Technique to Improve MPEG-4 Traffic Schedulers in IEEE 802.15.3 WPANs

Document Sample
Technique to Improve MPEG-4 Traffic Schedulers in IEEE 802.15.3 WPANs Powered By Docstoc
					  Technique to Improve MPEG-4 Traffic Schedulers
             in IEEE 802.15.3 WPANs
                                           Shahab Moradi and Vincent W.S. Wong
                                     Department of Electrical and Computer Engineering
                                    The University of British Columbia, Vancouver, Canada
                                           e-mail: {shahab, vincentw}@ece.ubc.ca


   Abstract— Video traffic, especially MPEG-4 streams, is en-        another objective performance metric is required for com-
visioned to be a dominant traffic type in the IEEE 802.15.3          paring the performance of different schedulers. We use the
high-rate wireless personal area networks (WPANs). The unique       decoding failure rate (DFR) criterion which is the proportion
hierarchical structure of MPEG-4 streams calls for special
measures at the MAC layer, in order to improve the quality of       of MPEG-4 frames that cannot be decoded at the receiver due
service (QoS). In this paper, we propose a frame-decodability       to the lack of other dependent frames. This metric will be
aware (FDA) technique to make the scheduling algorithms             explained in detail in Section II-B. To our knowledge, none
aware of the hierarchical structure and decoding dependencies       of the existing scheduling algorithms consider the hierarchical
in MPEG-4 streams. Simulation results show that the FDA             structure of MPEG-4 streams for performance improvement.
technique can significantly reduce the decoding failure rate of
various scheduling algorithms, under MPEG-4 traffic.                 In this paper, we propose a frame-decodability aware (FDA)
                                                                    technique, which exploits the MPEG-4 frames’ type (I, P or
                      I. I NTRODUCTION                              B) as well as the decoding dependencies among them.
                                                                       The contributions of this paper are as follows:
   By providing high-rate, low-power and short-range com-              • Using the FDA technique, the scheduler can determine
munication, ultrawide band (UWB) technology makes many                    whether the receiver of an MPEG-4 frame has all the
new applications possible in wireless personal area net-                  required information to decode that frame. The FDA
works (WPANs): home entertainment and real-time multime-                  technique minimizes the channel time wastage caused
dia streaming, to name a few. Although UWB technology is                  by scheduling undecodable frames, and consequently,
a breakthrough in physical layer, it cannot be fully exploited            improves the QoS.
without properly designed upper layers, (e.g., media access            • The FDA technique can be applied to different schedul-
control (MAC) layer). In the IEEE 802.15.3 standard for MAC               ing algorithms. Simulation results show that when FDA
[1], there are a few pivotal issues (e.g., scheduling and quality         technique is applied to the F-SRPT [2], EDD+SRPT [3],
of service (QoS) support), whose implementation details are               and PAP [4] schedulers, there are up to 37%, 40% and
not addressed.                                                            9% reduction in DFR, respectively.
   Multimedia traffic (especially video) is predicted to be one
                                                                       The rest of this paper is organized as follows. In Section
of the dominant traffic types that need to be handled in high-
                                                                    II, we give an overview of IEEE 802.15.3 standard, the
rate WPANs. In order to reduce bandwidth consumption, video
                                                                    hierarchical structure of MPEG-4 streams, and describe the
traffic is usually compressed with variable bit rate (VBR)
                                                                    related work. In Section III we describe our proposed FDA
encoders, among which MPEG-4 is the most widely used.
                                                                    technique. Performance comparisons are presented in Section
Besides the typical challenges to satisfy QoS requirements of
                                                                    IV. Conclusions are given in Section V.
multimedia traffic (e.g., strict delay and loss bounds), MPEG
traffic is bursty and has large peak-to-average ratio of the frame             II. BACKGROUND AND R ELATED W ORK
sizes. It also has a hierarchical structure with inter-dependency
among different frame types. Consequently, it is crucial to         A. IEEE 802.15.3 Standard for MAC
study the performance of proposed scheduling algorithms for            The IEEE 802.15.3 is designed for WPANs and aims to
IEEE 802.15.3 under MPEG-4 flows.                                    provide low cost, low power consumption, and high data
   MPEG-4 frames, like other realtime traffic, are delay sensi-      rate, within its area of operation called the piconet [1]. A
tive. The frames will be dropped at the receiver if their delay     piconet consists of a number of independent devices (DEVs)
exceeds the maximum tolerable delay. This is the base of            that communicate with each other under the control of the
job failure rate (JFR) criterion for evaluating performance of      piconet coordinator (PNC). The PNC provides basic timing,
schedulers in the MAC layer. JFR is defined as the percentage        performs scheduling, and manages the quality of service (QoS)
of frames that do not meet their transmission deadlines, and        requirements of the piconet.
