A Cross-Layer Optimization Framework
for Coded Wireless Video Multicast
Introduction
The current commercial deployments for video multicast are largely based
on managed content delivery or distributed networks [1-3] consisting of media
servers and caches/gateways with a high operational cost. On the other hand, innovative
techniques by P2P or P2P overlay such as CoolStreaming [4], NTUStreaming [5],
MediaGrid [6] and PPLive [7] are evolved to deliver streaming videos in public
IP networks but with less reliability assurance. Regardless of which of these
approaches is to become the predominant architecture, it is envisioned that the
existing real-time video multicast services will be extended to and offered in wireless
environments due to the emergence of broadband wireless access (BWA) technologies
such as IEEE 802.16 [8] and Long Term Evolution (LTE). The demand for real-time
wireless video multicast services will drive the growth of future carrier networks
while pushing bandwidth and accessibility limits.
To enable a reliable and efficient video multicast
service over BWA networks, there are a couple of legacy problems that need to
be carefully addressed and solved. The first one is video quality loss due to
packet loss. In a BWA network that supports video multicast services, the source end first digitalizes
analog video signals via quantization and compression, and then packetizes the compressed
video bitstream for transmission. Received packets at each multicast recipient
will be re-assembled into an orderly bitstream for reconstructing the video
with a potential quality loss (i.e. the end-to-end distortion due to quantization
and transmission loss). Due to the time-varying nature of wireless channels, possibly
user mobility, and scalable coding of video signals, transmission loss may
yield a significant portion of the total end-to-end distortion at a recipient
when the corresponding channel experiences temporary deafness, which calls for
a design of temporal diversity. Although repetitive transmission or re-transmission
may alleviate the transmission loss problem, it is neither efficient nor scalable
in the multicast scenario.
Multi-user diversity is another legacy problem. One of the most common
causes of multi-user diversity is user channel diversity, where each multicast recipient is subject to different channel
conditions due to their mobility, varying distances to the base station (BS) or
access point (AP), and respective channel fading. In addition to channel
diversity, the multi-user diversity could be due to the respective video
quality requirement stipulated in a service level agreement, where some premium
users expect to access more video bitstreams for better video qualities that
are not necessary to the other general receivers. With
multi-user diversity, the transmission rates that can be accommodated by these
receivers vary from one receiver to another. Such multi-user diversity among a
group of wireless receivers makes it a challenge in selection of an effective
transmission strategy at the BS/AP for each multicast transmission of video
bitstreams. Note that using a very conservative transmission policy or any
retransmission/relay approach may satisfy all the potential users at the expense
of much larger resource consumption, which impairs the system economic scale and
increases the network total cost and management overhead.
One conceptual solution proposed in the literature that partially tackled the above multi-user diversity problem is to use superposition coded (SPC) modulation coupled with scalable video coding [9], [10], [11]. For example, a 2-level SPC multicast scheme for delivering IPTV content over WiMAX was introduced in [9], where the video bitstreams of base and enhancement quality layers in each video frame are superimposed together within a single SPC multicast signal, so that most receivers can demodulate and decode at least the basic video quality with high probability, and meanwhile the receivers with good channel conditions can obtain both quality layers of video bitsreams from the same multicast signal. The employment of SPC can achieve the effect of multilevel modulation that exploits the progressive refinement feature of scalable video coding.
To tackle the transmission loss problem, a partial conceptual solution is to apply erasure coding at the transport layer such as Digital Fountain Code [12] or multiple description coding (MDC) at the application layer [13], [15], [16], [17], [18]. For example, the Digital Fountain Code was combined in [14] with a MAC layer multicast policy based on a statistical value threshold. Nonetheless, erasure coding and MDC have been considered so far without reference to SPC multicast.
In this research, we intend to establish a complete optimization framework in which scalable video coding, MDC, and SPC can be jointly designed and optimally integrated to overcome both the multi-user diversity and transmission loss problems so as to minimize the end-to-end distortion (EED) for intended receivers.
SECTION
1: RESEARCH GOALS, METHODOLOGY, AND PLANNING
1.1 OBJECTIVES
The long-term goal of the project is to develop a general optimization framework that can enable efficient and reliable video multicasting services in last-hop wireless networks, such as IEEE 802.16, IEEE 802.11, and 3GPP LTE, as described in the above. The short-term objectives are as follows:
· To establish closed-form formulas for the EED of multicast systems involving scalable coding and SPC with or without MDC, based on which our optimization framework will be formulated.
