Performance Analysis of a Distributed Resource
Allocation Scheme for D2D Communications
Marco Belleschi∗ , Gábor Fodor† , Andrea Abrardo∗
∗ Department of Information Engineering, University of Siena, Via Roma 56, 53100 Siena, Italy
E-mail: {belleschi | abrardo}@dii.unisi.it
† Ericsson Research, SE-164 80, Stockholm, Sweden
Email: gabor.fodor@ericsson.com
Abstract—Device-to-device (D2D) communications underlaying
a cellular infrastructure has recently been proposed as a means
of increasing the cellular capacity, improving the user through-
put and extending the battery lifetime of user equipments by
facilitating the reuse of spectrum resources between D2D and
cellular links. In network assisted D2D communications, when
two devices are in the proximity of each other, the network can
not only help the devices to set the appropriate transmit power
and schedule time and frequency resources but also to determine
whether communication should take place via the direct D2D
link (D2D mode) or via the cellular base station (cellular mode).
In this paper we formulate the joint mode selection, scheduling
and power control task as an optimization problem that we
first solve assuming the availability of a central entity. We
also propose a distributed suboptimal joint mode selection and
resource allocation scheme that we benchmark with respect to the
centralized optimal solution. We find that the distributed scheme
performs close to the optimal scheme both in terms of resource
efficiency and user fairness.
I. I NTRODUCTION
Device-to-device (D2D) communications supported by a
cellular infrastructure holds the promise of four types of gains.
The proximity of user equipments (UE) may allow for very
high bit rates, low delays and low power consumption [1].
Secondly, the hop gain refers to using a single link in the
D2D mode rather than using an uplink and a downlink resource
when communicating via the base station (BS) in the cellular
mode. Thirdly, the reuse gain implies that radio resources
may be simultaneously used by cellular as well as D2D links
thereby tightening the reuse factor even of a reuse-1 system [1].
Finally, the pairing gain refers to the degree of freedom of
selecting the UEs communicating with the BS and the UE
pairs using a direct D2D link that should use the same time
and frequency resources. Additionally, D2D communications
may also facilitate new types of wireless peer-to-peer services
[1], [2].
However, D2D communications utilizing cellular spectrum
poses new challenges, because relative to cellular communica-
tion scenarios, the system needs to cope with new interference
situations. For example, in an orthogonal frequency division
(OFDM) system in which D2D communication links may
reuse some of the OFDM physical resource blocks (PRB),
intra-cell interference is no longer negligible. Solution ap-
proaches to deal with this problem include power control
[3], [4], various interference avoiding multiple-input-multiple-
output (MIMO) techniques [5] that can be combined with
proper mode selection [6] and advanced (network) coding
schemes [7]. However, to the best of our best knowledge,
prior works have not proposed a global optimization model that
takes into account mode selection, OFDM resource assignment
and power allocation in an unified framework with the aim
of minimizing the overall power consumption in a multicell
OFDM system.
Therefore, the purpose of the current paper is to develop
a unified optimization problem that minimizes the used sum
power in an OFDM system that may reuse PRBs for D2D links.
Although the proposed centralized mathematical model turns
out to be strongly NP-hard and the corresponding resolution
methods might be cumbersome and not-practically useful, it
serves as a benchmark for the development of distributed
heuristic and power control schemes in D2D multicell systems.
By adopting the proposed framework, we gain insight in the
potential gains of using the direct D2D link as compared to
using cellular links between two communicating UEs (Tx UE
- Rx UE) when transmitting in both (i.e. cellular and D2D)
operational modes.
Although conceptually D2D links may use downlink (DL) as
well as uplink (UL) cellular resources, UEs transmitting in DL
spectrum may not be acceptable by regulatory bodies in certain
geographical regions. Therefore, in this paper we focus on an
OFDM system in which the same uplink PRBs may be used
simultaneously for a cellular and one (or possibly more) D2D
link(s) tightening the reuse factor below 1 (as in Figure 1). For
a particular UE pair, the sum power minimization problem is
combined with mode selection that determines whether the UE
pair (Tx UE - Rx UE of Figure 1) should use the direct D2D
link or they should communicate via the cellular BS. Therefore,
we compare the performance of these two communications
modes, i.e. cellular mode and D2D mode, and provide a
simulation analysis of the impact of resource sharing, when
the positions of both the D2D pair and the cellular UE vary
within the cells.
II. S YSTEM M ODEL
We consider the uplink of an OFDM cellular system with
N cells and a complete reuse of the available frequency
resources among different cells and possibly also within the
same cell due to D2D communications. In each cell a base
station (BS) controls a fixed number of UEs consisting of
cellular UEs (c-UEs) and D2D transmitters. While all c-UEs
 UE2 = Tx UE III. A CENTRALIZED JOINT MODE SELECTION , RESOURCE
ALLOCATION , AND POWER ASSIGNMENT FRAMEWORK
AP1 AP2
According to the underlying D2D infrastructure, we select
UE1
two possible transmission modes, i.e. q = 0 in which a given
UE communicates with the serving BS (cellular mode) and q =
UE3 1 in which a D2D pair communicates directly either sharing
Rx UE
PRBs or with an exclusive assignment of PRBs (D2D mode).
Figure 1. Illustration of D2D communications, when a user equipment (UE1) Obviously, for c-UEs, q is by definition always 0, while for
and a D2D pair (Tx UE - Rx UE) may use the same OFDM PRB. Due to D2D pairs mode selection can select q = 0 or q = 1. Let
the D2D link, intracell interference as well as intercell interference between
D2D and cellular links (UE3 to Rx UE) can be very high. (In this example xl,f (q) be a binary variable equal to 1 if user l is assigned
assuming that the D2D link uses cellular UL resources.) PRB f with mode q and 0 otherwise, and let Pl,f (q) ∈ R be a
positive real variable denoting the transmission power allocated
to user l on PRB f when adopting mode q.
Accordingly,
n
G if user l adopts mode q = 0 on PRB f
Gh(l),f (q) = Gn(l),f if user l adopts mode q = 1 on PRB f
d(l),f

