Professional Documents
Culture Documents
Power Control For D2D Underlaid Cellular Networks: Modeling, Algorithms and Analysis
Power Control For D2D Underlaid Cellular Networks: Modeling, Algorithms and Analysis
Abstract—This paper proposes a random network model for D2D underlaid cellular networks. In particular, we consider a
a D2D underlaid cellular system using stochastic geometry and hybrid random network model using stochastic geometry and
develops centralized and distributed power control algorithms. develop two different power control algorithms for the pro-
The goal of the centralized power control is two-fold: ensure the
cellular users have sufficient coverage probability by limiting the posed network model. With a carefully designed (centralized)
interference created by underlaid D2D users, while scheduling power control technique, we show that multiple D2D links
as many D2D links as possible. For the distributed power may communicate successfully while guaranteeing reliable
control method, the optimal on-off power control strategy is communication for the existing cellular link. This shows that,
proposed, which maximizes the sum rate of D2D links. Analytical with an appropriate power control technique, underlaid D2D
expressions are derived for the coverage probabilities of cellular,
D2D links, and the sum rate of the D2D links in terms of the links help to increase the network sum-throughput without
density of D2D links and the path-loss exponent. The analysis causing unacceptable performance degradation to existing cel-
reveals the impact of key system parameters on the network lular links.
performance. For example, the bottleneck of D2D underlaid
cellular networks is the cross-tier interference between D2D links A. Related Work
and the cellular user, not the D2D intra-tier interference when There has been considerable interest in power control
the density of D2D links is sparse. Simulation results verify the
exactness of the derived coverage probabilities and the sum rate
techniques for D2D underlaid cellular networks. A simple
of D2D links. power control scheme was proposed in [5] for a single-cell
scenario and deterministic network model, which regulates
I. I NTRODUCTION D2D transmit power to protect the existing cellular links. To
Device-to-device (D2D) communication underlaid with cel- maximize the sum rate of the network, a D2D transmit power
lular networks allows direct communication between mobile allocation method was proposed in [6] for the deterministic
users [1]–[3]. D2D is an attractive approach for dealing network model. A dynamic power control mechanism for a
with local traffic in cellular networks. The initial motivation single D2D link communication was proposed in [7], which
for incorporating D2D communication in cellular networks targets improving the cellular system performance by mit-
is to support proximity-based services, e.g. social network- igating the interference generated by D2D communication.
ing applications or media sharing [3]. Assuming there are The main idea was to adjust the D2D transmit power via
proximate communication opportunities, D2D communication base station (BS) to protect cellular users. In [8], a power
may also increase area spectral efficiency, improve cellular minimization solution with joint subcarrier allocation, adaptive
coverage, lower end-to-end latency, or reduce handset power modulation, and mode selection was proposed to guarantee
consumption [4], [2]. In spite of these potential gains, the the quality-of-service demand of D2D and cellular users. In
coexistence of D2D and cellular communication in the same prior work [1], [2], [5], [7]–[17], D2D power control strategies
spectrum is challenging due to the difficulty of interference are developed and evaluated in a deterministic D2D link
management [2]. Specifically, the underlaid D2D signal be- deployment scenario. For a random network model, spectrum
comes a new source of interference. As a result, cellular links sharing between ad hoc and cellular networks was studied in
experience cross-tier interference from the D2D transmissions [18]–[20] but power control – an essential component of D2D
whereas the D2D links need to combat not only the inter- underlaid cellular networks – has not been addressed. Power
D2D interference but also the cross-tier interference from the control has been studied in other random ad hoc networks
cellular transmissions. Therefore, interference management is without considering cellular networks (see e.g. [21]–[23]). In
essential to ensure successful coexistence of cellular and D2D our paper, we propose power control algorithms and analyze
links. their performance in a D2D underlaid cellular network.
Power control is an effective approach to mitigate interfer-
ence in wireless networks; it is broadly used in current wireless B. Contributions
systems. In this paper, we propose power control methods In this paper, we consider a D2D underlaid cellular network
for interference coordination and analyze their performance in in which an uplink cellular user intends to communicate with
2
the BS while multiple D2D links coexist in the common D2D link 8
D2D link 3 D2D link 10
propose a centralized and a distributed power control algo- D2D link 7 D2D link 4
rithm. The main idea of the centralized algorithm is to design D2D link 9
lular and D2D links and the sum rate of D2D links. One yk =hk,k dk,k sk +hk,0 dk,0 s0 +
2 2
hk,` dk,`
2
s` +nk , (1)
important insight obtained from the analysis is that on-off `=1,`6=k
K
power control strategy for the uplink user is actually optimal in −α/2
X −α/2
terms of the coverage probability of the cellular link, agreeing y0 =h0,0 d0,0 s0 + h0,k d0,k sk + n0 , (2)
k=1
with the finding in ad hoc networks [23]. Further, we derive
the optimal D2D transmission probability which maximizes where subscript 0 is used for the uplink signal to the BS and
the sum rate of D2D links when the distributed on-off power subscript k, k 6= 0, are used for D2D links; sk and s0 denote
control algorithm is used. In contrast to the centralized power the signal sent by D2D transmitter k and the uplink user; yk
control method, the distributed power control algorithm is not and y0 represent the received signal at D2D receiver k and the
sufficient to guarantee reliable cellular communication, though BS; nk and n0 denote the additive noise at D2D receiver k
it does improve the cellular network throughput by additional and the BS distributed as CN (0, σ 2 ); hk,` and h0,k represent
D2D communication. We verify the results by simulating two the distance-independent fading from D2D transmitter ` to
different D2D link deployment scenarios. receiver k and the channel from D2D transmitter k to the
The remainder of this paper is organized as follows. In BS, and are independently distributed as CN (0, 1). Here, we
Section II, the proposed model for D2D underlaid cellular assume the distance dependent path-loss model, i.e., d−α k,` for
networks is described. The proposed power control algorithms all k, ` where dk,j denotes the distance from transmitter j
are presented in Section III. In Section IV, the analytical to receiver k and α is the path-loss exponent. The transmit
expressions for the coverage probabilities of the cellular and power satisfies the peak power constraints, i.e., |s0 |2 ≤ Pmax,c ,
typical D2D link are derived and validated through comparison |sk |2 ≤ Pmax,d for k ∈ {1, 2, . . . , K}.
