Supporting Wireless Video Growth and Trends

Supporting Wireless Video Growth and Trends
TABLE OF CONTENTS
Executive Summary ...................................................................................................................................... 4
1. Introduction .............................................................................................................................................. 4
1.1 Overview of whitepaper ....................................................................................................................... 4
1.2 Market trends in video delivery/market analysis ................................................................................. 5
1.2.1 Usecases for video content delivery ............................................................................................. 6
1.2.2 Factors contributing towards growing demand for mobile video .................................................. 7
1.2.3 Trends for streamed video consumption ...................................................................................... 8
1.2.4 Ecosystem support for the growth in mobile video ..................................................................... 11
2. Types of video ........................................................................................................................................ 12
2.1 Elements of video content and summary of video delivery technologies .......................................... 12
2.2 Short form content, Long form content .............................................................................................. 13
2.3 Operator video and OTT video .......................................................................................................... 13
2.4 Best Effort video and Guaranteed Bit Rate video ............................................................................. 14
2.5 Video for entertainment and video for real time communications ..................................................... 14
2.6 HTTP progressive download ............................................................................................................. 14
2.7 RTP/RTSP delivered video ............................................................................................................... 15
2.8 HTTP Adaptive Streaming (HAS) ...................................................................................................... 16
2.8.1 Comparison between progressive download, streaming and adaptive streaming ..................... 18
2.9 Platforms for video delivery ............................................................................................................... 18
3. Challenges for mobile video delivery ..................................................................................................... 19
3.1 Challenges for video delivery due to air interface ............................................................................. 19
3.1.1 Spectral efficiency of 3G/4G cellular technologies and impact on mobile video ........................ 20
3.1.2 Air interface latency and impact on mobile video ....................................................................... 23
3.1.3 Interference and handover issues in 3G/4G cellular technologies ............................................. 24
3.1.4 Scheduling mechanisms at the air interface ............................................................................... 28
3.1.5 Signaling challenges for video .................................................................................................... 30
3.3 Challenges from the network ............................................................................................................. 32
1
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
3.3.1 Backhaul congestion ................................................................................................................... 32
3.3.2 Data plane/control plane encryption ........................................................................................... 34
3.4 Content based challenges ................................................................................................................. 34
3.4.1 Digital Rights Management (DRM) ............................................................................................. 34
3.4.2 Entitlement/Content Management Rights ................................................................................... 34
4. Video Quality of Experience ................................................................................................................... 35
4.1 Video usage behavior and expectation of experience ...................................................................... 35
4.2 Metrics for characterizing video QoE ................................................................................................ 35
4.3 Behavior of video chunks with TCP/IP .............................................................................................. 37
4.4 Monitoring of video QoE .................................................................................................................... 40
4.5 Specific attributes that contribute to improved video QoE ................................................................ 42
4.6 New/Upcoming codecs (HEVC) ........................................................................................................ 42
5. Architectures for mobile video delivery .................................................................................................. 43
5.1 Cellular (HET-NET) and Wi-Fi offload architectures ......................................................................... 43
5.1.1 Video users in a heterogeneous network ................................................................................... 45
5.2 LTE broadcast architecture ............................................................................................................... 46
5.3 Video delivery architecture ................................................................................................................ 49
5.3.1 Mobile Operators and their migration to IMS ................................................................................. 50
5.4 Content management in cloud environment ..................................................................................... 52
5.4.1 Content management architecture and delivery ......................................................................... 52
5.5 Policy management architecture ....................................................................................................... 53
6. Survey of optimization techniques/mitigation strategies ........................................................................ 53
6.1 Objectives of wireless video optimizations ........................................................................................ 53
6.2 End-to-End QoS support in LTE for video delivery ........................................................................... 54
6.2.1 LTE QCI and other network QoS techniques ............................................................................. 54
6.2.2 LTE guaranteed bit rate bearers ................................................................................................. 56
6.2.3 Maintaining constant video quality with GBR service ................................................................. 57
6.2.4 Differentiation and fairness with GBR service ............................................................................ 57
2
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
6.2.5 Limitations of GBR bearer .......................................................................................................... 58
6.2.6 Microbursting .............................................................................................................................. 58
6.3 RF specific optimizations for video delivery ...................................................................................... 58
6.3.1 Guaranteeing minimum video quality on the air interface .......................................................... 59
6.3.2 Broadcast/multicast of video ....................................................................................................... 59
6.4 Non-RF specific optimizations for video delivery .............................................................................. 60
6.4.1 DASH (HAS) ............................................................................................................................... 60
6.4.2 BIS Scalable Video Codecs (SVC) ............................................................................................. 60
6.4.3 Transcoding/Transrating/Rate capping ...................................................................................... 61
6.4.4 Congestion-aware optimization .................................................................................................. 62
6.4.5 Client optimizations for HAS ....................................................................................................... 62
6.4.6 Content caching in the network .................................................................................................. 63
6.4.7 Time shifting or sideloading supported by content caching in the UE ........................................ 64
6.4.8 Video usage behavior ................................................................................................................. 64
7. Recommendations ................................................................................................................................. 65
7.1 Recommendations for video application developers for efficient radio resource management ........ 65
7.2 Recommendations on bit rates for various codecs and screen sizes/device types .......................... 66
8. Summary and Conclusions .................................................................................................................... 67
List of Acronyms .......................................................................................................................................... 69
List of References ....................................................................................................................................... 70
List of Figures .............................................................................................................................................. 71
List of Tables ............................................................................................................................................... 72
Acknowledgements ..................................................................................................................................... 72
3
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
EXECUTIVE SUMMARY
Video is increasingly becoming one of the most pervasive technologies in terms of everyday usage, both
for entertainment and in the enterprise environments. Mobile video is responsible for a majority of the
growth seen in mobile broadband data volume. This white paper presents the expected growth of mobile
video based on current trends and user behaviors, describes the various types of video based on content
type or delivery strategy, and details the various challenges and solutions to deliver mobile video over a
wireless end-to-end network. A multipronged approach to address the various challenges, unique to
mobile video, is presented through optimization techniques in the radio and core networks (HTTP
Adaptive Streaming, etc.), support of multiple network architectures (HET-NETs, LTE broadcast, content
delivery networks), support of client optimizations and through enhanced codecs (HEVC). Metrics for
quantifying and attributes that contribute to improved video QoE are described in detail.
Recommendations on bit rates for various codecs and screen sizes, and recommendations for video
application developers to support the scalable growth of video in mobile networks are made after
examining unique aspects specific to mobile video delivery.
1. INTRODUCTION
1.1 OVERVIEW OF WHITEPAPER
New and emerging classes of mobile devices (smartphones, PC data cards, USB modems, consumer
devices with embedded wireless, machine-to-machine, etc.) are fostering the explosive growth of wireless
data usage and higher-bandwidth data applications and services by public and enterprise users. Users
would like to see various kinds of video content delivered to and uploaded from their mobile devices with
good quality of experience. With the growth of video traffic on the Internet and end users’ expectation of
having content delivered with freedom of movement, the reliable delivery of video over wireless networks
to mobile end user devices will become essential in the coming years.
Operators are beginning to roll out heterogeneous networks that include base station and core network
elements from various 2G, 3G and 4G technologies. Such networks, with macrocells, picocells and
femtocells and Wi-Fi offload capabilities help best utilize the limited air interface resources while serving
the users with the best Quality of Experience (QoE). By bringing the base station closer to the user, small
cells provide valuable radio resources to high bandwidth video application users that are generally low
mobility, thus resulting in overall spectral efficiency gains. However, even with the introduction of picocells
and femtocells, air interface resource limits will be reached at some point due to increasing demands.
Delivery of wireless video poses particular challenges for the network. A video stream requires a
sustained high bandwidth session lasting several minutes. The limited spectrum availability can result in
air interface congestion restricting bandwidth to only a few active users in a cell. The congestion problem
is compounded by cell handovers because of user mobility and interference from neighboring cell
users. These handover and interference issues affect packet throughput, loss and jitter, all of which
adversely affect video delivery. Throughput to individual users varies with time and location, based upon
cell load and proximity to the cell. Throughput at the cell edge can be significantly lower than near the cell
center. The broad range of emerging devices requires flexibility in video delivery to address variations in
screen sizes, screen resolution of end user devices, and impact of video usage on battery life. It is
therefore critical to minimize wireless congestion due to video traffic, by tailoring the video application to
the underlying network resources and device types, while still maximizing end user experience and
operating efficiency.
4
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
This paper addresses the various aspects of mobile video delivery from operator, vendor and device
perspectives. Market trends in video delivery, types of video content, challenges for mobile video delivery
and various optimization techniques to mitigate those challenges in the radio and core network are
described in detail. Architectures for mobile video delivery including the policy management architecture
and content management are discussed in detail. Metrics for measuring video quality are discussed to
provide insights into the performance of video in current networks. This paper concludes with
recommendations for techniques to deploy technology to support video and for application developers.
The whitepaper is structured as follows:

Section 1 describes industry trends and drivers that lead to increased use of video over wireless
networks. Summary from various industry reports are provided.

Section 2 addresses different types of video content and use cases which must be examined to
help define techniques that ensure robust delivery to the consumer.

Section 3 focuses the discussion on challenges of video delivery over a wireless access.

Section 4 addresses video quality of experience including metrics and mechanics of
measurements, codec technology and industry direction.

Section 5 gets into various architectures to support video delivery including HET-NETs, LTE
broadcast, and content management and delivery networks.

Section 6 describes some techniques used in wireless and supporting networks to mitigate the
challenge of delivering low latency, high volume video to consumers.

Section 7 provides recommendations to the reader, including application developers and device
manufacturers to help maximize video quality of experience.

Section 8 summarizes some of the main findings and suggested way forward in the industry to
enhance video QoE over wireless networks.
1.2 MARKET TRENDS IN VIDEO DELIVERY/MARKET ANALYSIS
The world of culture including TV, film, video and music is changing. This technology-induced change not
only improves people’s viewing experience, but it also changes the ways in which consumers
conceptualize culture and entertainment. Devices in the wireless market have evolved in the past two
years in video capabilities (resolution, rendering abilities), computing power (dual core, quad core) and
user interface. These developments coupled with the deployment of advanced wireless networks have
led to an increasing adoption of video on wireless devices.
Users like to see various kinds of video content delivered to and uploaded from their mobile devices.
Users would also like to participate in bidirectional conversational videos such as mobile-to-mobile video
calls, video teleconferences on their mobile devices, etc. New video applications are being spawned at a
faster rate that will further cause growth of mobile video. Mobile video traffic is already over 50 percent of
mobile data traffic and will account for 66 percent of global network demand by 2017 [1]. The growth in
video traffic could soon overwhelm the wireless network resources due to finite and limited spectrum
availability.
5
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
1.2.1 USECASES FOR VIDEO C ONTENT DELIVERY
Video content is delivered for various purposes, as noted below, and will be referenced further in
discussing trends in mobile video delivery.
1) Content may be downloaded on the device prior to viewing on the device. Since the viewing of
content is separate from the download of content, delays in download of video content may be
acceptable to within a limit.
2) Streamed video is driven by availability of content and degree of digital rights management
(DRM) in the content. With content being consumed immediately, there are tighter bounds on the
delay in delivering the video content.
3) Uplink video delivery can result in sporadic high volume of delivery, in addition to significant
downlink video, particularly at sporting events and stadiums when content is shared towards a
server.
4) Applications like Facetime are generating traffic with increasing popularity due to device
capabilities driving video volume.
The Ericsson Consumer Lab TV and Video Report 2012 [2] shows that smartphone and tablet users
spend several hours per week watching video. These will be delivered over cellular and Wi-Fi networks.
Several respondents browse the Internet while watching video and TV. The report notes that the use of
mobile video is increasing due to increased numbers of social forums and chats. A significant percentage
are actually discussing the things they are watching. This kind of multitasking behavior is increasing
because of the availability of different easy-to-use mobile devices such as smartphones, laptops and
tablets. This trend is expected to grow in the coming years.
Video represents the largest data traffic volume today and continues to increase. As an example, a user
on a data plan of around 15 GB consumed on average 2.3 GB per month of online video in Q1 2012,
while in Q4 2012 this figure was 3.5 GB per month [3]. According to [3], video traffic has yet to reach its
peak. One reason is that improved video quality and increased screen sizes will facilitate looking at longer
clips or watching videos for a longer period of time. The increasing availability of content will also result in
increased video usage. Generational shifts in viewing habits also contribute to the increased video usage.
Mobile devices, such as the smartphone, are becoming a natural part of the video and TV experience and
increasingly important. Many consumers state that their various mobile devices have actually begun
replacing their secondary TV screens around the home. While the main TV is still the preferred screen for
TV and video consumption, mobile devices are being used more and more, partly due to applications that
allow time shifting, e.g. Netflix and Hulu.
6
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 1: Number of hours per week spent watching TV or Video on various devices [4]
1.2.2 FACTORS CONTRIBUTING TOWARDS GROWING DEMAND FOR MOBILE VIDEO
Factors that have contributed to the growing demand for mobile video include the following:
1) With the growing capacity of the wireless network, through the introduction of HSPA+ and LTE, it
is possible to select and immediately view the content on demand. This is particularly popular for
short form content such as YouTube clips.
2) Social media makes it easy to share links that result in further increasing the demand. Mobile
devices are even becoming a partial replacement for linear TV in homes with catch up TV
services such as Hulu and iPlayer as well as TVEverywhere. Some recent services even link the
viewing of linear TV on a big screen with apps on tablets to get more information or to enable
interaction with the show.
3) Video streaming over the network is growing in popularity. Many mobile tablets and smartphones
have front facing cameras making them an ideal video phone terminal. Apple provides Facetime
enabling video calls between users of its devices. Skype has been extended to provide video as
well as voice. Video streaming over the network not only imposes the sustained bandwidth
requirements of on-demand video but also requires low latency.
7
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 2: Projection of mobile traffic trends in the United States [5]
Figure 2 shows a projection of mobile traffic trends in the United States from the Sandvine report [5]. It
indicates that real-time entertainment comprising of video and audio streaming applications will grow
significantly in the coming years and account for close to 70 percent of mobile usage by 2018.
1.2.3 TRENDS FOR STREAMED VIDEO CONSUMPTION
Streamed video consumption is growing significantly as summarized below based on various reports, e.g.
Sandvine report [5]. From Figure 3, YouTube video viewing constitutes a significant component of realtime entertainment traffic, expressed as a percentage of downlink mobile traffic in 2012.
8
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 3: Global Shares of YouTube and Real-Time Entertainment Traffic [5]
During peak traffic periods, streaming video entertainment traffic is by far the most dominant traffic
category, accounting for nearly half of all bytes sent and received on the network, continuing a trend that
is expected into the foreseeable future. Real-time entertainment is defined as applications and protocols
that allow for on-demand entertainment that is consumed (viewed or heard) as it arrives on the device.
Examples include streamed or buffered audio and video (RTSP, RTP, RTMP, Flash, MPEG). Both HTTP
video and P2P TV are types of online video where the user watches the film while it is being downloaded
(streamed). In the case of P2P file sharing the user first downloads the entire file, most often a movie, and
watches it offline. Examples of applications that use HTTP include YouTube, Hulu, Netflix, and BBC
iPlayer. Examples of P2P TV include PPStream and Octoshape. Octoshape is a service that lets a
streaming video site use your Internet connection to stream video to other people.
There is an increasing trend in consumption of streaming video during live events, much of which is on
mobile devices. Spikes in video usage are driven by live events and breaking news. Figure 4 shows
usage data captured from FLOTV in 2009. It shows that average minutes of daily viewing peaked around
live events such as the NCAA Tournament and college football. Breaking news such as inauguration day
and Michael Jackson's memorial drove high usage.
9
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 4: Streaming Video usage by users on different days in 2009 [FLO TV Incorporated]
During the first week of the 2012 Olympic Games, streaming traffic accounted for almost 22 percent of
total network traffic and between 8-12 percent of network traffic in the U.S. at its peak daily levels. At the
Feb 4, 2013 Super Bowl games, official statistics for live stream of XLVII on CBSSports.com and
NFL.com [6] indicated the following statistics:




Unique viewers: 3 million
Total minutes streamed: 114.4 million
Live video streams: 10 million
Engagement: 38 minutes per viewer
North America continues to lead in adoption of this traffic category, with almost two-thirds of downstream
traffic during peak period being streaming audio or video. The dominance of video streaming is due in
large part to the continued uptake of video-on-demand, which now accounts for 33 percent of peak period
downstream Internet traffic, some of which is mobile video traffic. In July 2012, an important threshold
was crossed with 1 billion hours of video streamed in a single month. It is expected that video
advertisement will become more and more prevalent and consume an increasingly higher portion of data
usage in the future.
In Europe, a major news provider offers regional and international catch-up TV service subscriptions. On
demand viewing of most programs broadcast on major TV channels for up to seven days after first airing
is allowed. Over 600 devices are supported from PCs, game consoles, connected TVs, smartphones and
tablets. The December 2012 statistics [7] show that 31 percent of iPlayer viewings are on smartphones
(28 million views per month) and tablets (26 million views per month). Most of this will be streamed via
Wi-Fi because of the low penetration of LTE in the UK. This shows the acceptance of people to use these
small screens to watch TV. The expectation is that mobile video streaming will increase with more
widespread availability of technologies like LTE.
Ooyala recently released the 2012 Video Index [8] that reveals key online video trends. Ooyala measures
the viewing habits of nearly 200 million unique viewers in 130 countries every month. The report reflects
10
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
the anonymous online video metrics of various publishers. Their data shows live streaming is the new
norm. Viewers watch live video longer on all devices. The share of tablet video viewing more than
doubled in 2012, as mobile, social and video converged on a single device. Smart TVs and gaming
consoles continue to change the way people watch TV. Xbox users spend more time watching video
than playing games. New findings in this Video Index show how viewing patterns change seasonally.
In summary, data from different sources consistently indicate a significant uptake of mobile video, be it
streaming traffic during live events, real time entertainment and You Tube traffic. This trend is only
expected to grow with newer video applications contributing to the mix, and with the introduction of newer
devices with larger screen sizes and increased capabilities.
1.2.4 ECOSYSTEM SUPPORT FO R THE GROWTH IN MOBILE VIDEO
To support the growing need for mobile video, multiple facets of the ecosystem have evolved as below:
1) There is an evolution in video delivery mechanisms with fixed, reliable connections (DSL, fiber,
etc.) being complemented by Wi-Fi and small cells cellular access. Initially, video was accessed
only from fixed access points using static devices like TVs and tethered devices. As mobile
broadband becomes more ubiquitous, video is being increasingly accessed at different types of
locations, and in crowded locations in dense urban and urban areas.
2) There is a dramatic increase in overall wireless data consumption as bandwidth intensive video is
increasingly being consumed on wireless. With the introduction of new wireless technologies
including HSPA+ and LTE, video has become mobile and is being viewed at locations such as
transport hubs, stadiums and arenas during games and in enterprise locations where users in
buildings are accessing their mobile office network.
With the ability to watch streamed multimedia applications (on-demand streaming, live streaming,
etc.) faster on LTE networks, subscribers are likely to watch longer duration videos more
frequently, consuming more data than ever before. In some cases, new LTE customers have
experienced as much as a 50 percent increase in their monthly broadband usage simply by
moving from a 3G to a LTE network.
3) There is increased complexity in playing of video content across devices, as users move across
accesses and expect continuity of sustained video quality. Streaming video to mobile devices is
increasing. Previously people would pre-load their portable video player with content. The choice
of device for access is now increasingly on highly portable devices such as iPads and other
tablets that allow for more visually dynamic, image rich content to be displayed along with audio.
4) There is an evolution in rendering quality (Standard Definition [SD], 720p, High Definition [HD],
etc.) on devices and content systems, as end user expectations demand increased quality in
video content. While video is currently delivered primarily as SD and below, it is expected that
users will, over time, demand that video be delivered as HD. Better viewing experience will
translate to increased user demand for more video content on their mobile devices.
5) There is an increased expectation on device clients and content servers to adapt to the changing
underlying transport capabilities due to the mobility aspects that are introduced by wireless
devices.
11
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
6) There is an evolution in business models for video as providers find new ways to monetize video
content. Videos can be informative, entertaining, educational, and each provide unique
opportunities to content and service providers, tool developers, etc. for new business avenues.
This paper addresses the various issues related to the mobile video delivery in the following sections.
2. TYPES OF VIDEO
The mobile video ecosystem can be broadly classified under two major dimensions as follows:
1) Video content types (encoding, container and delivery methodologies).
2) Video delivery strategy and technologies.
This section gives an overview of these two dimensions to help address the full scope of the mobile video
experience, including challenges, metrics, architectures and solutions.
2.1 ELEMENTS OF VIDEO CONTENT AND SUMMARY OF VIDEO DELIVERY TECHNOLOGIES
Videos are specialized content that comprise of three basic elements. They are:
1) Moving video component: This section actually contains the moving images that when played in
quick succession gives the feeling of watching a continuous video stream. Some popular video
codecs used in the industry today are H.264, VP6, VP8, etc.
2) Synchronized audio component: This section has the audio content corresponding to the video
being played back. Some popular audio codecs used in the industry are AAC+, AAC, mp3, etc.
3) Video container: Both the video and audio components are wrapped around a media object called
a video container. This has additional information about the audio and video components like, the
codecs used and the frame offsets. This information is used by the player to play back the video
smoothly and also to handle video seek requests from the user. Some popular video containers
used in the industry are MPEG4, Flash Video and WebM.
Both visual quality as well as the smoothness of the video playback influences the end-user’s quality of
experience. The quick start of the video playback is also often considered to influence the quality of
experience. Hence different techniques are used to protect the visual quality as well as ensure the
smoothness of the video playback in a dynamic network environment, like in mobile networks.
Video content may be classified in various ways depending on their duration, their type of quality, whether
they are delivered by the service provider or not, etc. These are described in detail in Section 2.2 through
2.4. There are four major video delivery technologies that have substantial presence in the mobile
network, which are summarized below and described in detail in sections 2.5 through 2.9.