communicate directly with the BS, D2D pairs can either exploit A Mixed Integer Linear Programming (MILP) formulation
for jointly optimizing mode selection, resource allocation and
the direct link or communicate via the BS (see Figure 1). We power assignment (JOMSRAP) can be derived as follows
represent the cellular network topology by a graph in which X
BSs belonging to the set B are labeled with n = 1, . . . , N ; minimize
l,f,q
Pl,f (q) (2)
UEs belonging to the set M with l = 1, . . . , L. We denote with subject to Pl,f (q) ≤ Kxl,f (q) ∀l, f, q (C1)
U ⊂ M and D ⊂ M the set of c-UEs and D2D transmitters X
Gh(l),f (q)Pl,f (q) − tgt
γl,f Gjh(l),f (q)Pj,f (q)
respectively. Hence, we define with h(l) the receiver of any j6=l,q
UE l, i.e., the serving BS n(l) for any c-UE l ∈ U and either ≥ γl,f
tgt
σ(1 − K(1 − xl,f (q))) ∀l, f, q (C2)
the serving BS n(l) or the D2D receiver d(l) for any D2D X
xl,f (0) ≤ 1 ∀n, f (C3)
transmitter l ∈ D. X
l∈n(l)

We consider the total frequency bandwidth W divided into xl,f (q) ≥ Rlmin
∀l (C4)
f,q
F PRBs each of bandwidth B = W/F . Each UE l can be tgt
γl,f σ
assigned to any of the f = 1, . . . , F PRBs belonging to the set Pl,f (q) ≥ xl,f (q) ∀l, f, q (C5)
Gh(l),f (q)
F. We model the effective channel gain between the transmitter X
l and its receiver when using PRB f as Gh(l),f . Let σ be the Pl,f (q) ≤ P max ∀l, q (C6)
f,q
thermal noise at the PRB operation bandwidth B = W/F and X
xl,f (q) ≤ 1 ∀l, f (C7)
Pl,f be the actual transmission power of user l on PRB f . q

