Abstract
IEEE 802.11e utilizes the Enhanced Distributed Channel Access (EDCA) protocol to provide quality of service (QoS) support over WLANs. However, under EDCA, all video data are allocated to the same access category (AC) queue, irrespective of their coding importance. Consequently, the quality of the transmitted video may be severely degraded under heavy traffic load conditions. The literature contains several queue mapping mechanisms based on the settings of the AC queues, such as contention window size, for improving the video transmission quality over IEEE 802.11e WLANs. However, these mechanisms fail to consider the effect of the dequeuing time on the video delivery performance. Accordingly, the present study proposes a new queue mapping algorithm, designated as the throughput-based mapping algorithm (TMA), for improving the video transmission quality and reducing the video delay by dynamically allocating the video data packets to different AC queues based on their length and throughput such that the probability of the dequeuing time being reduced is increased. In implementing the proposed approach, the throughputs of the AC queues are analyzed using two Markov chain models. The simulation results show that TMA outperforms previously proposed static and dynamic queue mapping schemes in terms of a reduced AC queue length, an improved PSNR (peak signal-to-noise ratio), and a shorter transmission delay.
1. Introduction
Wireless communications play an essential role in modern life in facilitating the use of notebooks, mobile phones, and other mobile devices to surf the Internet, chat, receive e-mail, watch video streams, listen to music, and so forth. Among the various services supported by wireless networks, multimedia transmissions, such as video streaming, digital picture or video on demand, and so on, are both one of the most common and the most challenging. In response to massive user demand, IEEE 802.11 WLANs [1, 2] have been extensively deployed all over the world. Such networks utilize distributed coordination function (DCF) as the primary medium access control technique. However, DCF provides only a best-effort delivery service. In other words, all of the packets are regarded as having the same priority for delivery over the network, irrespective of the application or service to which they belong. Accordingly, the IEEE 802.11e standard released in 2005 [3] attempts to achieve quality of service (QoS) support in WLANs by specifying various service classes in the medium access control (MAC) layer to perform the expedited delivery of high-priority packets such as voice or video data. In particular, the standard defines four access categories (ACs) which provide different opportunities for the packets to earn a transmission opportunity based on the settings assigned to the related MAC parameters.
Although the Enhanced Distributed Channel Access (EDCA) protocol utilized by IEEE 802.11e improves the QoS situation for differentiated services, all of the video data are assigned to the same AC (i.e., the same queue), irrespective of their coding priority. Consequently, the quality of the transmitted video may still be degraded as a result of burst losses, excessive delays, and lack of bandwidth. For example, under heavy network loads, the video packets may have to wait for a long time before being dequeued, or may even be dropped as a result of buffer overflows. Accordingly, Ksentini et al. [4] proposed a cross-layer architecture for enhancing the performance of video transmissions under heavy network loading by using a queue mapping algorithm to assign encoded H.264 [5] video data to different AC queues depending on their coding significance. The proposed mechanism provides an effective means of mitigating the loading imposed on the video queue by allocating the video packets to other AC queues when the network loading is heavy. However, the mapping mechanism is not adaptive. In other words, it performs well only under the designed network loading condition (i.e., overloaded). Thus, Lin et al. [6] proposed a more adaptive mapping mechanism which considered not only the video-coding significance but also the queue length. In particular, under normal traffic load conditions, the incoming video packets are assigned to the high-priority AC queue in accordance with the conventional EDCA protocol. However, under heavy traffic loads, when the high-priority queue becomes full, the video packets are redistributed over the lower-priority queues with a probability based on the relative importance of the video packets and the current lengths of the queues.
The queue mapping mechanisms in [4, 6] are designed to mitigate the loading of the queues in such a way that the video packets entering the queues can be serviced as quickly as possible. However, in practice, the speed at which the packets can be dequeued depends not only on the length of the queues to which the packets are assigned but also on their throughputs (departure rates). The queue throughput in IEEE 802.11 and IEEE 802.11e WLANs is commonly analyzed using a Markov chain model. For example, Bianchi [7] proposed an analytical model for evaluating the throughput performance of DCF under saturation conditions. Wu et al. [8] extended the model in [7] to take account of the packet retry limit. Tantra et al. [9] and Kong et al. [10] further refined Wu’s model to analyze the throughput of EDCA under IEEE 802.11e. Finally, Hwang et al. [11] proposed a model for analyzing the performance of EDCA with different virtual collision functions. Besides, a lot of related works applied the queue mapping mechanisms to vehicular ad hoc networks (VANETs) [12–15] to improve the transmission quality.