with the simulation results. For the distributed power control, Then the SINR at D2D receiver k and the BS are given by
the sum rate of D2D links is derived in Section V. Simulation |hk,k |2 d−α
k,k pk
results are provided in Section VI to compare the performance SINRk (K, p) = K
(3),
|hk,0 |2 dk,0 p0 + `6=k |hk,` |2 d−α
−α P 2
of the proposed algorithms, which are followed by our con- k,` p` + σ
clusions in Section VII. |h0,0 |2 d−α
0,0 p0
SINR0 (K, p) = PK , (4)
II. S YSTEM M ODEL k=1 |h0,k |2 d−α
0,k pk + σ
2
In this section, we present the system model and describe where p = [p0 , p1 , . . . , pK ]T denotes transmit power profile
network metrics that will be used in this paper. vector with pi being the transmit power of transmitter i.
We consider a D2D underlaid cellular network, as shown Note that our system model ignores out-of-cell interference
in Fig. 1. In this model, let the circular disk C with radius R from macro users in the other cells. Nevertheless, the proposed
3
model is able to capture the effect of dominant interference, where Gk,` = |hk,` |2 d−α k,` and ` ∈ {1, 2, . . . , } and k ∈
which mainly determines the network performance. For the {1, 2, . . . , }. This optimization problem is compactly written
case of the uplink user transmission, the dominant interferer is in a vector form as
the nearest D2D transmitter in the cell because the interference g0T p
power of the nearest D2D transmission is strong than the max cT
p g0 p + σ2
power of the out-of-cell interference with high probability.
subject to (I − F) p ≥ b
For the case of the typical D2D transmission, the dominant
interferer to any D2D link is either the uplink transmission 0 ≤ p ≤ pmax , (9)
or the nearest D2D transmission in the cell. Therefore, our where g0T = [G0,0 , 0, . . . , 0], g0c T = [0, G0,1 , G0,2 , . . . , G0,K ],
system model captures the dominant interference effect for pmax = [Pmax, c , Pmax, d , . . . , Pmax, d ]T , and the normalized
the both uplink and D2D links; it is possible to offer a tight channel gain matrix F and target SINR vector b are defined
upper performance of the D2D underlaid cellular system. as
We are interested in the coverage probability of the cellular (
link and D2D links. The cellular coverage probability is 0, k = `,
Fk,` = βk Gk,`
defined as Gk,k , k 6= `.
T
(C) (C) β0 σ 2 β1 σ 2 β 2 σ 2 βK σ 2
P̄cov (β0 ) = E[Pcov (p, β0 )] = E[P(SINR0 (K, p) ≥ β0 )], (5)
b = , , , ..., .
G0,0 G1,1 G2,2 GK,K
where β0 represents the minimum SINR value for reliable
uplink connection. Similarly, the D2D coverage probability is Since the objective function (linear-fractional function) is
defined as quasi-convex and the constraint set is convex (a polytope in
(D) (D)
particular) with respect to power profile vector p, the optimal
P̄cov (βk ) = E[Pcov (p, βk )] = E[P(SINRk (K, p) ≥ βk )], (6) solution is able to be obtained by using standard convex
where βk represents the minimum SINR value for reliable programming tools, provided that the feasible set is nonempty.
D2D link connections. Further, we define the ergodic sum rate Note that the matrix F = [f0 , f1 , . . . , fK ] is comprised of
of D2D links as nonnegative elements and is irreducible because all the active
"K # D2D links interfere each other. By the Perron-Frobenious
(7) theorem, the following well-known lemma proved in [24] gives
X
(D)
R =E log2 (1+SINRk (K, p)) .
k=1
a necessary and sufficient condition on the feasibility of the
optimization problem (9).
III. P OWER C ONTROL A LGORITHMS Lemma 1: [24] The constraint set in the optimization
When the global CSI is available at the central controller, problem (9) is nonempty if and only if the maximum modulus
a centralized power control algorithm is proposed, which eigenvalue of F is less than one, i.e., ρ(F) < 1, where ρ(·)
maximizes the SINR of the cellular link while satisfying the denotes the spectral radius of a matrix.
SINR constraints for both the cellular link and D2D links. We next describe our proposed centralized algorithm to
Further, when the transmitter has CSI of the direct link of the solve the optimization problem (9). First we assume that D2D
corresponding receiver only, a distributed on-off power control receivers can feedback all the perfect normalized channel gains
algorithm is proposed. G i,k and target SINR information βi to the BS. Using this
assumption, the BS then computes the transmit power used
A. Centralized Power Control for both D2D transmitters and the uplink user. Note that
A main difference between ad hoc networks and underlaid the feasible set should be nonempty to obtain the optimal
D2D cellular networks is that centralized power control is solution, i.e., ρ(F) < 1. Since the normalized channel gains
possible when the D2D links are managed by the BS. For Gi,k , however, are random variables (the locations of all the
other management strategies, centralized power control is able transmit nodes are random variables), there exists a non-
to provide an upper bound on what can be achieved with more zero probability that the power control solution is infeasible,
decentralized algorithms. i.e., P({ρ(F) ≥ 1}) 6= 0, especially when the number of
Suppose that the BS has global channel state information D2D links K is large. When the solution is infeasible, an
(CSI). Under this assumption, the centralized power control admission control method is needed in conjunction with the
problem is formulated as power control algorithm to provide a feasible solution to
the power control problem by selecting a subset of D2D
G0,0 p0
max PK links. This D2D link selection problem PK may be solved by
{p0 ,p1 ,...,pK } G p + σ 2 K
k=1 0,k k brute-force search, which requires r=1 r computations.
G0,0 p0 The computational complexity grows exponentially with K.
subject to PK ≥ β0
2 Instead of brute-force search, we propose an efficient D2D link
k=1 G0,k pk + σ
Gk,k pk selection algorithm with low computational complexity for this
PK ≥ βk , problem. The key idea is to drop D2D communication links
Gk,0 p0 + `6=k Gk,` p` + σ 2
successively that causes the maximum sum of the interference
0 ≤ p0 ≤ Pmax, c , power in the network until the feasibility condition is satisfied.
