You are on page 1of 5

QoS-Aware Scheduling and Resource Allocation for

Video Streams in e-MBMS towards LTE-A System


Chong Lou Ling Qiu
Personal Communication Network & Spread Spectrum Lab Personal Communication Network & Spread Spectrum Lab
Univ. of Science and Technology of China Univ. of Science and Technology of China
Hefei, Anhui, China Hefei, Anhui, China
Email: louchong@mail.ustc.edu.cn Email: lqiu@ustc.edu.cn

AbstractGiven that statistical multiplexing of video streams is is reduced to the problem of packet scheduling. Various packet
employed in enhanced Multimedia Broadcast/Multicast Service scheduling strategies for video transmission over wireless have
(e-MBMS), we address the issue of scheduling across the streams been discussed including [8]-[10]. Regarded as a delay-limited
towards Advanced-Long Term Evolution (LTE-A) system, in capacity problem, The Earliest Deadline First (EDF) strategy is
order to improve the quality of service (QoS) performances. In put forward to satisfy the delay constraints in [8]. Moreover, in
this paper, the optimization problem of scheduling and resource content-aware schemes, the importance of the scheduled packet
allocation for separate streams is first formulated. With practical for decoders is considered as well [9] [10], i.e., Minimization
considerations including the terminal power consumption and Cost (MC) strategy. Nevertheless, these strategies dont refer to
MBMS over a Single Frequency Network (MBSFN) operation,
e-MBMS system due to the following considerations. (I). each
hence a suboptimum QoS-aware two-layer scheduling scheme is
OFDM-based frame including multi-subcarriers is apt to be
proposed to allocate resources among the streams. By separating
streaming scheduling and packet sorting, the scheduler is aware scheduled to the data from more than one stream. Obviously, it
of queue state information, fairness among the streams, and the is inappropriate for multicast in view of power consumption,
frame weight. Simulations results with H.264/AVC coded video since each terminal needs to decode more frames including the
sequences show that significant gains could be observed by our data for its desired contents. [11]. (II). as for MBSFN, the data
scheme in terms of spectrum efficiency, QoS of packet loss, and entity is separated from the control entity. The control entity
video quality while maintaining the fairness among the streams. which is responsible for allocating resources has no idea of the
related factors used by packet scheduling [6].
Keywords-e-MBMS; statistical multiplexing; video streaming;
two-layer scheduling; QoS; H.264/AVC To resolve the problems above, we propose a suboptimum
scheduling scheme, called the QoS-aware two-layer scheduling.
The innovations lie in (I). it is up to specification of e-MBMS
I. INTRODUCTION that a frame is allocated to one stream, thus the terminal is
Wireless network has been on the verge of a 3rd growth, and enabled to turn into sleep mode during several frames, when its
it is predicted that the future traffic will be dominated by video. undesired streams are being transmitted [12]. (II). the process
Multicast/Broadcast plays an important role in delivering such of resource allocation is divided into two layers. In the long-
bandwidth-hungry video-based applications [1]. Following the term scheduling, we add the QoS-aware Scheduling Module
trend, Multimedia Broadcast/Multicast Service (MBMS) is first (QASM) to the control entity, and it is able to acquire the
defined as a point-to-multipoint service for UMTS. In order to information of queue state from the data entity, such as the
maintain its competiveness, 3GPP began the projects of LTE packet urgency and fairness, to help decide the transmission
and LTE-A. LTE systems specified e-MBMS to gain higher order of streams. After that, the data entity ensures the prior
spectrum efficiency, where the remarkable gain from macro transmissions of more important packets based on the frame
diversity leads to the deployment of MBSFN operation [2]. weight in the short-term scheduling. Simulations results show
that our scheme performs well in exhaustive QoS metrics
H.264/AVC is the currently specified codec in e-MBMS for including spectrum efficiency, packet loss, and video quality,
video streaming applications [3]. This poses a challenge of the while maintaining the adequate fairness among the streams.
efficient use of radio resources due to its higher traffic burst.
Therefore, the function of statistical multiplexing is proposed The remainder is organized as follows: Sec. II provides a
in e-MBMS system [4]. By multiplexing some parallel streams overview of e-MBMS architecture. The optimization problem
into a bundle, the traffic variability decreases obviously. Most that we consider is formulated in Sec. III and our scheme is
of the existing studies, as with [5] [6], all focus on the resource described in Sec. IV respectively. Simulation results are given
allocation schemes for the bundle. Aiming to further reduce the in Sec. V. Finally, Sec. VI summarizes our conclusions.
amount of discarded packets caused by expiration, the issue of
scheduling is first stressed [7], by which our work is motivated. II. SYSTEM OVERVIEW
Our main contribution is to devise such a scheduler towards
LTEA system, so as to improve the QoS performances. A conceptual architecture in this paper is provided in Fig. 1.
Multicast/Broadcast Service Center (BM-SC) takes charge of
In our study, we first formulate the optimization problem of service announcements and session management. It determines
resource allocation for separate streams. Then, we show that it