The present study proposes a new queue mapping algorithm, designated as the throughput-based mapping algorithm (TMA), for improving the video quality and minimizing the delay of video streaming over IEEE 802.11e WLANs. The proposed TMA mechanism improves the video quality by applying a differential level of probability to the queue mapping of the video packets based on both their coding significance and the AC queue throughput. Notably, the proposed approach not only improves the perceived quality of the delivered video but also maximizes the utilization efficiency of the AC queues and reduces network congestion.
The remainder of this paper is organized as follows. Section 2 presents the background to the problem considered in the present study and introduces the related work in the field. Section 3 introduces the proposed mapping algorithm and describes the Markov chain model used to analyze the AC queue throughputs. Section 4 presents and discusses the simulation results obtained for the throughout, transmission quality, and video delay under the proposed scheme. Finally, Section 5 provides some brief concluding remarks and indicates the intended direction of future research.
2. Background and Related Work
2.1. MPEG-4 Overview
MPEG-4 [16–18] is one of the most common video-coding techniques and is designed to minimize the amount of data transmitted over the Internet by predicting motion from one frame to the next in the temporal direction. In MPEG-4, the video stream is divided into consecutive groups of pictures (GoPs), where each GoP consists three different types of picture frame with a periodic sequence and a particular coding dependency between them (see Figure 1). The I-frame (intracoded picture) is encoded independently of all the other pictures and can thus be decoded without reference to any other frames in the GOP. By contrast the P-frames (predictive-coded pictures) are encoded and decoded based on motion differences from the preceding I-frame or P-frame. Finally, the B-frames (bidirectionally predictive-coded pictures) are encoded and decoded based on motion differences from both the preceding and the succeeding I-frame or P-frame. Based on these encoding relations, it is clear that the I-, P-, and B-frames have different impacts on the perceived quality of the transmitted video in the sense that the loss of an I-frame or a P-frame, for example, causes the following P-frame or B-frame to be undecodable.
2.2. IEEE 802.11e Enhanced Distributed Channel Access
IEEE 802.11e provides two channel access functions to support the QoS of differentiated services, namely Enhanced Distributed Channel Access (EDCA) and HCF (Hybrid Coordination Function) Controlled Channel Access (HCCA). This paper focuses on the EDCA function, which is a contention-based mechanism for wireless channel access. In contrast to DCF in IEEE 802.11, in which all of the network traffic is assumed to have the same priority, EDCA classifies the traffic packets into four distinct access categories (AC), each with particular settings of the arbitration interframe spacing (), minimum contention window (), maximum contention window (), and transmission opportunity limit () parameters in the MAC layer. As shown in Figure 2, each AC of a particular station is regarded as a virtual station for transmission purposes. To minimize collisions, each AC queue defers its transmission for a particular period. The maximum duration for which each AC queue is permitted to initiate transmissions toward the access point (AP) is denoted as . Thus, AC queues with smaller , and and longer gain access to the wireless channel with greater probability than the other AC queues. In the EDCA standard, the four ACs are assigned a higher or lower priority depending on the particular values given to the MAC layer parameters, and each AC is assigned to a particular class of traffic (e.g., voice, video, best effort, and background) depending on their characteristic QoS requirements (see Table 1). Internal collisions among the different ACs of the same station are handled by a virtual collision handler, which is aimed at preventing collisions of the higher-priority AC transmissions and minimize those of the lower-priority transmissions.
2.3. Related Works
2.3.1. Static Mapping Mechanism
As described above, EDCA classifies network traffic into four different level access categories (ACs), where video data is always assigned to AC2 (see Table 1). However, under heavy network loading conditions, AC2 cannot transmit data packets at the usual speed. As a result, the queue rapidly becomes full, and thus, the packets may be dropped with a high probability. The authors in [4] therefore proposed a cross-layer architecture (MAC layer and application layer) which attempted to avoid video data congestion at the AC2 queue by assigning video packets to different access categories depending on their coding priority. In particular, the I-frames in each GoP (having the highest importance among all the frames in the GoP) were assigned to AC2, but the P- and B-frames (with a lower importance from a coding point of view) were assigned to AC1 and AC0, respectively (see Figure 3).