0 ≤ pk ≤ Pmax, d , (8) For K given D2D links, we first test feasibility condition
4
TABLE I
P ROPOSED C ENTRALIZED P OWER C ONTROL A LGORITHM
Step Algorithm
Initialization Set initial F` for ` = 0, assuming K D2D links are all active.
Step 1 Test feasibility condition ρ(F` ) < 1. If this condition is satisfied, go to Step 5. Otherwise, go to Step 2.
Step 2 Pick the column of F` such that k̂ = arg maxk∈K/{0} kfk` k2 .
Step 3 Generate a reduced matrix F̂` by removing the k̂-th column and row vectors in F` .
Step 4 Update F`+1 = F̂` by increasing ` = ` + 1. Go to Step 1.
Step 5 Solve the optimization problem in (9).
of the optimization problem in (9). If ρ(F) > 1, i.e., the the randomness in the network modeling is not a key part of
feasibility set is empty, we select the k̂-th D2D transmitter the algorithm, rather it is a component of the analysis to show
such that it creates the maximum sum of interference power to that it works.
all other receivers, i.e., k̂ = arg maxk∈K/{0} kfk k2 , and then
accordingly we remove the k̂-th row and column to reduce IV. C OVERAGE P ROBABILITY A NALYSIS F OR
the size of matrix F. We keep reducing the size of matrix F D ISTRIBUTED P OWER C ONTROL
until the feasibility condition is satisfied. Table I summarizes In this section, we derive the cellular link coverage probabil-
the proposed D2D link selection method in conjunction with ity, propose an optimal power control strategy for the cellular
power control. link under the average transmit power constraint, and derive
the D2D link coverage probability. To analyze the coverage
B. Distributed On-Off Power Control Algorithm
probabilities using the tool of stochastic geometry, we assume
In this subsection, we provide a distributed power con- that the transmit power of each D2D transmitter is i.i.d. with
trol algorithm. The distributed power control is an effective distribution function Fpk (·) and that the transmit power of
interference mitigation method that requires no coordination the uplink user is independent and has distribution function
between transmitters; the signaling overheads for sharing CSIT Fp0 (·). Note that the coverage probability analysis we provide
is not needed. In the absence of coordination, each D2D in this section is valid for any distributed power control
transmitter chooses its transmit power to maximize its own algorithms that select its own transmit power independently
rate towards its intended receiver, disregarding the interference of the transmit power used at the other D2D transmitters.
caused to the others. The proposed on-off method is to select
the D2D transmit power from the decision set {0, Pmax, d } A. Cellular Link Coverage Probability
solely based on knowledge of the direct link information
Assume that the BS is located at the origin. The SINR of
and a nonnegative threshold Gmin that is fixed and known
the typical uplink is given by
by all users. Specifically, the power used by D2D pair k
is Pmax, d when the link quality is good in the sense that p0 |h0,0 |2 d−α
0,0
SINR0 = P . (12)
|hk,k |2 d−α
k,k > Gmin , and 0 otherwise. Mathematically, pk |h0,k |2 d−α 2
k∈Φ 0,k + σ
Pmax, d with Ps Further, since the cellular user’s location is distributed uni-
pk = (10)
0 with 1 − Ps . formly in the circle with radius R, the distribution function of
the distance d0,0 of the cellular link is given by
where Ps denotes the transmit probability given by
0 if r < 0;
Ps = P[|hk,k |2 d−α α
k,k > Gmin ] = exp −Gmin dk,k . (11)
Fd0,0 (r) = r2
if 0 ≤ r ≤ R;
R2
Note that the proposed power control method is distributed 1 if r ≥ R.
as each D2D transmitter decides its transmit power by the The following theorem provides an analytical formula for the
own channel gain |hk,k |2 and threshold Gmin . For a given uplink coverage probability.
distribution of the channel gain, selecting a proper threshold
Gmin (the transmission probability Ps ) plays an important role Theorem 1: The cellular link coverage probability is
in determining the sum rate performance of the D2D links.
2
On the one hand, choosing a large Gmin (a small Ps ), reduces
(C)
P̄cov (β0 ) = EX e−a1 X−a2 X α , (13)
the inter-D2D interference. On the other hand, larger Gmin 2 h 2i
(smaller Ps ) leads to smaller number of active D2D links πλβ α
where a1 = σ 2 β0 , a2 = sinc( 02 ) E pkα , X = p−1 α
0 d0,0 with
within the disk. Therefore, a good choice of Gmin balancing R 1 1
α
these two competing factors leads to achieve a high D2D sum cdf FX (x) = Fd0,0 (x α p α )dFp0 (p).
rate performance. This motives us to optimize the Gmin (Ps ) Proof: See Appendix A.
for maximizing the D2D sum rate performance. That problem
Theorem 1 provides an intuition that how important network
is tackled in Section V.
parameters affect the cellular link coverage probability. For
Remark: The proposed power control algorithm may be (C)
useful in non-random networks because it can be applicable example, we observe that P̄cov (β0 ) depends
h 2 i on two D2D-
in any realization of the proposed random network. Therefore, related network parameters: λ and E pkα . In particular,
5
(C)
P̄cov (β0 ) decreases as the density λ of D2D transmitters
1
increases, which is intuitive as higher D2D link density causes λ=0.00002 (Analytic)
more interference to the cellular link. Further, the random 0.9 λ=0.00005 (Simulation)
(C)
D2D power control pk affects P̄cov (β0 ) only through its α2 -th λ=0.00005 (Analytic)
0.8
moment. This implies that the system can control the λ=0.00002 (Simulation)
h 2impact
i
of D2D links on the cellular link by constraining E pkα and 0.7
P [SIR>β ]
0
link coverage probability.