· To gain deep understanding on how scalable coding, MDC, and SPC should interplay with each other for the purpose of providing efficient and reliable video multicast, i.e., minimizing the EED.
·
To develop algorithms for jointly designing scalable coding, MDC, and
SPC, where a number of key parameters, such as the power allocation to each quality
layer in SPC, and the redundancy added to the video bitstreams through MDC,
will be dynamically and optimally configured, so as to maximize
the received video quality for the intended receivers.
· To develop a software-based logical SPC modulation/ demodulation approach without employing additional hardware or modification of the existing protocols, and further prototype the proposed cross-layer framework via Mad WiFi and Linux 2.6 on IXP425 network processor.
· To promote knowledge and technology transfer through interactions and collaborations with the supporting companies (i.e., EION Wireless Inc. and RIM) along with publications in archival journals and premier conference proceedings, technical seminars, and invention disclosures and patents.
· To train graduate students and postdoctoral fellows in the aspects of coded video multicast over wireless links in terms of the knowledge and trainings on the wireless network operation principles and video source and channel coding techniques, etc., and of the experience in research excellence, new knowledge creation, and technology transfer.
The proposed research and solution approaches are expected to solidly contribute to the research community and telecommunication industry in developing the next-generation video multicast service provisioning infrastructure in BWA networks by offering a complete new and effective design paradigm for video multicasting over wireless links. Our research results will help our research partners and other Canadian companies to gain competitive technological advantage in advanced communications.
1.2 DETAILED RESEARCH PROPOSAL
1.2.1 Background – A System of Coded Wireless Video
Multicast
A system of coded video multicast has been proposed in [19], and has been studied via simulation-based analysis that demonstrated its promise as an effective solution to the problems of user diversity and fast channel fluctuation in a single stage. Since our proposed optimization framework will be based on the system, we begin with a brief review of the system.
1.2.1.1 Initiation
The original video data is encoded by a video codec into a scalable bitstream at the source end, which is split into segments corresponding to multiple quality layers. Before passing the bitstream to the BS for multicast transmissions, each segment is further encoded by a protection code into row(s) of protected units (PUs) with an extended size. PUs of different quality layers are queued in corresponding buffers and modulated strategically by corresponding modulation schemes before being superimposed altogether as a cross-layer coded multicast transmission block.
1.2.1.2 Source
Coding using MDC
Suppose there are exactly L quality layers (i.e., l = 1, 2, …, L) in the scalable
video bitstream of a group of frames (GoFs) encoded by layered video coding, which are indexed in
order of non-increasing importance such that layer l is protected with
an (N, Kl) code,
as shown in Fig. 1. Smaller Kl
brings a better protection for the bitstream data of layer l against loss/error of any PU of N bytes. The PUs of layer l are then packetized in a way that the i-th byte in the PU
will be assigned to the i-th MDC packet for transmission, where i =
1, …, N. All these N MDC packets of the PU are then regarded as
being equally important, where only the number of received MDC packets matters
in determining the reconstructed video quality of the GoF regardless of which
MDC packets are received. Thus, the layered MDC is consistent with the concept
of generic MDC, where the i-th MDC packet
constitutes the i-th description of the GoF, and contains a bitstream of
multiple quality layers in the GoF. A sequence of the i-th
MDC packets for all the GoFs of the video content constitutes the i-th
description for the whole media stream.
1.2.1.3 Superposition Coded Multicast (SPCM)
A 2-level
Superposition Coded Multicast (SPCM) was introduced in [9] for video
multicasting in WiMAX. Instead of using a single modulation scheme at a time,
each multicast signal is generated at
the transmitter by superposing video bitstreams of base layer modulated by BPSK
with those of enhancement layer modulated by a higher-order modulation such as
QPSK. A receiver can either obtain the base video quality of an IPTV channel by
partially decoding the multicast packet for those bitstreams modulated in BPSK
when the channel is not good enough; or obtain the full video quality from all
bitstreams modulated in BPSK and QPSK by successfully decoding the whole
superposition coded multicast packet when the channel is good. Thus, the
adoption of SPCM by a WiMAX BS can fully take advantage of the dependency of
base and enhancement layer data in the task of video multicasting.
