Video streaming transmission: performance modelling ... - IEEE Xplore

www.ietdl.org Published in IET Communications Received on 10th November 2010 Revised on 7th October 2011 doi: 10.1049/iet-com.2010.1013

ISSN 1751-8628

Video streaming transmission: performance modelling over wireless local area networks under saturation condition X.W. Yao1 W.L. Wang1 S.H. Yang2 1

College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou, 310023, People’s Republic of China 2 Department of Computer Science, Loughborough University, Loughborough, LE11 3TU, UK E-mail: [email protected]

Abstract: Transmitting delay-sensitive video streaming over IEEE 802.11e wireless local area networks (WLANs) is becoming increasingly popular. However, the transmission of real-time video streaming is very challenging because of the time-varying wireless channels and video content characteristics. The authors propose an accurate model to assess the perceived quality of video streaming over WLANs with enhanced distributed coordination function (EDCF) mechanism. The analytical model considers not only the packet loss caused by wireless interference and channel fading, but also the effects of loss from channel access competition. Based on the Markov chain, the authors then present the discrete probability distribution of medium access control (MAC) layer packet service time by using the signal transfer function of the generalised state transition diagram. Moreover, the coding relation of lost video frames is also explored in the performance analysis of the proposed model. Simulations based on Network Simulator 2 (NS-2) are conducted to verify the performance of the analytical model. The results show that the proposed model provides superior accuracy for the perceived quality of MPEG-4 video streaming over IEEE 802.11e EDCF-based WLANs.

1

Introduction

Wireless local area networks (WLANs) based on the IEEE 802.11 distributed coordination function (DCF) have been widely used in recent years because of their simple deployment and low cost. Since the current DCF can only support best effort traffic, to overcome the quality-of-service (QoS) problem, the IEEE 802.11 working group has designed a new protocol designated IEEE 802.11e [1] by classifying traffic through the introduction of four service classes. Enhanced distributed channel access (EDCA) provides service differentiation by allowing each class to have its own set of QoS parameters. Meanwhile, with the increase in network coverage and increase in data rates and bandwidth, more users are watching video streamings over wireless networks [2, 3]. Thus, accurate evaluation of video quality delivered over wireless networks becomes an important research issue. In recent years, based on Bianchi’s work [4], several papers have analysed the performance of IEEE 802.11e EDCF using analytical models. Hui and Devetsikiotis [5] provided the saturation throughput and delay by using a transmission probability matrix with a unified model. However, the model did not consider the backoff counter freezing before the end of arbitration inter frame space (AIFS). Zhu and Chlamtac [6] presented the throughput and delay under saturation conditions, but the AIFS difference was IET Commun., 2012, Vol. 6, Iss. 1, pp. 13 –21 doi: 10.1049/iet-com.2010.1013

considered with only the highest priority class. In [7], Lee and Lee provided the saturation throughput for all access categories (ACs) with arbitrary contention windows and AIFS values with a new Markov chain model. Previous studies on the performance analysis of 802.11e EDCF focused on the throughput, and generally evaluated the MAC access delay using only the mean of access delay [8 – 10], but lack of the research on the probability distribution of the packet service time at the MAC layer. To address the above issue, Zhai et al. [11] first characterised the probability distribution of the MAC layer packet service time, but the analysis model only focused on DCF mechanism, did not take the more complicated EDCF mechanism into consideration. However, the analyses of these papers focused on system performance and were not sufficient to evaluate the video quality over IEEE 802.11 EDCF [12]. For example, the significance of video frame type is different for compressed frames with MPEG-4 video coding [13]. In, Ziviani et al. [14] and Ke et al. [15] proposed a performance metric, decodable frame rate (DFR), to analyse the video quality of MPEG streaming. But, the wireless transmission is only assumed average packet loss rate in wireless network without considering the actual wireless channel loss. Lin et al. [16] proposed model considers not only the loss of the wireless channel but also transmission collision. However, these studies only focused on the DCF 13

& The Institution of Engineering and Technology 2012

www.ietdl.org mechanism and did not consider the effect of access delay to video transmission. To achieve a more realistic analysis, we propose an analytical model that evaluates the video quality of MPEG4 video streaming over IEEE 802.11e EDCF WLANs. In this paper, our proposed model considers both wireless channel loss and transmission collisions. Furthermore, we use the signal transfer function of the generalised state transition diagram to derive an approximate probability distribution of the MAC layer service time and take the video play-out time into consideration. Finally, the validity of the proposed model is verified by comparing analytical results with those of simulation results. The remainder of this paper is organised as follows: Section 2 models the packet loss of IEEE 802.11e EDCF including wireless loss and collision loss, and derive the probability distribution of access delay. The proposed new analytical model of video quality is discussed in Section 3. Section 4 illustrates the results using simulations designed to validate the proposed model. Moreover, we used more realistic parameters from analytical model to guarantee the QoS of MPEG-4 video streaming. Finally, Section 5 draws the conclusions and future works.