0.5
Example 1 (A Closed Form Expression): Let us consider
0.4
the case where the uplink user uses a constant transmit power
p0 = Pmax,c and ignore the noise σ 2 = 0. With path-loss 0.3
exponent α = 4, a standard value in terrestrial outdoor wireless
(C) 0.2
systems, the expression of P̄cov (β0 ) simplifies substantially as
Z R ! 0.1
(C) a2 2r
P̄cov (β0 ) = exp p r2 2
dr 0
0 Pmax,c R −5 0 5 10 15 20
SIR (dB)
√
πλ √β0 √
1 − exp − E[ pk ]R2 Fig. 2. Coverage probability performance of the uplink user with a set of
sinc(1/2) Pmax,c parameters dk,k = 50 m, R = 500 m, Pmax,c = 100 mW, Pmax,d = 0.2
= √
πλ √β0 √ . (14) mW, P[Gk,k > Gmin = 0.5], and λ ∈ {0.00002, 0.00005}.
E[ pk ]R2
sinc(1/2) Pmax,c
Further, if the D2D transmitters send the signal using power coverage probability of the cellular link is a function of
Pmax,d with probability Ps = 0.5, the coverage probability the transmit power p0 and the distance d0,0 of the uplink
becomes user. Conditioning on the location for the uplink user, i.e.
d0,0 = d, the cellular link coverage probability is reduced
Pmax,d √
2
q
1 − exp − π(λ/2)R
2
β α −1 2 −α
(C) sinc(1/2) Pmax,c 0
as φ(p0 ) = e−a1 d p0 −a2 d p0 . Under the average and peak
P̄cov (β0 ) = . (15)
Pmax,d √
q
π(λ/2)R2
β power constraints of the uplink transmission power p0 , the
sinc(1/2) Pmax,c 0
optimal distribution function of p0 , i.e., Fp0 , is obtained by
This expression explicitly shows that the coverage perfor- solving the following optimization problem:
mance of the cellular link is jointly determined by three Z
factors: 1) the average number of active D2D transmitters maximize φ(p0 )dFp0 (p0 )
E[K] = π(λ/2)R2 ,q2) the power ratio between the cellular Z
Pmax,d
and the D2D user Pmax,c , and 3) the target threshold β0 .
subject to p0 dFp0 (p0 ) = Pavg, c
To validate our analysis, we compare the coverage probability Z
expression in (15) with the simulation result. As illustrated dFp0 (p0 ) = 1
in Fig 2, the coverage probability performance of the uplink
p0 ≤ Pmax,c . (17)
user is well matched with the corresponding Monte Carlo
simulation over the entire range of β0 and different λ ∈ Note that (17) is an infinite-dimensional optimization problem
{0.00002, 0.00005}. because the distribution function Fp0 may have an infinite
(C)
We next provide a simple lower bound for P̄cov (β0 ), which number of degrees of freedom. Although this class of opti-
is useful for an arbitrary path-loss exponent value and noise- mization problem is not solvable in general, we are able to find
limited case. The lower bound simply depends on the certain the optimal solution of the distribution function Fp0 thanks to
moments of p0 and d0,0 (rather than the distributions). This a remarkably simple structure for the optimization problem in
lower bound is formalized in the following corollary. (17), which is stated in the following theorem.
(C)
Corollary 1: Cellular link coverage probability P̄cov can
Theorem 2: There exists a non-degenerate p? (d) ∈ (0, ∞)
be lower bounded as
that maximizes φ(p 0)
p0 . Further, conditional on d0,0 = d the
(C) (C)
P̄cov (β0 ) ≥ P̄cov,lb (β0 ) optimal power allocation strategy maximizing the cellular link
α
·E[p−1
2 2 −1
2 coverage probability is on-off power control.
0 ]−a2 ( 2+α ) R ·(E[p0 ])
2 2
= e−a1 2+α R
α α
. (16)
Proof: See Appendix C.
Proof: See Appendix B. From Theorem 2, the on-off power control strategy provides
the cellular user with the optimal coverage probability perfor-
mance and the optimal transmission power p?0 (d) is a maxi-
B. Optimal Cellular Link Power Control Strategy
mizer of the function φ(p 0)
p0 . Although the exact expression of
In this subsection, we provide an optimal power control ?
p0 (d) is difficult to obtain, we are able to find a closed form
strategy for the cellular link when the cellular user has location expression in the interference limited regime, i.e., σ 2 = 0.
information of distance d0,0 . As shown in Theorem 1, the
6
Corollary 2: For the interference limited regime (σ 2 = 0), target SIR β0 . In the regime of p̃0 (d) ≥ Pmax, c (i.e., the user is
the optimal transmit power of the cellular user placed at the located at around the cell edge), the cellular user sends a signal
Pavg,c
distance d0 = d with respect to the BS is with its maximum transmit power with probability Pmax,c due
to the peak power constraint; thus, the coverage probability
p?0 (d) = max {min {p̃0 , Pmax,c } , Pavg, c } . (18) 2
decreases linearly with respective to β0α again.
2 α2 We provide an example to help the understanding of three
2E[pkα ] d α
where p̃0 (d) = αsinc( 2 ) E[K]α/2 β0 R . different behaviors on the coverage probability performance.