are hence useless and get dropped. However, for an MPEG-               Within a piconet, the timing is based on superframes,
4 flow, due to its hierarchical structure, low JFR does not          which consists of three parts (see Fig. 1). The first part, bea-
necessarily indicate good user-perceived quality. As a result,      con, announces timing allocations, superframe duration, and
                                                                                                   TABLE I
                                                                             N UMBER OF FRAME DEPENDENCIES FOR EACH MPEG-4 FRAME

                                                                                   Frame type         Number of frames depending on it
                                                                                       I                       N + (M − 1)
                                                                            kth   P frame in GOP   N − 1 − (k − 1)M , k = 1, . . . , N/M − 1
                                                                                       B                             1
                     Fig. 1.   Superframe structure.
                                                                         receive the frame completely, either due to channel error or
                                                                         deadline expiration. A frame is indirectly lost when some
                                                                         frames that it depends on are directly lost. As shown in Fig. 2,
                                                                         all the frames in a GOP depend on the I frame in that GOP;
                                                                         therefore, if that I frame does not meet its deadline, the whole
                                                                         GOP is considered to be undecodable (i.e., direct loss of the
                                                                         I frame implies indirect loss of the rest of the other frames).
                                                                         Similarly, loss of P frames can degrade the quality of MPEG-4
                                                                         stream. Table I shows the number of frames depending on each
                                                                         frame inside a GOP. Consequently, MPEG-4 frame types can
Fig. 2. GOP structure (N = 9, M = 3). The arrows indicate the decoding   be ranked, from the most important to the least important, as
dependencies.
                                                                         I, P and B.
                                                                            An important consequence of hierarchical structure of
other piconet synchronization parameters. The second part,               MPEG-4 stream is that JFR by itself cannot accurately reflect
the contention access period (CAP), is used to communicate               the QoS given to an MPEG-4 flow at the MAC layer. Another
commands and asynchronous data. The third part, channel                  metric, which takes frame dependencies into account, is DFR.
time allocation period (CTAP), is composed of channel time               DFR is defined as the ratio of total number of undecodable
allocations (CTAs) and management CTAs (MCTAs). A DEV                    frames to the total number of transmitted frames [6]. DFR is
can use its CTA, which is assigned to it by the PNC, for either          an objective measure of user-perceived degradation of quality,
isochronous streams, asynchronous data transfer, or sending              and is the main performance metric in this paper.
commands. Channel access during CTAP uses time division
multiple access (TDMA) and there is no contention in this                C. Related Work
period. By sending beacon at the beginning of superframe,
the PNC announces the start time and duration of each CTA                   Recently, various channel time allocation algorithms have
as well as the DEVs that are allowed to use it.                          been proposed for delay-sensitive traffic in high-rate WPANs
   As the central coordinator, PNC is responsible for schedul-           [2]–[4], [7]–[12].
ing. Depending on the scheduling algorithm, PNC requires                    In [2], Mangharam et al. propose the fair shortest remaining
certain information such as the number of flows, their reserved           processing time (F-SRPT) scheduler. SRPT schedules different
rates, their queues’ status, and type and deadline of the frames         jobs in the system in order of their remaining processing time,
in their queues, in order to allocate CTAs to different flows.            from the shortest to the longest. Fair-SRPT is a variation of
                                                                         SRPT that maintains fairness among flows with different data
B. Characteristics of MPEG-4 Streams                                     rates, so that flows with smaller mean data rates will not
   A typical MPEG-4 encoder compresses video as a sequence               dominate channel access. In [3], the earliest due date (EDD)
of three types of compact frames: intra-coded (I), predictive            is used along with SRPT. EDD considers frames’ deadline in
(P ), and bidirectional (B) frames [5]. Because of the different         the system, and schedules them in the order of their deadlines,
compression schemes used to encode different frame types,                from the earliest on.