2

Video transmission model

In this section, we describe our proposed analytical model of video transmission in detail. To accurately evaluate the transmission of video frames over IEEE 802.11e EDCF WLANs, the analytical model is divided into two parts: first, the packet loss including wireless loss and collision loss; and second, the time distribution of video frame to be serviced. However, most of the previous studies did not take the video packet loss caused by long delay. Table 1 describes all the important notations used in the paper. We model the WLANs as a stochastic directed graph G(V, E), where V is the set of the vertices representing the nodes in a wireless network, and E is the set of directed edges representing the wireless links between the nodes. In the process of video transmission, the packet loss includes not only the packets dropped during the transmission, but also the overdue packets arriving at the receiver beyond the play-out deadline, especially the real-time video traffic. We characterise the link {i, j} [ E by (i) tij: the delay of link {i, j}; (ii) pijloss : the packet loss rate of link {i, j}. Assume that the path for delivering video packets be Ev , E. For each link, we define an index variable Xij to indicate whether the link {i, j} belongs to path Ev Xij =

1 if 0 if

{i, j} [ Ev {i, j} Ev

(1)

And the end-to-end packet delay of path Ev is given by T (Ev ) =

Xij tij

(2)

{i, j}[Ev

which means the delay of the path is equal to the sum of the delays of all its links. Assume that D denotes the value of play-out deadline, then the total end-to-end packet failure probability for the path can be computed as Pf (Ev ) = 1 −

ij (1 − Xij Ploss ) P(T (Ev ) ≤ D)

{i, j}[Ev

14


(3)

Table 1

Adopted notation used in our proposed analytical

model Notation tij T(Ev) D ij Ploss Pf (Ev) M N Wn,i CWmin,n pn dn tn ln Ts,n Z PTs,n (Z ) Ct,n(Z) St,n(Z) Dt,n(Z) Ndec/I/P/B Ntotal – send NGOP CI , CP , CB NP , NB

Definition the propagation delay of link {i, j} the end-to-end delay of path Ev the play-out deadline the packet loss ratio of link {i, j} the total packet failure probability for path Ev the number of ACs the number of stations the size of AC[n] contention window in backoff stage i + 1 minimum CW size of AC[n] the collision probability of AC[n] the state transmission probability of AC[n] the probability that AC[n] accesses the channel in a randomly chosen time slot the number of retransmission the MAC layer service time for a priority n packet Z-transform the probability generating function of MAC layer service time Ts,n the PGFs of a collision period the PGFs of a successful transmission period the PGFs of a decrement of the backoff timer expected number of decodable I, P and B frame the total number of sent video frames total number of GOPs in the video streaming mean number of packets from the I, P and B frame number of P and B frame in a GOP

where P(T (Ev) ≤ D) indicates the probability of propagation delay over the path Ev not beyond the play-out deadline. ij And the part calculates the {i, j}[Ev (1 − Xij Ploss ) probability of the video packets be received correctly all over the links. According to the pre-defined topology of a wireless network (i.e. fixed sender node, receiver node and other wireless nodes), with special routing algorithm (e.g. AODV or DSR etc.), the best routing path can be selected, viz. the variable Ev , {i, j} [ Ev and Xij can be determined. So, to obtain the end-to-end video packets failure probability Pf(Ev), we will calculate the packet loss ratio ij Ploss and the probability of overdue video frame PD ¼ P(T (Ev) ≤ D) in the following sections. 2.1

Analysis of packet loss over wireless network

In the IEEE 802.11e EDCF mode, the loss of a transmission frame can be caused by two factors: first, wireless loss caused by wireless interference and fading; and second, collision loss resulting from channel access competition. We assume that at each transmission attempt, regardless of the number of retransmissions, each frame has a constant and independent failure probability. So the frame failure transmission probability can be expressed as ij Ploss = 1 − (1 − Pcij )(1 − Peij ) = Pcij + Peij − Pcij Peij