HTTP Progressive Download (HTTP-PD): This is a popular video delivery methodology for
short form video content. User generated video content publishers like YouTube and Facebook
use this mode for their short form content.
HTTP Adaptive Streaming (HAS): This is an upcoming video streaming technique being
popularized by the likes of Netflix, for long form content. This technology empowers the clients to
determine the video quality based on the dynamics of the network condition. With HAS, the client
requests a higher quality video in case of good network condition and lower quality when the
network condition is poor.
12
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013


Non-HTTP, Real Time Streaming Protocols (RTSP, PPStream etc.): These protocols are
typically used by certain video content providers to stream live and long form video content to
proprietary clients. While HTTP is stateless, RTSP is a stateful protocol.
Video file downloads: In this scenario, the video content is delivered offline and viewed later by
the user using generic or proprietary video players.
The following sections describe the various types of video content and the various video content
delivery technologies in detail.
2.2 SHORT FORM CONTENT, LONG FORM CONTENT
Video content can be broadly classified into short form and long form content. Videos shorter than 10
minutes are typically classified as short form, while those that are longer than 10 minutes are classified as
long form video content.
Short form video content could be anything like video advertisements, user generated videos, product
videos, sports highlights, news extracts, etc. Some of this short form content is produced professionally
while the majority of them are not. Hence, there is a wide variation in the encoding quality of short form
videos. Due to the short time duration involved, video publishers typically use some kind of file download
mechanism (progressive download, for example) to deliver this video content. Players are typically
designed to play the video as it receives them from the servers. This progressive nature of the client
video playback ensures that users don’t have to wait for the complete download of the video before
playback begins. Since this is delivered as a simple file download, any bandwidth variation in the network
would adversely impact the smooth playback of these videos.
On demand version of TV shows and live telecast of events are the typical examples of long form content.
Long form content involves a more consistent viewing form. Since most of this content is produced
professionally and delivered to paying subscribers, the overall quality of video experience becomes
important. It is for this reason that most long form content publishers employ some form of real-time
streaming technology that modulates the video bit rate according to current network condition
dynamically. Due to the premium nature of the content, most content publishers protect the copyrights
through some digital rights management (DRM) mechanism. By being DRM protected, neither an innetwork entity, like a video optimization server, nor the end user can modify or duplicate the content for
shared viewership or later viewership.
2.3 OPERATOR VIDEO AND O TT VIDEO
Many mobile network operators are looking to extend their offers by offering content as a part of their
subscriber’s package. They host premium video content inside their network in a form optimized for
delivery to the devices they support. To provide a premium viewing experience, they will manage the
network in order to optimize the streaming of the video to the user device by caching content closer to the
edge of the network and using QoS on the RAN. The operator can implement various business models
including zero rating their content as well as introducing additional services and applications such as ad
pre-rolls, post-rolls and overlays.
The majority of the videos watched on mobile networks today are from Over-The-Top (OTT) providers.
OTT videos are typically delivered over HTTP-Progressive Download channel for short form content and
through HTTP-Adaptive Streaming for long form content. Mobile operators typically don’t have much
influence in defining the quality of service and quality of experience for the end user.
13
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
2.4 BEST EFFORT VIDEO AN D GUARANTEED BIT RATE VIDE O
Superior and uninterrupted online video experience relies a lot on the availability of high bandwidth
continuously for long durations. Unfortunately, this can never be guaranteed on a mobile network. Even
when the user is stationary, the effective bandwidth experienced by his device is likely to vary
dramatically within a 5 minute duration (average length of video on YouTube is 5 minutes) due to timevarying channel conditions. Mobile operators have mainly two options to address the dynamic network
variations and offer satisfactory service to their subscribers.
1) Best Effort Video Delivery: With the best effort delivery model, operators make it explicitly known
to the consumers that actual video experience is dependent on network characteristics. In most
cases, mobile operators promote/advertise an “up to” maximum network bandwidth that the
specific pricing plan provides. This is by no means a guarantee as this is simply a theoretical
maximum. Under lightly loaded network conditions, a good user experience may be realized.
2) Guaranteed Bit Rate Delivery: In this model, mobile operators guarantee a specific minimum
bandwidth as part of the data service. 4G networks do provide mobile operators the technical
capabilities to provide guaranteed bandwidth to users. It remains to be seen if this can be offered
for higher values (like 1-2 Mbps) at all times, at all cell sites, without significantly affecting the
throughputs delivered to the remaining users, particularly when there are many users.
2.5 VIDEO FOR ENTERTAINMENT AND VIDEO FOR REAL TIME COMMUNICATIONS
Today, the majority of video on the mobile network is intended for entertainment. Whether short form or
long form, this content is made available by video producers to entertain, amuse and educate. Typically,
there is a long time (days and even years) between production and consumption. Even for live streaming,
there can be a delay of 30-60 seconds or longer between the live event and it being displayed on the
screen. The content can be of sporting events, drama, films, cartoons, nature documentaries, etc. Such
content is demanding of the highest feasible quality for both the audio and video.
Video telephony and video conferencing are growing in popularity. Such communications need to have a
maximum latency of 150 msecs [9] to ensure good human interaction. Hence, they are often termed RealTime Communications. However, the video image is usually of talking heads so there is no requirement
for high fidelity audio and video. The WebRTC initiative from W3C will allow real-time communications to
be embedded in and enabled from web browsers. Some browsers like Chrome and FireFox support
WebRTC. This allows real-time communications to be added to web pages without the need for
additional plug-ins or applications.
2.6 HTTP PROGRESSIVE DOWNLOAD
Basic download is the process by which a content provider will encode video and put it up in a web server
used to host HTTP pages. Viewers will click on the URL of the video file and the video file is downloaded
into their computer for viewing and the videos can be watched after the download is completed.
Encoding involves two key steps. Before putting the raw file that is the output from the recording device,
the video is compressed into a format that the media player could play, based on the right quality chosen.
Two decisions need to be made here namely, the choice of video quality and the choice of video format.
14
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013


