Sleep Period Optimization Model For Layered Video Service Delivery Over eMBMS Networks

0.36 MB·English

by  Lorenzo Carlà

Sleep Period Optimization Model For Layered
Video Service Delivery Over eMBMS Networks
Lorenzo Carla`∗, Francesco Chiti∗, Romano Fantacci∗, Andrea Tassi‡
∗Department of Information Engineering, University of Florence, Firenze, Italy
‡School of Computing and Communications, Lancaster University, Lancaster, UK
Abstract—Long Term Evolution-Advanced (LTE-A) and the concentrates the data transmissions in short time intervals
evolved Multimedia Broadcast Multicast System (eMBMS) are and, (ii) signals to the user to turn off its radio interface
the most promising technologies for the delivery of highly
for a predetermined period of time, called sleep period. It
bandwidth demanding applications. In this paper we propose
is straightforward to note that longer sleep periods can help
5 a green resource allocation strategy for the delivery of layered
1 videostreamstouserswithdifferentpropagationconditions.The to reduce the user battery consumption but it also increases
0 goal of the proposed model is to minimize the user energy con- the time interval when the user cannot receive any data from
2 sumption.Thatgoalisachievedbyminimizingthetimerequired the network. For instance, J.-M. Liang et al. [5] attempts to
byeach usertoreceivethebroadcastdataviaan efficientpower maximise the sleep period duration of users involved in PtP
n transmissionallocationmodel.Akeypointinoursystemmodelis
a thatthereliabilityoflayeredvideocommunicationsisensuredby communications. That optimization provides a generic trade-
J meansoftheRandomLinearNetworkCoding(RLNC)approach. off between activity and sleep period duration suitable for a

4 Analytical results show that the proposed resource allocation set of independentflowswhich maybe receivedbyeach user.
1 model ensures the desired quality of service constraints, while Ontheotherhand,C. Wuetal.[6]proposetoreducetheuser
the user energy footprint is significantly reduced.
energyfootprintbydecreasingthetransitiontimesbetweenthe
]
I sleep period and active period. The aforementioned proposal
N I. INTRODUCTION
refers to a system model, where both PtP and Point-to-
s. Thediffusionofmultimediacapabledevices,suchassmart- Multipoint (PtM) data flows are conveyed to the users. We
c phonesandtablets,hasgeneratedarapidgrowthofmultimedia remark that the optimization model proposed in [6] refers
[
service demand. In particular, by 2018 the video traffic will to best effort non-multimedia data flows. Unfortunately, a

1 represent 69% of the mobile Internet data traffic. This surge little attention has been paid to the problem of reducing the
v in demand has been addressed in fourth generation (4G) user energy consumption during the reception of multimedia
0
LTE-Advanced (LTE-A) cellular networks through the adop- eMBMS flows.
1
tion of the evolved Multimedia Broadcast Multicast Service This paper focuses on the broadcasting of layered video
3
3 (eMBMS) framework [1]. In particular, a broadcast/multicast services over SFN-eMBMS networks. In particular, we refer
0 multimediacontentcanbedeliveredtotheusersasaneMBMS to video streams encoded by using the H.264 Scalable Video
. traffic flow in two modes: the Single Cell-eMBMS (SC- Codingcompressionstandard(H.264/SVC)[7].AH.264/SVC
1
0 eMBMS) mode or the Single Frequency Network-eMBMS video service consists of one base layer that provides a
5 (SFN-eMBMS) mode [1]. In the former mode, each base basic reconstruction quality and one or more enhancement
1 station independently selects the communication parameters layers. Each user can improve the quality of the recovered
:
v for the transmission of the multimedia content, such as, video stream by combining the basic layer with one or more
i the amount of radio resources to be allocated, the power enhancementlayers.Inparticular,weproposeagreenresource
X
transmission, etc. Conversely,in the SFN-eMBMS mode, two allocation framework aiming at minimizing the transmission
r
a or more space contiguous base stations, forming the SFN, time duration of a layered video stream and, hence, the user
transmit the same eMBMS flow in a synchronous fashion. energy footprint. That goal is achieved by optimizing the
Since basestation ofthesame SFN donotinterferewitheach power transmission associated to each video layer. Intuitively,
other, both the user Signal-to-Interference plus Noise Ratio ifthepowertransmissionassociatedtoavideolayerincreases,
(SINR) and the system spectral efficiency are significantly then the video service can be delivered in a shorter time
improved compared to the SC-eMBMS mode. intervalbecauseofthe adoptionofa highermodulationorder,
Among the issues related to the multicast/broadcastservice lowercodingrate,etc.Asaconsequence,theusersleepperiod
delivery, the amount of energy required by a user to receive are expected to increase and the user energy consumption
aneMBMSflowisofparamountimportance[2].Forinstance, decreases.