(4)

where Pcij and Peij represent the probability of collision loss and wireless loss, respectively. IET Commun., 2012, Vol. 6, Iss. 1, pp. 13– 21 doi: 10.1049/iet-com.2010.1013

www.ietdl.org According to the analysis of wireless loss in [16], the probability of wireless loss can be expressed as Peij = 1 − (1 − Pb )Ldata +Lack

(5)

where Pb is the bit error probability (BEP) in the wireless channel, Ldata and Lack are the length of a data frame and an ACK frame in bits, respectively. Then the frame failure transmission probability can be calculated as ij Ploss = Pcij (1 − Peij ) + Peij = 1 − (Pcij − 1)(1 − Pb )Ldata +Lack

(6) Under the same wireless channel conditions and the same modulation scheme [e.g. multiple quadrature amplitude modulation (MQAM)], which means that the value of BEP Pb is constant, a larger frame size results in higher frame loss probability. So there is a trade-off between frame error rate and transmission overhead when it comes to frame sizes. 2.2

Collision loss model of EDCF

With EDCF, traffic of different priorities is assigned to one of four transmission queues, which correspond to four ACs, that is, AC_VO (for voice traffic), AC_VI (for video traffic), AC_BE (for best effort traffic) and AC_BK (for background traffic). To simplify the notations, we rename four ACs as AC(3), AC(2), AC(1) and AC(0) from the high priority to low priority in the rest of this paper. Each queue transmits packets with an independent channel access function, which implements the prioritised channel contention algorithm. In other words, different channel access function uses different contention windows (the minimum and maximum contention windows) and backoff timers.

So, only when one access category of a station’s backoff timer drops to zero and others do not, a frame has the opportunity to be transmitted. However, collisions may occur if multiple active stations start transmissions simultaneously, which are called external collision. And, because of the multiple ACs of one station, collisions may also occur if multiple access categories start transmissions simultaneously, which are called internal collisions. To calculate the collision loss rate, we make the following assumption: I. Each priority queue is treated as an independent agent. II. External collision and internal collision are independent. Note that in this paper, we assume a fixed number of stations (N ). Each station has multiple ACs, and the system is in the saturation condition. In the model, time is considered to be slotted and each state represents an AC in a time slot. At the end of each time slot an event that triggers a transition to another state occurs as shown in Fig. 1. Let mn and ln denote the maximum number of backoff stage and retransmission. If after ln + mn retries, the AC still cannot access the medium successfully, then the packet is discarded, in other words, ln + mn is the retry limit. According to the analysis of Markov Chain under the assumption that pn (the collision probability of AC[n] are independent of the backoff procedure) as shown in [7, 10], we can calculate the probability tn that AC[n] accesses the channel in a randomly chosen time slot. (see (7)) Then we can obtain the collision probability of AC[n], which includes the external collision and internal collision. pn = 1 −

M −1

(1 − ti )N /(1 − tn )

(8)

i=0

To achieve differentiation, instead of using fixed distributed interframe space (DIFS) as in 802.11 DCF, EDCA assigns

Fig. 1 Markov chain model of the EDCA

⎧ 2dn (1 − 2pn )(1 − plnn +1 ) ⎪ ⎪ ⎪ , ⎪ l +1 ln +1 ⎨ CW )(1 − pn ) + (1 − 2pn )(1 − pnn ) min ,n (1 − (2pn ) tn = ⎪ 2dn · (1 − plnn +1 /1 − pn ) ⎪ ⎪ , ⎪ ⎩ l +1 l −m CWmin ,n (1 − (2pn )ln +1 /1 − 2pn ) + (1 − pnn /1 − pn ) + CWmin ,n (2pn )mn pn (1 − pnn n /1 − pn ) IET Commun., 2012, Vol. 6, Iss. 1, pp. 13 –21 doi: 10.1049/iet-com.2010.1013

ln ≤ mn (7) ln . mn

15


www.ietdl.org higher-priority ACs with smaller AIFS to influence the successful transmission probability in favour of highpriority ACs [1], where the AIFS of a given AC[n] is determined by the following equation AIFS[n] = SIFS + AIFSN[n]∗ aSlotTime