α
Example 2 (Three Different Behaviors of the Cellular
Proof: Let x = p10 . Then, for the interference limited Link Coverage Probability): In this example, let us consider
regime, the objective function for the uplink power optimiza- a set of typical parameters: the path-loss exponent α = 4, the
tion problem becomes φ(x)x = x exp −a2 d2 x2/α . From the cell radius R = 500m, the target SINR β0 = 6 dB, the average
first order optimality condition, i.e., ∂φ(x)x
∂x = 0, we obtain number of D2D links E[K] = λπR2 = 39, and the average
2
α πλβ0α 2
and maximum transmit power constraints of the cellular user
the maximizer x∗ = ( 2aα2 d2 ) 2 . Putting a2 = 2 E[pk ]
sinc( α )
α
=
2
Pavg, c = 0.1 W and Pmax, c = 0.2 W. Further, we assume that
2
E[K]β0α the D2D links use a constant transmit Pmax,d = 0.0001
power √
2
R2 sinc( α
E[pk ] and using p∗0 = x1∗ , we obtain the p̃0 (d) in
)
α
√ p
(18). Since the transmit power should satisfy the maximum W, which gives us E[ pk ] = Pmax, d = 0.0001. In this
transmit power constraint, we take the minimun value between set of parameters, the transmit power of the cellular link is
p̃0 (d) and Pmax,c . Note that since the cellular user uses binary expressed in terms of the distance d as
power control, the cellular user’s transmit probability becomes 1 !2 2
Pavg, c E[pk2 ] 2 d
p? (d) ≤ 1 . p̃0 (d) = E[K] β0
0
2sinc( 21 ) R
Using the optimal cellular user transmit power obtained in 2
Corollary 2, we have a closed form expression on the cellular d
' 0.375 × W. (21)
user coverage probability for the interference-limited regime. R
Corollary 3: For the interference limited regime (σ 2 = 0) If the cellular user is located in the half of cell ra-
and a given uplink user distance d, the cellular user coverage dius d = R2 , the optimal uplink transmission power
probability becomes as in (19). p?0 (R/2) = max {min {p̃0 , Pmax, c } , Pavg, c } = 0.1 W because
Proof: Recall that under the assumptions of interference of p̃0 (R/2) ' 0.093 < 0.1 W, implying that the average
limited regime and the fixed distance of uplink user, the transmit power is used in this regime. Thus, the cellular user
(C)
cellular user coverage
(C)
probability becomes P̄cov (d; β0 ) = coverage probability becomes P̄cov (R/2, β0 ) ' 0.743. Alter-
natively, if we consider that the cellular user is located at mid
2
2 −α
Ep0 e−a2 d p0
. Since the optimal power control strategy range of the cell edge with d = 0.7R, then the transmission
of the uplink user is the binary power control, i.e., p0 = p?0 (d) power of the cellular user becomes p?0 (0.7R) = 0.183 W,
P c P c which means that the uplink user opportunistically sends its
with probability p?avg, and p0 = 0 with probability 1 − p?avg, ,
0 (d) 0 (d)
the coverage probability expression is reduced as uplink signal using transmit power 0.183 W with probability
of 100
183 . Therefore, it gives a coverage probability performance
Pavg, c
(C)
(C)
P̄cov (d; β0 ) = ? exp −a2 d2 p?0 (d)−2/α . (20) P̄cov (0.7R, β0 ) ' 0.3298. For the regime of d = R, the
p0 (d)
optimal transmit power of the cellular link p?0 (R) = 0.2 W
2
E[K]β α 2 (the maximum transmit power); thus, the coverage probability
Using the solution of p?0 (d) in (18) and a2 = R2 sinc(02 ) E[pkα ], (C)
performance in this regime is P̄cov (R, β0 ) ' 0.216.
α
we obtain the desired coverage probability expression.
(C)
The coverage probability of the cellular user P̄cov (d; β0 ) C. D2D Link Coverage Probability
behaves in three different ways according to the location of We derive an expression for the coverage probability for
the user d. When p̃0 (d) < Pavg, c (i.e., the user is located the typical D2D link. Consider an arbitrary communication
at around the cell center), the cellular user uses the constant D2D pair k and assume that the D2D receiver is located at
transmit power Pavg, c ; this results in the coverage proba- the origin. Then,
2
bility decreases exponentially with respective to β0α . When
Pavg, c < p̃0 (d) < Pmax, c (i.e., the user is located at mid pk |hk,k |2 d−α
k,k
SINRk = P ,
range of the cell edge), the on-off power control strategy is x∈Φ\{k} pi |hk,i |2 ||xi ||−α + p0 |hk,0 |2 d−α
k,0 + σ
2
activated. In this regime, the cellular user increases its transmit (22)
power proportionally to dα , implying that the cellular user
where ||xi || = dk,i . Using the same approach we used to
should increase the transmit power according to the inverse of
prove
h Theorem 1, we
i need toh compute two Laplace transforms
path-loss, agreeing with intuition. Further, the uplink user is 2 −α P 2 −α
i
required to increase the transmit power linearly according to E e−sp0 |hk,0 | dk,0 and E e−s x∈Φ\{k} pi |hk,i | ||xi || to
α
E[K] 2 where K is the random number of D2D links in the derive the distribution of hSINRk . i
2 −α
coverage of the BS. From this on-off power control strategy, First let us focus on E e−sp0 |hk,0 | dk,0 . As we assume the
the coverage probability decreases linearly with respect to the uplink user and the D2D receiver are randomly positioned in
7
2
2
Pavg, c E[K]β α d 2 −2/α
Pmax, c exp − sinc( 20) E[pkα ] R Pmax,c , for p̃0 (d) ≥ Pmax, c ,
α
(
Pavg, c exp 2
−α )
, for Pavg, c < p̃0 (d) < Pmax, c ,
(C)
2
α/2
P̄cov (d; β0 ) = 2E[p α ] d α
(19)
k
αsinc( 2 )
E[K]α/2 β0 ( )
R
α
2
2
E[K]β0α d 2
−2/α
exp −
2 E[p ]
α
k R Pavg, c for p̃0 (d) ≤ Pavg, c .
sinc( α )
the disk with radius R, the pdf fdk,0 (r) is given by [25]
! 1
r
2 λ=0.00002 (Simulation)
2r 2 r r r
fdk,0 (r)= 2 cos−1 − 1− 2 , 0 ≤ r ≤ 2R. 0.9 λ=0.00002 (Analytic)
R π 2R πR 4R λ=0.00002 (Approximated)
0.8 λ=0.00005 (Simulation)
(23) λ=0.00005 (Analytic)
λ=0.00005 (Approximated)
Besides, |hk,0 |2 is a random variable with the exponential 0.7
distribution, i.e. |hk,0 |2 Exp(1) and p0 has cdf Fp0 (p). Noting 0.6
further that p0 , |hk,0 |2 , and dk,0 are independent, we have
P[SINR>β]
λ=0.00002
0.5
h i ZZ ∞Z 2R
−sp0 |hk,0 |2 d−α −α
E e k,0 = e−sphr −h fdk,0 (r)drdhdFp0 (p). 0.4
0 0
(24) 0.3
With a similar approach as in the previous subsection, it is
0.2 λ=0.00005
possible to derive the complementary cumulative distribution
function (ccdf) of SINRk , and the coverage probability for the 0.1
typical D2D link is given in the following theorem.