The perceived signal to noise and interference ratio (SINR) on xl,f (1) = 0 ∀l ∈ U , f (C8)
PRB f for user l is: xl,f (q) ∈ {0, 1} ∀l, f, q (C9)
Gh(l),f Pl,f The objective function of JOMSRAP minimizes the overall
γl,f (pf ) = (1)
Il,f (pf ) + σ transmission power. In Constraints (C1) and (C2), K is a
large positive number. Specifically, constraints (C1) force the
where pf = (P1,f , . . . , PL,f )T is the vector of powers al-
transmission power of user l on a given PRB f to be 0
located to UEs overall the P cellular network and transmitting
j in case xl,f (q) = 0. Constraints (C2) can be derived by
on PRB f , Il,f (pf ) = j6=l h(l),f Pj,f is the interference
G
direct inspection of Equation 1 and they state that if user l
experienced at the receiver h(l) of user l, and Gjh(l),f accounts is assigned to PRB f with mode q = 1, i.e. xl,f (q) = 1,
for the effect of distance-based attenuation, beamforming, its transmission power must guarantee that γl,f (q) ≥ γltgt
fading etc., between h(l) and the interferer j transmitting on (see Equation (1)). Otherwise, if xl,f (q) = 0, the large
PRB f . Neglecting the details of practical coding schemes, we constant K ensures that constraints (C2) are always satisfied.
employ the Shannon capacity as a measure of the achievable bit Constraints (C3) guarantee the exclusivity of PRBs assignment
rate. Specifically, given a certain SINR, the spectral efficiency for UEs selecting mode q = 0 and belonging to the same
in bit/s/Hz for user l employing PRB f is ηl,f = log2 (1 + cell. Constraints (C4) impose that at least Rlmin PRBs must be
γl,f (pf )) and the maximum achievable theoretical rate in bit/s allocated to each user l ∈ M . Constraints (C5) are redundant
is Rl,f = Bηl,f . For the sake of tractability, in this paper but enforce Constraints (C2). Constraints (C6) ensure that each
we disregard link adaptation (LA). Thus, we assume that user l ∈ M does not consume more than P max power to satisfy
PRB f allows user l to transmit with a spectral efficiency its rate request. Finally, Constraints (C7) and Constraints (C8)
tgt tgt
ηl,f = log2 (1 + γl,f ) if the experienced γl,f (pf ) exceeds a state that each user l can exploit PRB f by adopting one mode
tgt
fixed target γl,f . Otherwise, user l cannot transmit on PRB f . at most, and each UE l ∈ U is forced to adopt cellular mode
tgt
We impose equal γl,f for each user-PRB pair. q = 0.