duringthe receptionofhighdata rate videostreams, the radio A key point of our system model is that reliability of PtM
interface of a user has to be in an active state for a time communication is ensured by the adoption of the Random
interval (hereafter referred to as activity period) which could Linear Network Coding (RLNC) approach [8]. Usually, the
notbenegligible.Furthermore,astheactivityperiodincreases, eMBMS framework adopts Application Layer-Forward Error
the energy footprint on the user battery increases as well [3]. Correction(AL-FEC) codesbased on Raptor codes[9]. How-
In order to mitigate that issue in the case of Point-to-Point ever, this kind of codes are usually applied over large source
(PtP) communications, the serving base station [2], [4]: (i) messages. Hence, the adoption of Raptor or traditional LT
(3/345-"6"2,(+012&!"’(0 )%*+",,%-"*(#+./
,!4>2?@5=4A/B ,!4>2?@5=4A/B
05;23<54=/v1 05;23<54=/vS
#! !"#!$!"% !"#!$!"% #()* 012&!"’(
!" $%#&’ &!!""#’$%(
%# ()#$!"% ()#$!"% !"#
$
./0+"00%-$"*(#+*%v1 ./0+"00%-$"*(#+*%v2 ./0+"00%-$"*(#+*%v3
,-./0123/-415662754
# ⊗p1 g1,1p2⊗ g2,1 ⊗pK1gK1,1 &9!:*9! &’#($%)# Ffliogw. 2a:llLoTcaEt-iAonr.adioframemodelandthe consideredeMBMS
’ ⊕
&
coding sublayer, called MAC-RLNC, placed at the top of the
c c c&’#4!"8 standard LTE-A Medium Access Control (MAC) layer.
1,1 1,2 1,t
Weassumethateachvideolayerisassociatedtoanindepen-
dentIPpacketstream.AlltheLlayersthatcomposethevideo
+ &’
* stream enter the Packet Data Conversion Protocol (PDCP)
!
layer which produces a PDCP Protocol Data Units (PDCP
PDU) stream. Then the Radio Link Control (RLC) layer
Fig. 1: The considered LTE-A protocol stack.
segments/concatenateseach PDCP PDU in orderto producea
stream of RLC PDUs per video layer, which is forwarded
codes could lead to a significant communication delay, as
to the MAC layer. According to the RLNC principle, the
notedin[10].ThatissuecanbeovercomebyusingtheRLNC
MAC-RLNC layer segments the ℓ-th layer of each GoP into
over short source messages [11]. In this paper, the proposed
a stream of K information elements {x ,...,x }. Then,
optimization framework ensures that predetermined fractions ℓ 1 Kℓ
all the information elements defining the video layer ℓ are
of users can recover the desired set of video layers with a
linearly combined by the MAC-RLNC in order to generate a
target probability. In addition, the user energy efficiency goal
stream of N ≥ K coded elements. The j-th coded element
ℓ ℓ
ofourmodelisachievedbyjointlyoptimizingthetransmission
associated with the ℓ-th video layer of a GoP is defined
parameters and RLNC parameters. as c = Kℓ g ·x , where g is a coding coefficient
ℓ,j j=1 ℓ,j j ℓ,j
The rest of the paper is organized as follows. Sec. II
uniformlyselected atrandomovera finite field of size q [12].
describes the considered system model. Sec. III defines the P
The stream of coded elements are mapped onto one or more
proposed optimization framework. In addition, Sec. III pro-
MAC PDUs, which are delivered to the Physical layer (PHY)
vides also an efficient heuristic strategy that can derive a
forthetransmission.ThelayerℓofaGoPisrecoveredassoon
good quality feasible solution to the proposed optimization
as a user recovers K linearly independentcoded elements.