I frames tend to be larger (less compressed) than P and                     In [9], Rhee et al. use the application layer information at
B frames, and P frames tend to be larger than B frames.                  MAC layer. Each DEV informs the PNC of the maximum size
MPEG-4 encoders generate the three frame types according                 of its I, P , and B frames. The CTAs are allocated based on
to a predefined pattern called group of pictures (GOP). This              these values and the GOP pattern of the flow. Kim and Cho
pattern is characterized by two parameters (N, M ), where                propose a scheduling algorithm designed for MPEG-4 flows
N is the I-to-I frame distance, and M is the I-toP frame                 [4]. Each MPEG frame type is scheduled with a pre-assigned
distance [6]. This pattern is generally fixed for a given video           priority (PAP) based on its importance (in I, P , and B order).
sequence, and N is a multiple of M . Fig. 2 illustrates the                 The scheme proposed in [10] focuses on reducing the
hierarchical structure of MPEG-4 streams, as well as decoding            average waiting time by using an M/M/c queuing model
dependencies among the frames. For a frame to be decodable               for channel time allocation at the PNC. It also proposed
at the receiver, all other frames that it depends on (e.g., I            a command aggregation scheme in order to reduce MAC
and P frames), must be available at the receiver. A frame is             overhead when sending short command frames.
undecodable if it is directly or indirectly lost [6]. A frame               Energy efficiency is considered by the scheduler proposed in
is considered to be directly lost when the receiver does not             [11]. It defines different service categories in order to balance
between energy efficiency and QoS. It also uses the application          For each flow i ∈ F , the scheduling eligibility table maintains the
                                                                        following info:
layer information to assign priorities to the buffered frames at           • li is the remaining number of bytes for the MPEG-4 frame in the
the source of a flow.                                                          queue of flow i.
   Similar to other previous work, our scheme uses the infor-              • di is the deadline of the MPEG-4 frame in the queue of flow i.
                                                                           • yi is the type of the MPEG-4 frame in the queue of flow i,
mation such as the frame size, deadline, and the type of the                  yi ∈ {I, P, B}.
frame. However, one can also determine the decodability of                 • δi is set to 0 if any I or P frame on which the current frame of
the queued frames in the system at no extra signalling cost.                  flow i depends is directly lost; and is equal to 1 otherwise.
                                                                           • ei indicates whether the current frame of flow i is eligible for
To the best of our knowledge, none of the previous work have                  being scheduled.
explored this opportunity, which is readily available at the
                                                                        (A) At the beginning of each superframe, perform the following steps
PNC, for improving scheduling performance under MPEG-                       for each flow i ∈ F :
4 traffic. This is the main motivation of our proposed FDA                   li ← li − (number of bytes transmitted by flow i in the previous
technique, which will be described in the next section.                     superframe);
                                                                            di ← (di − 1);
                                                                            yi is not changed;
     III. F RAME -D ECODABILITY AWARE T ECHNIQUE
    In this section, we first state the assumptions and then                 if (di = 0) and (li > 0), then
                                                                               if (yi = I) or (yi = P ), then δi ← 0;
describe the FDA technique in detail. Without loss of gen-                        else δi is not changed.
erality, we assume that the transmission deadline of MPEG-
4 frames is constant, and is equal to the frame inter-arrival               if the DEV corresponding to flow i has sent information about a
                                                                               new arrival in the previous superframe, then
time. Thus, at any given time, the queue in each flow holds                            new
                                                                               li ← li ; di ← dnew ; yi ← yi ;         new
                                                                                                       i
no more than one MPEG-4 frame. We also assume that the                                  new
                                                                               where, li , dnew and yi
                                                                                               i
                                                                                                          new denote the length, deadline and

length of a superframe is a constant. In general, since the size               type of the new frame, respectively. This information is contained
                                                                               in the message sent to the PNC from the DEV.
of an MPEG-4 frame is larger than the maximum allowable                            new = I, then δ ← 1;
                                                                               if yi                 i
size for MAC protocol data unit (MPDU) [1], segmentation is                       else δi is not changed.
performed at the MAC layer.                                                 if (li > 0) and (di > 0) and (δi = 1),
    Let F denote the set of flows. In FDA, the PNC creates a                 then set ei = 1 (eligible for being scheduled);
scheduling eligibility table. The number of entries is equal to             else set ei = 0 (ineligible for being scheduled).
the number of flows |F |. The entry for flow i ∈ F includes the           (B) After updating the scheduling eligibility table, feed all the
following information: the remaining number of bytes for the                eligible flows to the scheduler.