1.2.1.4 Interplay of Source and
Channel Coding Techniques
To demonstrate the idea of our cross-layer design with a simple case
study but without loss of generality, we assume that there are 2 quality layers
in the video bitstream of a video source symbol for simplicity. Note that that a
source symbol can be a large block of source symbols that represent a
macro-block in a video frame. The PUs of bitstreams in layer 1 and 2 are
generated by layered MDC using Reed Soloman (RS) codes with K1= 3,
The transmitter repeats the same superposition process for the next set of PUs belonging to the 2nd description in all buffers for the next transmission timeslot (e.g. at time t = 2) and so forth, until all the descriptions of the video source symbol are transmitted. The same process is then continued with the next video source symbol until the end of the video. The transmitter could launch the video bitstreams with a certain scheduling policy under the given transmission resources. Also, the network operators could select the appropriate power and K value of each layer for a particular optimization purpose. An illustration on the proposed cross-layer framework is given in Fig. 4.
By assuming that the bottom layer of SPC can be decoded by most receivers
except for those under very deep fading, different descriptions (or subset) of
MDC packets with embedded redundancy are transmitted across consecutive SPC
multicasting blocks, which not only provides differentiated video qualities to
the receivers with different channel qualities according to their instantaneous
channel conditions, but also tackles the short-term deafness due to the fast
and deep fading. This is considered an effective approach in maintaining the
optimal long-term video quality perceivable by each receiver.
Fig. 4. An
illustration on the proposed framework for coded video multicast that interplays
among scalable video coding, MDC, and SPC.
Section
1.2.2 End-to-end Distortion (EED) Analysis and Algorithm Design
Although the above system was demonstrated by simulation in [19] to be effective in the setting of conducted simulation for wireless video multicast, it is not clear at this point how scalable coding, MDC, and SPC should be jointly designed and optimally integrated in order to provide efficient and reliable video multicast. Previous research endeavors such as [21] have studied optimal power allocation for minimizing the EED in a layered Gaussian broadcast system with scalable coding of Gaussian sources. However, these studies only focused on asymptotic behavior in an idealistic system setting, where a Gaussian source is encoded progressively and then transmitted with the compression rate at each layer perfectly matched with the capacity of the corresponding channel realization; there was no consideration on the impact of channel symbol errors on the EED and MDC was not in the system configuration either. As such, little insight could be gained from these previous research works into how scalable coding, MDC, and SPC should interplay with each other to minimize the EED.
To establish an optimization framework within which scalable coding, MDC, and SPC can be jointly designed and optimally integrated, a key step is to develop formulas for the EED of multicast systems involving scalable coding and SPC with or without MDC, via which the impact of channel error probability on the EED can be evaluated. In our previous work [20], [21], for a point to point tandem source-channel coding system, we derived a closed-form formula for its EED:
|
|
(1) |
where DQ is the conventional quantization distortion of the source code with N codewords, s2 is the variance of the source, SQ is a newly discovered quantity of the source code called the scatter factor, which represents the extent how far codewords of the source code are from the mean of the source, and perr is the average channel symbol error rate. The approximation in Eq. (1) is accurate for relatively large N, which is the case for high rate coding. Under the assumption of random index mapping between source codewords and channel inputs (or channel codewords), the above formula is valid for any source (not necessary a Gaussian source), any source code with single resolution, and any coded or uncoded channel. The measure of EED expressed in Eq. (1) has been shown in [21] to provide appropriate approximations even for any deterministic mapping between source codewords and channel inputs for noisy channels. From Eq. (1), the impact of channel error probability on EED can be clearly seen and evaluated.
However, since Eq. (1) is derived for point-to-point communications, it is not applicable to multicast systems with scalable coding and SPC for at least two reasons. First, multi-resolution source codes are more structured than single resolution source codes, and hence one does not have the same flexibility in selecting codewords for different layers (resolutions) as in the case of single resolution. Second, the impact of error probability on the EED now depends on which layer the transmission errors occur. Hence, in this research, we will first extend Eq. (1) to multicast systems involving scalable coding and SPC with or without MDC. The EED formula with MDC will be different from that without MDC; the former will be a further extension of the latter.
Based on these new EED formulas, it is then possible to investigate in our proposed optimization framework how scalable coding, MDC, and SPC should optimally interplay with each other so as to reduce the impact of packet loss on the end-to-end distortion in the layered broadcast system. In particular, we will further investigate the following three design components:
(1) design of quantization techniques for general multi-resolution source codes subject to inter-layer decoding dependency and channel symbol error for each layer, which serves as an extension to the single layer noisy channel quantizer design studied in [20];
(2) derivation of closed-form expressions to yield optimal configuration(s) of the (N, K) parameters in the design of a suitable MDC for the layered broadcast system ; and
(3) identification of channel rate and modulation selection appropriate in scenarios of heterogeneous channel gains of different layers of video data superimposed in a common channel symbol using SPC and the allocation of power between layers.