This work is supported by the National Basic Research Program of China


(973) under Grant No.2007CB310602, National Major Science & technology
Project of China under Grant No.2008ZX03003-004, and 2009ZX03002-001.

978-1-4244-8327-3/11/$26.00 2011 IEEE


as Least Channel Gain (LCG) scheme. i.e., transmitting power
is determined by the worst user who has the minimum
subchannel gain. Function f (.) demonstrates the power with the
unity channel gain, which depends on the specific MCS [14].

f (cn ,,t ) n , k ,t
max Pn,t , t (2)
k
sS , k =1,2,, K
hn2, s ,t
The weight wk is introduced to generalize the problem. By
changing the value of wk, (RAVe) could cover different packet
scheduling strategies. We assume that the objective function is
a weighted sum of rate for all the video streams. The problem
Figure 1. e-MBMS System Architecture can be formulated as follows
T K N
which streams can be multiplexed. MBMS-Gateway (MBMS-
GW) ensures that the same content is sent to all the evolved
(RAVe) max
n ,k ,t , cn ,t
w [ (
t =1 k =1
k
n =1
n, k ,t cn,t )] (3)
Node Bs (eNB) synchronously within the MBSFN Area, and
s.t.
performs statistical multiplexing. At the beginning of a MCH
K
Scheduling Period (MSP), Multi-cell/Multicast Coordination
Entity (MCE) is responsible for allocating the resources to all k =1
n, k , t = 1, n, t (4)
the eNBs uniformly, such as the transmission order1 and the N
concrete Modulation & Coding Scheme (MCS). Inside the eNB
itself, multi-streams with the same QoS requirements could be
P
n =1
n ,t PT , t , (5)

multiplexed onto one Multicast Channel (MCH). After that, where (4) ensures that each pair of RBs can be used for only
actual scheduling for data is employed in MAC layer based on one stream in each TTI. PT represents the total power.
the amount of its stored data and results of resource allocation
[12] [13]. Suppose that the difficulties in synchronization are The problem is known to be NP-hard for nonlinear integer
resolved effectively in LTE-A system, we could easily deduce programming. To simplify the defined problem, we consider
that the action of each eNB is identical. MBSFN operation. Observation 1 implies a reduced form.

In LTE system, QoS Class Identifier (QCI) may be used for Observation1.Besides the strict constraints of synchronization
scheduling. However, as static QoS profile, it is insufficient for among the eNBs, MBSFN transmission requires that all the
the multiplexed streams, because they have the same QCI. To RBs belonging to one MCH are dedicated to all the users with
improve the performance of scheduling, we consider adding the same MCS [12]. In other words, it could be expressed as
the QASM to MCE as Fig. 1. It acquires the related dynamic cn , t = c , n , t (6)
QoS information from eNB through the existing interface (M2).
The detailed operation will be described later in Sec. IV. With the help of the observation 1, the control variable cn,t
is replaced by a constant c eventually. If the transmit power is
assigned to each pair of RBs equally, the constraint (5) could
III. PROBLEM FORMULATION be removed. In general, the reduced problem becomes
A. General Problem Definition T K N
(RAVe-sfn) max wk [ (n , k ,t c)] (7)
We consider a general problem of Resource Allocation for n ,k ,t
t =1 k =1 n =1
Video streams in e-MBMS (RAVe) in one eNB. K streams are
s.t.
included in the bundle and there are totally N pairs of Resource K
Blocks (RB) in a Transmission Time Interval (TTI), of which
one RB consists of 12 consecutive subcarriers and 1 slot in k =1
n, k , t = 1, n, t. (8)
frequency and time domain respectively. TTI is the minimum When RB n in TTI t is allocated to the streaming k, Wn,k,t,
time unit for scheduling that is comprised of two slots, so one the achievable weighted rate of streaming k is equivalent to
subchannel consists of a pair of RBs [12]. T TTIs in one MSP
have been determined to be allocated. The binary indicator Wn, k ,t = wk c (9)
n,k,t is used to show that if streaming k utilizes the pair n of Therefore, (RAVe-sfn) becomes one of finding the largest
RBs in TTI t or not. value of Wn,k,t when allocating resources. i.e., it degenerates as
1,if the pair n of RBs is used by streaming k in t a problem of packet scheduling as equation (10). To resolve
n , k , t = (1) the problem, packet scheduling is considered in the following.
0, otherwise
k * = arg max wk . (10)
Let Sk be the set of users for streaming k. hn,s,t denotes the k =1,2, K