(9)

where AIFSN[n] is value of AIFSN and determined by the special AC and physical settings, and aSlotTime is the duration of a time slot. The AC with the smallest AIFS has the highest priority, and a node needs to defer for its corresponding AIFS interval. The smaller the parameter values, the greater the probability of gaining access to the medium. So, when a collision occurs among different ACs in the same station, the higher priority AC is granted the opportunity to transmit, whereas the lower-priority AC suffers from a virtual collision. In the EDCA, after the AIFS period, the backoff counter decreases by one at the beginning of the last slot of the AIFS (as illustrated in Fig. 2). So we cannot use the probability of idle at some time for the state transmission probability dn directly. In fact, the backoff counter decreases to a sufficient condition that the channel is idle for at least AIFS slot times. So we can obtain the state transmission probability dn =

AIFS[n] pidel

=

M −1

AIFS[n] (1 − ti )

N

(10)

i=0

Based on the combination of the above non-linear equations (7) – (10), the channel access probabilities (t0 , t1 , . . . ,tM21) and collision probabilities ( p0 , p1 , . . . , pM21) of each AC are determined. According to the IEEE 802.11e specification, there are four ACs, that is, M ¼ 4. When transmitted over an IEEE 802.11e wireless network, video packets are placed in AC [2], which has a better opportunity to access the channel than lower priority. So we obtain the probability of collision loss: Pcij = p2 .Then, ij through (4) and (5), the frame failure probability Ploss over link {i, j} is determined. ij As expressed by (4), Ploss is the loss probability of a frame transmitted over IEEE 802.11e wireless networks. However, retransmission increases the successful received probability of a transmission frame. In the proposed analytical model, we assume that each frame can be retransmitted by a maximum number of ln times, before being discarded by the sender. The effective loss probability of a frame with a maximum retransmission time is ij Ploss

=

ln l n

i=0

i

× pi2 × ((1 − p2 )Peij )ln −i

Fig. 2 IEEE 802.11e EDCA access procedure 16


(11)

The parameter i represents the number of failed transmission of a frame, when a collision or a channel error event occurred. The probability of frame lost from collision is denoted by p2 = Pcij ; the probability of frame lost caused by wireless error is denoted by ((1 − p2 )Peij ). According to the analysis of packet loss over wireless network in Section 2.1, the effective packet loss probability with a maximum retransmission can be rewritten as ij Ploss =

ln ln i=0

i

× pi2 × ((1 − p2 )(1 − (1 − Pb )Ldata +Lack ))ln −i (12)

2.3

Transmission delay model of EDCF

Under the 802.11e EDCA scheme, we consider the effect of access delay in terms of the MAC layer packet service time, and derive the probability distribution of the MAC service time for using the method proposed in [11]. The MAC layer service time for a priority n packet, denoted by Ts,n , is the time interval from the time instant that a packet starts to contend for transmission, to the time instant that the packet either is acknowledged for correct reception by the intended receiver or is dropped. To derive the probability distribution of the MAC layer service time, we choose to calculate the probability generating function (PGF), which is the Z-transform of the probability distribution function denoted by PTs,n (Z). Specifically, we denote the PGFs of a collision period, a successful transmission period and the decrement of the backoff timer as Ct,n(Z ), St,n(Z ) and Dt,n(Z ). Based on the request to send/clear to send (RTS/CTS) mechanism [11], the collision period associated with a priority n AC is RTS + SIFS + CTS + AIFS[n], then Ct,n(Z ) can be obtained as Ct,n (Z) = Z RTS+SIFS+CTS+AIFS[n]

(13)

Similarly, we can obtain St,n(Z ) as St,n (Z) = Z RTS+CTS+3 SIFS+DATAn +ACK+AIFS[n]

(14)

where DATAn is the average packet transmission time in a successful transmission period for AC[n]. As described in Section 2.2, given the collision probability pn and the probability of one AC transmits successfully ps,n , we show one simple example for the state transition of Markov chain model in Fig. 3 (s indicates the time slot). When the medium has been idle for AIFS period, the backoff timer will decrease by one at the beginning of the last AIFS slot, that is, the state transition from the state ‘k ’ to ‘k 2 1’. And the PGF of the duration of time taken for the state transition from the state ‘k ’ to ‘S ’(successfully transmitted) is ps,n St,n(Z ). When detecting an ongoing successful transmission, the backoff timer will be suspended, which means the state