2
From JOMSRAP, we note that each PRBs can be assigned
to at most one c-UE within a cell in order to avoid intra-
cell interference among c-UEs, while we let D2D users decide
whether share PRBs or not. Hence, while in classical OFDMA
cellular networks the multiple access interference (MAI) can
only affect users belonging to different cells, in this scenario
MAI can jeopardize the system performances even within a
cell. To this end, we note that in classical uplink cellular
systems, interfering users are located at a distance that amounts
at least to the radius of the cell, while in D2D communications
MAI can operate at any distance thus leading to a potentially
much heavier impairment of performances.
Proposition 1: JOMSRAP is NP-hard in the strong sense.
Proof: The proof follows by reducing the strongly NP-
hard 3-partition problem [8] to JOMSRAP.
IV. A DISTRIBUTED SUBOPTIMAL MODE SELECTION AND
RESOURCE ALLOCATION SCHEME
As showed in Proposition 1, the resolution of JOMSRAP
by means of a brute-force approach can be demanding in
terms of computational complexity that renders it infeasible
in most scenarios in practice. This calls for the development
of more practical distributed algorithms that can be applied
in a real-world system. To this end, we reduce JOMSRAP to
a single-cell mode selection and resource allocation problem
(S-MSRAP) in which each BS is in charge of selecting the
proper mode and PRB to each mobile.
X
minimize Pl,f (q)xl,f (q) (3)
l,f,q
X 1: ∆ < 1
subject to xl,f (0) ≤ 1 ∀f (C10)
n(l)
X redSlots(cell) = 0 then
xl,f (q) ≥ Rlmin ∀l (C11)
f,q
X redP rob ←
xl,f (q) ≤ 1 ∀l, f (C12)
q 5:
xl,f (q) ∈ {0, 1} ∀l, f, q
Similarly to related constraints in Problem (2), con-
straints (C10) impose the exclusivity of PRB assignment for
UEs, constraints (C11) impose that at least Rlmin PRBs must
be allocated to each user in the cell, constraints C12 state that
each user l can exploit PRB f by adopting at most one mode.
The computation burden of JOMSRAP is distributed among
the BSs that have now to solve a simple polynomial algorithm.
Proposition 2: The linear programming (LP) problem ob-
tained by relaxing constraints (C13) has an integer optimal
solution
Proof: Constraints (C10) - (C12) can be grouped as
Ab ≤ c
where b = [x1,1 (0), . . . , xL,1 (0), . . . , xL,1 (1), . . . , xL,F (1)]
and c = [1, . . . , 1, . . . , −Rlmin , . . . , −Rlmin , 1, . . . , 1]. The
proof follows from the total unimodularity of the constraint
matrix A, i.e. every square submatrix of A has determinant
equal to 0,1, or -1 [9].
In a multicell scenario the optimal solution of Problem 3 is
compromised by the intercell interference and PRB sharing.
Much like non-cooperative games, each BS is a player and the
payoff function is the utility function in which PRB costs, i.e.
the power vector, are iteratively computed at each iteration (t)
from the interference perceived in each PRB at the iteration
(t−1) while the allocation strategies of all the other players are
kept fixed. In this framework the strategy profile is the set of
all possible mode selection and PRB assignment that each BS
can adopt. The overall objective is to reach a Nash equilibria
in which each player, given the strategy to the other players,
does not gain any reward by making individual changing in its
own strategy. Unluckily, a Nash equilibrium is not guaranteed
to exist when equal SINR target are fixed [10], [11], [12]. The
natural consequence is that the resource allocation procedure
outputs unstable PRBs assignment.
To this end, we propose a load control (LC) policy that pro-
gressively reduces the load of the network (see Algorithm 1).
LC runs for each PRB f and iteration t. In particular LC
reduces the SINR target of those users that consume the most
and in turn interfere the most and cause higher instability.
Hence, LC tends to penalize some users leading to unfair
traffic configuration. To prevent this effect, the LC mechanism
is activated by taking into account 1) the amount of instability
that the cell induces and 2) the user to be penalized is the one
that can potentially experience the highest benefit in terms of
SINR reduction.
The resource allocation and LC strategy are repeated until a
steady PRB assignment and power allocation are reached. The
derivation of ∆Pl and ∆ηl are provided in the Appendix.
Algorithm 1 Load control for S-MSRAP.
2: if allocation is still changing after Nini slots and
unst ← Set  of users in cell still
3:
P  oscillating
P 
(cell) (cell)
4: l∈unst Pl / l Pl
 ← rand
(C13) 6: if  < redP rob then
7: Compute for each l ∈ cell ∆Pl ← |Pl0 − Pl | and ∆ηl ←
|ηl0 − ηl |
8: Compute for each l ∈ cell the benefit hl ← ∆Pl /∆ηl
9: Select h? ← max {h}
? ← γh? ∆
γhtgt tgt
10:
11: redSlots(cell) ← NredSlots
12: end if
13: else
14: redSlots(cell) ← max {redSlots(cell) − 1, 0}
15: end if
V. N UMERICAL RESULTS
(4) We study an OFDM system featuring a total uplink band-
width of W = 5MHz with N = 7 cells. Both JOMSRAP (2)
and S-MSRAP with LC have been coded in the M ATLAB-
based RU DIMENTARY N ETWORK E MULATOR (RUNE) that
has been extended to incorporate a D2D-OFDMA system.
For a more accurate analysis, we consider the realistic 3GPP
macro-cell propagation environment for urban and suburban
scenarios presented in [14]. In Table I, we provide the input
parameters of our simulation study.
Figures 2 - 3 illustrate the potential benefit of a D2D network
3
 0.08 0.08
100

0.06 0.06
Power tot. [W]

Power tot. [W]

50

D2D Power Gain [%]

0.04 0.04 Gain≈ 99.9%
0

0.02 0.02
−50

0 0 −100
400 400 400
Loss≈ 137% 300
300 200 300 200
150 150 −150 200
200 200
100 100 0 100
100 100 50
50 50 100
150 0
0 0 0 0 200 RUE [m]
RUE [m] RD2D [m] RUE [m] RD2D [m]
RD2D [m]

(a) Power consumption in cellular mode. (b) Power consumption in D2D mode. (c) Gain of D2D over cellular mode.
Figure 2. Comparison between cellular and D2D mode: the percentage gain of D2D against cellular mode is about 100% for RD2D = 20m and RUE = 50m,
while progressively reduces as RD2D and RUE increase. The advantage becomes negligible and not effective for RD2D ≥ 120m.