subchannel gain of user s over the pair n of RBs in TTI t. eNB


decides the power level Pn,t and the number cn,t of bits to carry B. Conventional Packet Scheduling
over the pair n of RBs in TTI t. With these notations, the In OFDM systems, the conventional scheduling strategies
transmitting power meets inequality (2), where it is expressed always take the time and frequency selectivity into account.

1
transmission order is generated by MCE to rule the action of eNB, when one
MSP starts. It implies that, multi-streams are scheduled one by one according
to such order. It is efficient in reducing signaling overhead and terminal power
consumption [13].
In MBSFN operation, there is no actual multi-user diversity. The fairness is earliest proposed in unicast systems [17],
Therefore, EDF in e-MBMS (MEDF) could be generalized as we adopt it into the e-MBMS system. scheduled_totalk is the
throughput of streaming k during a period. received_totalk is
1 the amount of received packets in this period for streaming k.
(MEDF) wk _ MEDF ( m) = , (11)
deadlinek , m t
After acquiring DT and FP, QASM would determine the
where deadlinek,m denotes the decoding line of mth video frame streaming weight as
of streaming k. Similarly, MC in e-MBMS, i.e., MMC, could
be generalized as received_totalk delayHoL
wsk = exp( ). (17)
scheduled _ totalk PDBk
1
(MMC) wk _ MMC ( m) = I k , m (12) Finally, the transmission order in the bundle is determined
deadlinek , m t
along with QCI for non-multiplexed streams. The stream in
( N P + 1 m) the bundle with a larger wsk is prior transmitted.
( N + 1) for I or P frame
P
I k ,m = , (13) B. Packet Sorting
1
for B frame To improve the video quality at receivers, the scheduler in
( N P + 1)( N B + 1)
eNB performs the packet sorting at TTI-level after streaming
where NP and NB denote the number of P frames and B frames scheduling, called Cost-based Sorting Strategy (CS).
in one Group of Pictures (GoP) respectively. Ik,m denotes the
The binary indicator k is used to show whether the
frame weight of mth frame of streaming k. We assume that
streaming k is scheduled completely or not, that is, whether
eNB could identify the frame type with the method described
there is any packet left in the queue to be scheduled.
in [15]. Therefore, Ik,m could be determined based on the frame
type (I, P, or B) and its position in a GoP as equation (13)[16]. 0, if the streaming k is scheduled completely
k = . (18)
According to our previous discussions, it is easy to see that 1, otherwise
such strategies at TTI-level are not applicable combining the The corresponding streaming k*, which is to be scheduled
fact. We consider dividing wk into two parts that are executed could be determined by the following equation
at the MSP-level and TTI-level respectively. Accordingly, a
novel scheduling scheme is presented in Sec. IV. k * = arg max wsk k . (19)
k =1,2, K