The above three design aspects will be formulated into an optimization problem as one minimizing the EED for the average typical receiver among a large group of receivers over a long-term period, with multi-resolution quantization parameters, MDC redundancy configuration, and power allocation across different layers as variables. Furthermore, algorithms will be developed for the optimal joint configuration of all three design aspects to maximize the received video quality quantified using the practical measure of EED.
1.2.3 Logical SPC Modulation and Demodulation
The
advantages in using SPC on video multicast are clear. However, nowadays very
few commercially available wireless systems and industry standards have adopted
the SPC modulation. The absence of SPC modulation in video multicast is likely due
to the requirement of additional system support, in which dedicated hardware
components and circuitry are needed. Besides, the recipients have to go through
signal-interference-cancellation (SIC) to decode the received SPC signals,
which is most likely implemented via a specialized hardware device. By
envisioning the prevalence of bandwidth-intensive video multicasting services via
the emerging BWA networks, it is becoming crucial to define a practical
implementation of SPC video multicast that offers the minimal barrier for
industry acceptance.
This project is committed
to develop a novel logical SPC modulation scheme for sending scalable video
bitstreams, which is essential to prototype our proposed
optimization framework. The proposed scheme is characterized by requiring no
additional hardware on the existing wireless systems and standards in
generating a PHY layer symbol that is
logically equivalent or close to that in a constellation formed by the
conventional hardware-based SPC modulation. On the other hand, the recipients
only rely on a standard demodulator to decode the received SPC modulated signal
without going through SIC. Particularly, the process involves in strategic
mapping of information bits from base and enhancement layers of scalable video
bitstreams that are merged to form a logical SPC signal through dynamic power
allocation and phase shift keying. At the recipients, instead of using specialized
SPC demodulators, our scheme decodes the received signals using a standard
demodulator on some proper modifications in the MAC layer, which can be done in
driver installation at the receiver devices.
1.2.3.1
Logical SPC Modulation
For a conceptual demonstration, it is assumed that an SPC modulated signal contains information bits of two quality layers from a scalable video bitsream. We observed that the SPC video multicast can be simply taken as a multilevel modulation process with a cross-layer mapping by the upper successive refinement relation among the bits in the video bitstreams, where a PHY symbol carries both base layer bits and the corresponding enhancement layer bits. To achieve a specific protection on the base layer bits of the symbol, dynamic phase shift keying and power allocation can be performed for generating the corresponding SPC modulated signal. By considering SPC on BPSK and QPSK modulated signals, each symbol block contains 1 bit from the base layer and 2 bits from the enhancement layer, and the symbol is mapped by a constellation diagram with 8 points, as shown in the lower portion of Fig. 5.
For example, in case a
superposed signal with BPSK and QPSK of power E1 and E2, respectively, is
desired, it can be created by performing vector addition of E1 and E2 in the
constellation, as shown in Fig. 6. By superposing QPSK of 4 points and BPSK of
2 points in the constellation, totally 8 points will be resulted, each with a
specific amplitude and angle. With the set of amplitude and angle for the 8
points, we can instruct the corresponding chipset to generate a modulated
signal of SPC modulation for BPSK and QPSK with power E1 and E2, respectively.
Fig. 6. An illustration of logically summing up two signals with QPSK and
BPSK with different energies.
The mapping of the 3-bit symbol block to the 8-point constellation depends on the knowledge regarding the information bits of the scalable video bitstreams in the application layer. For a symbol referring ‘0’ in the base layer with BPSK and a symbol referring ‘10’ in the enhanced layer with QPSK, a corresponding 3-bit symbol block containing “010” (i.e. “0”+”10”) can be formed and mapped to the symbol ‘0,10’ in the 8-point constellation diagram. The abovementioned cross-layer mapping and SPC modulation mechanism is further illustrated in Fig. 7.
