A Scalable Multiuser Framework for Video over

OFDM Networks: Fairness and Efficiency

Guan-Ming Su, Student Member, IEEE, Zhu Han Member, IEEE, Min Wu, Member, IEEE, and
K. J. Ray Liu Fellow, IEEE

Abstract
In this paper, we propose a framework to transmit multiple scalable video programs over downlink
multiuser OFDM networks in real time. The framework explores the scalability of the video codec and
multi-dimensional diversity of multiuser OFDM systems to achieve the optimal service objectives subject
to constraints on delay and limited system resources. We consider two essential service objectives, namely,
the fairness and efficiency. Fairness concerns the video quality deviation among users who subscribe the
same quality of service, and efficiency relates to how to attain the highest overall video quality using the
available system resources. We formulate the fairness problem as minimizing the maximal end-to-end
distortion received among all users and the efficiency problem as minimizing total end-to-end distortion
of all users. Fast suboptimal algorithms are proposed to solve the above two optimization problems. The
simulation results demonstrated that the proposed fairness algorithm outperforms a time division multiple
(TDM) algorithm by 0.5∼3dB in terms of the worst received video quality among all users. In addition,
the proposed framework can achieve a desired tradeoff between fairness and efficiency. For achieving
the same average video quality among all users, the proposed framework can provide fairer video quality
with 1∼1.8dB lower PSNR deviation than a TDM algorithm.

Index Terms
Scalable video coding, multiuser OFDM networks, dynamic resource allocation, multiuser video
communications.

Copyright (c) 2006 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to [email protected].
Manuscript received January 6, 2005; revised December 23, 2005 and July 24, 2006. This work was supported by the U.S.
National Science Foundation under Award #CCR-0133704 and MURI F496200210217. Some preliminary results of this work
were presented in the IEEE Global Telecommunications Conference 2004. This paper was recommended by Associate Editor
S. Li.
G.-M. Su is with ESS Technology, Fremont, CA 94538 USA {email: [email protected]}.
Z. Han is with the Department of Electrical and Computer Engineering, Boise State University, Boise, ID 83725 USA {email
: [email protected] }.
M. Wu and K. J. Ray Liu are with the Department of Electrical and Computer Engineering, University of Maryland, College
Park, MD 20742 U.S.A {email: [email protected]; [email protected] }
Digital Object Identifier:
 1

I. I NTRODUCTION

With the advancement of video compression technology and wide deployment of wireless local area

networks (WLAN), transmitting multiple compressed video programs over band-limited wireless fading

channel has become an emerging service. A multiuser video transmission system should consider not only

the reconstructed video quality of each individual user but also different perspectives from network-level

point of view. We consider two essential service objectives, namely, the fairness and efficiency. The first

objective regards whether the received video qualities are fair or not for the users who subscribe the same

video quality level. The second objective is efficiency, namely, how to achieve the highest overall users’

received video qualities with a limited amount of system resources. If the users pay the same price for

a certain video quality, the received qualities for these users should be similar. The challenge to attain

each objective is how to effectively allocate radio and video resources to each video stream. To facilitate

resource management, a system with highly adjustable radio and video resources is preferred. For the radio

resource, the wireless communication system should provide high data rates to accommodate multimedia

transmission and equip multi-dimensional diversity so that radio resources can be dynamically distributed

according to users’ needs and channel conditions. For the video source coding, the video codec should

have high scalability to aid rate adaptation to achieve the required quality. In this paper, we address the

above issues and present a framework to reach a desired tradeoff between fairness and efficiency.

To provide high data transmission rate, orthogonal frequency division multiplex (OFDM) system

is a promising modulation scheme and has been adopted in the current technology, such as Digital

Audio Broadcasting (DAB), Digital Video Broadcasting (DVB), WLAN standard (IEEE 802.11 a/g), and

Wireless Metropolitan Area Networks (WMAN) standard (IEEE 802.16a). Compared to the traditional

OFDM system, a multiuser OFDM system has higher adjustability for dynamic allocation of resources

such as subcarrier, rate, and transmission power. Therefore, a multiuser OFDM system can explore

time, frequency, and multiuser diversities to improve system performances, such as throughput [1], [2].

Allocating resources in a multiuser OFDM system is often formulated as an optimization problem. If

DRAFT
 2

the objective function and system resource constraints are continuous and convex, multiuser iterative

waterfilling is an effective solution to maximize system’s utility [3]–[5]. However, the multiuser OFDM

system often has resources with both continuous and integer valued parameters and systems may also have

non-linear or/and non-convex constraints. Thus, obtaining the optimal solution is often NP hard. Through

Lagrangian relaxation, an algorithm satisfying users’ minimal quality requirement and minimizing the

overall transmission power was proposed in [6]. To alleviate the high computational complexity, several

suboptimal but computationally efficient algorithms for transmitting generic data were proposed in [7]–

[10]. Unlike generic data, compressed video sources exhibit different characteristics, for example, there is

highly bursty rate from frame to frame and different compression complexity from one scene to another

scene. Furthermore, a streaming video system has a strict delay constraint that belated video data is useless

for its corresponding frame and will cause error propagation for the video frames encoded predictively

from this frame. Therefore, the radio resource allocation problem for transmitting video is more difficult

than the problem for transmitting generic data. A real-time low-complexity algorithm for transmitting

wireless video is desired.

To transmit video programs over wireless networks, a system should be able to adjust the video source

bit rates according to the varying channel conditions. A highly scalable video codec is desired since it

provides flexibility and convenience in reaching the desired visual quality or the desired bit rate. The

Fine Granularity Scalability (FGS) coding and Fine Granular Scalability Temporal (FGST) coding in the

MPEG-4 video coding standard can provide high flexibility. However, their overall qualities are worse

than the non-scalable coding results, and there remains a non-scalable base layer. The development of

3-D subband video coding [11]–[16] provides an alternative to compress video with full scalability,

namely, spatial scalability, temporal scalability, and SNR scalability. Unlike the motion compensated

video codec based on block matching (such as H.263 and MPEG-4), the 3-D subband coding explores

the spatiotemporal redundancies via a 3-D subband transform. Extending the bit allocation ideas from

the EBCOT algorithm for image compression [17], the 3-D embedded wavelet video codec (EWV)

DRAFT
 3

[16] outperforms MPEG-4 for sequences with low or moderate motion and has comparable performance

to MPEG-4 for most high-motion sequences. Moreover, the rate-distortion (R-D) information can be

predicted during the encoding procedure and provide a one-to-one mapping between rate and distortion

such that we can achieve the desired perceptual quality or the targeted bit rate. Thus, we adopt the EWV

codec in the proposed framework as an example. We can easily incorporate other codecs with similar

coding strategy into the proposed scheme.

A wireless system transmitting a single video program has been widely studied in the literature [18]–

[20]. To improve the overall system performance, joint source and channel coding has been shown as

an effective approach [21]–[27]. When we consider a system transmitting heterogeneous video programs

simultaneously, the system has another dimension of diversity to explore since different video scenes have

different content complexity: at a given encoded bit rate, some video scenes may have unnecessarily high

perceptual quality, while others may have low perceptual quality. It has been shown that joint multiple

video source coding can leverage different video content complexities to achieve more desired quality

[28]–[32]. Thus, for a multiuser wireless system, the main challenges to achieve the highest system

performance are how to allocate limited and shared radio resources to multiple users, how to jointly

select video source and radio parameters, and how to deliver video streams to multiple users in real time.

A simple solution for a multiuser wireless video system was proposed by assigning subcarriers according

to the length of terminal’s queue [33]. In this paper, we overcome the aforementioned challenges by

allocating resources through a multiuser cross-layer optimization, namely, we formulate the whole system

as optimization problems by jointly exploring the diversity of video and radio resources in a cross-layer

fashion to optimize the network-level service objectives.

Motivated by the above advantages of multiuser OFDM system and EWV video codec, we propose

a framework to provide multiple video streams to different users using dynamic distortion control. The

proposed framework has the following features. First, the system dynamically gathers the information

of system resources from different components to capture the time-heterogeneity of video sources and

DRAFT
 4

time-varying characteristics of channel conditions. Subject to delay constraint, the system explores multi-

dimensional diversity among users and across layers, performs joint multiuser cross-layer resource al-

location optimization, and then distributes the system resources to each user. The benefit for such joint

consideration is the higher utilization of system resources. The simulation results demonstrated that the

proposed fairness algorithm outperforms a time division multiple (TDM) algorithm based on traditional

WLAN technology by 0.5∼3dB in terms of the worst received video quality criterion. Second, extremely

fair allocation in such a heterogeneous environment will cause low overall video qualities when some

users are trapped in bad channels. On the other hand, optimizing the system efficiency will only cause

unfairness among users. To reach the tradeoff between fairness and efficiency, our proposed framework

first achieves baseline fairness among all users and then pursuits the high overall system’s efficiency.

Compared to the TDM algorithm, the proposed framework can provide fairer video quality with 1∼1.8dB

lower PSNR deviation among all users for achieving the same overall video quality.

This paper is organized as follows. The system architecture for transmitting 3-D EWV over multiuser

OFDM networks is described in Section II. In Section III, we concentrate on fairness issue among users

and formulate the proposed system as a min-max problem. In Section IV, we focus on system efficiency.

The tradeoff between fairness and efficiency and potential solution to increase efficiency through unequal

error protection are addressed in Section V. Simulation results are presented in Section VI and conclusions

are drawn in Section VII.

II. S YSTEM D ESCRIPTION

There are three major subsystems in the proposed wireless video system, namely, the video source

codec subsystem, the multiuser OFDM subsystem, and the resource allocator subsystem. We first review

the video source codec subsystem along with the corresponding R-D characteristics, and describe the

multiuser OFDM subsystem with adaptive modulation and adaptive channel coding. Then, we present the

proposed framework for transmitting multiple scalable video bitstreams over multiuser OFDM networks.