100

0.08 100
80
Mode Selection Gain [%]

Mode Selection Gain [%]

80
0.06
60
Power tot. [W]

60
0.04
40 40
Gain ≈ 99.8% Gain ≈ 86%
0.02 20
20

0
0 0 Gain ≈ 3.6% 400
400 0 Gain ≈ 3.5%
300 200 300 200
50
200 150 100 200 150
100 350 400 100
100 150 250 300 100
50 200 50
100 150
RUE [m] 0 0 200 50 RUE [m] 0 0
R [m] R [m]
D2D RD2D [m] RUE [m] D2D

(a) Power consumption. (b) Mode selection vs. Cellular Mode. (c) Mode selection vs. D2D Mode.
Figure 3. Power consumption when both cellular and D2D mode are available: compared with Figure 2, mode selection allows to drastically improve the
system performances.

22.2

22
Tot. Spectral efficiency [b/s/Hz]

0.25
1
21.8
0.2
21.6
Power tot. [W]

0.15 0.995
Jain´s Fairness Index

21.4
0.1
21.2
0.99
0.05 21

0 20.8
400 0 0.985 0
300 200 50
200 150 100 100
100 0.98
100 150 350 400
50 250 300
200 150 200 50 100 150 200 200
RUE [m] 0 0 50 100 250 300 350 D2D radius [m]
RD2D [m] 400
D2D radius [m] RUE [m]
R [m]

(a) Power consumption. (b) Total spectral efficiency. (c) Jain’s fairness index [13].
Figure 4. Performances of S-MSRAP with LC.

when the optimal centralized JOMSRAP (2) is used. To solve an accurate exploitation of the multi-user diversity and mode
JOMSRAP, RUNE has been integrated with the optimization selection. Specifically, the gain of mode selection strategy
solver I LOG C PLEX 12.2. Hence, we perform extensive sim- over a conventional cellular communication can be signifi-
ulations varying the distance RUE between each c-UE and cant for RD2D ≤ 150m and become less effective only for
the serving BS, as well as the proximity RD2D between higher distances. The impact of mode selection is analyzed
a D2D pair. For each {RUE , RD2D }, we average over 500 in Figure 5 and Figure 6. In particular, Figure 5 illustrates
feasible Monte Carlo experiments and we consider the channel the percentage of selecting cellular mode by D2D pairs, while
gains constant during the JOMSRAP procedure. In Figure 2, Figure 6 indicates the occurrence of resource sharing in D2D
we solve JOMSRAP by forcing cellular mode, i.e. q = 0 communications.
∀l (see Figure 2(a)), and D2D mode, i.e. q = 1 ∀l (see The performances of the distributed S-MSRAP with LC heuris-
Figure 2(b)), and compare the gain of D2D over conventional tic are depicted in Figure 4. Compared with the optimal results
cellular communications (see Figure 2(c)). Figure 3 depicts the provided by JOMSRAP (Figure 3(a)), the distributed scheme
advantage of the jointly mode selection, PRB assignment and achieves suboptimal performances in terms of power consump-
power allocation scheme. Such an advantage is triggered by tion (Figure 4(a)), but rather than relying on a central entity

4
 Table I
S YSTEM PARAMETERS .
Parameters Summary 80

System bandwidth W 5MHz

Impact of cellular mode [%]

60
Carrier frequency 2GHz
40
Gain at 1 meter distance -37dB ≈ 72%
BS antenna height 15m 20

D2D antenna height 1.8m 0

400
Max Tx power 500mW 300
≈ 5%
200
200 150
Lognormal shadow fading std. dev. 6dB 100
50
100

Lognormal fading corr. distance 110m R

UE
[m] 0 0
RD2D [m]

Time delay 5e-7s

Figure 5. Percentage of selecting cellular mode by D2D pairs.
Number of multi-paths 8
Cell radius 500m
Number of cells 7
Target Spectr. Eff. η tgt 4.77dB 80

≈ 75.9
RcellU E 50m,100m,. . . ,400m 70

Impact of PRB sharing [%]

60
RD2D 20m,40m,. . . ,200m 50

40

30

20

to solve JOMSRAP the computational burden is distributed 10 ≈ 19.5

0
among the BSs in the network. As shown in Figure 4(b), 50
100 400
LC comes into play when the distances {RUE , RD2D } are 300 350
150 250
150 200
200 50 100
RD2D [m] RUE [m]
relatively high involving a reduction on the total spectral
efficiency. Nevertheless, the method is able to guarantee fair Figure 6. Percentage of selecting cellular mode by D2D pairs.
share of the overall network capacity (see Figure 4(c)).