2.3.2. Dynamic Mapping Mechanism
Although the mapping mechanism described above provides better performance than IEEE 802.11e EDCA under heavy network loads, it employs a static mapping approach and hence wastes the transmission advantages of AC2 by continuing to map the P- and B-frames to the lower-priority queues even when the network loading is light. As a result, the mechanism results in unnecessary packet losses and delivery delay. Accordingly, the authors in [6] proposed a dynamic mapping algorithm which dynamically allocates the video to the most appropriated AC at the MAC layer according to both the significance of video type and the network traffic load. Based on the MPEG-4 coding dependence, the channel access priorities used to prioritize the transmission opportunity at the MAC layer are set with the I-frame as the highest, the P-frame below I-frame, and the B-frame set 4at the lowest priority. To allocate important video data into higher-priority AC queue in the 802.11e MAC layer, the authors [6] give different downward mapping probabilities, defined as , to different video frame types according to its coding significance. The means the probability of the video packet allocated to the lower-priority AC queue, such as AC1 or AC0. The downward mapping relationship of MPEG-4 video frame types is . Furthermore, to support dynamic adaptation to changes in network traffic loads, the dynamic mapping algorithm uses the MAC queue length as an indication of the current network traffic load. If the queue length is less than low_threshold (light loading), the video packet is assigned to the AC2 queue. However, if the queue length is higher than high_threshold (heavy loading), the video packet is mapped to a lower-priority AC queue in order to avoid the packet being dropped or suffering a transmission delay. Finally, if the queue length lies between low_threshold and high_threshold (medium loading), the video packet is assigned to either the AC2 queue or a lower-priority AC queue with a probability determined by the current AC2 queue length and the video frame type (I, P, or B). The integrated function to introduce these two parameters in the algorithm is in the following expression [6]:
Figure 4 presents a schematic illustration of the proposed dynamic mapping mechanism. In this function, the will be adjusted according to the current queue length of AC2 and threshold values. The final result of the downward mapping probability (Prob_New) will be calculated as Equation (1). The higher the , the greater the opportunity for the packet to be mapped into a lower-priority queue.
2.4. Contributions of Present Study
The present study proposes a throughput-based mapping algorithm (TMA) for improving the quality of video transmissions over IEEE 802.11e WLANs. The literature contains many proposals for the queue mapping of data frames in IEEE 802.11e EDCA. For example, as described above, the method in [4] allocates the video data to different AC queues based on their coding significance and priority of AC queues, while that in [6] allocates the video packets based on a joint consideration of both the packet importance and the length of the AC queues. However, both schemes in [4, 6] suffer certain disadvantages. For example, the algorithm in [4] performs well only under overloaded network conditions, while that in [6] ignores the effects of the dequeuing time on the transmitted video quality. Accordingly, the TMA mechanism proposed in this study improves the video quality by applying a differential level of probability to the queue mapping of the video packets based on both their coding significance and the AC queue throughput. Notably, the proposed approach not only improves the perceived quality of the delivered video but also maximizes the utilization efficiency of the AC queues and reduces network congestion.
3. Throughput-Based Mapping Algorithm
3.1. TMA Mapping Algorithm
The queue mapping mechanisms in [4, 6] are aimed at mitigating the loading of the AC queues and minimizing the dequeuing time. However, in practice, the dequeuing time depends not only on the queue length but also on the throughput (i.e., departure rate). Accordingly, the TMA mechanism proposed in this study considers both the network load and the throughput of the AC queues in dynamically mapping the video packets across the various queues at each station. As shown in Figure 5, when a new packet arrives, TMA calculates to predict how fast the packet can be dequeued at each of the different AC queues. TMA then assigns the packet to the queue with the shortest predicted dequeuing time. In the event that the same predicted dequeuing time is obtained for two different queues, TMA assigns the packet to the queue with the higher priority in order to minimize the service waiting time. (Note that the various notations used in the TMA algorithm are summarized in Table 2.)
3.2. Analytical Model of System Throughput in TMA
In modelling the throughput of each station in the proposed TMA mechanism, the present study draws on the analysis of the IEEE 802.11e EDCA mechanism reported in [7, 9, 10]. Two Markov chain models are proposed, namely one model for AC3 to AC1 and a second model for AC0. Figure 6 shows the model for AC3 to AC1, where the state represents the backoff stage of , with a backoff counter value of . The contention window size of backoff stage for is expressed as where is the backoff stage and is the retry limit.