DRAFT
 5

A. Video Source Codec Subsystem

The EWV encoder consists of four stages [16], namely, 3-D wavelet transform, quantization, bit plane

arithmetic coding, and rate-distortion optimization. At the first stage, we collect a group of frames (GOF)

as an encoding unit and apply 1-D dyadic temporal decomposition to obtain temporal subbands. The

2-D spatial dyadic decomposition is applied in each temporal subband to obtain wavelet spatiotemporal

subbands (or “subbands” for short). At the second stage, a uniform quantizer is used for all wavelet

coefficients in all subbands. At the third stage, fractional bit plane arithmetic coding is applied to each

subband. Except that the most significant bit plane (MSB) has only one coding pass, every bit plane is

encoded into three coding passes, namely, significance propagation pass, magnitude refinement pass, and

normalization pass. Each coding pass can be treated as a candidate truncation point and the EWV decoder

can decode the truncated bitstream containing an integer number of coding passes in each subband. The

more consecutive coding passes of each subband a receiver receives, the higher decoded video quality

we have. The coding passes among all subbands can be further grouped into several quality layers such

that the received video quality can be refined progressively by receiving more layers. At the last stage,

the encoder determines which coding passes are included in the output bit stream subject to quality or

rate constraint.

To maintain the coding efficiency, the R-D curve in each subband should be convex [17]. Some coding

passes in a subband cannot serve as feasible truncation points to maintain the convexity and they will be

pruned from the truncation point list. To facilitate the discussion, we call all the coding passes between

two truncation points as a coding pass cluster.

Consider now there are a total of B subbands for the k th user and the subband b has Tkb,max coding

pass clusters. We can measure the rate and the corresponding decrease in normalized mean squared

distortion of the tth coding pass cluster in subband b for the k th user [17], and denote them as ∆rt,b,k

and ∆dt,b,k , respectively. We divide the whole duration for transmitting a total of L quality layers into

L transmission intervals with equal length. The lth quality layer is transmitted at the lth transmission

DRAFT
 6

Fig. 1. Illustration of the relationship among coding pass, subband, and quality layer.

interval. The received distortion Dkl and rate Rkl for quality layers 0 to l can be expressed as:
b,l
X TX
B−1 k −1

Dkl = Dkmax − ∆dt,b,k , (1)

b=0 t=0
l
X
Rkl = ∆Rkq . (2)
q=0

Here Dkmax is the distortion without decoding any coding pass cluster,
b,l
X TX
B−1 k −1

∆Rkl = ∆rt,b,k , (3)

b=0 t=T b,l−1
k

and Tkb,l is the total number of coding pass clusters of subband b in the quality layers 0 to l, which

satisfies:
0 ≤ Tkb,l−1 ≤ Tkb,l ≤ Tkb,max , ∀b and 0 < l < L. (4)

Define the number of coding pass clusters for subband b in quality layer l as ∆Tkb,l = Tkb,l − Tkb,l−1 and

for all subbands

∆Tlk = [∆Tk0,l , ∆Tk1,l , . . . , ∆TkB−1,l ]. (5)

We also define a matrix ∆Tl whose k th row is ∆Tlk . Thus, in each transmission interval l, the source

coding part of our system determines the coding pass cluster assignment ∆Tlk and packetizes them as a

quality layer for each user. We use Figure 1 to illustrate the relationship among coding pass, subband,

and quality layer. Note that owing to different content complexities and motion activities shown in video

sources, the R-D information should be evaluated for each GOF of each user to capture the characteristics

of the corresponding bitstream.

DRAFT
 7

B. Multiuser OFDM Subsystem

We consider a downlink scenario of a single-cell multiuser OFDM system in which there are K users

randomly located. The system has N subcarriers and each subcarrier has bandwidth of W . We use an

indicator akn ∈ {0, 1} to represent whether the nth subcarrier is assigned to user k . Note that in a single-
PK−1
cell OFDM system, each subcarrier can be assigned to at most one user, i.e., k=0 akn ∈ {0, 1}, ∀n.

The overall subcarrier-to-user assignment can be represented as a matrix A with [A]kn = akn . Let rkn be

the k th user’s transmission rate at the nth subcarrier and the total rate for the k th user can be expressed
PN −1
as n=0 akn rkn . The overall rate allocation can also be represented as a matrix R with [R]kn = rkn .

In mobile wireless communication systems, signal transmission suffers from various impairments such

as frequency-selective fading due to multipath delay [40]. The continuous complex baseband representa-

tion of user k ’s wireless channel impulse response is expressed as

X
gk (t, τ ) = υk,i (t)δ(τ − τk,i ), (6)
i

where υk,i (t) and τk,i are the gain and the delay of path i for user k , respectively. In Rayleigh fading,

the sequence υk,i (t) is modelled as a zero-mean circular symmetric complex Gaussian random variable

with variance συ2k,i proportional to d−α , where d is the distance and α is the propagation loss factor. All

υk,i (t) are assumed to be independent for different paths. The root-mean-square (RMS) delay spread is

the square root of the second central moment of the power delay profile:
q
σk,τ = τk2 − (τ¯k )2 , (7)
P 2 2
P 2
i συk,i τk,i i συk,i τk,i
where τk2 = P and τ¯k = P .
i συk,i i συk,i
2 2

After sampling at the receiver, the channel gain of OFDM subcarriers can be approximated by the

discrete samples of the continuous channel frequency response as

Z ∞
Ghkn = gk (t, τ )e−j2πf τ dτ |f =nW,t=hTf , (8)
−∞

where Tf is the duration of an OFDM symbol and h is the sampling index. This approximation does not

consider the effect of the smoothing filter at the transmitter and the front-end filter at the receiver.

DRAFT
 8

We assume a slow fading channel where the channel gain is stable within each transmission interval.1

The resource allocation procedure will be performed in each transmission interval. To facilitate the

presentation, we omit h in the channel gain notation. The channel parameters from different subcarrier

of different users are assumed perfectly estimated, and the channel information is reliably fed back from

mobiles users to the base station in time for use in the corresponding transmission interval. Denote Γkn

as the k th user’s signal to noise ratio (SNR) at the nth subcarrier as:

Γkn = Pkn Gkn /σ 2 , (9)

where Pkn is the transmission power for the k th user at the nth subcarrier and σ 2 is the thermal noise

power that is assumed to be the same for each subcarrier of different users. Further, let [G]kn = Gkn be

the channel gain matrix and [P]kn = Pkn the power allocation matrix. For downlink system, because of the

practical constraints in implementation, such as the limitation of power amplifier and consideration of co-
PK−1 PN −1
channel interferences to other cells, the overall power is bounded by Pmax , i.e., k=0 n=0 akn Pkn ≤

Pmax .

The goal of the proposed framework is to provide good subjective video quality of the reconstructed

video. Since the distortion introduced by channel error is typically more annoying than the distortion

introduced by source lossy compression, the system should keep the channel-induced distortion at a

negligible portion of the end-to-end distortion so that the video quality is controllable by the source

coding subsystem. This can be achieved when we apply an appropriate amount of channel coding to

keep the bit error rate (BER) after the channel coding below some targeted BER threshold [31], which is

10−6 in our system and achievable in most 3G/4G systems. In addition, joint consideration of adaptive

modulation, adaptive channel coding, and power control can provide each user with the ability to adjust

each subcarrier’s data transmission rate rkn to control video quality while meeting the required BER.

We focus our attention on MQAM modulation and convolutional codes with bit interleaved coded

modulation (BICM) as they provide high spectrum efficiency and strong forward error protection, re-

1
In practice, the duration of a transmission interval can be adjusted shorter enough so that the channel gain is stable within
a transmission interval.

DRAFT
 9

TABLE I

R EQUIRED SNR AND TRANSMISSION RATE USING ADAPTIVE MODULATION AND CONVOLUTIONAL CODING RATES [35]

Rate Modulation Convolutional SNR ρk (dB) for SNR (dB) for

k νk Coding Rate BER ≤ 10−6 BER ≤ 10−5
1 1W QPSK 1/2 4.65 4.09
2 1.33W QPSK 2/3 6.49 5.86
3 1.5W QPSK 3/4 7.45 6.84
4 1.75W QPSK 7/8 9.05 8.44
5 2W 16QAM 1/2 10.93 10.04
6 2.66W 16QAM 2/3 12.71 12.13
7 3W 16QAM 3/4 14.02 13.29
8 3.5W 16QAM 7/8 15.74 15.01
9 4W 64QAM 2/3 18.50 17.70
10 4.5W 64QAM 3/4 19.88 18.99
11 5.25W 64QAM 7/8 21.94 21.06

spectively. We list the required SNRs’ and the adopted modulation with convolutional coding rates to

achieve different supported transmission rates under different BER requirement in Table I based on the

results in [35]. Given a targeted BER, there is a one-to-one mapping between the selected transmission

rate and the chosen modulation scheme with convolutional coding rate when the required SNR is satisfied.

In this case, determining rkn is equal to determine the modulation and channel coding rate. For each rate

allocation [R]kn , the corresponding power allocation [P]kn should maintain the SNR in (9) larger than

the corresponding value listed in Table I to achieve the BER requirement. To facilitate our discussion, we

define the feasible set of the transmission rate in Table I as ν = {ν0 , ν1 , ν2 , . . . , νQ } and the corresponding

set of the required SNR for BER ≤ 10−6 as ρ = {ρ0 , ρ1 , ρ2 , . . . , ρQ }. Here, ν0 = 0 and ρ0 = 0, and Q

represents the number of combinations for different modulation with convolutional coding rates, which

is 11 in our case. All transmission rate rkn s’ should be selected from the feasible rate set ν .