0
−5 0 5 10 15 20
Theorem 3: The coverage probability of the typical D2D SIR (dB)
link is given by
Fig. 3. Coverage probability performance of the typical D2D link with a set
2 of parameters dk,k = 50 m, p0 = 100 mW, pk = 0.1 mW, R = 500 m,
(D) −b1 Z−b2 Z α
P̄cov (β) = EZ e L̂Y (βZ)] , (25) and λ ∈ {0.00002, 0.00005}.
2 1 1
2
where b1 = σ β, b = E[pk ], Z = p−1
πλβ α α
2
an approximation of E 1+ dακ ' , which is ob-
k dk,k with
α κ2/α
2 2
sinc( α ) 1+ E[d
k,0
k,0]2
1 1
(x p i)dF (p), Y = p0 |hk,0 |2 d−α
R
cdf FZ (z) =h F dk,k α α
pk k,0 and
tained from numerical observations, we have an approximated
−sp0 |hk,0 |2 d−α expression on the coverage probability as
LY (s) = E e . k,0
!
2
(D) πλβ α 2 1
Proof: See Appendix D. P̄cov (β) ' exp − 2 dk,k 2/α d2 ,
sinc( α )
To shed further light on the significance of the expression 1+ β ppk0 k,k
E[dk,0 ]2
derived in Theorem 3, it is instructive to consider a special 2
!
πλβ α 2 1
case where all D2D transmitters communicate with their = exp − 2 dk,k 2/α ,
sinc( α ) d2k,k
corresponding receivers with a fixed distance dk,k using a fixed 1 + β ppk0 (128R/(45π))2
transmit power pk . When the BS uses also a constant transmit (27)
power p0 , we are able to derive a closed from expression
of the coverage probability for the typical D2D link in the where the equality follows from the first moment of dk,0 ,
interference limited regime σ 2 = 0 as E[dk,0 ] = 128R
45π given in [25].
2
! d α To validate our analysis, we compare the analytic expres-
(D) πλβ α − α2 α2 2 p
−β p 0 dk,k |hk,0 |2 sions (exact and approximated) in (26) and (27) with simula-
P̄cov (β) = exp − p p d E e k k,0
sinc( α2 ) k k k,k tion results. Fig. 3 depicts the analytical expressions alongside
2
! the result of the corresponding Monte Carlo simulation for
πλβ α 1 the entire range of β and different λ. The agreement is
= exp − d2 E α ,
sinc( α2 ) k,k
1+β p0 dk,k accurate. Further, the approximated expression on the coverage
pk dk,0
probability in (27) provides a very precise approximated
(26) performance, especially when α = 4 case.
where the second equality comes from the fact that |hk,0 |2 ∼
exp(1) and the expectation in (26) is over dk,0 . V. S UM R ATE A NALYSIS OF D2D L INKS
We further consider a simple but approximated expression In this section, we analyze the sum rate of D2D links when
of the coverage probability for the typical D2D link. With the proposed on-off power control is applied and characterize
8
the optimal threshold of the on-off power control, which approximated transmission capacity of D2D links given as
maximizes the sum-rate of D2D links.
R(D) (β) = λ̃πR2 log2 (1 + β)P[SIRk ≥ β]
A. Sum Rate of D2D Links 2
!
π λ̃β α d2
k,k log2 (1 + β)
Let us denote the normalized inter-D2D P link interference ' λ̃πR2 exp − (32)
sinc( α2 )
2
power at the k-th D2D receiver as Ik = `6=k |hk,` |2 d−α 1+ κβ α
k,` for
2
!
k, ` ∈ {1, 2, . . . , |S|} where |S| denotes the number of active 2
πλPs β α d2k,k log2 (1 + β)
links selected by the proposed on-off power control algorithm, = λPs πR exp − . (33)
sinc( α2 )
2
1+ κβ α
i.e., |S| = λP[|hk,k |2 d−α 2 2
k,k ≥ Gmin ]πR = λ̃πR . Further,
Pmax, c
let p̃ = Pmax, d denote the transmission power ratio between To this end, we first compute the optimal transmission prob-
the D2D transmitter and the uplink user. Assuming Gaussian ability Ps by solving the optimization problem:
signal transmission from all the active links, the distribution of
the interference becomes Gaussian. Then, the achievable sum max R(D) (β)
rate of D2D links is written as subject to 0 < Ps ≤ 1 (34)
|hk,k |2 d−α
"K !#
k,k
X
(D)
R =E log2 1+ Although the objective function is not concave, the optimal
Ik + |hk,0 |2 d−α
k,0 p̃
k=1 solution of Ps can be obtained by using the first order
= |S|E [log2 (1 + SIRk )] , optimality condition since the objective function has an unique
= λ̃πR2 × R̄d2d . (28) optimum point. The first order optimality condition yields
2
Using the SIR distribution of the typical D2D link given in πλβ α d2k,k
(26), the ergodic rate of the typical D2D link can be rewritten 1− Ps = 0, (35)
sinc( α2 )
as
Z ∞
2
sinc( α )
R̄d2d = log2 (1+x)P[SIRk ≥ x]dx from which we have Ps? = min 2 , 1 . Finally, since
0 πλβ α d2k,k
Z ∞ (D)
Pcov (x) Ps = P[|hk,k |2 d−α
k,k > Gmin ], the optimal on-off threshold can
= dx be obtained as
0 1+x
! − ln(Ps? )
2
∞ π λ̃x α d2k,k G?min =
Z
1 1 . (36)
= exp − E α dx dα
0 1+x sinc( α2 ) d
1+xp̃ k,k
k,k
dk,0
(29) Using the solution of Ps? , the approximated transmission
capacity in (33) can be expressed as
is taken over dk,0 . Using an approxi-
where the expectation
2
1 1 πλβ α d2
mation of E 1+ ακ ' κ2/α
, we have an approximated λπR2 exp − sinc( 2 k,k
log2 (1+β)
, for β < β̃,
d 1+ E[d 2
k,0
k,0 ]
2 (D) α) 1+κβ α
expression of the ergodic rate of the typical D2D link in (29) R (β) ' 2
sinc( α ) R β − α2 log2 (1+β)
2
for β > β̃.