With the help of equation (19), for a determined streaming


IV. TWO-LAYER SCHEDULING SCHEME
k*, CS could be described as (20)
In this section, a QoS-ware two-layer scheduling scheme
is devised, where wk is divided into the streaming weight wsk m* = arg max I k * , m , (20)
m
and frame weight Ik,m. In the first layer streaming scheduling, where the more important packet with a higher frame weight
streaming weight is determined by MCE at MSP-level. And is prior allocated in CS during the MSP. Generally, QoS-
then, eNB performs the packet sorting based on the results of aware Two-Layer Scheduling can be described as follows
streaming scheduling at TTI-level.
QoS-aware Two-layer Scheduling Scheme
A. Streaming Scheduling 1) Initialization
With the help of QASM, the following parameters offered a) Set k = 1 for all k {1,2, ,K}.
by eNBs at the end of MSP, would help MCE to decide the b) Set n,k,t = 0 for all n {1,2, ,N}, k {1,2, ,K}, and
transmission order for the next MSP. A certain eNB is enough t{1,2, ,T}.
since the action of each eNB is identical. Considering the cost c) Set i = 1 and j =1.
of additional signaling, Delay Tolerance Factor (DT) and 2) Determine wsk in MCE, where wsk is defined as (17)
Fairness Penalty Factor (FP) are included to guarantee the for all k {1,2,,K}. Then, MCE informs eNBs of the
throughput and fairness among the streams. Such scheduling is results of resource allocation.
called Time-out-Based Scheduling Strategy (TBS) here. 3) eNBs receive the MCH Scheduling Information (MSI).
delayHoL 4) While j T or k = 0, k, in eNBs
(DT) DTk = (14) a) While i N
PDBk i. Find k *where it is defined as (19).
delayk , HoL = t Tk , HoL , (15) ii. Find the m* as (20) for a given k *, then the
1 selected packet is allocated to the pair i of RBs
(FP) FPk = . (16) in TTI j.
scheduled _ totalk
iii. Update k, k , according to (18),
received_totalk
iv. Update i = i+1.
where delayk,HoL is the period from the time spot Tk,HoL, i.e., b) Update j =j+1.
when the head of line (HOL) packets arrived at the queue, to c) Set i =1.
the current time spot t for the streaming k . PDBk is the Packet
5) The procedure of resource allocation is complete.
Delay Budget for video streaming k indicated by QCI.
Since the interval time of scheduling is enlarged, our B. QoS of Packet Loss
scheme is suboptimum in the case of conventional scheduling The QoS performance of Packet Error Loss Rate (PELR) is
strategies at TTI-level. However, from the view of realization, considered in this section. Fig. 3 shows a comparison in PELR.
it ensures that one TTI is allocated to one stream. Moreover, Obviously, PELR increases as transmission capacity decreases.
differing from the current semi-dynamic scheduling in LTE For a given streaming scheduling strategy, slight gains could
system, QoS of packet delay, fairness and the frame weight be observed by FIFO when compared with CS, but the gap is
are considered in the long-term and short-term scheduling negligible. Overall, our scheme outperforms the current
respectively, to aim to approach the performances achieved by scheme. Suppose that the constraint of PELR is 2% or 5%, our
the conventional strategies. Comparisons are shown in Sec. V. scheme could save at least one MBSFN subframe in one radio
frame when compared with the current scheme. Similarly,
V. PERFORMANCE ANALYSIS under the constraints of the same amount of radio resources,
For the purpose of comparisons, the current semi-dynamic the QoS performance improves significantly by our scheme.
scheduling in LTE system is introduced, i.e., FIFO and FIX,
where transmission order is set to be FIX, and FIFO means 1

that the packets arrived earlier in the queue are scheduled MEDF
earlier in eNB. In this section, our scheme (TBS and CS) is 0.9 PROPOSED
compared with (FIX and FIFO), as well as MEDF and MMC.

Spectrum Efficiency(bit/s/Hz)
CURRENT
The detailed parameters are listed in Table I [1] [18]. To 0.8
simplify our analysis, we assume that all the transmitted
packets are received correctly by the receivers.
0.7
In order to combat the effect of frame losses, the decoders
MEDF
employ the simple whole-frame error concealment technology, MMC
0.6
[19]. There are five video sequences of YUV QCIF (176144) CURRENT
(News, Claire, Container, Foreman and Highway) to evaluate FIX+CS
the performances [20]. H.264 JM encoder and decoder are 0.5 TBS+FIFO
used, Quantization Parameter (QP) is 20 and the frame rate is PROPOSED
30f/s. We assume that the GOP pattern is IPPPPPPPPPPP. 0.4
PDB is 200ms as TV programs of Live Sports in e-MBMS [3], 3 4 5 6 7 8 9 10
Number of MBSFN Subframes in One Radio Frame
which are common to all the video sequences.
Figure 2. Comparison of Spectrum Efficiency