C. EWV Video over Multiuser OFDM

The block diagram of the proposed wireless video system is shown in Figure 2. The upper and lower

parts show the modules located in the server side and the mobile user side, respectively. For the server

side, the server buffers each user’s incoming video frames in the user’s frame buffer. After collecting a

GOF with H frames for each user, the server moves those raw video frames to a wavelet video encoder for

DRAFT
 10

All Coding Pass

frame buffer bit stream

Video
Video Program 0 0
Encoder
Resource
Allocator
1
Downlink
Video Data
Video Program 1 2 Stream
Encoder Distortion OFDM
Control system
A
P
Bit Loading
Power R
Control
Video
Video Program K-1 N-1
Encoder

R-D information Channel Feedback

Channel Condition Estimator

Server

Received
coding pass frame buffer
bit stream

OFDM Video Video

Channel
receiver system Decoder Output

User k

Fig. 2. System block diagram

compression as a coding pass bit stream. The selected coding pass clusters will be transmitted during the

next GOF transmission time of H/F second long, where F is the video frame sampling rate. To capture

the varying channel conditions and video content characteristics, the resource allocator should obtain the

channel information for each transmission interval from the channel estimator and the R-D information of

each GOF from the video coder. With the estimated channel conditions, the resource allocator can predict

how many data rates with BER ≤ 10−6 the wireless networks can support in the next transmission interval.

With the R-D information, the resource allocator can estimate the qualities of the reconstructed videos

after decoding at each mobile terminal. By jointly considering the R-D information and the estimated

channel conditions, the resource allocator performs resource optimization and distributes video and radio

resources to each video stream in each transmission interval. According to the allocated resources, the

source coding subsystem will group the selected coding pass clusters into a quality layer for each user

and pass them to the transmission system; and the multiuser OFDM subsystem will load the video data

to be transmitted to different subcarriers at a controlled amount of power. On the mobile user side, an

OFDM receiver buffers the received data until the end of the current GOF transmission time. Then, those

received data are moved to a wavelet video decoder for decoding and the decoded frames are sent for

display during the next GOF transmission time.

DRAFT
 11

Since we only know the channel conditions provided by the channel estimator in the near future

within the next transmission interval and the GOF bitstreams are transmitted across L transmission

intervals, it is necessary to break down the optimization problem into a sequential optimization problem

and solve each problem in each transmission interval. There are two different objectives we want to

achieve in each transmission interval: fairness and efficiency. To ensure the fairness among all users, we

formulate the problem as a min-max optimization problem to minimize the maximal (weighted) end-to-

end distortion among all users. Maintaining short-term fairness in each transmission interval ensures the

long-term fairness for GOFs [34]. To achieve high resource-allocation efficiency in terms of a high overall

video quality, we formulate the problem as an optimization problem to minimize the overall end-to-end

distortion among all users. We will discuss the fairness and efficiency problems in Section III and Section

IV, respectively. The tradeoff between efficiency and fairness will be addressed in Section V.

III. O PTIMIZATION IN R ESOURCE A LLOCATOR : F OCUSING ON FAIRNESS

In this section, we consider how to achieve fair video quality among all users in a transmission interval

and formulate this problem as a min-max problem. Given the integer programming nature of the problem,

we propose a three-stage suboptimal algorithm to solve the optimization problem in real time.

A. Formulation of the Fairness Problem

At the beginning of the lth transmission interval, according to the channel information and subject to

the transmission delay constraint as one transmission interval long, the resource allocator minimizes the

maximal distortion received among all users as follows:

min max wk · f (Dkl ) (10)
A,R,∆Tl k
 PK−1




Subcarrier Assignment: k=0 akn ≤ 1, akn ∈ {0, 1}, ∀n;


 Subcarrier Rate: r ∈ ν, ∀k, n;
kn
subject to PN −1

 User Rate: 0 ≤ ∆Rkl ≤ n=0 akn rkn , ∀k;



 P P
 Power: K−1 N −1
k=0 n=0 akn Pkn ≤ Pmax ;

where wk is the quality weighting factor and f (·) the perceptual distortion function. We solve this

optimization problem by selecting the values of subcarrier assignment matrix A, rate assignment matrix

DRAFT
 12

Obtain GOF R-D of all unsent coding

pass clusters for all users.
Step 1

Fairness Efficiency
Perform Fairness or
Efficiency algorithm

Fairness: bisection Efficiency: 2-D

search for feasible waterfilling for
subcarrier and rate subcarrier and rate
assignment assignment
Step 2

Coding pass cluster

assignment
Step 3

No The last
transmission interval for current
GOF?

Yes

Done

Fig. 3. Flowchart of the proposed algorithm

R, and coding pass cluster assignment ∆Tl subject to four constraints: the first constraint is on subcarrier

assignment that a subcarrier can be assigned to at most one user; the second one restricts the subcarrier

rate to be selected only from the feasible rate set ν ; the third one is that the user’s overall assigned rate

in (3) should be no larger than the overall assigned subcarrier rate; and the fourth one is on the maximal

power available for transmission. Note that the system can provide differentiated service by setting {wk }

to different values according to the quality levels requested by each user. As a proof of concept, we

consider the case of wk = 1, ∀k and f (Dkl ) = Dkl for providing uniform quality among all users here.

The proposed solution can be easily extended to other quality weighting factors and quality functions, and

we will demonstrate the ability for providing differentiated service in Section VI. The problem in (10)

is a multi-dimension generalized assignment problem, which is an NP hard problem [36]. In a real-time

system, a fast approximation algorithm with good performance is needed and will be designed next.

B. Proposed Algorithm for Fairness

We propose a three-stage fast algorithm to solve the optimization problem (10). As illustrated by the

flowchart in Figure 3, at the first stage, we obtain continuous GOF R-D functions that provide a distortion-

to-rate mapping to facilitate the resource allocation. At the second stage, we determine the subcarrier

assignment matrix A and rate assignment matrix R to find the largest distortion reduction that the OFDM

system can support. This goal can be achieved through a bisection search on the R-D functions obtained

DRAFT
 13

TABLE II

GOF R-D USED IN EACH TRANSMISSION INTERVAL

a) Sort all {λt,b,k } for all t ≥ Tkb,l−1 in a decreasing order for user k
b) For each (t, b) for user k, we have indices
Ik (t, b) ∈ {1, 2, ..., Mkl } and Ik−1 (m) ∈ {(t, b)}
s.t. λi,b,k > λj,c,k for Ik (i, b) < Ik (j, c)
c) Set Dkl [0] = Dkl−1 and Rkl [0] = 0
For m = 1, ..., Mkl
Dkl [m] = Dkl [m − 1] − |∆dI −1 (m),k |
k
Rkl [m] = Rkl [m − 1] + ∆rI −1 (m),k
k
d) Construct the continuous GOF R-D function
l l
Dk [m+1]−Dk [m]
Dkl (γkl ) = l
l [m+1]−Rl [m] (γk − Rkl [m]) + Dkl [m],
Rk k
for Rkl [m] ≤ γkl ≤ Rkl [m + 1] and m = 0, ..., Mkl − 1.

at Stage 1. At the third stage, the coding pass cluster assignment ∆Tl is decided subject to the allocated

subcarrier and rate assignment at Stage 2. We explain the details of each stage below:

1) Stage 1: At this stage, a continuous GOF R-D function of the unsent coding pass clusters for each

user is obtained. The goal for determining the GOF R-D function is to provide a one-to-one mapping

between rate and distortion such that we can know the amount of rate increment necessary for a given

amount of reduction in distortion.

Suppose there are Mkl unsent coding pass clusters for user k at the beginning of the transmission

interval l. Define the distortion-to-length slope for a coding pass cluster with the rate increment ∆rt,b,k

and the distortion reduction ∆dt,b,k as

λt,b,k , |∆dt,b,k | /∆rt,b,k . (11)
The distortion-to-length slope represents how much distortion a coding pass cluster can reduce by given

one unit of rate. We can sort all distortion-to-length slopes of all unsent coding pass clusters (λt,b,k ,

where Tkb,l−1 ≤ t) in a decreasing order and obtain the corresponding mapping indices Ik (t, b) and

inverse indices Ik−1 (m) ∈ {(t, b)}. For example, if the sorting result is λt1 ,b1 ,k > λt2 ,b2 ,k > · · · , we

assign Ik (t1 , b1 ) = 1, Ik (t2 , b2 ) = 2, and Ik−1 (1) = (t1 , b1 ), Ik−1 (2) = (t2 , b2 ). Then, a decreasing

discrete R-D function (Rkl [m],Dkl [m]) for quality layer l can be obtained according to this sorting result,

as shown in Table II (c). To facilitate the distortion-to-rate searching, we relax the constraints on integer

value of rate and integer number of coding pass clusters to allow them to be real numbers; and construct

DRAFT
 14

a continuous R-D function through linear interpolation of the discrete R-D function as follows:
Dl [m + 1] − Dkl [m] l
Dkl (γkl ) = kl (γ − Rkl [m]) + Dkl [m], (12)
Rk [m + 1] − Rkl [m] k

for Rkl [m] ≤ γkl ≤ Rkl [m + 1] and m = 0, ..., Mkl − 1,

where γkl is the required bit rate. We can calculate the least required rate, γkl , to achieve the targeted

distortion, D, by finding the inverse function of Dkl (·). We summarize the algorithm used in this stage

in Table II. The complexity of this stage for each user is O(Mkl log(Mkl )) due to sorting.

2) Stage 2: At this stage, the goal is to minimize the maximal distortion supported by the OFDM

subsystem through a bisection search procedure. By checking the continuous GOF R-D functions obtained

in (12), the resource allocator can calculate the minimum transmission rates {γkl } necessary to achieve the

same targeted distortion among all users. Then the resource allocator checks whether these requested rates

can be supported by the OFDM subsystem under current channel conditions. If the requested rates are

feasible, the resource allocator tries to further decrease the targeted distortion by increasing the requested

rates. Otherwise, the resource allocator increases the targeted distortion to reduce the requested rates and

checks the feasibility again. A bisection search algorithm is deployed to find the minimal distortion D

that the OFDM subsystem can support.