Figure 7 shows the Markov chain model for AC0. In the transmission period, AC0 must wait for an additional slot time compared to the other AC queues. Consequently, in constructing the analytical model, the waiting time (d) is added to the state, i.e., the state is represented as . State thus models the situation where AC0 waits for the period, while state models the case where the backoff counter is frozen by the AC0 queue when transmissions by higher-priority queues occur before the of AC0 elapses. Referring to Figure 7, and are the probabilities of time slots without transmissions by non-AC0 queues and any AC queue, respectively. When a transmission occurs during the countdown of the AC0 backoff counter, the queue freezes the counter, and the state is changed to . This type of transmission event occurs with probability . The AC0 queue freezes its backoff counter until an period is encountered, which implies that no higher-priority stations transmit during that period. The backoff counter is thus restarted with probability .
Let be the stationary distribution of Figure 6 with state for , where , , and . Owing to the regularity of the Markov chain, can be represented as
Imposing the normalization condition, , Equation (4) can be formulated as
Let be the stationary distribution for () with state , where , , and . Due to the regularity of the Markov chain, can be expressed as
Imposing the normalization condition, , Equation (6) can be written as
Let and denote the probabilities of a slot being idle and busy, respectively. The idle channel probability is given as where the busy channel probability PB is . The probabilities and that no packets are transmitted from stations after a busy slot and idle slot, respectively, are expressed as
Meanwhile, the probability that of any station accesses the channel in a particular slot is given by
The probability depends on the collision probability that a packet from encounters a collision, which is given by
The collision probability of , where , is conditioned on whether the previous slot is a busy slot or an idle slot. If the previous slot is busy, , where , can transmit in the current slot. However, if the previous slot is idle, , where , must wait for an additional slot time. Equations (10) and (11) form a set of nonlinear equations that can be computed numerically. When a packet is transmitted from , , the transmission is successful if the packet suffers no internal or external collisions. The successful transmission probability of , , is conditioned on the fact that a packet is transmitted from at least one AC buffer, i.e.
The virtual collision function handles any internal collisions which may occur among different ACs within the same station. In IEEE 802.11e, higher-priority ACs are preferentially granted an opportunity for transmission, and hence, lower-priority ACs suffer a greater risk of internal collisions.
The saturation throughput of is given by where is . In Equation (13), is the average length of the packet payload, is the length of each time slot, and and are the average lengths of successful transmission and collision resolution in the basic access method, respectively. In other words, and are given as
4. Performance Evaluation
4.1. Experimental Environment and Parameter Settings
The performance of the proposed TMA mechanism was compared with that of three existing mapping methods, namely IEEE 802.11e EDCA, the static mapping algorithm in [4], and the dynamic mapping algorithm in [6]. The simulations were performed using NS-2 [19] and were integrated with EvalVid [20] in order to evaluate the video transmission quality. The simulations considered two YUV QCIF () video sources, namely Foreman and Coastguard [21]. Foreman is a static-type video sequence with 400 frames, while Coastguard is a motion-type sequence with 300 frames. For both video sources, each video frame was fragmented into small-size packets before transmission. The maximum transmission packet size was set as 1000 bytes. The remaining simulation parameters were set as shown in Table 3.
The simulations considered two different scenarios. In the first scenario, the wireless network consisted just two nodes, namely a video sender and a video receiver (see Figure 8). Moreover, the network traffic was assumed to consist of only video traffic. In other words, the simulations were used to compare the dynamic variation of the AC queue lengths under the different mapping schemes. The second scenario considered networks with different numbers of nodes (). Each node was assumed to transmit four types of network traffic, namely voice traffic (64 k, in AC3), video traffic (in AC2), TCP traffic (in AC1), and UDP traffic (in AC0). The simulations were designed to evaluate the video quality and transmission delay of the four different mapping algorithms under different network loading conditions (low, medium, and high). Figure 9 shows a typical topology considered in the second scenario.
4.2. Performance Analysis of TMA Mechanism
Figure 10 presents the simulation results obtained in scenario 1 for the dynamic variation of the AC queue lengths under the four mapping schemes. In the figures, the queue length means the number of packets in the queue. Under the IEEE 802.11e EDCA mechanism, all of the video packets are mapped to AC2, as shown in Figure 10(a). Hence, the AC2 queue length varies dynamically over time, while the AC0, AC1, and AC3 queues remain empty. However, under the static mapping algorithm, the I-frame packets are mapped to AC2, while the P- and B-frame packets are mapped to AC1 and AC0, respectively (see Figure 10(b)). In implementing the dynamic mapping algorithm, the low_threshold and high_threshold parameters were set as 10 and 40 packets, respectively. As shown in Figure 10(c), at a simulation time of 10.18 s, the queue length of AC2 exceeds the low_threshold value, and hence, some of the following video packets are mapped to AC1. However, at 10.63 s, the AC2 and AC1 queue lengths are higher than the high_threshold and low_threshold settings, respectively, and consequently, some of the subsequent packets are mapped to the lowest-priority queue (AC0). In implementing the proposed TMA mechanism, the throughput proportion of the AC2, AC1, and AC0 queues was assumed to be 4 : 2 : 1. Thus, as shown in Figure 10(d), the queue length proportion of AC2, AC1, and AC0 was also equal to 4 : 2 : 1.