ℓ
framework in a finite number of steps. Analytical results are
Thefrequency-structureofaLTEradioframeconsistsof10
providedinSec.IV.Finally,wedrawourconclusioninSec.V.
subframes, where the time duration of each subframe is one
TTI, namely, t =1ms. In addition, as shown in Fig. 2, the
II. SYSTEMMODEL TTI
LTE-Astandardimposesthatatmost6outof10subframeper
WeconsideranetworkscenariocomposedbyBˆ contiguous radioframecanbeusedtodelivereMBMSflows.Inparticular,
basestations,whereB outofBˆ basestationsformtheSFN we assume that the ratio of eMBMS-capable subframes per
SFN
area. In addition, in the rest of the paper, we assume that a radio frame is T = 0.6. The basic data units delivered
MBMS
H.264/SVC video stream is transmitted to a group of users to a user in each subframeare called TransportBlocks (TBs).
U over the SFN. According to the H.264/SVC compression Each TB is transmitted by using a specific Modulation and
standard, the video stream can be modelled as a set of Group Coding Scheme(MCS) and consists of one or moreResource
of Pictures (GoPs), each of which has a fixed time duration Block Pairs (RBPs) [1]. In this paper, we assume that: (i)
of t seconds [7]. The video stream is composed by L each TB is composedby the same numberof RBPs, (ii) a TB
GoP
layers{v1,...,vL},wherev1 isthebaselayerandvℓ (forℓ= can hold data associated to one video layer and, (iii) at most
2,...,L) the ℓ-th enhancement layer. In addition, throughout one TB associated with the same video layer can be mapped
the paper, we define the user Quality of Service (QoS) as the onto the same eMBMS-capable subframe (see Fig. 2). In our
number of consecutive video layers that can be successfully analysis,weassumethattheuserradiointerfaceisactiveuntil
recovered by a user, starting from the base layer. each layer of a GoP is delivered. Hence, we define the user
sleep period ξ as follows:
A. Layered Video Transmission over eMBMS Network
ξ =d − max (t ) (1)
GoP ℓ
This paper refers to the LTE-A system design proposed ℓ=1,...,L
in [11] and sketched in Fig. 1, where the reliability of the where d = tGoP·TMBMS is the time duration of a GoP
GoP tTTI
service deliveryis improvedby the adoptionof the RLNC. In expressedin terms of numberof TTIs;while t is the number
ℓ
particular,alltheRLNCrelatedoperationsareperformedbya of TB transmissions associated to the layer ℓ of a GoP.
B. Performance Evaluation eMBMS service delivery approach, the average number of
coded elements associated to the ℓ-th layer of a GoP that are
In order to efficiently estimate the QoS level experienced
received by user u is
by a user, it is useful to model the data rate associated with
the reception of a TB stream by means of the Shannon’s rate N (P ,t )= tTTI ru(Pℓ) tℓ (6)
formula: ℓ ℓ ℓ L
$ ℓ %
r(σ)=αW log2(1+βσ) (2) where Lℓ is the bit size of a coded element. Furthermore, the
probabilitythatuseruisabletorecovertheℓ-thlayerofaGoP,
where W is the bandwidth spanned by a TB and σ is the
i.e., the probability that user u is able to collect K linearly
SINR associated with the considereduser. The terms α and β ℓ
independent coded elements out of N , can be expressed as
are non-negative correction factors which adapt the Shannon ℓ
follows [13]:
rate expression to fit to the actual user reception rate of a TB
stream. In particular, parameters α and β can be found by Kℓ−1 1
solving the mathematical regression problem presented in the gu(Pℓ,tℓ)= 1− qNℓ(Pℓ,tℓ)−j . (7)
Appendix. j=0 (cid:20) (cid:21)
Y
In a SFN-eMBMS network, the BSFN base stations belong- In the following section, we will show how the pair (Pℓ, tℓ)
ing to the SFN are coordinated in order to deliver the same can be optimized in order to ensure that a predetermined
physical signal. Hence, the signals coming from all the base fraction of users can achieve the QoS level ℓ with at least
stations of the SFN area are treated by a user as well as a target probability.
multipathcomponentsofasinglebasestationtransmission[1].