Note that the newly generated 8-point constellation has each pair of
adjacent points without an equal Euler distance to reflect the fact that the
enhancement layer bits are dependent on the bit of the base layer, which is
different from the design of conventional 8QAM constellation where all bits are
taken as equal important. Thus, a set of two-level logical SPC modulations can
be defined through superimposing two standard modulated signals. For example,
QPSK + 16QAM can form a constellation diagram with 64 points, which is
equivalent to a 64QAM modulated signal. Certainly, our logical SPC modulation
framework can be extended to the general case with and without MDC.
1.2.3.2
Implementation Issues
To implement the proposed logical SPC modulation at the transmitter, a new software module is required in the existing MAC layer to obtain the dependency knowledge of the bits of the two quality layers from the scalable video source buffered in the corresponding queues at the transmitter, as shown in Fig. 7. The modified MAC software then interacts with the modulation DSP chipset in the PHY layer through a set of primitives. The primitives act as the passage for the MAC software to designate a specific logical SPC modulation scheme and input parameters, such as the amplitude and phase keying corresponding to each point in the resultant constellation.
In addition to the MAC software module, the modulation DSP chipset has to be equipped with more functions so that some service access points (SAPs) are defined in order to receive and recognize the parameters passed from the upper MAC software. Furthermore, the modulation DPS chipset should be able to generate the logical SPC modulated signals accordingly along with the associated allocated energies. The locations of the points in the constellation diagram should be dynamically determined by the given amplitude and phase keying.
1.2.3.3
Demodulation by Leveraging Existing Receiver Demodulators
The design in the CPE side should be simple, cheap, and easily configurable through plug-and-play. The proposed logical SPC demodulation allows the decoding of the base layer information directly using a standard modulation scheme no matter the channel is good or bad. In the example of 8-point constellation, when the channel condition is poor, the first bit in the symbol block (that carries the base layer information) will have a very high likelihood to be decoded as “0” if the recipient can simply tell the received symbol is on the left hand side of constellation diagram. When the channel condition is good, instead of subtracting the base layer bit for decoding the enhancement layer symbol using SIC as the conventional SPC demodulation, the recipient simply decodes the three bits by using a standard 8QAM demodulator.
Since no signal subtraction at the hardware level is required, the received logical SPC modulated signals can be decoded using a standard modulation scheme already implemented in commercially available hardware chipsets. Another advantage is that any receiver device that is not equipped with the proposed logical demodulation software can still decode the logical SPC multicast signals.
1.2.3.4 Proposed Research on Logical SPC Modulation
Based on the proposed logical SPC modulation/demodulation scheme, the following three tasks will be tackled in the project:
(1) We will develop a software-based controller, which serves as a virtual MAC, in order to enable to the proposed cross-layer mapping between scalable video bitstreams at the source and multilevel modulated symbols at the channel. Our goal is to minimize the changes that have to be made to the standard MAC in commercial WiMAX or 3GPP LTE systems.
(2) An analytical model on symbol error rate (SER) and bit error rate (BER) will be developed for each possible SPC modulation and demodulation scheme, and the corresponding end to end distortion analysis will be carried out accordingly within our proposed optimization framework. We expect that the proposed logical SPC will yield similar performance when compared with the conventional schemes.
(3) Based on the optimized parameters we obtain from the above EED analysis for scalable coding, MDC, and SPC, we will construct a prototype for the proposed logical SPC modulation scheme using Mad-WiFi and IEEE IXP425 network processor and for the whole optimized multicast system. Mad-WiFi is one of a few IEEE 802.11 drivers coming with publicly available open source code.
With scalable video bitstreams fed in the application layer, we will first implement the cross-layer mapping by reporting the Mad-WiFi driver in the Linux 2.6 environment, in order to examine the obtained parameters for achieving the proposed logical SPC modulation framework. Next, MDC and SPC will be launched together to verify the performance of the whole optimized video multicast system. The completion of the prototyping task is considered an important milestone for our industry partners to proceed with further involvement and research momentum beyond the project.
REFERENCES
[1] J.
Ni and D. Tsang, “Large-scale cooperative caching and application-level
multicast in multimedia content delivery networks”, IEEE Communications Magazine, vol., no. 5, pp. 98–105, May 2005.
[2] L.
Kontothanassis, R. Sitaramanz, J. Weinz, D. Hongz, R. Kleinbergzx, B. Mancusoz,
D. Shawz and D. Stodolsky, “A transport layer for live streaming in a content
delivery network”, Proc. of the IEEE,
vol. 92, no. 9, pp. 1408 – 1419, Sep.
2004.
[3] E.