Video quality: higher the bit rates result in higher video quality. For instance, a 100 kbps
streaming video is phone-video quality but a 10 Mbps streaming video is DVD quality. Higher bit
rates requests consume higher bandwidths.
Video format: There are different file types supported by different media players, such as H.264
(MP4), FlashVideo (FLV), QuickTime (QT), Windows Media (WMV) etc.
A progressive download is the transfer of digital media files from a server to a client, typically using the
HTTP protocol when initiated from a computer. Progressive download allows a user to access a file's
contents before the download has been completed. In a progressive download, the video usually does not
start playing until the buffer has grown large enough to ensure stall-free playback. Progressive
downloading transmits a media file with metadata in the header of the downloaded file. A media player
capable of progressive downloading reads this metadata and begins playback of the media file after a
specified minimum portion of the data has been downloaded. Media players also require a buffer to hold
data for use at a later time. The amount of data held in the buffer before playback is determined by both
the producer of the content (in encoder settings) and by additional settings imposed by the media player.
2.7 RTP/RTSP DELIVERED VIDEO
Traditional streaming uses special media protocols, such as RTSP (Real-Time Streaming Protocol),
RTMP (Real-Time Messaging Protocol) and RDT (Real Data Protocol). With RTSP, traditional streaming
will send the video as a series of small packets in sequence in RTP packets. The RTP packet size is
around 1 Mb for 1 Mbps quality video and these packets will get transmitted over UDP transport (UDP
RFC 768). The protocol is designed so that it is better to have a momentary glitch in audio or video than
for the playback to stop altogether and wait for missing data. The packets are streamed at a specified bit
rate. The client can tell the server this rate using RTSP or derive it from the encoding rate of the content.
Forward Error Correction (FEC) can be applied to the UDP stream to help mask packet loss. Alternatively,
the client can request the retransmission of the missing packets. This is done using RTP Control Protocol
(RTCP RFC3611). RTCP can also be used to indicate the quality of the communication such as jitter and
round-trip-time delay (RTT). This can be used to help decide whether to change the codec or bit rate.
RTP/RTSP video is most often used in wireline IPTV deployments. Strict admission control before starting
a new stream ensures there will be sufficient bandwidth dedicated for the stream. It also works best over
low loss networks so the FEC or packet retransmission overheads are minimized.
In contrast to progressive download described in Section 2.6, during a streaming media session, the file is
never completely downloaded to a local storage device as it is with progressive download. This difference
makes streaming media a more secure option that reduces the risk of piracy.
Issues with traditional streaming that motivate the need for HTTP adaptive streaming (in Section 2.8)
include the following:

Since dedicated media servers are needed, existing web servers cannot be leveraged. The
transmission of media packets is through UDP protocols.

Once the video is encoded at certain bitrates, the client will receive that quality of video content
regardless what kind of CPU condition and bandwidth network the client has. This is similar to
progressive download.
15
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013

Traditional streaming is harder to scale since, as a stateful protocol, traditional media servers
maintain a one-to-one persistent connection with each client.
2.8 HTTP ADAPTIVE STREAMING (HAS)
The available bandwidth over the Internet is highly variable. Wireless technology adds to this variability
due to the time and spatially varying radio propagation characteristics of the Radio Access Network
(RAN). To handle this variability in bandwidth and to ensure best image quality when bandwidth is good,
an adaptive streaming technology is required. For professional content streamed over the Internet, a
common approach for streamed content is HTTP Adaptive Streamed (HAS). There are several variants
including MPEG DASH, Apple HLS, Microsoft Smooth Streaming but they all work in the same way.
Figure 5 shows the operation of HAS and Figure 6 shows the message exchanges to execute HAS.
.
Figure 5: HTTP Adaptive Streaming
16
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Time,
seconds
0
2
4
6
8
10
Quality levels (nominal rates)
Video
400 kbps 850 kbps 1.5 Mbps
Chunk # Congested
Moderate
Uncongested
0
HTTP GET
1
HTTP GET
2
HTTP GET
3
HTTP GET
4
HTTP GET
5
HTTP GET
Figure 6: Message exchanges during HTTP Adaptive Streaming
Video gets encoded into different formats and video qualities. HTTP Adaptive Streaming uses normal
web servers and transmits data via HTTP protocol. Video files are divided into small non-overlapping
video fragments (or intervals) in sequence, between 1 and 10 seconds long. Each interval is then
encoded in multiple bit rates. The bit strings associated with these encodings are referred to as chunks
(also called segments in MPEG DASH). The HAS client uses HTTP to request chunks – the first one
usually at a low bit rate. If chunks are consistently delivered in a time shorter than the interval length, the
Rate Determining Algorithm (RDA) in the client selects a higher bit rate for the next chunk (see Figure 5).
In that way, the RDA continually senses the available throughput and adapts the video rate accordingly.
In order to absorb mismatches between video rate and throughput, the video client maintains a play-out
buffer. A larger playout buffer means a longer period of poor radio conditions can be tolerated. But it also
increases the startup time. Assuming that the RDA is working properly, there should be no stalls while it
automatically adjusts to the available bandwidth.
Using HTTP has several advantages. It allows reuse of standard web technologies in the distribution and
delivery of video. This helps minimize costs and avoids the need for special network technologies or
configurations. Furthermore, HTTP operates over TCP, which automatically provides flow control and
recovery from packet loss. Another advantage of HTTP-based video delivery is its ability to traverse
firewalls through port 80. Finally, HTTP uses caching mechanisms on the web. Cached data is generally
closer to the viewer and hence more easily retrievable. This capability should both decrease the total
bandwidth costs associated with delivering the video, since more data can be served from web-based
caches rather than the origin server, and improve quality of service.
Adaptive streaming enables viewers to start video play-back while the content is being downloaded.
Consider two common sources for video streaming traffic in the Internet namely, Netflix and YouTube.
Users can view Netflix and YouTube videos either on PCs, using a Web browser, or on mobile devices,
using a Web browser or a mobile application. A mobile application is the native Netflix or YouTube
application running on mobile devices.
17
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Netflix and YouTube use TCP to stream videos. A number of professional video content providers, such
as Netflix, Hulu and BBC iPlayer, currently use HAS to deliver video content over unmanaged networks.
HAS allows video delivery to adapt dynamically to the available bandwidth of a network while providing
adequate quality of experience (QoE) for subscribers.
Below are some common implementations in the industry:



Microsoft Smooth Streaming: http://www.iis.net/download/SmoothStreaming
Adobe Dynamic Streaming for Flash
Apple HTTP Adaptive Streaming
Apple’s HTTP Live Streaming (HLS), Microsoft’s Smooth Streaming, and Adobe’s HTTP-based Dynamic
Streaming (HDS) use Hypertext Transfer Protocol (HTTP) delivery protocol and standard HTTP web
servers to deliver streaming content, thus obviating the need for a streaming server. HAS technology is
standardized in MPEG DASH (Dynamic Adaptive Streaming over HTTP). The International Standard was
published in April 2012 as ISO/IEC 23009-1. DASH is discussed further in Section 2.9.
2.8.1 COMPARISON BETWEEN PROGRESSIVE DOWNLOAD, STREAMING AND ADAPTI VE STREAMING
With HTTP- Progressive Download and traditional streaming protocols (RTP/RTSP), once the client starts
streaming a video, the bit rates of that video clip will stay unchanged. Even if the client has a very bad
network condition later on, it still needs to load the video with high bit rates. This is in contrast to HTTP
Adaptive Streaming (HAS), where a client could enjoy a smooth viewing experience because of video bit
rates that get switched seamlessly according to the network bandwidth availability and possibly the
client’s CPU condition.
In addition, HAS utilizes HTTP protocol which leverages the scalability of the whole Internet. There are a
lot of intermediary proxies or gateways between client and the server. Since video files get chunked into
small file sizes, it is easier to cache in all these intermediaries. Therefore, HAS works well with content
delivery networks (CDNs) to enable media content providers scale in a very cost-effective manner.
2.9 PLATFORMS FOR VIDEO DELIVERY
Adaptive streaming involves producing several instances of a live or on-demand source files and making
them available to various clients depending upon their delivery bandwidth and CPU processing power. By
monitoring CPU utilization and/or buffer status, adaptive streaming technologies can change streams
when necessary to ensure continuous playback or to improve the experience.
One key difference between the various adaptive streaming technologies is the streaming protocol
utilized. For example, Adobe’s RTMP-based Dynamic Streaming uses Adobe’s proprietary Real-Time
Messaging Protocol (RTMP), which requires a streaming server and a near-continuous connection
between the server and player.
There are multiple video delivery platforms that are being used. The important point to note is that most
video delivery platforms are building in some form of bandwidth and device awareness into them so as to
enable them to provide superior video experience. Proprietary adaptive streaming platforms rely on
cooperative video players that implement their proprietary transport protocol. The popular platforms
belonging to this category are Apple HLS, Adobe Dynamic Streaming, Microsoft Smooth Streaming, etc.
18
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Recently, major video delivery platforms have gotten together and are formulating a dynamic streaming
standard called DASH. MPEG DASH (Dynamic Adaptive Streaming over HTTP) is a standard for
adaptive streaming over HTTP that has the potential to replace existing proprietary technologies like
Microsoft Smooth Streaming, Adobe Dynamic Streaming, and Apple HTTP Live Streaming (HLS). A
unified standard could be helpful to content publishers, who could produce one set of files that play on all
DASH-compatible devices. DASH is an attempt to combine the best features of all HTTP-based adaptive
streaming technologies into a standard that can be utilized from mobile to OTT devices.
All HTTP-based adaptive streaming technologies have two components: the encoded audio/video
streams themselves and manifest files that identify the streams for the player and contain their URL
addresses. For DASH, the actual audio/video streams are called the Media Presentation, while the
manifest file is called the Media Presentation Description. The Media Presentation is a collection of
structured audio/video content that incorporates periods, adaptation sets, representations and segments.
The Media Presentation defines the video sequence with one or more consecutive periods that break up
the video from start to finish. The DASH manifest file, called the Media Presentation Description, is an
XML file that identifies the various content components and the location of all alternative streams. This
enables the DASH player to identify and start playback of the initial segments, switch between
representations as necessary to adapt to changing CPU and buffer status, and change adaptation sets to
respond to user input, like enabling/disabling subtitles or changing languages. With several big players
like Microsoft, Adobe, and Google participating actively, it is possible that DASH will be widely adopted in
the marketplace.
From the client standpoint, most browsers today start a separate app/plug-in to play the selected video.
With the advent of HTML5, browsers would be capable of playing videos natively. This vertical integration
of video rendering into the browsers provides web authors the added ability to control the look and feel of
the video area.
3. CHALLENGES FOR MOBILE VIDEO DELIVERY
Video delivery is characterized by the need to send large amounts of data over long time durations. The
sections below describe the challenges of delivering video in a mobile environment over the air interface
and within the core network.
3.1 CHALLENGES FOR VIDEO DELIVERY DUE TO AIR INTERFACE
Mobile video can create a multitude of problems on mobile networks due to their highly demanding
requirements on bit rate, latency, delay and jitter. Although actual mobile networks do comprise both
wired and wireless interfaces, it is the air interface and possibly the backhaul that suffer the most from
highly strained operational conditions for video delivery. Air interface actually represents the weakest link
in any network architecture due to several reasons:

The high error rate potentially present in mobile environments must be overcome through the use
of multiple techniques, such as adaptive modulation and coding, retransmissions at different
levels, and limitation of the block sizes to be transferred to the users. Some of these techniques
increase the effective delay incurred in the radio access part to transfer a given application-level
information block. An increased delay results in a higher Bandwidth-Delay product (BxD), and this
in turn may impair the net throughput of TCP connections if receive window sizes are not
appropriately dimensioned.
19
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013

The scarce auctioned spectrum must be shared among the mobile users at the radio interface,
and fair network operation must ensure that users have at least a minimum access to resources
(at least for normal tariff plans).

Typical mobile channels involve several degradation phenomena such as multipath, large-scale
and small-scale signal fading, Doppler spreading, and increased noise, to name a few. In addition
to these, there appear other interference issues between cells caused by single-frequency
operation in technologies such as UMTS, WiMAX or LTE.

Limitations in devices usually incur some penalty on processing power capabilities, resulting in a
lower spectral efficiency for a given radio condition. Limitations in the RF circuitry of devices
introduce additional degradations such as poorer transmission mask, lower sensitivity, increased
noise figure or reduced antenna efficiency, to name a few. All these limitations directly translate
into poorer air interface capacity.