On the other hand, the remaining Bˆ − B base stations III. SLEEPPERIOD RESOURCE ALLOCATION FRAMEWORK
SFN
interferewiththeuserreception.Thechannelgainexperienced
Inthissectionweproposeanenergy-efficientradioresource
by user u that is receiving from base station i is denoted by
allocationstrategycalled“MaximumSleepPeriod”(MSP)that
h and can be defined as follows [11]:
u,i aims at maximizing the user sleep period by minimizing the
h =G +G −PL −δ I (3) number of TB transmissions needed for the transmission of a
u,i Tx Rx u,i i u,i
GoP.Inaddition,theproposedmodelensuresthatapredefined
where G and G are the antenna gains at the transmitter
Tx Rx fraction of users can recover the first ℓ layers with a given
and receiver side, respectively. The term PL is the path-
u,i probability, for ℓ = 1,...,L. To this end, for each video
loss experienced by user u and associated with base station
layer the MSP model (i) minimizes the maximum number of
i (see Table I), while N is the noise power. The terms I
u,i TB transmissions per video layer and, (ii) selects the power
denotes the Inter-Cell Interferenceand δ ∈{0,1} is a binary
i transmission P , for for ℓ = 1,...,L, such that the overall
parameter indicating if the i-th base station belongs to the ℓ
transmission power is less than or equal to a threshold value
SFN or not. In particular, we have that δ =1 if base station
i Pˆ. Hence, the MSP model is defined as follows:
i belongs to the SFN, otherwise δ = 0. For these reasons,
i
theSINRvalueexperiencedbyauserreceivingtheℓ-thvideo (MSP) minmax tℓ (8)
ℓ∈{1,...,L}
layer can be expressed as follows:
U
Hu subject to ∆ g (P ,t )≥Φˆ ≥θˆU ℓ∈{1,...,L} (9)
u ℓ ℓ ℓ
BSFN uX=1 (cid:16) (cid:17)
z }|hu,i { Kℓ ≤tℓ ≤dGoP ℓ∈{1,...,L}(10)
σu,ℓ =Bˆ−BSFNXi=1  Pℓ =Hu Pℓ (4) L Pℓ ≤Pˆ (11)
 hu,j Pt+N Xℓ=1
 Xj=1  Pℓ ∈R+,tℓ ∈N ℓ∈{1,...,L} (12)
where Pℓ is the transmission power used to broadcast each where ∆(s) is an indication function such that ∆(s) = 1 if
TB related to the ℓ-th layer of a GoP. The terms hu,i and thestatementsistrue,otherwise∆(s)=0.Theconstraint(9)
hu,j denote the channel gains associated with the i-th base ensures that the fraction of users recovering layer ℓ with, at
station belonging to the SFN and the j-th interfering base least, a probabilityof Φˆ isatleast equalto θˆ. Constraint(10)
ℓ
station, respectively. Finally, Pt is the transmission power of ensures that the number of TB transmissions associated with
any interfering base station. layer ℓ does not exceed the time duration of a GoP. Finally,
From (2), (4) and the regression model presented in the constraint (11) imposes that the overall transmission power
Appendix, the rate experienced by the u-th user during the does not exceed the threshold Pˆ. Unfortunately, the coupling
reception of a TB related to the ℓ-th layer can be restated as constraint (11) turns MSP into a computationally complex
a function of Pℓ: mixed integer non-linear optimization problem. In order to
mitigate this issue, in the rest of the section, we propose the
0, if H P <σ
u ℓ min
Heuristic-MSP(H-MSP) strategythat providesa goodquality
r (P )= αW log (1+βH P ), if σ ≤H P ≤σ (5)
u ℓ  2 u ℓ min u ℓ max feasible solution to MSP, in a finite number of steps.
αW log2(1+βσmax), if HuPℓ>σmax. For the sake of our analysis, we define from MSP the
For this reason, from (5) and according to the network coded UnconstrainedSleep Period(USP) problembysimplyremov-
Procedure 1 Heuristic Maximum Sleep Period Strategy. TABLE I: Main simulation parameters.