The feasibility of the requested rates depends on two factors. First, the OFDM subsystem should be

able to transmit the requested rates {γkl } for all users. Second, the overall transmission power, Psum ,

cannot exceed Pmax . We develop a fast suboptimal algorithm shown in Table III to allocate the bits

and power to satisfy the rate constraint first and then the power constraint. There are three steps in the

proposed algorithm for feasibility checking: initialization, minimal rate assignment, and power reduction.

First, the subcarrier assignment matrix A, the rate assignment matrix R, and the power assignment matrix

P are initialized to zeros. Next, the system tries to satisfy the requested rates. In each round, we allocate

an unassigned subcarrier to a user. If Gk̂n̂ has the maximal value in current G, subcarrier n̂ is assigned

to user k̂ and we update Gkn̂ = 0, ∀k to prevent this subcarrier from being assigned again. We then

determine the modulation schemes and the coding rates for all subcarriers currently allocated to user k̂

such that the requested data rate can be accommodated and the required transmission power is minimized

DRAFT
 15

TABLE III

R ESOURCE A LLOCATION AND F EASIBILITY C HECK IN OFDM SUBSYSTEM

a) Initialization: Get ∆Rkl , ∀k and set A = 0, R = 0, and P = 0.

b) Minimal Rate Assignment:
PN −1
While n=0 rkn akn ≥ ∆Rkl , ∀k not satisfied
1) Find k̂, n̂ = arg maxk,n [G]kn .
2) Assign subcarrier n̂ to user k̂. Set Gkn̂ = 0, ∀k. Waterfill all
subcarriers of user k̂ to minimize power with rate ∆Rl .
k̂
3) If the requested rate of user k̂ is achieved, set Gk̂n = 0, ∀n.
4) If no subcarriers left and not all requested rates are satisfied,
report infeasibility and exit.
c) Power Reduction:
While there are subcarriers left
1) Assign user k who has the highest average power per subcarrier
with a remaining subcarrier having the largest Gkn for user k.
2) Minimize the transmission power among the subcarrier set
assigned to this user.
3) Calculate the overall power. If not greater than Pmax ,
calculate A, R, and P. Exit and report feasibility.
Calculate the overall power. If greater than Pmax , report infeasibility.

in the meantime. This can be implemented by the well-known waterfilling algorithm with Table I and

(9). If the requested rate of user k̂ can be allocated, user k̂ is removed from future assignment list in this

step by assigning Gk̂n = 0, ∀n. This step is repeated until all users’ requested rates are satisfied. If all

subcarriers are already assigned and not all requested rates can be allowed, infeasibility is reported and

the resource allocator has to reduce the requested rates. In the third step, we try to reduce the overall

transmission power Psum to be below Pmax by assigning the remaining subcarriers. In each round, we

select a user who has the highest average power per subcarrier and assign him/her with one of the

remaining subcarriers in which this user has the largest channel gain. Then we minimize the transmission

power among the subcarrier set assigned to this user. The overall transmission power is calculated and

if it is greater than Pmax , the power reduction procedure is repeated again. Otherwise, we calculate A,

R, and P for OFDM subsystem, report feasibility and exit. An infeasibility is reported if there is no

subcarrier left and Psum > Pmax . Since the required power for each user in each subcarrier can be

pre-calculated, the complexity of checking feasibility in each iteration is O(N ). The overall number of

iterations, which is typically fewer than 20 in our experiment, is bounded by the bisection search.

DRAFT
 16

TABLE IV

C ODING PASS CLUSTER ASSIGNMENT

a) Find m̂k = arg max{Rkl [m] ≤ ARkl , ∀m}.

Allocate coding pass clusters with indices Ik (t, b) ≤ m̂k .
b) Calculate unused bandwidth U Rkl = ARkl − Rkl [m̂k ].
While U Rkl ≥ 0
1) Search the coding pass clusters set, S, whose element
satisfies ∆rt,b,k ≤ U Rkl and t = Tkb,l in all subbands.
2) If set S is empty, leave the loop.
3) Select the coding pass cluster, b̂, with largest λt,b,k in set S.
4) Update Tkb̂,l = Tkb̂,l + 1 and U Rkl = U Rkl − ∆rt,b̂,k .

3) Stage 3: At this stage, we perform the coding pass cluster assignment for each user individually.
PN −1
Denote the assigned rate from Stage 2 for the k th user as ARkl = n=0 akn rkn . Due to the discrete

rate provided by the OFDM subsystem (rkn ∈ ν, ∀k, n), the assigned transmission rate ARkl is generally

larger than the requested rate, i.e., ARkl ≥ γkl . Therefore, we have rate budget ARkl to allocate the coding

pass clusters. We formulate the problem as minimizing the distortion subject to the rate constraint by

allocating the coding pass cluster:

min Dkl subject to ∆Rkl ≤ ARkl . (13)
∆Tlk

Since for each unsent coding pass cluster we need to decide whether we select it or not, the problem (13)

is a binary knapsack problem [36], which is NP hard. To ensure the real-time performance, we apply a

two-step greedy algorithm here. First, among all values of Rkl [m] that are not larger than ARkl , we find

the largest one, Rkl [m̂k ]. We will include all coding pass clusters whose indices Ik (t, b) are not larger

than m̂k in the current quality layer. Notice that the sorting order {Ik (t, b)} has ensured the decoding

dependency of coding pass clusters in each subband. This is because if user k receives the tth coding

pass cluster in subband b, user k must receive coding pass cluster 0 to t − 1 since λt0 ,b,k > λt,b,k , ∀t0 < t

or Ik (t0 , b) < Ik (t, b), ∀t0 < t due to the convexity of R-D in subband b. Second, a round of refinement

is performed to utilize the unused bandwidth, U Rkl = ARkl − Rkl [m̂k ]. We search all unsent coding pass

clusters that follow the currently selected truncation points and pick those with rates not larger than

the unused bandwidth. The coding pass cluster with the largest distortion-to-length slope is selected for

transmission during current transmission interval. The system updates the unused bandwidth and unsent

DRAFT
 17

coding pass clusters; and then repeats the above search procedure until there is no coding pass cluster

with size not larger than the unused bandwidth. Since the first step directly uses the result in (12), the

refinement step consumes most computation power in the whole coding pass cluster assignment and the

complexity to search a feasible coding pass cluster is O(B). We recap the algorithm used in this stage

in Table IV.

IV. O PTIMIZATION IN R ESOURCE A LLOCATOR : F OCUSING ON E FFICIENCY

In this section, we study the efficiency problem in which the overall distortion of all users is mini-

mized in a transmission interval. We first formulate the efficiency problem as an optimization problem.

Then, similar to the fairness case, we also propose a three-stage algorithm to determine the subcarrier

assignment, rate assignment, and coding pass cluster assignment to achieve the optimization goal.

A. Formulation of the Efficiency Problem

We formulate the efficiency problem as to minimize the overall (weighted) end-to-end distortion among

all users subject to constraints on subcarrier assignment, subcarrier rate, user rate, and power:
K−1
X
min l wk · f (Dkl ) (14)
A,R,∆T
k=0
 PK−1




Subcarrier Assignment: k=0 akn ≤ 1, akn ∈ {0, 1}, ∀n;


 Subcarrier Rate: r ∈ ν, ∀k, n;
kn
subject to P −1

 User Rate: 0 ≤ ∆Rkl ≤ N

 n=0 akn rkn , ∀k;

 P P
 Power: K−1 N −1
k=0 n=0 akn Pkn ≤ Pmax .

The constraints are similar to the fairness case presented in Section III-A. Similar to the fairness case,

the delay constraint is implicitly imposed in the problem (14) so as the transmission delay is restricted

within a transmission interval.

B. Proposed Algorithm for Efficiency

To solve this minimization problem, we propose a three-stage algorithm shown in Figure 3. The first

stage is to obtain the continuous R-D functions of all unsent coding pass clusters for the current GOF.

The second stage is to perform subcarrier assignment and rate assignment through a 2-D waterfilling

DRAFT
 18

procedure; and the third stage is the coding pass cluster assignment. The first and third stage are the

same as what have been discussed in Section III-B. Here, we focus on the second stage and consider the

case of wk = 1, ∀k and f (Dkl ) = Dkl .

Having the continuous R-D functions, the problem (14) can be simplified as follows:
K−1
X
min Dkl (γkl ) (15)
A,R
k=0
 PK−1




Subcarrier Assignment: k=0 akn ≤ 1, akn ∈ {0, 1}, ∀n;
subject to Subcarrier Rate: rkn ∈ ν, ∀k, n;


 Power: PK−1 PN −1 a P ≤ P

k=0 n=0 kn kn max ;
PN −1
where γkl = n=0 akn rkn . To solve this problem, a two-step suboptimal algorithm is proposed by first

determining the subcarrier assignment matrix A and then deciding the rate assignment matrix R.

1) Subcarrier Assignment: In this step, we relax the power constraint by assuming the maximal

transmission power is unlimited and thus each subcarrier can be loaded with maximum rate νQ to fully

utilize the available bandwidth. Then, the problem (15) has only the subcarrier assignment constraint and

the goal is to find the subcarrier-to-user assignment that can reduce most distortion by using the least

amount of power. This problem can be solved by an iterative greedy algorithm. In each iteration, we

evaluate which user can achieve the most distortion reduction by using the least power if we assign an

unused subcarrier to him/her. There are two factors affecting this evaluation, as reflected by ψkn and

defined below: µ ¶Ã !
Dkl (γkl ) − Dkl (γkl + νQ ) νQ
ψkn , ρQ σ 2
. (16)
νQ
Gkn
γkl is the accumulative allocated rate for user k in the current iteration. The first term of (16) evaluates

the gradient of reduced video distortion with respect to the allocated rate, i.e., how much distortion we

can reduce by assigning a unit of rate for user k . The second term of (16) evaluates the gradient of the

allocated rate to the required transmission power (calculated using (9)), i.e., how many bits this system

can transmit at BER≤ 10−6 per unit of power. If both factors of user k at subcarrier n are large, it implies

that assigning subcarrier n to user k can use the same amount of power to reduce more distortion. Since

the second term is only a function of channel gain, we can further simplify (16) and have a matrix Ψ

DRAFT
 19

as [Ψ]kn = ψkn :
ψkn = (Dkl (γkl ) − Dkl (γkl + νQ ))Gkn . (17)

Once we obtain Ψ, we can find its entry (k̂, n̂) with maximal value and assign subcarrier n̂ to user k̂ .