as exp(1) dk,k 2
1+κβ α
Z ∞ 2
π λ̃x α d2k,k
! (37)
1 1
R̄d2d ' exp − 2 2 dx
0 1+x sinc( α ) 2
d α2
1+(xp̃) α E[dk,k
k,0 ]
sinc( α
2
)
where β̃ = πλd2k,k
. The transmission capacity of the D2D
(30)
Z ∞ 2
2
! links behaves differently depending on the relative relationship
1 π λ̃x α dk,k 1 between the target SINR value β and network parameters:
= exp − 2 2 dx, (31)
0 1+x sinc( α ) 1+ κx α path-loss exponent α and the density of D2D links λ, and the
2 2 distance of D2D link dk,k . In the case where β is smaller
Pmax,c α dk,k
α
where κ = Pmax,d 128R/(45π) . Interestingly, the approxi-
sinc( α
2
) 2
than πλd2 , all D2D transmitters are scheduled, which
mated expression of the ergodic rate of the typical D2D link in k,k
(31) is determined by two factors: (1) the Laplace transform of leads to achieve the same performance with that of no power
the total interference power control. Meanwhile, when β is large enough, the D2D links
created2
by all active links on the are scheduled with the transmission probability Ps? , which
π λ̃x α d2k,k
entire network, i.e., exp − sinc( 2 ) and (2) the approximated results in mitigating the inter-D2D interference. In particular,
α α
effect of the uplink interference 1
. sinc( α
2
) 2
1+κx α
2 in the case of β > πλd2 , the transmission capacity of
k,k
TABLE II
T HE S UM R ATE P ERFORMANCE OF D2D L INKS
1
800 0.7
0.6
600
P [SIR>β]
0.5 D2D Link
400
Uplink user 0.4
P <1
s
200 Centralized PC
D2D Rx 0.3
On−Off PC
D2D Tx
0 0.2 No PC
Centralized PC
BS
0.1 On−Off PC
−200
No PC
0
−400 −5 0 5 10 15 20
SIR (dB)
−600
Fig. 5. Coverage probability performance of both the cellular and D2D links
−800
−800 −600 −400 −200 0 200 400 600 800
according to different power control methods when the D2D links are sparse,
i.e., λ = 0.00002.
Fig. 4. A snap shot of link geometry for a D2D underlaid cellular
network when the dense D2D link deployment scenario, i.e., λ=0.00005.
In our simulation, we consider out-of-cell interference created by the D2D
transmission for the warp-around effect.
is isotropically dropped at a fixed distance dk,k = 50m away
from the D2D transmitter. Since the D2D communication are
By integration of the transmission capacity in (37) with supposed to be of short range compared to the cellular link,
respect to β, we have the sum rate of D2D links as we assume that the average transmit power of the cellular
2
! user and D2D links are equal to Pmax, c = 100 mW and
πλx α d2k,k
Z β̃
(D) λπR2 Pmax, d = 0.1 mW. Since the number of D2D links K is
R ' exp − dx a random variable and we evaluate the coverage probability
sinc( α2 )
2
0 (1+ κx α )(1 + x)
Z ∞ 2 2 and sum rate performance of the proposed algorithms by
sinc α2
x− α R averaging 1000 independent realizations. Further, the optimal
+ 2 dx. (38)
β̃ (1+ κx α )(1 + x) exp(1) dk,k transmission scheduling parameter Gmin is obtained as in (36).
To validate our analysis, we compare the analytical result of The parameters used in the simulations are summarized in
the D2D sum rate with that obtained through Monte Carlo Table III.
simulation. Table II shows the sum rate performance of the 2) Coverage Probability Comparison in Sparse D2D Link
D2D links under different D2D line densities when α = 4, Deployment: Suppose the sparse D2D link deployment sce-
and dk,k = 50 m, R = 500 m, Pmax,c = 100 mW, and nario where the average number of D2D links in the cell equals
Pmax,d = 0.1 mW. It can be seen that the analytcial results E[K] = πR2 λ = 15.7. In this scenario, we compare the cov-
well match the simulation results. erage probability of the cellular link and the D2D links under
different D2D power control algorithms. As shown in Fig. 5,
we observe that the proposed power control methods improve
VI. S IMULATION R ESULTS the cellular user coverage probability. The proposed power
In this section, we provide numerical results for the D2D control methods also provide increased D2D link coverage
underlaid cellular system. From our simulation results, we first probability compared to the no power control case, especially
show the performance gain of the proposed power control in the high target SINR regime. This implies that the power
methods compared to the no power control case in terms of control methods are efficient to mitigate both intra-D2D and
the cellular user and the D2D user coverage probability. cross-tier interference when D2D links communicate with a
1) Simulation Setup: Fig. 4 shows one snap shot of the high data rate. In particular, one remarkable observation is that
cell geometry. As illustrated, the BS is located at the center the centralized power control achieves nearly perfect cellular
position (0, 0) in R2 plane and the cellular user is uniformly user coverage probability performance, i.e., (no outage) in
dropped within the range of R = 500 m. The D2D transmitters the low target SINR values, while successfully supporting
are dropped according to PPP with the density parameter a large number of active D2D links (48 %) when target
λ ∈ {0.00002, 0.00005} in a ball centered at the origin and SNIR β = 3 dB. Meanwhile, the on-off distributed power
the radius of R + 250 m so that the average number of D2D control method yields performance gains for both cellular
links are E[K] = πR2 λ ∈ {15, 39} while removing cell and D2D links compared to that of no power control case
edge effect on the D2D link performance. Further, for a given when the target SINR is larger than 12 dB. This is because
D2D transmitter’s location, the corresponding D2D receiver the proposed on-off power control method provides the same
10
TABLE III
S IMULATION PARAMETERS
Paramers Values
Cell radius (R) 500 (m)
The D2D link range (dk,k ) 50 (m)
D2D link density (λ) 0.00002 and 0.00005
Average number D2D links (K) E[K] = πR2 λ ∈ {15, 39}
Path-loss exponent (α) 4
Target SINR threshold (β) from -6 to 21 (dB)
The maximum transmit power of the cellular user Pmax, c = 100 mW
The maximum transmit power of the D2D transmitters Pmax, d = 0.1 mW
Noise variance (σ 2 ) for 1MHz bandwidth -143.97 (dBm)
The number of realizations 1000 geometry drops
1
greater than 3 dB. Although the D2D user coverage probability
Centralized PC
performance decreases in the dense scenario, the total number
0.9
On−Off PC of successful D2D transmissions is large than that of the
0.8 No PC sparse D2D link deployment scenario. For example, when
Centralized PC
0.7 the target SINR is 3 dB, the total numbers of successful
On−Off PC
No PC D2D transmissions in both sparse and dense scenarios are
0.6 (D)
about |S|sparse = E[KPcov (3)] = 15 × 0.5 ' 7.5 and
P[SIR>β]
Cellular Link
0.5 (D)
|S|dense = E[KPcov (3)] = 39 × 0.27 ' 10, respectively.