MPEG frame to be transmitted li , the deadline of the frame
                                                                                     Fig. 3.   Pseudo-code of the FDA technique.
di , and the type of the MPEG frame yi .
    Since the superframe length is constant, we save di in units
of superframe size. So, di indicates the number of superframes        when the frame of flow i has been transmitted and the queue of
left until the deadline of the frame expires. The type of an          flow i becomes empty. In addition, di is reduced by one unit.
MPEG-4 frame yi ∈ {I, B, P }.                                         If di becomes 0 and the queue of flow i is not empty (li > 0),
    The FDA technique determines the decodability of each             then the frame in that queue has expired and is no longer
frame based on the deadline and type information. For each            eligible for transmission. This would impact the decodability
flow i ∈ F , we define a binary variable δi . The value of δi is        of the subsequent frames of flow i if yi is either I or P . In
set to 0 if any I or P frame on which the current frame of flow        either case, we set δi to 0 in order to indicate that an I or
i depends is directly lost; otherwise, δi is set to 1. According      P frame on which the upcoming frames of the current GOP
to the hierarchical structure of MPEG-4 encoding, a frame             depend is directly lost. Note that if the expired frame is of
depends on its preceding I and P frames up to the most recent         B type, then δi will not be changed since the decodability of
I frame. Therefore, when yi = I or P , δi indicates if the frame      subsequent frames is not affected by the loss a B frame.
of flow i is decodable, (i.e. neither directly nor indirectly lost).      In case of new frame arrivals, the PNC should also update
If δi is equal to 0, the subsequent frames up to the next I frame     the entries of the flows that have new frames according
(i.e., beginning of the next GOP) are indirectly lost. However,       to the information sent from the corresponding DEVs. The
when yi = B, δi only indicates that the frame of flow i is not         parameters li , di and yi are set to the length, deadline and
indirectly lost. The entry of each flow also includes a variable       type of the new frame arrived at flow i. If the new frame
ei , which shows whether or not the frame of flow i ∈ F is             is of type I, then a new GOP has begun. The decodability
eligible for being scheduled.                                         status of the frames preceding the new I frame does not affect
    The PNC updates the scheduling eligibility table at the           the decodability of the I frame and its subsequent frames.
beginning of each superframe before making the scheduling             Therefore, when a new I frame arrives for flow i, we set δi
decisions. The pseudo-code of the table update for entry i            to 1 regardless of the previous value of δi .
is shown in Fig. 3. The table is updated by the following                Finally, after updating the values of li , di , yi and δi , the FDA
procedures: If the MPEG-4 frame of flow i ∈ F is partially             technique determines if the frame of flow i is eligible for being
transmitted in the previous superframe, then li is reduced by         scheduled. The frame is marked as eligible (i.e., ei = 1) if the
the number of transmitted bytes. The value of li becomes 0            following three conditions are valid: (1) it is not completely
  i     li     di      yi   δi   ei         i     li     di      yi   δi   ei                                 TABLE II
  1    18k     1       B    1    1          1    18k     0       B    1    0                          S IMULATION PARAMETERS
  2    23k     1       P    1    1    ➪     2    8k      0       P    0    0
  3    28k     0        I   0    0          3    21k     3       B    0    0                             Parameter               Value
  4    14k     0       B    0    0          4    37k     3        I   1    1                 Channel data rate                100 Mbps
  5    32k     2       P    1    1          5    0k      1       P    1    0                 Superframe duration                 8 ms
                 (a)                                       (b)                               Number of MPEG-4 flows              2 – 15
                                                                                             MPEG-4 GOP pattern                 (12, 3)
Fig. 4. An illustration of how the FDA technique updates the scheduling ta-                  Mean data rate of MPEG-4 flows     8 Mbps
ble. Scheduling table after being updated at the beginning of (a) superframe n,              Frame inter-arrival time           1/30 s
(b) superframe n + 1.                                                                        Maximum tolerable delay            1/30 s
                                                                                             Maximum MPDU size               2048 bytes
transmitted (li > 0), (2) the frame has not expired (di > 0),                                Simulation duration                 500 s
and (3) all the frames that the frame of flow i depends on are                                Start time separation           1, 9, 17 ms
available at the receiver side of the flow (δi = 1).