To prevent this subcarrier from being assigned again, we set Gkn̂ = 0, ∀k . Then, we update the current

allocated rate of user k̂ by setting γk̂l = γk̂l + νQ and subcarrier assignment matrix by setting ak̂n̂ = 1

and ak0 n̂ = 0 for k 0 6= k̂ . This procedure is repeated until all subcarriers are assigned. The complexity

of this step is O(N ).

2) Rate Assignment: Based on the subcarrier assignment in the previous step, we determine how

much rate should be assigned to each subcarrier. To facilitate our discussion, let θkn ∈ {0, 1, . . . Q} be

the subcarrier usage index, which indicates the selected row in Table I for user k at subcarrier n. For

example, θkn = q represents that user k has loaded rkn = νq bits in subcarrier n and the required SNR to

achieve BER≤ 10−6 is ρq . Further, we define a set of incremental rate ∆ν = {∆ν1 , ∆ν2 , . . . , ∆νQ } and

a set of incremental power ∆ρ = {∆ρ1 , ∆ρ2 , . . . , ∆ρQ }, where ∆νq = νq − νq−1 and ∆ρq = ρq − ρq−1

for q = 1, ..., Q, respectively. We solve this rate assignment problem using a 2-D discrete waterfilling

algorithm. At the beginning, we set all required rate {γkl } and all subcarrier usage indices {θkn } to

zeros. In each iteration, similar to Step 1, we select the subcarrier setting that can achieve the most

distortion reduction by using the least power when we evaluate the results of filling each subcarrier an

incremental rate ∆νθkn +1 . This procedure is repeated until all subcarriers are fully loaded or the overall

required power reaches the maximal available amount. The evaluation of distortion-to-power ratio for all

subcarriers and users can be quantified as a matrix Φ with [Φ]kn = φkn :

Dkl (γkl ) − Dkl (γkl + ∆νθkn +1 ) ∆νθkn +1
φkn , ( )( ). (18)
∆νθkn +1 ∆ρθkn +1 σ 2 /Gkn
The first term of (18) represents how much distortion user k can reduce with an extra unit of rate and

the second term of (18) represents how many bits to transmit for user k at subcarrier n with a unit of

power. The overall φkn represents how much distortion user k will reduce at subcarrier n with one extra

unit of power.

After obtaining Φ, we select the entry (k̂, n̂) with largest value. If subcarrier n̂ does not belong to

DRAFT
 20

TABLE V

P ROPOSED ALGORITHM TO MINIMIZE OVERALL DISTORTION

a) Initialization: Psum = 0, Exit = False.

b) Subcarrier Assignment:
Set γkl = 0, ∀k.
While not all subcarriers are assigned
Calculate Ψ using (17).
Find k̂, n̂ = arg maxk,n [Ψ]kn .
Set ak̂n̂ = 1 and ak0 ,n̂ = 0 for k0 6= k̂. Set Gk,n̂ = 0 ∀k.
Update user’s rate γl = γl + νQ .
k̂ k̂
c) Rate Assignment:
Set γk = 0, ∀k; and set Exit as False.
While Exit == False
Calculate Φ using (18).
Set Found as False.
While Found == False
Find k̂, n̂ = arg maxk,n [Φ]kn .
If φk̂n̂ == 0, set Exit as True and Found as True.
Else If ak̂n̂ == 0, set φk̂n̂ == 0.
Else
2
∆ρθ +1 σ
If rk̂n̂ + ∆νθ +1 ≤ νQ and Psum + k̂n̂
Gk̂n̂
≤ Pmax
k̂n̂

Set Found as True,

Update system using (19)(20)(21).
Else, set φk̂n̂ = 0.

user k̂ , i.e., ak̂n̂ = 0, we set φk̂n̂ = 0 and search the highest value in Φ again. If so, we update user k̂ ’s

subcarrier usage index

θk̂n̂ = θk̂n̂ + 1, (19)

the overall transmission rate of user k̂

γk̂l = γk̂l + ∆νθk̂n̂ , (20)

and the overall transmission power

∆ρθk̂n̂ σ 2
Psum = Psum + . (21)
Gk̂n̂
If subcarrier n̂ is overloaded, (i.e., rk̂n̂ > νQ ), or the overall required power exceeds the Pmax , we need

to pick other entry by setting φk̂n̂ = 0 and search the highest value in Φ until a valid one is found.

The search algorithm terminates if no more valid assignment is found. The whole algorithm is presented

in Table V. Since the accumulative rate γk̂l is updated for the selected user k̂ only, the complexity of

updating Φ in each iteration is O(nk̂ ), where nk̂ is the number of subcarriers assigned to user k̂ . The

DRAFT
 21

maximal number of iteration in the rate assignment is N Q; and the actual number of iteration depends

on the transmission power level and channel condition.

V. E XTENDED F UNCTIONALITIES IN THE P ROPOSED F RAMEWORK

In this section, we discuss the extended functionalities based on the proposed algorithms in Section

III and IV. We first address how to achieve a desired tradeoff between system fairness and efficiency.

Then we investigate how to incorporate unequal error protection in the proposed framework to increase

system’s efficiency.

A. Tradeoff between Fairness and Efficiency

We have proposed two solutions to ensure fairness and improve efficiency in each transmission interval,

respectively. If we apply fairness algorithm in all transmission intervals (L intervals), the received video

qualities for all users will be similar to each other. However, the users whose video programs require

more rates to achieve the same video quality or who are in bad channel conditions will become a

bottleneck in the whole system and degrade the overall video qualities. If we apply efficiency algorithm

in all transmission intervals, the system will achieve the highest overall video qualities. Nevertheless,

the users in good channel conditions with low video content complexity will be assigned more system

resources. Consequently, some users will have unnecessarily good video qualities while others will have

very bad qualities. In other words, a system achieving more efficiency will suffer from more unfairness.

We are interested in how to design a system with partial fairness and partial efficiency. To achieve

this tradeoff, for each GOF, we propose to apply fairness algorithm in the first several transmission

intervals (x intervals) to ensure the baseline fairness, and then apply the efficiency algorithm in the rest

transmission intervals (y = L−x intervals) to improve the overall video qualities. We denote this strategy

as Fx Ey algorithm. Note that FL E0 algorithm is the pure fairness algorithm and F0 EL algorithm is the

pure efficiency algorithm.

DRAFT
 22

B. Unequal Error Protection

It has been shown that the unequal error protection (UEP) can improve the expected video qualities

[37]–[39]. Relaxing the requirement from lower targeted BER to higher targeted BER but sending the

same bit rate, the required power can be reduced. In other words, if the overall transmission power is

fixed, the overall bit rate using higher targeted BER can be higher than the one with lower targeted

BER. It is potential to improve the overall expected video qualities. The UEP takes the advantage of

different priorities within a video bit stream by using different targeted BER. For the video bit stream with

higher priority, the UEP adopts stronger error protection (lower targeted BER) to increase the probability

of successful transmission. For the video bit stream with lower priority, the UEP applies weaker error

protection (higher targeted BER) to utilize a larger effective bandwidth for statistical performance gain.

Because the EWV bit stream exhibits a strong decoding dependency, all received coding pass clusters

in a subband should be adjacent to each other and also a truncated version of the original bit stream

starting from MSB. Assuming all bits in the quality layer 0 ∼ l −1 are received correctly, given a targeted

BERi , the expected distortion of the quality layer l can be represented as:
b,l
B
X −1 TX
k −1

Ei [Dkl ] = Dkl−1 − i
pt,b,k ∆dt,b,k . (22)
b=0 t=Tkb,l−1

Here pit,b,k is the probability that the receiver can correctly receive all coding pass clusters from Tkb,l−1

to t in subband b and can be expressed as follows:

pit,b,k = (1 − BERi )∆Rt,b,k . (23)

Here ∆Rt,b,k is the cumulative number of bits from coding pass cluster Tkb,l−1 to t in subband b and can

be expressed as t
X
∆Rt,b,k = ∆rt0 ,b,k . (24)
t0 =Tkb,l−1
Quality layer l has higher priority than quality layer k if l < k since both layers may have coding

passes in the same subband so that coding passes in quality layer k have decoding dependency on the

ones in quality layer l due to (4). We incorporate the unequal error protection strategy in the proposed

framework by considering the priorities of quality layers in different transmission intervals. In the first

DRAFT
 23

L − 1 transmission intervals, we solve the original problem (15) using the proposed algorithm shown

in Table V with the strongest error protection. At the last transmission interval, we solve the problem

(15) but replacing Dkl with Ei [Dkl ] using several different BERi as shown in Table I (BER0 = 10−6 and

BER1 = 10−5 in our case). We pick the BER setting that achieves the lowest overall expected distortion.