Many connection-oriented protocols are not optimized for operation in mobile networks because
they usually assume that a packet loss should probably be caused by a network congestion issue
instead of a link error. As an example, TCP congestion control reacts to a packet loss by reducing
the size of the transmission window, thereby decreasing the peak data rate. High delays
experienced by packets when the user suffers from poor signal conditions should be matched by
appropriate TCP receive window sizes, which are not usually optimized for the actual radio
interface performance [11].
3.1.1 SPECTRAL EFFICIENCY OF 3G/4G CELLULAR TECHNOLOGIES AND IMPAC T ON MOBILE VIDEO
Limitations in the spectral efficiency of radio access technologies, which account for typical aggregate
throughput values in a given amount of spectrum, have a clear impact on video performance. The higher
the spectral efficiency, the higher the air interface throughput and hence the improved user experience for
mobile video. Table 1 shows the data consumed by different streaming applications [10]. This table
indicates that different devices need different bit rates for acceptable Quality of Experience. Moreover,
even a modest use (e.g. with content encoded in AVC/H.264, 500kbps on smartphone for 30 minutes per
day 20 days a month) will add up to a high monthly data usage totals for subscribers. It is apparent that
medium- and high-quality video applications demand net throughput values of several Mbps, which
directly results in minimum required values for the average spectral efficiency of the weakest network link
(the air interface). Unique ways to minimize congestion to the network due to video application download,
utilization of radio resources for video download during non-peak hours, mobile video delivery through
broadcast solutions, and influencing subscriber behavior strategies will be needed in order to make
mobile video delivery over cellular networks efficient and scalable for large number of users. These will be
discussed in the detail in the following sections.
20
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Table 1: Data Consumed by Different Streaming Applications [10]
Application
Throughput MByte/Hour Hrs./day GB/month
(Mbps)
Stereo Music
0.1
58
Small Screen Video (e.g. Feature
Phone)
0.2
90
Medium Screen Video (e.g.
Smartphone)
0.5
Medium Screen Video (e.g. Tablet)
1
Larger Screen Video (e.g. High
Quality on Tablet)
2
225
0.5
1
2
4
0.5
1
2
4
0.5
0.9
1.7
3.5
6.9
1.4
2.7
5.4
10.8
3.4
450
1
2
4
0.5
6.8
13.5
27
6.8
1
2
4
0.5
1
2
4
13.5
27
54
13.5
27
54
108
900
In order to satisfy the requirements for these typical video services, typical air interface performance
values are given in Table 2. Table 2 shows a comparison of various technologies in terms of downlink
and uplink speeds.
Instead of considering peak data rates (which can serve for benchmark comparison between
technologies), typical average aggregate throughput values can give a better idea of realistic performance
in average cell and user conditions. This is particularly true for video users, since they require high
sustained rates for long periods of time (tens of minutes to hours), and are likely to be moving across
multiple cells with varying radio and hence interference conditions during that time.
Growth of video is likely to be facilitated with recently introduced technologies such as LTE and LTEAdvanced. A study of various 3GPP technologies [10] indicates that introduction of OFDMA in LTE
allows very high throughput rates in uplink and downlink with all communications handled in IP domain.
LTE-Advanced introduces carrier aggregation that recognizes the asymmetry of uplink/downlink traffic
patterns and thus increases the downlink bandwidth by aggregating multiple carriers while keeping uplink
operation unchanged.
21
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Table 2: Performance comparison of 3GPP mobile technologies [10]
Technology
Name
Type
Characteristics
Typical
Downlink
Speed
200 to 300 kbps
200 to 300 kbps
1 Mbps to
4 Mbps
500 kbps
to 2 Mbps
UMTS
CDMA
3G technology providing voice and
data capabilities. Current
deployments implement HSPA for
data service.
HSPA
CDMA
Data service for UMTS networks.
An enhancement to original UMTS
data service.
CDMA
Evolution of HSPA in various
stages to increase throughput and
capacity and to lower latency.
LTE
OFDMA
New radio interface that can use
wide radio channels and deliver
extremely high throughput rates.
All communications handled in IP
domain.
LTEAdvanced
OFDMA
Advanced version of LTE designed
to meet IMT-Advanced
requirements.
HSPA+
1.9 Mbps to
8.8 Mbps
in 5/5 MHz
3.8 Mbps to
17.6 Mbps with
dual carrier in
10/5 MHz
6.5 to 26.3
Mbps in
10/10 MHz
Typical Uplink
Speed
1 Mbps to
4 Mbps
in 5/5 MHz or in
10/5 MHz
6.0 to 13.0 Mbps
in
10/10 MHz
HSPA and HSPA+ throughput rates are for a 5/5 MHz deployment. N/M MHz means 5 MHz is used for
the downlink and M MHz is used for the uplink. From the above table it is clear that only HSPA, HSPA+,
LTE and LTE-Advanced technologies may cope with the huge bit rate requirements of present wireless
video applications in smartphones, tablets and laptops. Legacy devices with limited features will definitely
suffer from reduced performance in the same network conditions.
Air interface performance can conveniently expressed in terms of the spectral efficiency, which is defined
as the throughput value per unit of occupied bandwidth (bps/Hz). LTE is more spectrally efficient with
wider bandwidths (10, 15 and 20 MHz), due to the significant overhead incurred by layer-1 control
signaling in narrow bandwidths. Figure 7 shows the LTE spectral efficiency as a function of system
bandwidth, thus highlighting the ability of LTE to support mobile video delivery with good Quality of
Experience for the user.
22
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
100
% Efficiency Relative to 20 MHz
90
80
70
60
50
40
30
20
10
0
1.4
3
5
10
20
MHz
Figure 7: LTE Spectral Efficiency as a Function of the System Bandwidth [10]
Typically, the need for downlink capacity is more important particularly for video. A study of various 3GPP
technologies indicates that spectral efficiency is usually lower in uplink, which is aligned with typical
usage patterns of mobile users. However, with the advent of WebRTC and apps such as Skype and
Facetime, it is expected that the uplink capacity will also come under pressure as people increasingly use
video telephony.
Higher peak rates are usually obtained only when the mobile is in good stationary conditions, with high
and stable signal to noise ratios and little Doppler spread. In these conditions, video delivery can be more
efficient due to the feedback mechanisms that usually exist between the mobile and the base station.
Proper link adaptation can only be achieved through use of some sort of signal quality feedback from the
users, such as the Channel Quality Indicator (CQI) reports employed in HSPA and LTE.
3.1.2 AIR INTERFACE LATENC Y AND IMPACT ON MOBI LE VIDEO
Closely related to the spectral efficiency is air-interface latency. Video quality of experience is facilitated
with the introduction of LTE and LTE-Advanced due to their high bandwidth and low latency
characteristics. Higher bandwidth allows for users to receive video with higher quality without resulting in
packet loss due to congestion in the radio interface. Lower latency allows for problems with TCP delivery
to be minimized, which is critical for the delivery of video through HTTP Adaptive Streaming (HAS).
Air interface introduces latencies due to mainly three mechanisms:

The link adaptation function in the base station must choose the preferred modulation and coding
scheme for each particular user, based on the amount of information to be transferred and the
23
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
signal quality reported by the users. Information blocks of different sizes are sent over the radio
interface thus requiring different amounts of time to transmit a given higher-level packet (such as
an IP packet). This results in different air interface latencies that depend heavily on radio
conditions.

The air interface usually introduces a hybrid retransmission mechanism to overcome channel
errors, sometimes known as Hybrid Automatic Repeat Request (HARQ). These retransmissions
involve additional latencies depending on the HARQ round-trip-time (HARQ-RTT), as e.g. 8 ms in
LTE and 12 ms in HSPA (for the downlink).