Ps<1
0.4
the distance d0,0 = d from the cellular transmitter to BS, we and consider the following relaxed conditional cellular link
next compute the second term (39).
h ToPthis end, we2 need−αthei coverage optimization problem:
Laplace transform LΦ (s) = E e−s( k∈Φ pk |h0,k | ||xk || )
Z
max φ(p)dFp0,0 (p)
given as
Z
2 π2 2
2
− α
sin( 2 π)
E pkα λs α subject to pdFPC (p) = Pavg, c . (47)
LΦ (s) = e α . (40)
p
Let dL(p) = Pavg, dFp0,0 (p). Then the above optimization
Using LΦ (s) yields c
problem is equivalently formulated as
2
2 π2
2
2 Z
− α λβ0α E pkα d2 p− α φ(p)
(C) −σ 2 β0 p−1 dα sin( 2 π)
max Pavg, c · dL(p)
P̄cov|p0 =p,d0,0 =d =e e α .
p
(41) Z
subject to dL(p) = 1. (48)
De-conditioning with respect to p0 and d0,0 yields the uplink
(C)
coverage probability P̄cov . The last step is to derive the whose optimal solution is L? (p? (d)) − L? (p?− (d)) = 1 and
probability distribution of X = p−1 α
0 d0,0 : L? (p) = 0 for p 6= p? (d). Therefore, we conclude that the
binary power control strategy is optimal.
FX (x) = P(p−1 dα ≤ x)
Z 0 0,0
1
= P(d0,0 ≤ (xp) α )dFp0 (p) D. Proof of Theorem 3
Z To prove Theorem 3, we need to derive the ccdf of SINRk .
1 1
= Fd0,0 (x α p α )dFp0 (p). (42) To this end, using Slivnyak’s theorem [26], it is easy to see
that
h P 2 −α
i
LΦ\{k} (s) = E e−s x∈Φ\{k} pi |hk,i | ||xi || |k ∈ Φ = LΦ (s)
B. Proof of Corollary 1
2
2 2
πλ
−a1 x−a2 x α − E pkα s α
Let φ(x) = e . We compute the first and second =e sinc( 2 )
α . (49)
derivative of φ(x) as follows:
It follows that
0
−a1 x−a2 x α
2
2 2 −1
φ (x) = −e a1 + a2 x α , (43) P(SINRk ≥ β) (50)
α
pk |hk,k |2 d−α
!
2
2
2 2 k,k
00
φ (x) = e−a1 x−a2 x α a1 + a2 x α −1 =P P 2 ||x ||−α +p |h 2 −α
≥β
α p |h
x∈Φ\{k} i k,i | i 0 k,0 | dk,0 + σ2
2
−a1 x−a2 x α 2 2 2
+e a2 1− x α −2 . (44) X
α α =P|hk,k |2 ≥ βp−1 α
k dk,k pi |hk,i |2 ||xi ||−α+p0 |hk,0 |2 d−α
k,0 +σ
2
00
x∈Φ\{k}
As α > 2, φ (x) ≥ 0 for x ≥ 0 and thus φ(x) is convex for (51)
x ≥ 0. Applying Jensen’s inequality, we obtain h −1 α 2 −α 2 −α 2
i
= E e−βpk dk,k ( x∈Φ\{k} pi |hk,i | ||xi || +p0 |hk,0 | dk,0 +σ )
P
2
2
(C)
P̄cov = EX e−a1 X−a2 X α ≥ e−a1 E[X]−a2 E[X] α , (45) h i
(52)
2 −1 α
= Ep−1 dα e−σ βpk dk,k LΦ (βp−1 α −1 α
k dk,k )LY (βpk dk,k ) ,
where E[X] = E p−1 = E[p−1
α
α k k,k
0 d0,0 0 ]E[d0,0 ] due to the (53)
α
independence of PC and DC . Finally, E[d0,0 ] can be computed
explicitly. which completes the proof.
Z
E[d0,0 ] = rα dFd0,0 (r)
α R EFERENCES
[1] K. Doppler, M. Rinne, C. Wijting, C. Ribeiro, and K. Hugl, “Device-
Z R to-device communication as an underlay to LTE-Advanced networks,”
2r
= rα 2 dr IEEE Communications Magazine, vol. 47, no. 12, pp. 42–49, December
0 R 2009.
2 [2] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklós,
= Rα . (46) and Z. Turányi, “Design aspects of network assisted device-to-device
2+α communications,” IEEE Communications Magazine, vol. 50, no. 3, pp.
170–177, March 2012.
[3] 3GPP, “3rd generation partnership project; technical specification group
C. Proof of Corollary 2 SA; feasibility study for proximity services (ProSe) (release 12),” 3GPP
Note that φ(p) is positive-valued and continuous when p > TR 22.803 V1.0.0, August 2012.
[4] M. Corson, R. Laroia, J. Li, V. Park, T. Richardson, and G. Tsirtsis,
0. Also, limp→0+ φ(p)p → 0 and limp→∞ φ(p) p → 0. These “Toward proximity-aware internetworking,” IEEE Wireless Communi-
facts imply that there exists a non-degenerate p? (d) ∈ (0, ∞) cations, vol. 17, no. 6, pp. 26–33, 2010.
φ(p) [5] C. H. Yu, O. Tirkkonen, K. Doppler, and C. Ribeiro, “On the perfor-
that achieves
R the maximum value of p . If we ignore the mance of device-to-device underlay communication with simple power
constraint dFp0,0 (p) = 1 and p ≤ Pmax,c for the time being control,” in IEEE Vehicular Technology Conference, 2009, pp. 1–5.
12