VI. S IMULATION R ESULTS

The simulations are set up as follows. The OFDM system has 32 subcarriers over a total 1.6MHz

bandwidth. The delay spread in root mean square is 3 × 10−7 s. An additional 5µs guard interval is used

to avoid inter-symbol interference due to channel delay spread. This results in a total block length as

25µs and a block rate as 40K per second. The Doppler frequency is 10Hz and the transmission interval

is 33.33ms. The mobile is uniformly distributed within the cell with radius of 50m and the minimal

distance from mobile to the base station is 10m. The noise power is 5 × 10−9 Watts, and the maximal

transmission power is 0.1 Watts. The propagation loss factor is 3 [40]. The video sampling rate is 30

frames per second with CIF resolution (352x288). The GOF size is 16 frames and each GOF is encoded

by the codec [16] using Daubechies 9/7 bi-orthogonal filter with 4-level temporal decomposition and

3/2/1 spatial decomposition in low/mid/high temporal subbands, respectively.

A. Performance Evaluation of the Fairness Algorithm

We first demonstrate how the proposed algorithm F16 E0 achieves pure fairness among all users when

all users request uniform quality. We consider a four-user system, where each user receives 10 GOFs from

one of the four video sequences, Foreman, Hall Monitor, Mother and daughter, and Silent, respectively.

Figure 4(a) shows the received video quality of the first GOF in terms of mean squared error in every

transmission interval. As we can see, all four users have similar video quality in each transmission interval

and the received video quality is improved by receiving more quality layers till the last transmission

interval. We also show the corresponding subcarrier assignment of the first GOF in each transmission

interval in Figure 5(a). As the source coding rate of each user is allocated in different time and frequency

slots according to the channel conditions and source characteristics, the diversities of frequency, time,

DRAFT
 24

4−User System 4−User System with Differentiated Service

90 450
User 0 User 0
User 1 User 1
80 User 2 400 User 2
User 3 User 3

70 350

60 300
Mean Squared Error

Mean Squared Error

50 250

40 200
User 3

30 150 User 2

User 1
20 100 User 0

10 50

0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Transmission interval index (unit:33.33ms) Transmission interval index (unit:33.33ms)

(a) F16 E0 system (b) F16 E0 system with differentiated service

Fig. 4. Comparison for the F16 E0 system providing uniform quality and differentiated service.

5 5

User 0 User 0
Subcarrier index

Subcarrier index

10 10

15 User 1 15 User 1

20 20 User 2
User 2
25 25

User 3 User 3
30 30

2 4 6 8 10 12 14 16 2 4 6 8 10 12 14 16

Transmission interval index (unit: 33.33ms) Transmission interval index (unit: 33.33ms)

(a) F16 E0 system (b) F16 E0 system with differentiated service

Fig. 5. Subcarrier assignment for the F16 E0 system in each transmission interval (a) Uniform quality. The system assigns more
subcarriers to user 0 at most intervals due to the required rate of video sequence 0 to achieve the same quality is higher than
other sequences. (b) Differentiated service. The system assigns more subcarriers to user 3 due to the highest requested quality.

and multiuser are jointly exploited. Figure 6(a) shows the frame-by-frame PSNR along 10 GOFs for all

users. The average PSNR along the received 160 frames for each user is 39.52, 39.71, 39.46, 39.54dB,

respectively. The deviation of users’ received quality is small and within 0.25dB. Thus, the pure fairness

algorithm, F16 E0 , can provide similar visual qualities among all users during the whole transmission

time.

DRAFT
 25

Frame by Frame PSNR for 4−user system Frame by Frame PSNR for 4−user system with differentiated service
50 50
User 0 User 0
User 1 User 1
48 User 2 48 User 2
User 3 User 3 User 3
User 2
46 46

44 User 1 44
User 0

42 42
PSNR (dB)

PSNR (dB)
40 40

38 38

36 36
User 3
34 34
User 2
User 1
32 32 User 0

30 30
20 40 60 80 100 120 140 160 20 40 60 80 100 120 140 160
Frame Index Frame Index

(a) F16 E0 system (b) F16 E0 system with differentiated service

Fig. 6. Frame-by-frame PSNR for the F16 E0 system with uniform quality and with differentiated service.

As we have mentioned that the proposed framework can provide differentiated service by appropriately

setting the quality weighting factor {wk } in (10). We repeat the above experiment with a new set {wk }

as w0 = 0.25, w1 = 0.5, w2 = 1, and w3 = 2. The PSNR difference between user i and i + 1 is

expected to be 3dB. Figure 4(b) shows the mean squared error of the first GOF received by each user

in every transmission interval. As we can see, the video qualities received by all users maintain the

desired quality gap in every transmission interval. The differentiated service is achieved when we receive

all quality layers. Figure 5(b) shows the corresponding subcarrier assignment of the first GOF in each

transmission interval. Compared to Figure 5(a), User 3 occupies more subcarriers in the system with

differentiated service than in the system with uniform quality. Figure 6(b) shows the frame-by-frame

PSNR along 10 GOFs for all users. As we have expected, User 3 has the highest received video quality

and User 0 has the lowest PSNR. The average PSNR received by each user is 35.06, 38.16, 40.91,

43.92dB, respectively. The PSNR differences between user i and i + 1 for i = 0, 1, 2 are 3.10, 2.75,

3.01dB, respectively, which is close to the design goal of 3dB differentiated service.

DRAFT
 26

B. Performance Evaluation of the Fx Ey Algorithm Family

Next, we evaluate the proposed algorithm Fx Ey with different values of x. Here we also compare the

proposed algorithm with the TDM algorithm 2 . Instead of allowing subcarriers in a transmission interval

to be allocated among multiple users, the TDM algorithm assigns all subcarriers in one transmission

interval to only one user whose current end-to-end distortion is the largest. Thus, the multiuser and

frequency diversity is not explored in this TDM algorithm.

We concatenate 15 classic CIF video sequences to form one testing video sequence of 4064 frames. The

15 sequences are 288-frame Akiyo, 144-frame Bus, 288-frame Coastguard, 288-frame Container, 240-

frame Flower, 288-frame Foreman, 288-frame Hall Monitor, 288-frame Highway, 288-frame Mobile,

288-frame Mother and daughter, 288-frame MPEG4 news, 288-frame Paris, 288-frame Silent, 256-frame

Tempete, and 256-frame Waterfall. The video for the k th user is 160 frames long and from frame 256×k +1

to frame 256×k +160 of the testing sequence.

Two performance criteria are used to measure the proposed algorithm and TDM algorithm. We first

calculate the average received video quality of all 160 frames for each user and denote it as PSNRk for

the k th user. To measure the efficiency, we average the PSNRk for all users, i.e.
K−1
1 X
avePSNR = PSNRk . (25)
K
k=0

The higher avePSNR is, the higher system efficiency of overall video quality we have. To measure the

fairness, we take the standard deviation for each user’s average received video quality, i.e.
v
u K−1
u1 X
stdPSNR = t (PSNRk − avePSNR)2 . (26)
K
k=0

The lower stdPSNR is, the fairer quality each user receives. If stdPSNR is high, it implies some users

receive video programs with high quality and the other users receive video programs with poor quality.

2
The current IEEE 802.11 medium access control (MAC) protocol supports two kinds of access methods: distributed
coordination function (DCF) and point coordination function (PCF). In both mechanisms, only one user occupies all the bandwidth
at each time, which is similar to TDM technology.

DRAFT
 27

4−user system 8−user system

5 6

F0E16
4.5 TDM
TDM F7E9
5
4 F4E12

3.5 F11E5 F0E16

4 F7E9

3 F13E3 F4E12
std PSNR (dB)

std PSNR (dB)

F14E2
F11E5
2.5 3

F15E1 F13E3
2 F14E2

2
1.5 F15E1

1
Proposed FxEy 1
Proposed FxEy
TDM TDM
0.5
F16E0
F16E0
0 0
30 31 32 33 34 35 36 37 26 27 28 29 30 31 32 33
ave PSNR (dB) ave PSNR (dB)

(a) 4-user system (b) 8-user system

12−user system 16−user system
6 6
Proposed FxEy
TDM TDM
TDM

5 5

4 F0E16 4
F7E9
F0E16
F7E9
std PSNR (dB)

std PSNR (dB)

F4E12

3 F11E5 3 F4E12
F13E3
F13E3
F11E5
F14E2

F15E1 F14E2
2 2
F15E1

1 1
Proposed FxEy
TDM
F16E0 F16E0
0 0
24 25 26 27 28 29 30 31 32 24 25 26 27 28 29 30 31
ave PSNR (dB) ave PSNR (dB)

(c) 12-user system (d) 16-user system

Fig. 7. avePSNR and stdPSNR results of the Fx Ey algorithm family and TDM algorithm.

Figure 7 shows the fairness and efficiency results for the proposed algorithms and TDM algorithm. We

first compare the performances for eight settings of the Fx Ey algorithm family, including F0 E16 , F4 E12 ,

F7 E9 , F11 E5 , F13 E3 , F14 E2 , F15 E1 , and F16 E0 . We see that the pure fairness algorithm F16 E0 achieves the

lowest PSNR deviation among all algorithms but has the lowest average PSNR; and the pure efficiency

algorithm F0 E16 achieves the highest average PSNR but has the highest PSNR deviation. The tradeoff

between avePSNR and stdPSNR can be adjusted by selecting different number of transmission intervals

for fairness algorithm. As revealed from Figure 7, the Fx Ey algorithm has higher average received video

quality but higher quality deviation than the Fx−1 Ey+1 algorithm. The second comparison included in

Figure 7 is between the Fx Ey algorithm family and the TDM algorithm. As shown in Figure 7, for

achieving the same avePSNR, the proposed Fx Ey algorithm family has about 1∼1.8 dB lower deviation

in PSNR than the TDM algorithm. In other words, the proposed algorithm provides fairer quality than

DRAFT
 28

User 0 at a 4−user system User 1 at a 4−user system

50 50
Proposed F0E16
Proposed F14E2
TDM F14E2 Proposed F16E0
F0E16 TDM
45 45

40 40
PSNR (dB) F16E0

PSNR (dB)
35 35

30 F14E2 30

Proposed F0E16
25 Proposed F14E2 25
TDM
Proposed F16E0 F16E0
TDM
F0E16

20 20
20 40 60 80 100 120 140 160 20 40 60 80 100 120 140 160
Frame index Frame index

(a) User 0 (b) User 1

User 2 at a 4−user system User 3 at a 4−user system
50 50
Proposed F0E16 Proposed F0E16
Proposed F14E2 Proposed F14E2
Proposed F16E0 Proposed F16E0
TDM TDM
45 45

F0E16

TDM
F0E16
40 40
PSNR (dB)

PSNR (dB)
35 F16E0 F14E2 35

30 30

25 25

TDM F14E2
F16E0

20 20
20 40 60 80 100 120 140 160 20 40 60 80 100 120 140 160
Frame index Frame index

(c) User 2 (d) User 3

Fig. 8. Frame-by-frame PSNR for different algorithms of a 4-user system.

the TDM algorithm. This is because the proposed scheme employs additional diversity in frequency and

multiuser.