Fig. 3 Signal flow graph for one step backoff IET Commun., 2012, Vol. 6, Iss. 1, pp. 13– 21 doi: 10.1049/iet-com.2010.1013

www.ietdl.org transition from the state ‘k ’ to ‘C’(collision). Thus, by using the well-known Mason formula, the PGF of the MAC service time PTs,n (Z) can be obtained from the signal transfer function of the generalised state transition diagram as described in [11, 17] PTs,n (Z) = (1 − pn )St,n (Z)

ln

(pn Ct,n (Z))

j=0

+ (pn Ct,n (Z))ln +1

j

j

GSk,n (Z)

k=0

ln

GSk,n (Z)

k=0

(15) where

GSk,n (Z) =

⎧ k 2 Wn,0 −1 ⎪ Dt,n (Z) ⎪ ⎪ , ⎪ ⎨ 2k W i=0

(0 ≤ k ≤ mn )

i,0

⎪ 2k Wn,0 −1 ⎪ Dt,n (Z) ⎪ ⎪ , ⎩ 2mn Wi,0 i=0

(16) (mn ≤ k ≤ ln )

Note that PTs,n (Z) is a function of the collision probability pn . Once the PTs,n (Z) is obtained, both the mean and variation of the MAC service time can be derived by taking the derivative with respect to Z ¼ 1. At the same time, we can obtain the probability of overdue frame PD .

3

Our proposed analytical model

In this section, we describe our proposed analytical model in detail. To accurately evaluate the perceived quality of video streaming over IEEE 802.11e EDCF-based WLANs, the analytical model considers the transmission delay of video frames caused by wireless channel competition. Moreover, the loss effect of video frames on the video quality is also considered in terms of coding dependencies of different frame type. Finally, the video quality can be assessed by mapping the packet loss ratio and transmission timeout ratio of the wireless network to analytical model of video quality. 3.1

Analytical model for video transmission

Today, the development of video encoding standards has made video transmission over wireless network possible and efficient, especially the latest international hierarchical video coding standard H.264 and MPEG-4, which provides a good video quality at substantially lower bit rates without increasing the complexity of the design. The encoding method is based on the idea that coding will be in the form of quality hierarchy where the lowest layer of hierarchy contains the minimum information for intelligibility, succeeding layers of the hierarchy adds increasing quality to

the scheme. In this paper, we focus on the MPEG-4 standard, which defines three types of video frames for the compressed video stream, that is, intra-coded frame (I-frame), predictive-coded (P-frame) and bi-directionally predictive-coded (B-frame). An I-frame is in effect a fully specified picture, like a conventional static image file. A P-frame (Predicted picture) holds only the changes in the image from the previous frame. A B-frame saves more space by using differences between the current frame and both the preceding and following frames to specify its content. Thus, the I-frame is just a frame coded as a still image, without any relationship to any previous or successive frames. Which means: I-frame is encoded independently and decoded by itself. The P-frame is encoded using prediction from the preceding I-frame or P-frame in the video sequence. Thus, the P-frame requires the information of the most recent I-frame or P-frame for encoding and decoding. The B-frame is encoded using predictions from the preceding and succeeding I-frame or P-frames. That is, the B-frame is predicted from the two closest I-frame or P-frames from the past and the future. In MPEG-4 coding, the video sequence can be decomposed into smaller units, group of pictures (GOP). A GOP pattern G(N, M ) is characterised by two parameters: where N is the I-to-I frame distance and M is the I-to-P frame distance, that is, there are one I-frame, (N/M 2 1) P-frames and (N . ((M 2 1)/M )) B-frames in a GOP (as shown in Fig. 4 (N ¼ 12, M ¼ 3)). According to the above analysis of coding relation, the most important video type is the I-frame in MPEG-4 video stream, with the P-frame being more important than B-frame. The impact of a lost B-frame is only for itself, but a loss of I or P-frame may cause more than one frame or even a whole GOP to be destroyed. For simplicity, we do not consider any error concealment method in this paper, and set the decoded threshold (DT) to 1.0, that is, the video frame is considered as decodable only when all packets belonging to the frame are received. Therefore the analysis of our model provides the worst-case boundary of video quality degradation because of the errors in transmission, since one packet loss leads to an un-decodable frame. In this paper, the commonly used metric (i.e. decoded frame rate (DFR) metric [14]) is adopted to access the quality of video streams transmitted over wireless packetswitched and lossy networks. DFR is defined as the sum of the decodable frames over the total number of frames sent by a video source DFR =

Ndec Ntotal−I + Ntotal−P + Ntotal−B

(17)

where Ndec ¼ Ndec2I + Ndec2P + Ndec2B is the summation of the successfully decoded frames in the video stream. Based on the probability theory, we calculate for each frame type