VI. C ONCLUSIONS R EFERENCES

In this paper we developed a model for a joint optimization
of mode selection, resource assignment and power allocation
for D2D communications underlaying a cellular infrastruc-
ture. Resource management in such a scenario is particularly
challenging, since intracell and intercell interference need
to be managed between the cellular and D2D layers. We
used this model to develop a distributed single-cell mode
selection and resource allocation algorithm (S-MSRAP) that
we benchmarked relative to the centralized optimal multi-
cell algorithm. We implemented the S-MSRAP algorithm in
a system simulator. The numerical results indicate that the
heuristic approach shows consistently good performance with
respect to the optimum solution. The results also indicate that
when D2D communications can reuse the cellular spectrum
resources under network control, it can improve the system
capacity, especially when joint mode selection and power
control are used along with proper resource block allocation.
The S-MSRAP algorithm effectively extends the range for
which D2D communications is useful, it helps to protect the
cellular layer from interference from D2D links and reduces
the overall power consumption in the network.
A PPENDIX
Derivation of ∆P in Algorithm 1. For a given PRB f and
iteration t, we compute
0 tgt
Pl,f = ∆γl,f (Il,f (pf ) + σ)/Gh(l),f
The SINR reduction by ∆ involves a power reduction of
∆Pl = |Pl0 −Pl |. Similarly, it is possible to derive the reduction
∆ηl = |ηl0 − ηl | in the spectral efficiency.
[1] K. Doppler, M. Rinne, C. Wijting, C. B. Ribeiro, and K. Hugl. Device-
to-Device Communication as an Underlay to LTE-Advanced Networks.
IEEE Communication Magazine, 47(12):42–49, Dec. 2009.
[2] M. S. Corson, J. Li, V. Park, T. Richardson, and G. Tsirtsis. Toward
Proximity Aware Internetworking. IEEE Wireless Communications,
17(6):26–33, Dec. 2010.
[3] C. H. Yu, O. Tirkkonen, K. Doppler, and C. Ribeiro. On the Performance
of Device-to-Device Underlay Communication with Simple Power Con-
trol. In Proc. of IEEE VTC Spring, Barcelona, Spain, Apr. 2009.
[4] C. H. Yu, O. Tirkkonen, K. Doppler, and C. Ribeiro. Power Optimization
of Device-to-Device Communication Underlaying Cellular Communica-
tion. In Proc. of IEEE ICC, Dresden, Germany, Jun. 2009.
[5] P. Jänis, V. Koivunen, C. B. Ribeiro, K. Doppler, and K. Hugl.
Interference-Avoiding MIMO Schemes for Device-to-Device Radio Un-
derlaying Cellular Networks. In Proc. of IEEE PIMRC, Tokyo, Japan,
Sept. 2009.
[6] K. Doppler, C. H. Yu, C. B. Ribeiro, and P. Jänis. Mode Selection
for Device-to-Device Communication Underlaying an LTE-Advanced
Network. In Proc. of IEEE WCNC, Sydney, Australia, Apr. 2010.
[7] K. Doppler and M. Xiao. Innovative Concepts in Peer-
to-Peer and Network Coding. In WINNER+/CELTIC
Deliverable CELTIC/CP5-026 D1.3, May 2008.
http://projects.celtic-initiative.org/winner+.
[8] M. R. Garey and D. S. Johnson. Computers and Intractability: A Guide
to the Theory of NP-Completeness. Freeman, 1979.
[9] C. H. Papadimitriou and K. Steiglitz. Combinatorial Optimization:
Algorithms and Complexity. Dover Publications, 1998.
[10] P. Soldati and M. Johansson. A Low Signalling Scheme for Distributed
Allocation in Multi-Cellular OFDMA Systems. In Proc. of IEEE
Globecom, New Orleans, LA, Nov. 2008.
[11] C.U. Saraydar, N.B. Mandayam, and D.J. Goodman. Efficient Power
Control via Pricing in Wireless Data Networks. IEEE Trans. on
Communications, 50(2):291–303, Feb. 2002.
[12] G. Scutari, D.P. Palomar, and S. Barbarossa. Optimal Linear Precoding
Strategies for Wideband Noncooperative Systems Based on Game The-
ory - Part I: Nash Equilibria. Transactions on Signal Processing, 56(3),
Mar. 2008.
[13] R. Jain, D. M. Chiu, and W. Hawe. A Quantitative Measure of
Fairness and Discrimination for Resource Allocation in Shared Systems.
Technical Report DEC Research Report TR-301, 1984.
[14] 3GPP TR 25.942 v10.0.0 - Radio Frequency (RF) System Scenarios
(Release 10), Apr. 2011.