TABLE I. E-MBMS SIMULATION PARAMETERS SETTING

Parameters Setting 0.8


Transmission Bandwidth 5MHz MEDF
Total number of pairs of available RBs for 25 0.7 MMC
e-MBMS system (12 subcarriers/RB) CURRENT
Transmission Time Interval(subframe) 1ms 0.6 FIX+CS
Packet Error Loss Rate

Radio Frame Duration 10ms CURRENT TBS+FIFO


MCH Scheduling Period 80ms2 0.5 PROPOSED
Total number of symbols per MBSFN subframe 6
Modulation and Coding Scheme 16QAM 1/2 0.4

A. Spectrum Efficiency 0.3


PROPOSED
Fig. 2 dedicates a relationship between spectrum efficiency 0.2
(SE) and the number of MBSFN subframes in one radio frame MEDF
with different solutions. When the total number of MBSFN 0.1
subframes is less than 5, the performances of SE are almost 0.05
0.02
the same. The reason is that the buffers in eNB keep almost 3 4 5 6 7 8 9 10
full under the limits of the number of MBSFN subframes, i.e., Number of MBSFN Subframes in One Radio Frame
the transmission capacity. With the increase of transmission
capacity, the gap of performance appears until it exceeds the Figure 3. Comparison of QoS of Packet Loss
achievable rate. MEDF outperforms MMC, because MMC
guarantees the transmission of I frames, while it causes larger
delay of the lower priority frames. Our proposed scheme C. Video Quality
outperforms current scheme (FIX+FIFO), the performance Video quality is accepted as another important metric of
even approaches to that of MEDF. The reason is that TBS performance for video transmission. We evaluate the results of
guarantees the delay constraint of such service, in order to Peak Signal to Noise Ratio (PSNR) with different solutions in
improve the throughput. Meanwhile, larger delay of the Fig. 4. Obviously, MMC outperforms other solutions. As for
streams, which are always scheduled later in FIX, could cause our proposed scheme, significant improvements could be
more packets discarded due to expiration. observed compared with the current scheme. The reason is

2
MSP is configurable, eg.80~320ms, where it is associated with the scheduling
capability of eNB, such as cache capability and CPU processing capability [13]
In this paper, we assume the MSP is 80ms in simulations.
that the performance of PELR decreases by TBS compared for separate video streams in the bundle in e-MBMS system is
with FIX. Furthermore, CS ensures the prior transmissions of concerned. With practical considerations, a suboptimum QoS-
I frames, which contribute more to PSNR. Consequently, the aware two-layer scheduling scheme is proposed towards LTE-
performance of PSNR is improved by our scheme. A system. It is divided into the streaming scheduling and
packet sorting by introducing dynamic QoS-related factors,
D. Fairness such as the packet urgency and fairness among the streams.
Based on the fact that multiplexed streams are always with Streaming scheduling determines the transmission order of the
the same QoS requirements, the performance of fairness multi-streams in MCE. And then, packet sorting ensures the
among the streams is considerable. The variance of PELR is transmissions of more important packets in eNB. Simulation
shown in Fig. 5. The results illustrate that the variance of results show that significant gains could be observed by our
PELR of TBS is much smaller than that of DBS, no matter scheme in terms of the spectrum efficiency, packet loss, video
what packet sorting strategy is used. TBS even obtains better quality and the fairness among the streams when compared
performance gains in fairness than MEDF. The reason is that with the current scheduling strategy.
TBS utilizes the FP as the fairness constraint. The stream
where the fewer packets have been transmitted will be prior REFERENCES
scheduled, so as to balance the opportunities of transmission [1] O. Oyman and J. Foerster, Toward Enhanced Mobile Video Services
among the streams. over WiMAX and LTE, IEEE Commun. Mag., vol. 48, no. 8, pp. 68-76,
August 2010.
[2] L. Rong, O. Ben Haddada, etc., Analytical Analysis of the Coverage of
40
a MBSFN OFDMA Network, in Proc. IEEE GLOBECOM, pp. 1-5,
38 December 2008.
[3] 3GPP TS 26.246, Multimedia Broadcast/Multicast Service (MBMS);
36
MMC Protocols and codecs (Release 9), v9.4.0, September 2010.
PROPOSED
[4] 3GPP R3-091300, ABBR based Statistical Multiplexing Solution for e-
34
MBMS, Alcatel-Lucent, May 2009.
Average PSNR(dB)