   When the scheduling eligibility table has been updated by
                                                                                  technique. The scheduling algorithms that we studied are the
FDA, the scheduling algorithm will be applied to the eligible
                                                                                  F-SRPT [2], EDD+SRPT [3], and PAP [4] algorithms. The
frames. Note that the FDA technique is independent of the type
                                                                                  performance metrics are the decoding failure rate (DFR) and
of scheduling algorithm being invoked. The FDA technique
                                                                                  job failure rate (JFR).
only gives the list of eligible frames to the scheduler. By
                                                                                     For the simulation, we use the real MPEG-4 trace of
using this technique, the scheduler does not need to allocate
                                                                                  Jurassic Park movie with GOP pattern of (N =12, M =3). The
channel time to the frames that are indirectly lost. As a result,
                                                                                  frame rate is 30 frames per second. The average rate of each
some CTAs are available for other frames that are decodable
                                                                                  flow is 8 Mbps. The maximum tolerable delay for MPEG-4
at the receiver. In Section IV, we study the impact of the FDA
                                                                                  frames is equal to the frame inter-arrival time, which is 1/30 s.
technique on several scheduling algorithms.
                                                                                  The number of flows vary from 2 to 15. The MAC overheads
   Fig. 4 shows an example of updating the scheduling eligibil-
                                                                                  such as headers and inter-frame spacings are set as specified
ity table with 5 MPEG-4 flows in the system. At the beginning
                                                                                  in the IEEE 802.15.3 standard [1]. The channel rate is equal to
of superframe n (see Fig. 4(a)), the frames of flows 3 and 4
                                                                                  100 Mbps. The size of a superframe is 8 ms. The simulation
expire. The flows 1, 2 and 5 are eligible for being scheduled.
                                                                                  time is equal to 500 s. Table II summarizes the simulation
Suppose that flows 2 and 5 are scheduled in superframe n,
                                                                                  parameters.
and transmit 15 kB and 32 kB, respectively. At the beginning
of the next superframe n + 1 (see Fig. 4(b)), the frame of flow                    B. Simulation Results
1 expires. Since this frame is of type B, it does not affect δ1 .
                                                                                     Fig. 5 shows the DFR and JFR for the three schedulers with
The frame of flow 2 also expires and is directly lost. Since
                                                                                  (and without) the use of FDA technique. The SRPT scheduler
this frame is of type P , δ2 becomes 0. A new frame arrives
                                                                                  tends to favor the frames with smaller size. Since in general B
for flow 3 with length, deadline and type equal to 21 kB, 3
                                                                                  frames are smaller than P frames, and P frames are smaller
and B, respectively. This frame is indirectly lost because its
                                                                                  than I frames, the B and P frames have a higher chance
previous I frame is directly lost, as indicated by the value of
                                                                                  for being scheduled than the I frames. As a result, many
δ3 . A new I frame arrives at flow 4. The value of δ4 will be
                                                                                  of the completed jobs (i.e., scheduled frames) by F-SRPT
set to 1. Finally, the queue of flow 5 becomes empty as its
                                                                                  are undecodable at the receiver, because they rely on other
frame has been completely transmitted in superframe n. The
                                                                                  more important frames that were deprived of channel time
frame of flow 4 is the only frame eligible for being scheduled
                                                                                  in the presence of small B (or P ) frames. This explains
in superframe n + 1. Although d3 > 0, the frame of flow 3 is
                                                                                  the significant gap between DFR and JFR for F-SRPT (see
not eligible for being scheduled because it is indirectly lost.
                                                                                  Fig. 5(a)). In other words, this gap is due to the channel
By preventing the undecodable frames being scheduled, the
                                                                                  time wastage by scheduling undecodable frames. The gap
FDA technique improves the utilization of the channel time.