1: Initialize Pℓ∗←P˜ℓ andt∗ℓ ←t˜ℓ,forℓ=1,...,L Parameter Value
L BaseStations (SFNArea) 19(4)
2: whileXPℓ∗≥Pˆ do InteSry-SstietemDBisatnadnwceid(tIhSD) 52000MmHz
ℓ=1 TransmissionScheme SISO
3: forℓ←1,...,Ldo
4: ift∗ℓ +1≤dGoP then PAennetetrnantiaonGaLionsss S20eedTBab(SleeeAT.2a.b1l.e1.A2-.22.[11.15.]5-2[15])
5: Pℓ′←Lℓ(t∗ℓ +1) Path-Loss SeeTableA.2.1.1.5-2[15]
6: t′ℓ←t∗ℓ +1 Duplexing Mode FDD
7: else Channel Model PedA
8: t′ℓ←∞ CarrierFrequency 2.0GHz
9: endif NoisePower −168dBm/Hz
10: endfor PowerTransmission 20−80W(TableA.2.1.1-2[15])
11: ifo′ℓ=∞ then Φˆ 0.1
12: returnnosolution TMBMS 0.6
13: endif q 28
1154:: ti∗i←←art∗igm+i1n{t′1,...,t′L} NumbedrGooPfRBPs 63,209,T1T2I
16: Pi∗←Pi′ VideoA pˆℓ={29.94,34.78,40.73}[dB][16]
17: endwhile L=3 rˆℓ={117.1,402.5,1506.3}[kbps]
18: return (Pℓ∗,t∗ℓ),forℓ=1,...,L θˆℓ={0.3,0.6,0.9}
VideoB pˆℓ={29.45,32.30,34.52,38.41}[dB][16]
ing the constraint (11). The solution of the aforementioned L=4 rˆℓ={160.0,300.0,560.0,1150.0}[kbps]
problemcanbeeasilyfoundbydecomposingtheUSPproblem θˆℓ={0.3,0.5,0.6,0.9}
intoLindependentsubproblems,wheretheℓsubproblemscan
be expressed as follows: solution of MSP.
(ii) Otherwise, the while-loop body (lines 2-17 of Proce-
(USP-ℓ) min tℓ (13) dure 1) for each layer increases the value of t∗ and sets
ℓ
U P′ =L (t∗+1). Then, the solution component charac-
subject to ∆ gu(Pℓ,tℓ)≥Φˆ ≥θˆℓU (14) teℓrizedbℓyℓthesmallestvalueoft′ isaltered(lines14-16).
uX=1 (cid:16) (cid:17) (iii) The while-loop body iterates until the constraint (11) is
Kℓ ≤tℓ ≤dGoP (15) met.
Pℓ ∈R+,tℓ ∈N (16) Itisworthnotingthatthewhile-loopbody(lines2-17)returns
From the analysis presented in [14], we understand that a feasible solution of MSP in a numberof steps which is less
the optimum solution of USP-ℓ, if it exists, belongs to the than or equal to Lℓ=1(dGoP−Kℓ).
.
set L = (P ,t ) ∈ R+ × N K ≤ t ≤ d ∧
ℓ ℓ ℓ ℓ ℓ GoP PIV. NUMERICAL RESULTS
U ∆ gn(P ,t ) ≥ Φˆ) ≥ θˆ(cid:12)U . Let L (t ) be the
u=1 u ℓ ℓ ℓ(cid:12) ℓ ℓ In this section, we investigate the performance of the MSP
(cid:12)
tPransmiss(cid:16)ion power(cid:17)value such that (Loℓ(tℓ),tℓ) ∈ Lℓ. Hence, and H-MSP models in terms of the normalized sleep period
the optimum solution of USP-ℓ is represented by the pair
associated with the delivery of each GoP, defined as follows:
(L (t˜),t˜) whichischaracterizedbytheminimumvalueoft
ℓ ℓ ℓ ℓ
ξ
amongallthepossiblepairsinLℓ[14].Theoptimalsolutionto ǫ= (17)
the USP model is given by {(L (t˜ ),t˜),...,(L (t˜ ),t˜ )}, dGoP
1 1 1 L L L
for ℓ=1,...,L. In addition, we also evaluate the user performance in terms
of the maximum Peak Signal-to-Noise Ratio (PSNR) p that
Because of the definition of the USP model, it is worth u
user u can achieve:
noting that its optimal solution could be infeasible from the
point of view of the MSP problem, i.e. the constraint (11) p = max pˆ g (18)
u ℓ u
could not be met. However, starting from the USP solution, ℓ=0,...,L
n o
we define the H-MSP. The idea behind H-MSP is that of where pˆℓ is the PSNR associated with the reconstruction of
turning the optimal USP solution, which is infeasible from the first ℓ layers of the video service [7].