32 [5] X. Wang Y. Zhang, Optimal Video Stream Multiplexing in MBSFN,


CURRENT
in Proc. IEEE ICCTA, pp.360-365, October 2009.
30 [6] Y. Chen Statistical Multiplexing for LTE MBMS in Dynamic Service
Deployment, in Proc. IEEE VTC, pp.2805-2809, May 2008.
28 MEDF
[7] 3GPP R3-102207, Considering MBR larger than GBR in e-MBMS,
MMC
26 Ericsson, August 2010.
CURRENT
[8] M. Andrews, Probabilistic End-to-End Delay Bounds for Earliest
FIX+CS
24 Deadline First Scheduling, in Proc. IEEE INFOCOM, March 2000.
TBS+FIFO
[9] G. Liebl, K. Kalman, and B. Girod, Deadline-Aware Scheduling for
22 PROPOSED
Wireless Video Streaming, in Proc. IEEE ICME, April 2005.
20 [10] P. V. Pahalawatta, R. Berry, etc., Content-aware Resource Allocation
3 4 5 6 7 8 9 10
and Packet Scheduling for Video Transmission over Wireless Networks,
Number of MBSFN Subframes in One Radio Frame
IEEE J.Select. Areas Commun., vol. 25, no. 4, pp. 749759, May 2007.
[11] P. Hosein and T. Gopal, Radio Resource Management for Broadcast
Figure 4. Comparison of Average PSNR Services in OFDMA-Based Networks, in Proc. IEEE ICC, pp.271-275,
May 2008.
[12] 3GPP TS 36.300, Evolved Universal Terrestrial Radio Access (E-
1 UTRA) and Evolved Universal Terrestrial Radio Access Network (E-
MEDF UTRAN); Overall description; Stage 2 (Release 10), v10.0.0, June
0.9 CURRENT 2010.
0.8
FIX+CS [13] 3GPP R3-091856,e-MBMS Dynamic Service Multiplexing and
Variance of Packet Loss Rate

TBS+FIFO Signaling on M2 Interface, Alcatel-Lucent, August 2009.


0.7 PROPOSED [14] C. Suh and J. Mo, Resource Allocation for Multicast Services in
Multicarrier Wireless Communications, IEEE Trans. Wireless
0.6
Commun., vol. 7, no. 1, pp.27-31, January 2008.
0.5
[15] H. Qi, L. Qiu, etc., An Adaptive Scheduling Algorithm for
Conversational Video Services in LTE System, submitted to Proc.
0.4 IEEE CHINACOM.
[16] T. Wiegand, J. Sullivan, etc., Overview of the H.264/AVC Video
0.3
Coding Standard, IEEE Trans. on Circuit and System for Video
0.2 Technol., vol. 13, no. 7, July 2003.
[17] P. Svedam, S. Wilson, etc., A QoS-aware Proportional Fair Scheduling
0.1 for Opportunistic OFDM, in Proc. IEEE VTC, pp.558-562, September
2004.
0
3 4 5 6 7 [18] 3GPP TS 36.331, Evolved Universal Terrestrial Radio Access (E-
Number of MBSFN Subframes in One Radio Frame UTRA); Radio Resource Control (RRC); Protocol specification (Release
9), v9.3.0, June 2010.
[19] S. Belfiore , M. Grangetto etc. ,Concealment of Whole-frame Losses
Figure 5. Comparison of Fairness
for Wireless Low bit-rate Video based on Multiframe Optical Flow
Estimation, IEEE Trans. Multimedia, vol. 7, no. 2, pp.316-329, April
2005.
VI. CONCLUSION [20] [Available on Line]: http://trace.eas.asu.edu/yuv/index.html.
In this paper, scheduling and resource allocation problem

You might also like