(a)
(b)
(c)
(d)
Figure 11 shows the results obtained in scenario 2 for the variation of the AC queue lengths in networks of different scales () when using the IEEE 802.11e EDCA mechanism. For a small network topology (5 nodes), each queue has a length of zero (see Figure 11(a)) since the network loading is light, and hence, all of the traffic packets can be successfully transmitted. As the number of nodes increases to 10, the lengths of the AC2, AC1, and AC0 queues increase (see Figure 11(b)). However, the AC3 queue length remains equal to zero since the queue has the highest priority among all the queues, and hence, all the packets assigned to it can be successfully transmitted. As the number of nodes increases to 15, the network loading becomes heavy, and hence, even the AC3 queue length increases slightly (see Figure 11(c)). As the number of nodes is further increased to 20, the queue length of AC3 increases accordingly, as shown in Figure 11(d).
(a)
(b)
(c)
(d)
Figures 12(a) and 12(b) present the PSNR variations of Foreman and Coastguard, respectively, under each of the network topologies () and mapping algorithms. For light network loading (), all of the queue mapping algorithms perform well. When the number of nodes increases to 10, EDCA outperforms the static mapping algorithm since the latter algorithm causes unnecessary delays and losses due to its policy of allocating less important video packets to lower-priority queues. However, under heavy network loading (i.e., 15 and 20 nodes), the static mapping algorithm achieves a better video quality than EDCA due to its utilization of the lower-priority AC queues. For all of the network topologies, the dynamic mapping algorithm outperforms both EDCA and the static mapping algorithm since it considers not only the importance of the video packets but also the queue length. However, for both video sources and all network topologies, the proposed TMA algorithm achieves the highest PSNR (i.e., the best video quality) of the four mapping algorithms since it considers both the queue length and the queue throughput in mapping the video packets to the different queues.
(a)
(b)
Figures 13(a) and 13(b) show the video delay variations of the Foreman and Coastguard sequences, respectively, under the four mapping algorithms. For all four algorithms, the video delay increases with an increasing number of network nodes due to the corresponding increase in the network load. The dynamic mapping algorithm achieves a lower delay than the static scheme and EDCA in all of the considered topologies. Moreover, the static mapping algorithm outperforms EDCA as the network load becomes heavy (i.e., ). However, TMA achieves a comparable performance to the dynamic mapping algorithm under light loading () and outperforms all of the existing mapping algorithms under medium () and heavy () loads.
(a)
(b)
Overall, the results presented in Figures 12 and 13 confirm that the proposed TMA algorithm achieves an improved video quality and reduced transmission delay under all network loading conditions due to its policy of considering both the queue length and the queue throughput in reaching an assignment decision for the video packets.
5. Conclusions
This study has proposed a throughput-based mapping algorithm (TMA) to enhance the quality of video transmissions over IEEE 802.11e WLANs. Notably, compared to existing algorithms presented in the literature [4, 6], the TMA algorithm considers not only the current length of the AC queues but also their throughput and therefore minimizes the packet dequeuing time. Two analytical Markov chain models have been proposed for analyzing the throughputs of the AC queues in the IEEE 802.11e EDCA protocol. The numerical results have shown that the TMA algorithm yields a better video quality (higher PSNR) and lower video delay than the IEEE 802.11e EDCA mechanism or the static mapping and dynamic mapping schemes proposed in the literature.
Future studies will explore the potential for using the proposed TMA algorithm to improve the performance of various types of network, including wireless mesh networks (WMNs) [22–24] and vehicular ad hoc networks (VANETs) [25–27]. In addition, the effect of the wireless channel error on the video transmission quality should be considered into the proposed TMA algorithm for determining the AC queue mapping [2, 7, 15].
Data Availability
The data used to support the findings of this study are available from the author upon request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported by the Ministry of Science and Technology (MOST), Taiwan, under contract No. MOST 108-2221-E-242-001 and MOST 108-2221-E-992-029.