the point of view of MSP, into a feasible MSP solution. The The proposed optimization framework is also compared
aforementioned transformation is performed by altering one against an Uniform Power Allocation (UPA) strategy. The
componentoftheoptimalUSPsolutionatatime.Inparticular, considered UPA strategy transmits each video layer by using
from (7), we have that L (t˜ +1) ≤ L (t˜). Hence, during the same transmission power P = Pˆ, for ℓ = 1,...,L. In
ℓ ℓ ℓ ℓ ℓ L
eachiteration,theH-MSPstrategyincreasesthesmallestvalue addition,forthe sakeof comparison,we assume thatthe UPA
of t among the video layers. As a consequence, the total model relies on the MAC-RLNC coding sublayer. For these
ℓ
transmissionpowerisreducedaftereachiteration.Thegeneral reasons, the UPA model can be defined as follows:
definitionofH-MSPisprovidedinProcedure1anditinvolves
(UPA) min t (19)
ℓ
the following steps:
K ≤t ≤d ℓ∈{1,...,L} (20)
ℓ ℓ GoP
(i) We set {(P∗,t∗),...,(P∗,t∗)} equal to the solution of
1 1 L L P ∈R+,t ∈N ℓ∈{1,...,L} (21)
the USP model. If the constraint (11) is met, then the ℓ ℓ
procedure returns the solution of USP. In other words, Analyticalresults have been derivedby consideringa LTE-
the solution of the USP model is equal to the optimal A network composed by 19 base stations deployed in two
0.8 MSP
H-MSP
UPA
6RBP
Center of the Cell II 9RBP
12RBP
0.6
SFN base station
SFN cell sector
Interfering base station ε 0.4
Interfering cell sector
0.2
Center of the Cell I
0
20 25 30 35 40 45 50 55 60 65 70 75 80
Power [W]
Fig. 5: Normalized sleep period durationvs. overalltransmis-
Fig. 3: A part of the considered SFN-eMBMS scenario.
sion power threshold, for video B.
0.8 MSP
H-MSP
U6PRABP 40
0.6 912RRBBPP Max. PSNR [dB]2300 =1 =2 =3
MH−SMPSP ˆθ ˆθ ˆθ
ε 0.4 109 0 UPA 140 190 240 290 340 390 440
Distance [m]
(a)VideoA
0.2
40
02 0 25 30 35 40 45Pow5e0r [W]55 60 65 70 75 80 Max. PSNR [dB]2300 =1 =2 =3 =4
MH−SMPSP ˆθ ˆθ ˆθ ˆθ
Fig. 4: Normalizedsleep period duration vs. overalltransmis- UPA
109 0 140 190 240 290 340 390 440
sion power threshold, for video A. Distance [m]
(b)VideoB
concentric rings, with an inter-site distance of 500 m. Each
Fig. 6: Maximum achievable PSNR of video A and video B
base station controls three hexagonal cell sectors. As shown
vs. distance from the centre of cell I, for Pˆ =40 W.
in Fig. 3, we refersto a SFN composed by 4 contiguousbase
stations, which is surrounded by the remaining 15 interfering Fig. 4. In particular, the proposed MSP and H-MSP ensure
basestations.Forwhatconcernstheuserdistribution,werefer sleep periods that are up to 40% greater than those achieved
toascenariocharacterizedbyahighheterogeneityintermsof by the UPA model.
propagationconditions.Inparticular,asreportedinFig. 3,we Fig. 6 compares all the considered strategies in terms of
assume thatusersare placedon the symmetryaxisof a sector the maximum user PSNR, in the case of TBs composed by 9
of cell I of the SFN. In particular, we refer to U =80 users, RBPsandPˆ =40W.Sinceusersareregularlydistributedand
which are spaced 2 m apart with the first user being placed equallyspacedamongeachother,theexpressionofp can be
u
90 m from the centre of cell I. Table I summarizes the main equivalently expressed as a function of the distance between
simulation parameters and the considered H.264/SVC video useruandthecentreofcellI.Forthesamereason,parameters
streams, namely video A and video B. θˆ ,...,θˆ can be equivalently interpreted as distances from
1 L
Fig. 4 compares performance of MSP, H-MSP and UPA the centre of cell I. In particular, the vertical dashed lines of
strategies in terms of the normalized sleep period associated Fig. 6 mark the correspondingvalues of θˆℓ, for ℓ=1,...,L.