Kusmierek, Y. Dong and D. H. C. Du, “Loopback:
exploiting collaborative caches for large-scale streaming“, IEEE Transactions on Multimedia, vol. 8,
issue 2, pp. 233 – 242, Apr. 2006.
[4] S.
Xie, B. Li, G. Y. Keung, and X. Zhang, “Coolstreaming: Design, Theory,
and Practice”, IEEE Transaction of
Multimedia, vol. 9, No. 8, pp. 1661-1671, Dec. 2007.
[5] M.-T.
Lu, J.-C. Wu, K.-J. Peng, P.
Huang, J. J. Yao, and H. H. Chen, “Design and Evaluation of a P2P IPTV System
for Heterogeneous Networks”, IEEE
Transaction of Multimedia, vol. 9, no. 8, pp. 1568-1579, Dec. 2007.
[6] Y. Huang,
Y.-F. Chen, R.
Jana, H. Jiang, M. Rabinovich, A. Reibman, B. Wei and Zhen Xiao, “Capacity analysis of
MediaGrid: a P2P IPTV platform for fiber to the node (FTTN) networks”, IEEE
JSAC, vol. 25, no. 1, pp. 131 – 139, Jan. 2007.
[7] [Online].
Available: http://www.pplive.com/PPLive
[8] IEEE
Standard 802.16-2004, “IEEE Standard for Local and Metropolitan Area
Networks-Par 16:Air Interface for Fixed Broadband
Wireless Access Systems,” 2004.
[9] J.
She, F. Hou, P.-H. Ho and L.-L. Xie, “IPTV over WiMAX: Key Success Factors, Challenges, and Solutions”,
IEEE Communications Magazine, vol.
45, no 8, pp.87-93, Aug. 2007.
[10] Y. S. Chan, J. W. Modestino,Q. Qu and X. Fan, “An End-to-End Embedded Approach for
Multicast/Broadcast of Scalable Video
over Multiuser CDMA Wireless Networks”, IEEE
Transactions on Multimedia, vol. 9, no. 3, pp. 655-667, Apr. 2007.
[11] T. M. Cover, “Comments on broadcast
channels”, IEEE Trans. Information Theory, vol.44, pp.2524-2530, October
1998.
[12] J. Byers et al., “A digital fountain approach to asynchronous reliable
multicast,” IEEE Journal on Selected
Areas in Communications, vol. 20, no. 8, pp. 1528-1540, Oct. 2002.
[13] V.
K. Goyal, “Multiple description coding: Compression meets the network”, IEEE
Signal Processing Magazine, pp. 74–93, Sep 2001.
[14] W.
Ge, J. Zhang and S. Shen, “A Cross-Layer Design Approach to Multicast in
Wireless Networks", IEEE Trans. on Wireless Communications, vol.
6, no. 3, pp. 1063-1071, 2007.
[15] P.
A. Chou, H. J. Wang, and V. N.
Padmanabhan, “Layered multiple description coding,” Proc. PV 13th
Int. Packet Video Workshop Nantes, Apr. 2003.
[16] A. El Gamal and T. M. Cover,
“Achievable rates for multiple descriptions”, IEEE Trans. Information Theory,
vol.28, pp.851-857, November 1982.
[17] L. Ozarow, “On a source-coding problem
with two channels and three receives”, Bell Syst. Tech. J., vol. 59, pp.
1909-1921, 1980.
[18] R. Venkataramani, G.
Kramer, and V. K. Goyal, “Multiple description coding with many channels”, IEEE
Trans. Information Theory, vol.49, pp.2106-2114, September 2002.
[19] J. She, X. Yu, P. –H. Ho, and E. –H.
Yang, “A
Cross-Layer Design Framework for Robust IPTV Services over IEEE 802.16 Networks”, IEEE Journal of Selected Areas on
Communications (JSAC), Vol. 27, No. 2, Feb. 2009, pp. 235 – 245.
[20] X. Yu, H. Wang, and E.-H. Yang, "Optimal quantization
for noisy channels with random index assignment", Proc. of the 2008
IEEE Intern. Symp. Inform. Theory,
Toronto, Canada, July 6-11, 2008.
[21]
C. T. K. Ng, D. Gunduz, A. J. Goldsmith, and E.
Erkip, “Distortion Minimization in Gaussian Layered Broadcast Coding with
Successive Refinement,” IEEE Trans.
Inform. Theory, vol. 55, no. 11, pp. 5074-5086, Nov. 2009.