FranceTelecom and the Ph.D. degrees in Information Engineering
R&D (today’s Orange Labs), Issy-les-Moulineaux from The Chinese University of Hong Kong in 1998.
France. He joined the School of Electric and Electronic
Her working domains are the signal detection, channel estimation, broadcast Engineering, Nanyang Technological University in
channel and D2D communications underlaying cellular network for the mobile December 2000 as an assistant professor and was
systems. Between 1992 and 1996, she was in charge of the studying and promoted as an associate professor in October 2005.
specifying a CDMA Modem with variable flows. Between 1996 and 2000, At present he is a professor with the National Labo-
she studied the DECT and PHS radio link systems for the 64 kbits/s and ratory of Communications, University of Electronic Science and Technology
2 Mbits/s data transmission. Between 2000 and 2003, she was in charge of of China.
studying on the UMTS-TDD physical layer performance. Between 2003 and Dr. Feng has extensive research experience and has published widely in
2005, she worked on the MC-CDMA system within the European Matrice computer networking and wireless networking research. His research interests
and 4More projects. Now she is working for the European Metis and Sharing include resource management in wireless networks, wireless network coding,
projects on the multiuser MIMO and the D2D subjects. She is a senior research energy efficient wireless networking, etc. Dr. Feng is a senior member of
expert of Orange Labs since 2004. IEEE.

0018-9545 (c) 2015 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2479258, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. Y, MONTH 2015 13

Shaoqian Li received his M.E. degree in Informa-


tion and Communication Systems from University
of Electronic Science and Technology of China
(UESTC), Chengdu, China, in 1984 and his B. Eng
in Wireless Engineering from Northwest Institute
of Telecommunication Engineering (current Xidian
University), Xi’an, China, in 1981. He joined U-
niversity of Electronic Science and Technology of
China in 1984. Currently, he is Professor of UESTC
and Director of National Key Laboratory of Science
and Technology on Communications (formerly Na-
tional Key Lab of Communications) of UESTC. His research interest include
mobile and wireless communications, anti-jamming techniques for wireless
communications, frequency-hopping techniques, cognitive radio and spectrum
sharing technologies.
He has co-authored 2 books and published more than 40 referred journal
papers and 200 conference papers. He is inventor of more than 20 issued
patents and more than 50 filed Chinese Patents. He has received the 2nd
class National Award for Technological Invention of China in 2008 and the
2nd class National Award for Science and Technology Progress of China
in 2007. He received the Innovation and Excellent Award for contribution
to National Information Industrial, by Ministry of Industry and Information
Technology of P. R. China and Excellent Award for personal contribution to
National High-Tech Development Program (863) of P. R. China from 2001-
2005.
Since 1990s, he has served as Senior member of China Institute of Commu-
nications (CIC) and Chinese Institute of Electronics (CIE). And since 2001,
he had served for Intl. Conf. on Commun., Circuits, and Systems (ICCCAS),
in 2007, in 2008, in 2009, in 2010, in 2012, as Chair of Steering Committee,
and/or General (co-) Chair, respectively. He also served as consultant Member
of Board of Communications and Information Systems of Academic Degrees
committee of the State Council, P. R. China, and Member of expert group of
Key Project on Next-Generation Mobile Broadband Wireless Communications
Systems towards 2020, Ministry of Industry and Information Technology of
P. R. China. He is now IEEE Senior Member.

0018-9545 (c) 2015 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.

You might also like