with the delivery of a GoP of video A as a function of For what concerns video A, Figure 6a shows that the UPA
the overall transmission power threshold Pˆ. In particular, we model delivers the base layer and all the video layers up
consider different TB sizes, namely, 6, 9, and 12 RBPs. The to a distance (from the centre of cell I) which is 33 m
effectivenessoftheproposedH-MSPstrategyisprovedbythe and 32 m smaller than the distance provided by both the
fact that the gap between the MSP and the H-MSP models is MSP and H-MSP strategies, respectively. On the other hand,
negligible. Furthermore, we note that as the value of Pˆ and Figure 6b shows the same performance metric in the case of
the number of RBPs increase, the H-MSP and MSP models video B. Also in this case, we understand that the service
always provide resource allocation solutions characterized by coverage provided by the UPA model is smaller than that of
sleep periodsthatare up to 10%greater than those associated the developed strategies.
with the UPA strategy. The normalized sleep period results ConsideragainFig. 6, the maximumPSNR valuesreported
relative to video B are provided by Fig. 5. The figure shows fordistancesgreaterthan272mrefertoauser(excludedfrom
a performance behaviour which is similar to what shown by the optimization process) which moves on the extension of
the considered cell-sector symmetry axis, towards the centre 2.5
Simulated Rate
of cell II (see Fig. 3). It is worth noting that the resource Fitting Function
2
astotlrloabtceeagtriieeocsneievsnoeslduurtwieohn(iasletpalrenoavsutis)deetrdhmeboybvaebsseotflhraoytmehrecoeMfllSvIPidtoeaoncdeAllHaI-InM.dOSBPn σ) [Mbps]1.15 .dB=633 .dB=3132
tUhPeAotshterartheganydc,anneiatchheirevinetthheecsaasmeeokfivniddeoofAsernvoicrevcidoenotinButihtye. r(0.5 σmin σmax
V. CONCLUSIONS 0
5 10 15 20 25 30 35
In this paper we address the problem of minimizing the σ [dB]
userenergyconsumptionforvideodeliveryoverSFN-eMBMS Fig. 7: eMBMS user reception rate derived by computer
networks. To this end, we propose an optimal and heuristic simulations and by (2) vs. σ.
radio resource allocation strategy, namely MSP and H-MSP
resulting from the LAD regression model, is a good approxi-
strategies, which maximize the user sleep period and improve
mation of the actual eMBMS rate values.
the reliability of communications by means of an optimized
RLNC approach. In this way, not only the the user energy REFERENCES
consumption is reduced but also the developed strategies can
[1] S.Sesia,M.Baker,andI.Toufik,LTE-TheUMTSLongTermEvolution:
meet the desired QoS levels. Results show that the developed FromTheorytoPractice. JohnWiley&Sons,2011.
H-MSP strategy provide a good quality feasible solution to [2] M. Gupta, S. Jha, A. Koc, and R. Vannithamby, “Energy Impact of
Emerging Mobile Internet Applications on LTE Networks: Issues and
the MSP model in a finite number of steps. In addition, the
Solutions,” IEEECommun.Mag.,vol.51,no.2,pp.90–97,2013.
proposed strategies are characterized by sleep periods that [3] S.JinandD.Qiao,“NumericalAnalysisofthePowerSavingin3GPP
are up to 40% greater than those provided by the considered LTEAdvancedWirelessNetworks,”IEEETrans.Veh.Technol.,vol.61,
no.4,pp.1779–1785, 2012.
Uniform Power Allocation (UPA) strategy.
[4] G. S.Kim,Y. H.Je, and S.Kim, “AnAdjustable PowerManagement
for Optimal Power Saving in LTE Terminal Baseband Modem,” IEEE
APPENDIX Trans.Consum.Electron.,vol.55,no.4,pp.1847–1853, 2009.
REGRESSION MODELFORTHERATE APPROXIMATION [5] J.-M. Liang, J.-J. Chen, H.-H. Cheng, and Y.-C. Tseng, “An Energy-
Efficient Sleep Scheduling With QoS Consideration in 3GPP LTE-
Let rˆ1,...,rˆT be the actual rate values associated to the Advanced Networks for Internet of Things,” Emerging and Selected
reception of a TB stream recovered by a user, for different Topics in Circuits and Systems, IEEE Journal on, vol. 3, no. 1, pp.
SINR values σˆ ,...,σˆ , respectively. By means of a Least 13–22,2013.