Fig. 4 Prediction encoding of MPEG-4, GOP (N ¼ 12, M ¼ 3) IET Commun., 2012, Vol. 6, Iss. 1, pp. 13 –21 doi: 10.1049/iet-com.2010.1013

17


www.ietdl.org Table 2

Parameter setting used in EDCF

Priority 3 2 1 0

AC

Designation

AIFSN

CWmin

CWmax

TXOPlimit

AC_VO AC_VI AC_BE AC_BK

voice video best effort background

2 2 3 7

7 15 31 31

15 31 1023 1023

0.003008 0.006016 0 0

(i.e. I-frame, P-frame and B-frame) the probability to be successfully decoded by using the model proposed in [15, 16]. The I-frames in a GOP are successfully decoded only if all the packets that belong to an I-frame received intact and the end-to-end delay is not beyond the playout time D. Therefore the expected number of correctly decoded I-frames for the whole stream is Ndec−I = (1 − Pf (Ev ))CI · NGOP

(18)

Similarly, the expected number of correctly decoded P-frames and B-frames which are received without errors and timeout for the whole video stream are Ndec−P = (1 − Pf (Ev ) ) × CI

NP

(1 − Pf (Ev )) jCP × NGOP

j=1

(19) Ndec−B = (1 − Pf (Ev ))CI +NP CP +

NP

(1 − Pf (Ev ))jCP

× (1 − Pf (Ev ))

4.1 Model validation in different network conditions

× NGOP (20)

Based on the combination of the above equations, the proposed analytical model (10) can provide an accurate evaluation of the perceived quality of MPEG-4 video over the 802.11e EDCF-based WLANs (Pf(Ev) can be calculated according to the parameters listed in Table 2).

4

packet size over the simulated network is 1024 bytes. All the parameters of the above two trace files are shown in Table 3. Moreover, the transmission is in unicast mode. The data rate of the wireless link is 1 Mbps, and the queue size of all ACs are limited to a maximum of 50 packets.

j=1

× (1 − Pf (Ev ))CB × (M − 1) CI +CB

Fig. 5 Simulation topology

Performance evaluation and discussions

In order to evaluate the performance of the proposed analytical model, we set up simulations using the widely adopted network simulation tool NS-2 and integrated with Evalvid [18]. In the simulations, we consider a wireless local area network with n stations operating at saturation conditions, which means that each station always has a packet available for transmission. The network topology is shown in Fig. 5. Video traffic with the second priority is sent from a randomly selected node and received by another randomly chosen destination node. Meanwhile, other nodes transmit UDP flows as a CBR flow with different level priorities. The test video traces selected in the simulations are YUV QCIF (176∗ 144 pixels) ‘Star Wars IV’ sequence and ‘Mr.Bean’ sequence. The adopted video codec is ‘ffmpeg’. Through video coding, each video frame is fragmented into packets before transmission, and the maximum transmission 18


In the simulations, we adapt the standard physical layer parameter of IEEE 802.11e. Assume that the BEP Pb is given according to the modulation scheme, in order to obtain the accurate loss probability of video frame as shown in (6), we calculate the collision loss probability based on the Markov chain including internal and external collision of the station. Fig. 6 shows the comparison between internal and external collision probability of the analytical model under different number of active stations and various retransmissions. Intuitively, as the number of active nodes increases, the performance of AC [2] degrades dramatically, especially, in terms of rapid increasement of collision probability. At the same time, a suitable value of retransmission can relieve the total collision probability. When increasing the number of wireless nodes, the curve of AC [2] collision probability is rigid monotone increasing, whereas the curve of its internal collision probability is strictly monotone decreasing, which means collision mainly comes from external collision. To validate the proposed model, Fig. 7 shows the comparison of simulation results obtained from the model described in [16] with results obtained from our proposed analytical model, and without considering the access delay. Table 3

Parameter settings of the video streaming

Video_Source

NGOP

NP

NB

CI

CP

CB

Star Wars IV Mr.Bean

7500 7428

3 3

4 4

3.836 6.832

2.015 3.8064

1.4929 2.736

IET Commun., 2012, Vol. 6, Iss. 1, pp. 13– 21 doi: 10.1049/iet-com.2010.1013

www.ietdl.org

Fig. 8 DFR of video streaming under different number of AC [3] traffic and BEP (Source file: StarWarsIV; Play-out: 0.008 s)

Fig. 6 Comparison between AC [2] collision probability and its internal collision probability

Fig. 9 DFR of video streaming under different number of AC [1] traffic and BEP (Source file: StarWarsIV; Play-out: 0.008 s)