                                                                                  also confirms that JFR is not an accurate performance metric
   The implementation of FDA requires the signaling to pass
                                                                                  for scheduling algorithms. Using the FDA technique, F-SRPT
the info from the DEV to the PNC. For example, FDA can use
                                                                                  scheduler does not waste the channel time by scheduling
the signaling method proposed in [12], which is compatible to
                                                                                  undecodable frames. In effect, none of the undecodable frames
the 802.15.3 standard. It allocates one MCTA to each flow,
                                                                                  are being scheduled. Thus, JFR and DFR are identical when
at the end of the superframe. Compared to using CAP or
                                                                                  FDA technique is used. As illustrated in Fig. 5(a), dropping
allocating MCTAs at the beginning of CTAP, this scheme gives
                                                                                  undecodable frames can reduce DFR of F-SRPT scheduler up
more recent information to the PNC and reduces the average
                                                                                  to 37%.
response time of the system.
                                                                                     The EDD+SRPT scheduler gives higher priority to the
                    IV. P ERFORMANCE E VALUATION                                  frames with more stringent deadline. This approach reduces
                                                                                  both JFR and DFR when compared to F-SRPT. Results
A. Simulation Model                                                               from Fig. 5(b) show that the FDA technique improves the
  In this section, we evaluate the performance of various                         EDD+SRPT scheduler by up to 40% reduction in DFR.
scheduling algorithms with (and without) the use of the FDA                          The FDA technique has the least impact on the PAP
                                                     (a) F−SRPT Scheduler
                              0.8                                                                                                             0.35




                                                                                                        Average Decoding Failure Rate (DFR)
                                    DFR                                                                                                              F−SRPT with FDA
                              0.7   DFR with FDA                                                                                               0.3   PAP with FDA
                                    JFR with FDA                                                                                                     EDD+SRPT with FDA
                              0.6
       Average Failure Rate
                                    JFR                                                                                                       0.25
                              0.5
                                                                                                                                               0.2
                              0.4
                                                                                                                                              0.15
                              0.3
                                                                                                                                               0.1
                              0.2

                              0.1                                                                                                             0.05

                               0                                                                                                                0
                                2   3   4   5      6   7    8    9  10 11    12   13   14   15                                                   2     3      4      5       6     7       8   9   10
                                                    Number of MPEG−4 Flows                                                                                        Number of MPEG−4 Flows

                                                   (b) EDD+SRPT Scheduler
                              0.8                                                                Fig. 6. Comparison of improved F-SRPT, EDD+SRPT and PAP schedulers
                                    DFR
                              0.7   DFR with FDA                                                 with FDA in terms of average DFR.
                                    JFR with FDA
                              0.6
       Average Failure Rate




                                    JFR

                              0.5
                                                                                                 FDA technique can be applied to various types of schedulers
                              0.4
                                                                                                 and does not incur any extra signalling overhead in the
                              0.3
                                                                                                 system. We used both DFR and JFR as performance metrics
                              0.2
                                                                                                 when evaluating MPEG-4 traffic schedulers. We analyzed the
                              0.1
                                                                                                 effect of FDA technique on F-SRPT, EDD+SRPT and PAP
                               0
                                                                                                 schedulers, and showed up to 37%, 40% and 9% reduction in
                                2   3   4   5      6   7    8    9  10 11    12   13   14   15
                                                    Number of MPEG−4 Flows                       DFR, respectively.
                                                      (c) PAP Scheduler                                                                                      ACKNOWLEDGMENT
                              0.8
                                    DFR
                              0.7   DFR with FDA
                                                                                                   This work is supported by Bell Canada and the Natural Sci-
                              0.6
                                    JFR with FDA                                                 ences and Engineering Research Council of Canada (NSERC)
       Average Failure Rate




                                    JFR

                              0.5
                                                                                                 under grant number CRDPJ 320552-04.
                              0.4
                                                                                                                                                                  R EFERENCES
                              0.3
                                                                                                  [1] IEEE, “Part 15.3: Wireless medium access control (MAC) and physical
                              0.2                                                                     layer (PHY) specifications for high rate wireless personal area networks
                              0.1                                                                     (WPANs),” IEEE Std 802.15.3-2003, Sept. 2003.
                               0
                                                                                                  [2] R. Mangharam, M. Demirhan, R. Rajkumar, and D. Raychaudhuri, “Size
                                2   3   4   5      6   7    8    9  10 11
                                                    Number of MPEG−4 Flows
                                                                             12   13   14   15        matters: size-based scheduling for MPEG-4 over wireless channels,” in
                                                                                                      Proc. of SPIE Multimedia Computing and Networking, Jan. 2004.