1 T [6] C.Wu,X.Sun,andT.Zhang,“APower-SavingSchedulingAlgorithm
Absolute Deviation (LAD) regression model we derive the
for Mixed Multicast and Unicast Traffic in MBSFN,” in Computing,
values of α and β. Communications and Applications Conference (ComComAp), 2012,
The LTE-A standard imposes to switch to a different MCS 2012,pp.170–174.
[7] P. Seeling and M. Reisslein, “Video transport evaluation with h.264
as soon as the TB error rate experienced by a user is greater
videotraces,”Communications SurveysTutorials, IEEE,vol.14,no.4,
than 10%. Hence, for the regression model, we consider only pp.1142–1165, Fourth2012.
thosevaluesofσˆ (fori=1,...,T)suchthattheTBerrorrate [8] A. Tassi, I. Chatzigeorgiou, and D. Vukobratovic´, “Resource Allo-
i cation Frameworks for Network-coded Layered Multimedia Multicast
is lessthan orequalto 0.1.TheLAD modelcan beexpressed Services,” IEEEJ.Sel.AreasCommun.,2014.
as follows [17]: [9] D. Lecompte and F. Gabin, “Evolved Multimedia Broadcast/Multicast
Service (eMBMS) in LTE-Advanced: Overview and Rel-11 Enhance-
T
ments,”IEEECommun.Mag.,vol.50,no.11,pp.68–74,2012.
(LAD) min rˆi−r(σˆi) (22) [10] E.Magli,M.Wang,P.Frossard,andA.Markopoulou,“NetworkCoding
subject to rˆα,≤β rXi(=σˆ1)(cid:12)(cid:12), (cid:12)(cid:12) i=1,...,T (23) pMpe.e1ts19M5–u1lt2im12e,di2a0:1A3.Review,”IEEETrans.Multimedia,vol.15,no.5,
i i(cid:12) (cid:12) [11] C.Khirallah,D.Vukobratovic´,andJ.Thompson,“PerformanceAnalysis
α>0, β >0 (24) andEnergyEfficiencyofRandomNetworkCodinginLTE-Advanced,”
IEEETrans.WirelessCommun.,vol.11,no.12,pp.4275–4285, 2012.
where the objective function (22) minimizes the sum of the [12] C. Fragouli, “Network Coding: Beyond Throughput Benefits,” Special
absolute value of the differences between rˆ and the value IssueonNetworkCodingattheProceedings oftheIEEE,2011.
i [13] O.Trullols-Cruces, J.Barcelo-Ordinas, andM.Fiore,“ExactDecoding
of the fitting function r(σˆi). Constraint (23) ensures that the Probability Under Random Linear Network Coding,” IEEE Commun.
functionr(σ)doesnotexceedthesimulatedTBreceptionrate. Lett.,vol.15,no.1,pp.67–69,Jan.2011.
We define σ to be the minimum value of SINR below [14] F.Chiti,R.Fantacci,F.Schoen,andA.Tassi,“OptimizedRandomNet-
min workCodingforReliableMulticastCommunications,”Communications
that a user cannot recover the data flow transmitted with the
Letters, IEEE,vol.17,no.8,pp.1624–1627, 2013.
smallest MCS. Conversely, a user that experiences a SINR [15] 3GPP TR 36.814 v9.0.0 (Release 9), “Further Advancements for (E-
equal to or greater than σ can always recover a data flow UTRA),”2010.
max [16] D.Munaretto,D.Jurca,andJ.Widmer,“BroadcastVideoStreamingin
transmitted with the highest MCS. By means of computer
Cellular Networks: AnAdaptation Framework for Channel, Video and
simulations we set σ =6.33dB and σ =31.32dB [18]. AL-FEC Rates Allocation,” in Proc. of WICON 2010, Singapore, CN,
min max
Then, the correcting factors, obtained by means of the Least Mar.2010,pp.1–9.
[17] P. Bloomfield and W. L. Steiger, Least Absolute Deviations: Theory,
Absolute Deviation (LAD) regression model, are α = 0.17 Applications, andAlgorithms. Birkha¨user Boston,1983.
and β =0.06. [18] S.Schwarz, J.Ikuno,M.Simko,M.Taranetz,Q.Wang,andM.Rupp,
Fig. 7 comparesboth the simulated eMBMS user reception “Pushing the limits of LTE: A survey on research enhancing the
standard,” IEEEAccess,vol.1,pp.51–62,2013.
ratewith theratevaluesobtainedfrom(2), asa functionofσ.
That figure clearly shows that the Shannon’s rate expression,