Fig. 7 Quality of video stream obtained with different BEP rates (every priority has one traffic flow)

Assume that there are four traffic flows with different-level priorities. Our proposed model matches the simulation results much closer than the model in [16]. In the model described in [16], the authors do not consider the differentiation of traffic flows, which is not likely in any practical network environment, especially for the real-time video transmission, for example, video surveillance. Conversely, the proposed model based on IEEE 802.11e EDCA mechanism, which provides service differentiation by setting different QoS parameters, can accurately evaluate the quality of delivered video streams under different wireless channel conditions. To evaluate the effect of wireless channel and traffic conditions on transmission of video streaming, we first consider the effect of traffics with different-level priorities (i.e. AC [3] and AC [1], which represents the highest priority and lower priority, respectively) as shown in Figs. 8 – 11. Assume that the value of play-out time is 0.008 s. For the priority of AC [3] is higher than the priority of video traffic (i.e. AC [2]), with increasing the number of AC [3] traffic flows, the probability of internal collision raises rapidly, even under the better wireless channel conditions (i.e. low BEP). Although the priority of IET Commun., 2012, Vol. 6, Iss. 1, pp. 13 –21 doi: 10.1049/iet-com.2010.1013

Fig. 10 DFR of video streaming under different number of AC [3] traffic and BEP (Source file: Mr.Bean; Play-out: 0.008 s)

AC [1] is lower than the priority of video traffic, the DFR of delivered video stream will hold the line (almost 0.3 in Fig. 11) with increasing the number of AC [1] traffic flows. Moreover, from the view of Figs. 9 and 11, when the priority of background traffic flow is lower than the priority of AC [2], the BEP of wireless channel is the principal factor to degrade the video quality. Conversely, if the priority of background traffic flow is higher than the priority of AC [2], the internal collision coming from the competition of accessing channel is the main factor for bad video quality as shown in Figs. 8 and 10. It is also observed that when the wireless channel is very noisy 19


www.ietdl.org

Fig. 11 DFR of video streaming under different number of AC [1] traffic and BEP (Source file: Mr.Bean; Play-out: 0.008 s)

(BEP ¼ 1022), the video quality is degraded even if there is very light load. This is because most of the lost video frames are caused as a result of wireless errors during transmissions. 4.2

Fig. 13 Model validation in large play-out buffer size (Source file: Star Wars IV)

Model validation in different delay condition

In this section, we evaluate the effect of the play-out buffer size of the receiver on the video streaming quality. Which means when the end-to-end transmission time is beyond the constrained delay, the video frame will be un-decodable even all the packets are received correctly. As shown in Figs. 10– 13, the results of our proposed analytical model match the simulation results closely. Specially, for video sequence ‘Star WarsIV’, when the number of traffic load with higher priority increases, the node has less chance to access the channel to successfully transmit video frames for intense channel competition, which leads to the unwanted bad video quality for the users. As shown in Fig. 12, when there are more than three background traffic flows in the WLANs and the play-out time is 0.005 s, the DFR of the reconstructed video is less than 50%. However, with the increasement of play-out buffer size, almost all video packets are received correctly at the receiver except for the loss packet for channel error (BEP ¼ 5 × 1023). Similarly, under the same wireless channel condition and the same play-out time, the quality of the reconstructed video sequence ‘Mr.Bean’ is monotone decreasing with increasing the background traffic flows (Figs. 14 and 15).

Fig. 14 Model validation in small play-out buffer size (Source file: Mr.Bean)

Fig. 15 Model validation in large play-out buffer size (Source file: Mr.Bean)

5 Fig. 12 Model validation in small play-out buffer size (Source file: Star Wars IV) 20


Conclusion

In this paper, we investigate the perceived quality of MPEG-4 video streaming over IEEE 802.11e EDCF-based WLANs. The contributions of this paper are summarised as follows. IET Commun., 2012, Vol. 6, Iss. 1, pp. 13– 21 doi: 10.1049/iet-com.2010.1013

www.ietdl.org (I) Consider not only the effect of packet loss such as collision loss resulting from channel access competition but also wireless loss caused by wireless interferences. (II) Calculate the collision loss probability of 802.11e EDCA by modelling the exponential backoff process as a Markov chain. (III) In addition to modelling the saturation performance as in existing work, we focus on evaluating the delay performance and characterise the probability distribution of the packet service time at the MAC layer. (IV) Based on the probability distribution model of the MAC layer packet service, we then study the effect of packet delay on the video streaming under different traffic loads and wireless channel condition. The proposed model is validated by comparing our performance results of analytical models with results obtained from simulation.