To evaluate the received video quality along the time axis, we show the frame-by-frame PSNR using

TDM algorithm and the Fx Ey algorithm family for each user in a four-user system in Figure 8. We

choose three algorithms from Fx Ey algorithm family, namely, the pure efficiency algorithm (F0 E16 ), the

pure fairness algorithm (F16 E0 ), and one example of the partial fairness-efficiency algorithm (F14 E2 ). As

we can see, the F0 E16 algorithm has higher PSNR than F14 E2 , F16 E0 for all users except User 1. This

is due to two factors: one is that the video content of User 1 requires higher rate to achieve the same

video quality than the other users and the other reason is that the channel condition for User 1 is the

worst among all users. Therefore, the F0 E16 algorithm assigns more rates to the other users than User 1

to achieve higher average received video quality of all users.

DRAFT
 29

Numer of users v.s. minimal quality received among all users

32
Proposed F16E0
TDM

30

28

PSNR (dB)
26

24

22

20
4 6 8 10 12 14 16
Number of users

Fig. 9. Performance comparison for the worst quality received among all users using the proposed algorithm and TDM
algorithm.

Figure 9 shows the average value of the worst received PSNR among all users from 10 different

terminals’ locations for different number of users in the system. We can see that the proposed algorithm,

F16 E0 , can improve the minimal PSNR better than that of the TDM algorithm. There is about 0.5∼3dB

gain for different number of users. The performance gap increases when the number of users increases

owing to the multiuser diversity.

In Table VI, we show the performance gain that the unequal error protection scheme outperforms the

equal error protection (EEP) for different numbers of users in the system using F0 E16 algorithm. For the

UEP strategy, the targeted BER of the first 15 transmission intervals is set to 10−6 . The targeted BER of

the last transmission interval is chosen from {10−5 , 10−6 }, depending on which BER setting achieving

better expected distortion using (22). For the EEP, the targeted BER for all transmission intervals is 10−6 .

The video content and channel setting are the same as before. For each setting, we run the simulation

10,000 times. As revealed in Table VI, the UEP can improve the expected average PSNR per user only

about 0.05∼0.13dB.

This small improvement using the UEP is due to several reasons. First, although a system with higher

BER has higher bit rate throughput, the distortion introduced by the channel becomes significant. Thus,

the increased effective bandwidth is limited, which limits the reduction of video distortion. Second, the

EWV codec has highly compression ratio at very low bit rate but its R-D curve becomes flatter at

DRAFT
 30

TABLE VI

P ERFORMANCE GAIN OF USING UNEQUAL ERROR PROTECTION VERSUS USING EQUAL ERROR PROTECTION

Number of users 4 8 12 16
average PSNR gain per user (dB) 0.13 0.09 0.07 0.05

higher bit rate due to the distortion-to-length slope sorting. So, for a system that has already received a

large amount of video data at the last transmission interval, the improved distortion due to the increased

effective bandwidth using the UEP is limited. Third, the joint multiple video coding has explored the

video content complexities for all videos and the system has assigned more system resources to users

who are in good channel conditions with simple video content complexity to achieve the highest system’s

efficiency. Thus, the extra bit rate budget benefited from the UEP will be distributed to users who are

in bad channel conditions with complex video complexity, which can only improve a limited amount of

overall distortion. Further, the selection of targeted BER is based on the expected distortion calculated

from (22). If the targeted BER is selected as 10−6 , the UEP is equivalent to the EEP and no performance

gain can be obtained. We also observe that the more users the system has, the smaller performance

improvement we have. It is because the increased bandwidth due to higher targeted BER is roughly a

constant and is shared by all users. When the number of user increases, the increased bandwidth assigned

to each user will reduce and the quality improvement will reduce.

VII. C ONCLUSION

In this paper, we have constructed a framework sending multiple scalable video programs over multiuser

OFDM networks. By leveraging the frequency, time, and multiuser diversity of the OFDM system and

the scalability of the 3-D embedded wavelet video codec, the proposed framework can allocate system

resources to each video stream to achieve the desired video quality. Two service objectives are addressed:

fairness and efficiency. For the fairness problem, we formulate the system to achieve fair quality among

all users as a min-max problem. For the efficiency problem, we formulate the system to attain the

lowest overall video distortion as a minimization problem. To satisfy the real-time requirement, two fast

algorithms are proposed to solve the above two problems.

DRAFT
 31

The simulation results have demonstrated that the proposed fairness algorithm outperforms TDM

algorithm by about 0.5∼3dB for the worst received video quality criterion. The proposed Fx Ey algorithm

family can achieve a desired tradeoff between fairness among users and overall system efficiency. At the

same average video quality among all users, the proposed algorithm has about 1∼1.8dB lower PSNR

deviation among all users than the TDM algorithm. So, the proposed algorithm can provide better system

efficiency and stricter fairness. In addition, the proposed fairness algorithm can allow differentiated service

by appropriately setting values for quality weighting factors. We also extend the proposed framework to

incorporate unequal error protection. In summary, the proposed framework is a promising solution for

broadband multiuser video communication.

R EFERENCES

[1] H. Rohling and R. Grunheid, “Performance comparison of different multiple access schemes for the downlink of an OFDM
communication system,” Proc. IEEE Vehicular Technology Conference, pp.1365-1369, 1997.
[2] M. Wahlqvist, H. Olofsson, M. Ericson, C. Ostberg, and R. Larsson, “Capacity comparison of an OFDM based multiple
access system using different dynamic resource allocation,” Proc. IEEE Vehicular Technology Conference, vol.3, pp.1664-
1668, May 1997.
[3] R. S. Cheng and S. Verdu, “Gaussian multiaccess channels with ISI: capacity region and multiuser water-filling,” IEEE
Transactions on Information Theory, vol.39, no.3, pp.773-785, May 1993.
[4] D. N. C. Tse and S. V. Hanly, “Multiaccess fading channels. I. Polymatroid structure, optimal resource allocation and
throughput capacities,” IEEE Transactions on Information Theory, vol.44, no.7, pp.2796-2815, Nov. 1998.
[5] S. Ulukus and R. D. Yates, “Iterative construction of optimum signature sequence sets in synchronous CDMA systems,”
IEEE Transactions on Information Theory, vol.47, no.5, pp.1989-1998, Jul. 2001.
[6] C. Y. Wong, R. S. Cheng, K. B. Lataief, and R. D. Murch, “Multiuser OFDM with adaptive subcarrier, bit, and power
allocation,” IEEE Journal on Selected Areas in Communications, vol.17, no.10, pp.1747-1758, Oct. 1999.
[7] D. Kivanc, G. Li, and H. Liu, “Computationally efficient bandwidth allocation and power control for OFDMA,” IEEE
Transactions on Wireless Communications, vol.2, no.6, pp.1150-1158, Nov. 2003.
[8] P. S. Chow, J. M. Cioffi, and J. A. C. Bingham, “A practical discrete multitone transceiver loading algorithm for data
transmission over spectrally shaped channels,” IEEE Transactions on Communications, vol.43, no.2/3/4, pp.773-775,
Feb./Mar./Apr. 1995.
[9] Z. Han, Z. Ji, and K. J. R. Liu, “Low-complexity OFDMA channel allocation with Nash bargaining solution fairness,”
IEEE Globel Telecommunication Conf., 2004.
[10] Z. Han, Z. Ji, and K. J. Ray Liu, “Power minimization for multi-cell OFDM networks using distributed non-cooperative
game approach”, IEEE Globel Telecommunication Conf., 2004.
[11] D. Taubman and A. Zakhor, “Multirate 3-D subband coding of video,” IEEE Transactions on Image Processing, vol.3,
no.5, pp.572-588, Sep. 1994.