                                                                                                  [3] A. Torok, L. Vajda, A. Vidacs, and R. Vida, “Techniques to improve
  Fig. 5.                       Effect of FDA technique on different scheduling algorithms.           scheduling performance in IEEE 802.15.3 based ad hoc networks,” in
                                                                                                      Proc. of IEEE Globecom, St. Louis, MO, Nov. 2005.
                                                                                                  [4] S.-M. Kim and Y.-J. Cho, “Scheduling scheme for providing QoS to real-
scheduler, and reduces DFR up to 9%. Since PAP schedules                                              time multimedia traffics in high-rate wireless PANs,” IEEE Transactions
I and P frames before B frames, most of the I and P frames                                            on Consumer Electronics, vol. 51, no. 4, pp. 1159–1168, Nov. 2005.
are scheduled, and they have low expiry rate. As explained                                        [5] M. Krunz and S. K. Tripathi, “On the characterization of VBR MPEG
                                                                                                      streams,” in Proc. of ACM SIGMETRICS, June 1997.
in Section II-B, indirect loss is the result of expiry of I or P                                  [6] A. Ziviani, B. E. Wolfinger, J. F. Rezende, O. C. Duarte, and S. Fdida,
frames. Therefore, PAP scheduler results in a small number of                                         “Joint adoption of QoS schemes for MPEG streams,” Multimedia Tools
indirectly lost frames, which reduces the channel time wastage                                        and Applications, vol. 26, no. 1, pp. 59–80, May 2005.
                                                                                                  [7] X. Shen, W. Zhuang, H. Jiang, and J. Cai, “Medium access control
caused by scheduling such frames. Thus, as shown in Fig. 5(c),                                        in ultra-wideband wireless networks,” IEEE Transactions on Vehicular
the gap between DFR and JFR for PAP scheduler is less than                                            Technology, vol. 54, no. 5, pp. 1663–1677, Sept. 2005.
the other two schemes. Since FDA technique uses the wasted                                        [8] C. Hu, H. Kim, J. Hou, D. Chi, and S. Shankar, “Provisioning qual-
                                                                                                      ity controlled medium access in ultrawideband-operated WPANs,” in
channel time to reduce DFR, there is little room for improving                                        Proc. of IEEE Infocom, Barcelona, Spain, Mar. 2006.
PAP.                                                                                              [9] S. H. Rhee, K. Chung, Y. Kim, W. Yoon, and K. S. Chang, “An
   Fig. 6 compares the improved DFR of the three studied                                              application-aware MAC scheme for IEEE 802.15.3 high-rate WPAN,”
                                                                                                      in Proc. of IEEE Wireless Communications and Networking Conference
scheduling algorithms. Results show that PAP performs worse                                           (WCNC), Atlanta, Georgia, Mar. 2004.
than F-SRPT and EDD+SRPT when the number of flows is                                              [10] R. Zeng and G.-S. Kuo, “A novel scheduling scheme and MAC enhance-
small. However, when the number of flows becomes large,                                                ments for IEEE 802.15.3 high-rate WPAN,” in Proc. of IEEE Wireless
                                                                                                      Communications and Networking Conference (WCNC), New Orleans,
its performance lies between F-SRPT and EDD+SRPT. On                                                  LA, Mar. 2005.
average, EDD+SRPT scheduler gives a lower DFR than the                                           [11] X. Chen, Y. Xiao, Y. Cai, J. Lu, and Z. Zhou, “An energy Diffserv
other two schedulers.                                                                                 and application-aware MAC layer scheduling for multiple VBR video
                                                                                                      streaming over high-rate WPANs,” Elsevier Computer Communications,
                                                                                                      vol. 29, pp. 3516–3526, Nov. 2006.
                                                V. C ONCLUSIONS                                  [12] X. Liu, Q. Dai, and Q. Wu, “Scheduling algorithms analysis for MPEG-4
   In this paper, we proposed an FDA technique to improve                                             traffic in UWB,” in Proc. of IEEE Vehicular Technology Conference
                                                                                                      (VTC-Fall), Los Angeles, CA, Sept. 2004.
the performance of schedulers for MPEG-4 video traffic. The