6

References

1 IEEE Std. 802.11e-2005: ‘Part 11: wireless medium access control (MAC) and physical layer (PHY) specifications: medium access control (MAC) quality of service enhancements’. 2005 2 Zhou, L., Geller, B., Wang, X., Wei, A., Zheng, B., Chao, H.C.: ‘Multiuser video streaming over multiple heterogeneous wireless networks: A distributed, cross-layer design paradigm’, J. Internet Technol., 2009, 10, (1), pp. 1– 11 3 Tsai, M.F., Chilamkurti, N., Shieh, C.K.: ‘A network adaptive forward error correction mechanism to overcome burst packet losses for video streaming over wireless networks’, J. Internet Technol., 2010, 11, (4), pp. 473–481 4 Giuseppe, B.: ‘Performance analysis of the IEEE 802. 11 distributed coordination function’, IEEE J. Sel. Areas Commun., 2000, 18, (3), pp. 535–547 5 Hui, J., Devetsikiotis, M.: ‘A unified model for the performance analysis of IEEE 802.11e EDCA’, IEEE Trans. Commun., 2005, 53, (9), pp. 1498– 1510

IET Commun., 2012, Vol. 6, Iss. 1, pp. 13 –21 doi: 10.1049/iet-com.2010.1013

6 Zhu, H., Chlamtac, I.: ‘Performance analysis for IEEE 802.11e EDCF service differentiation’, IEEE Trans. Wirel. Commun., 2005, 4, (4), pp. 1779– 1788 7 Lee, J.Y., Lee, H.S.: ‘A performance analysis model for IEEE 802.11e EDCA under saturation condition’, IEEE Trans. Commun., 2009, 57, (1), pp. 56–63 8 Banchs, A., Vollero, L.: ‘A delay model for IEEE 802.11e EDCA’, IEEE Commun. Lett., 2005, 9, (6), pp. 508 –510 9 Huang, C., Liao, W.: ‘Throughput and delay performance of IEEE 802.11e enhanced distributed channel access (EDCA) under saturation condition’, IEEE Trans. Wirel. Commun., 2007, 6, (1), pp. 136–145 10 Kong, Z., Tsang, D.H.K., Bensaou, B., Gao, D.: ‘Performance analysis of IEEE 802.11e contention-based channel access’, IEEE J. Sel. Areas Commun., 2004, 22, (10), pp. 2095– 2106 11 Zhai, H., Kwon, Y., Fang, Y.: ‘Performance analysis of IEEE 802.11 MAC protocols in wireless LANs’, Wirel. Commun. Mobile Comput., 2004, 4, (8), pp. 917 –931 12 Dapeng, W., Yan, Z., Muqing, W., Xiaojing, Z.: ‘Medium access control access delay analysis of IEEE 802.11e wireless LAN’, IET Commun., 2009, 3, (6), pp. 1061–1070 13 Xie, G., Swamy, M., Ahmad, M.O.: ‘Joint optimal multipath routing and rate control for multidescription coded video streaming in ad hoc networks’, IEEE Trans. Multimedia, 2008, 10, (8), pp. 1687– 1697 14 Ziviani, A., Wolfinger, B.E., Rezende, J.F., Duarte, O.C.M.B., Fdida, S.: ‘Joint adoption of QoS schemes for MPEG streams’, Multimedia Tools Appl., 2005, 26, (1), pp. 59– 80 15 Ke, C.H., Lin, C.H., Shieh, C.K., Hwang, W.S., Ziviani, A.: ‘Evaluation of streaming MPEG video over wireless channels’, J. Mobile Multimedia, 2007, 3, (1), pp. 47–64 16 Lin, C.H., Chilamkurti, N.K., Zeadally, S., Shieh, C.K.: ‘Performance modeling of MPEG-4 video streaming over IEEE 802.11 using distribution coordination function’, Wirel. Commun. Mobile Comput., 2011, 11, (9), pp. 1312– 1322 17 Chen, X., Zhai, H., Fang, Y.: ‘Enhancing the IEEE 802.11e in QoS support analysis and mechanisms’. Conf. Quality of Service in Heterogeneous Wired/Wireless Networks, Lake Vista, FL, August 2005, pp. 8 –23 18 Network Simulator-2, http://www.isi.edu/nsnam/ns/

21