DRAFT
 32

[12] J.-R. Ohm, “Three-dimensional subband coding with motion compensation,” IEEE Transactions on Image Processing,
vol.3, no.5, pp.559-571, Sep. 1994.
[13] C. I. Podilchuk, N. S. Jayant, and N. Farvardin, “Three-dimensional subband coding of video,” IEEE Transactions on
Image Processing, vol.4, no.2, pp.125-139, Feb. 1995.
[14] S.-J. Choi and J. W. Woods, “Motion-compensated 3-D subband coding of video,” IEEE Transactions on Image Processing,
vol.8, no.2, pp.155-167, Feb. 1999.
[15] B.-J. Kim, Z. Xiong, and W. A. Pearlman, “Low bit-rate scalable video coding with 3-D set partitioning in hierarchical
trees (3-D SPIHT),” IEEE Transactions on Circuits and Systems for Video Technology, vol.10, no.8, pp.1374-1387, Dec.
2000.
[16] J. Hua, Z. Xiong, and X. Wu, “High-performance 3-D embedded wavelet video (EWV) coding,” IEEE Fourth Workshop
on Multimedai Signal Processing, 2001.
[17] D. S. Taubman, “High performance scalable image compression with EBCOT,” IEEE Transactions on Image Processing,
vol.9, no.7, pp.1158-1170, Jul. 2000.
[18] P. J. Cherriman, T. Keller, and L. Hanzo, “Orthogonal frequency-division multiplex transmission of H.263 encoded video
over highly frequency-selective wireless networks,” IEEE Transactions on Circuits and Systems for Video Technology, vol.9,
no.5, pp.701-712, Aug. 1999.
[19] B. Girod and N. Farber, Wireless video, in Compressed video over networks, M-T. Sun and A.R. Reibman, Eds. New
York, Marcel Dekker, Inc, 2001.
[20] R. Yan, F. Wu, S. Li, R. Tao, and Y. Zhang, “Error resilience methods for FGS enhancement bitstream,” The First IEEE
Pacific-Rim Conference on Multimedia (PCM), pp.148-151, Dec. 2000.
[21] J. Xu, Q. Zhang, W. Zhu, X.-G. Xia, and Y.-Q. Zhang, “Optimal joint source-channel bit allocation for MPEG-4 fine
granularity scalable video over OFDM system,” IEEE International Symposium on Circuits and Systems vol.2, May 2003.
[22] H. Zheng and K. J. R. Liu, “The subband modulation: a joint power and rate allocation framework for subband image
and video transmission,” IEEE Transactions on Circuits and Systems for Video Technology, vol 9, no 5, pp.823-838, Aug.
1999.
[23] J. Song and K. J. R. Liu, “An integrated source and channel rate allocation scheme for robust video coding and transmission
over wireless channels,” EURASIP Journal on Applied Signal Processing, vol.2004, no.2, pp.304-316, Feb. 2004.
[24] H. Zheng and K. J. R. Liu, “Robust image and video transmission over spectrally shaped channels using multicarrier
modulation,” IEEE Transactions on Multimedia, vol.1, no.1, pp.88-103, Mar. 1999.
[25] J. Song and K. J. R. Liu, “Robust progressive image transmission over OFDM systems using space-time block code,”
IEEE Transactions on Multimedia, vol.4, no.3, pp.394-406, Sep. 2002.
[26] Z. Ji, Q. Zhang, W. Zhu, J. Lu, and Y.-Q. Zhang, “Video broadcasting over MIMO-OFDM systems,” IEEE International
Symposium on Circuits and Systems, vol.2, pp.844-847, May 2003.
[27] H. Cai, B. Zeng, G. Shen, Z. Xiong, and S. Li, “Error-resilient unequal error protection of fine granularity scalable video
bitstreams,” EURASIP Journal on Applied Signal Processing, vol.2006, Article ID 45412, 2006.
[28] L. Wang and A. Vincent, “Bit allocation and constraints for joint coding of multiple video programs,” IEEE Transactions
on Circuits and Systems for Video Technology, vol.9, no.6, pp.949-959, Sep. 1999.
[29] X. M. Zhang, A. Vetro, Y. Q. Shi, and H. Sun, “Constant quality constrained rate allocation for FGS-coded videos,” IEEE
Transactions on Circuits and Systems for Video Technology, vol.13, no.2, pp. 121-130, Feb. 2003.

DRAFT
 33

[30] G.-M. Su and M. Wu, “Efficient bandwidth resource allocation for low-delay multiuser MPEG-4 video transmission”,
IEEE Trans. on Circuits and Systems for Video Technology, vol. 15, no.9, pp.1124-1137, Sep. 2005.
[31] Z. Han, G.-M. Su, A. Kwasinski, M. Wu, and K. J. R. Liu, “Multiuser distortion management of layered video over
resource limited downlink MC-CDMA”, accepted by IEEE Trans. on Wireless Communication, Jan. 2006.
[32] G.-M. Su, Z. Han, M. Wu, and K. J. R. Liu, “ Dynamic distortion control for 3-D embedded wavelet video over multi-user
OFDM networks,” IEEE Globel Telecommunication Conf., vol.2, pp.650-654, 2004.
[33] J. Gross, J. Klaue, H. Karl, and A. Wolisz, “Subcarrier allocation for variable bit rate video streams in wireless OFDM
systems,” IEEE Vehicular Technology Conference, Fall. 2003
[34] C. E. Koksal and H. Kassab and H. Balakrishnan, “An Analysis of Short-Term Fairness in Wireless Media Access
Protocols,” ACM Sigmetrics 2000, Jun. 2000.
[35] M. H. Ahmed, H. Yanikomeroglu, and S. Mahmoud, “Fairness enhancement of link adaptation techniques in wireless
networks,” IEEE Vehicular Technology Conference, vol.4, pp.1554-1557, Fall 2003.
[36] S. Martello and P. Toth, “Knapsack Problems: Algorithms and Computer Implementations,” Wiley, West Sussex, England,
1990.
[37] A. Albanese, J. Blomer, J. Edmonds, M. Luby, and M. Sudan, “Priority encoding transmission,” IEEE Transactions on
Information Theory, vol.42, no.6, pp.1737-1744, Nov. 1996.
[38] A. E. Mohr, E. A. Riskin, and R. E. Ladner, “Unequal loss protection: graceful degradation of image quality over packet
erasure channels through forward error correction,” IEEE Journal on Selected Areas in Communications, vol.18, no.6,
pp.819828, Jun. 2000.
[39] P. G. Sherwood and K. Zeger, “Error protection for progressive image transmission over memoryless and fading channels,”
IEEE Transactions on Communications, vol.46, no.12, pp.1555 1559, Dec. 1998.
[40] T. S. Rappaport, Wireless Communications: principles and practice: 2nd edition, Prentice Hall, 2002.

Guan-Ming Su (S’04) received the B.S.E. degree from National Taiwan University, Taipei, Taiwan, in 1996
and the M.S. degree from University of Maryland, College Park, in 2001, both in electrical engineering. He
PLACE
is currently working toward the Ph.D. degree in the Department of Electrical and Computer Engineering
PHOTO
at the University of Maryland, College Park.
HERE
He is with ESS Technology, Inc., Fremont, CA. His research interests are multimedia communications
and multimedia signal processing.

DRAFT
 34

Zhu Han (S’01-M’04) received the B.S. degree in electronic engineering from Tsinghua University, in
1997, and the M.S. and Ph.D. degrees in electrical engineering from the University of Maryland, College
PLACE
Park, in 1997 and 2003, respectively.
PHOTO
From 1997 to 2000, he was a Graduate Research Assistant at the University of Maryland. From 2000
HERE
to 2002, he was an Engineer in the R&D Group of ACTERNA, Maryland. From 2003 to 2006, he
was a Research Associate at the University of Maryland. During Summer 2006, he was a visitor in
Princeton University. Currently He is an assistant Professor in Electrical and Computer Engineering Department at Boise State
University, Idaho, USA. His research interests include wireless resource allocation and management, wireless communications
and networking, game theory, and wireless multimedia.
Dr. Han is a member of the Technical Programming Committee for the IEEE International Conference on Communications
of 2004, 2005, and 2007, the IEEE Vehicular Technology Conference, Spring 2004, the IEEE Consumer Communications and
Networking Conference 2005, 2006, and 2007, the IEEE Wireless Communications and Networking Conference 2005, 2006
and 2007, and the IEEE Globe Communication Conference 2005 and 2006, as well as Session Chair of the IEEE Wireless
Communications and Networking Conference 2004, 2005, 2006 and the IEEE Globe Communication Conference 2005.

Min Wu (S’95-M’01) received the B.E. degree in electrical engineering and the B.A. degree in economics
(both with the highest honors) from Tsinghua University, Beijing, China, in 1996, and the Ph.D. degree
PLACE
in electrical engineering from Princeton University in 2001.
PHOTO
Since 2001, she has been on the faculty of the Department of Electrical and Computer Engineering
HERE
and the Institute of Advanced Computer Studies at the University of Maryland, College Park, where she
is currently an Associate Professor. Previously she was with the NEC Research Institute and Panasonic
Laboratories. She co-authored two books, Multimedia Data Hiding (Springer-Verlag, 2003) and Multimedia Fingerprinting
Forensics for Traitor Tracing (EURASIP/Hindawi, 2005), and holds five U.S. patents. Her research interests include information
security and forensics, multimedia signal processing, and multimedia communications.
Dr. Wu is an Associate Editor of the IEEE Signal Processing Letters and a member of the IEEE Technical Committee
on Multimedia Systems and Applications. She received an NSF CAREER award in 2002, a University of Maryland George
Corcoran Education Award in 2003, an MIT Technology Review’s TR100 Young Innovator Award in 2004, and an ONR Young
Investigator Award in 2005. She is a co-recipient of the 2004 EURASIP Best Paper Award and the 2005 IEEE Signal Processing
Society Best Paper Award.

DRAFT
 35

K. J. Ray Liu (F’03) is Professor and Associate Chair, Graduate Studies and Research, of Electrical
and Computer Engineering Department, University of Maryland, College Park. His research contributions
PLACE
encompass broad aspects of wireless communications and networking, information forensics and security,
PHOTO
multimedia communications and signal processing, bioinformatics and biomedical imaging, and signal
HERE
processing algorithms and architectures.
Dr. Liu is the recipient of numerous honors and awards including best paper awards from IEEE Signal
Processing Society (twice), IEEE Vehicular Technology Society, and EURASIP; IEEE Signal Processing Society Distinguished
Lecturer, EURASIP Meritorious Service Award, and National Science Foundation Young Investigator Award. He also received
various teaching and research awards from University of Maryland including Poole and Kent Company Senior Faculty Teaching
Award and Invention of the Year Award.
Dr. Liu is Vice President - Publications and on the Board of Governor of IEEE Signal Processing Society. He was the
Editor-in-Chief of IEEE Signal Processing Magazine and the founding Editor-in-Chief of EURASIP Journal on Applied Signal
Processing.

DRAFT