The scheduler in the base station ensures proper access to the available shared resources for
coexistence of multiple users in a given cell. If the cell is in low load conditions, the scheduling
process is easy and the users are eventually given full cell resources (if allowed by the applicable
Quality of Service policy). In high load conditions, the scheduler must maximize cell throughput
while ensuring fair use of resources among users. This results in higher queue sizes and
increased packet delay.
A study of the latency performance in various 3GPP technologies [10] indicates that LTE deployments are
likely to result in the lowest possible latency values according to the physical layer capabilities. Round trip
time refer to the measured time delay value between the subscriber unit and a node immediately external
to the wireless network and does not include Internet latency. As an example, trial results reported by
LTE/SAE Trial Initiative (LSTI) measured LTE round-trip times ranging from 18 to 28 msec [12].
Air interface latency and jitter significantly impact the performance of mobile video. An increased latency
represents a drawback for the deployment of wireless interactive services, such as high-quality
videoconferencing or online gaming based on streaming protocols like RTP, RTSP or RTCP. Other video
services like progressive download video (as opposed to streaming video) are not so sensitive to latency
due to their limited interactivity with the user. With progressive download video, the contents are
transferred via standard HTTP protocol and appropriate receive buffers at the client minimize the impact
of throughput and latency impairments. Note that the air interface latency is in addition to the expected
latency incurred by the chain of network nodes that the information must traverse.
3.1.3 INTERFERENCE AND HAN DOVER ISSUES IN 3G/4 G CELLULAR TECHNOLOGIES
Interference and handover issues can affect the video quality of experience, depending on the user
mobility, type of video being accessed etc. Metrics for video QoE are discussed in Section 4.2.
3.1.3.1 IMPACT OF RADIO INTERFERENCE ON VIDEO DELIVERY
Mobile video users are likely to experience significantly varying interference levels, both as they consume
video over long periods of time and as they move across cells. Higher interference translates to lower
signal to noise ratio and hence lower bandwidth available for video, in addition to resulting in various
video impairments described in Section 4. It is therefore important to consider the impact of interference
over the quality of video as viewed by the users, and on the allocation and optimization of network
resource to address the challenges of spatial and temporal variations in interference as described below.
24
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Interferences in wireless networks appear when the spectrum in use is shared among users and cells. By
sharing the same frequency in all cells, effective mechanisms must be designed to cope with increased
interference between users, cells and services (especially at the cell edges).
Soft handover in WCDMA allows the user to be connected to more than one base station simultaneously
because the mobile can coherently add downlink signals in order to improve coverage. This feature
enables a more continuous coverage when moving between cells, especially for control channels where
correct demodulation is critical to maintaining synchronization. In a stationary situation, the user in soft
handoff can typically be connected to two or three cells. Soft handover is not present in HSDPA channel
for the downlink, due to the fast allocation mechanisms that would be involved and the need for tight
phase synchronization of the cells. For the rest of the downlink channels, as well as all uplink channels,
soft handover is employed.
LTE does not allow such soft combining (except for multicast/broadcast data transmissions). However,
the flexibility of assigning time and frequency resources with such low granularity enables a highlyefficient utilization of radio resources in each cell, at the cost of increasing terminal complexity through
advanced interference-cancellation techniques.
Interference is an impairment that ultimately limits performance in cellular networks. It also prevents base
stations from delivering the highest data rates to the users if their perceived signal-to-noise ratios are not
high enough. In the context of traditional homogeneous networks (comprising a collection of macrocells),
interference represents the dominant issue at the boundaries between cells and forces handovers and
cell reselections to keep the best radio conditions at every location. Interference and handover could
result in reduced data throughput for video delivery. Therefore, the TCP and the lower layer protocols
must cope with increased interference when crossing the boundaries between cells.
Heterogeneous networks comprising a mixture of cells with different powers, frequencies and (possibly)
radio access technologies, make interference issues more prevalent due to increased percentage of
locations being in the boundaries between cells.
Figure 8: HET-NET architecture [13]
HET-NET deployments pose their own challenges as compared to homogeneous networks. A number of
techniques have been devised in order to cope with increased interference in the boundaries between
25
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
macrocells and small cells. These issues, as well as mitigation techniques, can significantly impact
mobility mechanisms in these networks and hence the performance of mobile video users.
3.1.3.2 IMPACT OF MOBILITY IN WIRELESS CELLULAR NETWORKS ON VIDEO DELIVERY
Mobility presents unique challenges in delivery of video. Mobility allows a user to get the impression of
being seamlessly connected to the Internet and/or the circuit-switched operator’s network, irrespective of
his/her actual location. Mobility can be accomplished through a set of radio access network and core
network features dealing with the maintenance of data session over the radio access.
Mobility management at the air interface must cope with eventual variations in the best server due to user
movement. The best server represents the most suitable base station to which the user must be
connected to in order to get the best radio conditions. Soft handover operation extends this best server
concept to multiple cells, but a single serving cell always exists that handles control signalling to/from the
user. Radio resource management (RRM) algorithms are defined by the different radio access
technologies to enable mobility through handovers (changes in connected-mode best server during the
course of an ongoing connection) and cell reselections (cell changes in idle-mode during inactivity
periods). Although the basic framework for performing radio resource management is standardized by
each technology (in terms of the reporting quantities to be considered by both devices and base stations),
actual RRM algorithms are implementation-dependent thus introducing an additional level of uncertainty
in performance. Depending on the network configuration and equipment in use, significant performance
differences can be observed as a consequence of RRM and mobility issues. Even highly-optimized radio
access networks must cope with increased signaling traffic coming from mobility management due to
handovers, information exchange among radio access network nodes (such as RNCs), cell updates,
location area updates, etc.
3.1.3.2.1 HANDOVERS AND IMPACT ON MOBILE VIDEO USERS
Handover mechanisms allow users to move from one cell to another while maintaining continuity of
service. Understanding of the handover mechanisms allows for identification of appropriate techniques to
address the challenges of mobile video delivery.
In hard handovers (like in HSDPA data channel), connection with the serving cell is lost when switching
from one cell to another. When delivering mobile video, this interruption should be absorbed by
appropriate receive buffers at the client’s player irrespective of the type of video service in use.
Furthermore, video users tend to be generally low mobility users which could result in fewer handover
failures by permitting sufficient time for the handover signaling messages to reach the UE successfully as
the user is riding through the strong interference region. Soft handovers (present in UMTS Rel-99
channels) are characterized by an active set where a number of cell connections are simultaneously kept
for the user, hence allowing for softer cell transitions and improved reception (at the cost of increased
radio resources). With soft handover, there is no interruption in the video service.
If the player’s buffer is not sufficiently large, handovers can cause excessive delays for some packets due
to retransmissions at the MAC or TCP levels, therefore increasing jitter and significantly impairing video
performance. If retransmissions are not present (as in video over UDP), the effect of handovers would
result in a reduced available bit rate during the handover time, thus resulting in significant video artifacts.
Video over UDP (i.e. non HTTP based protocols) may benefit from solutions like “bicasting” that allows
duplication of RLC unacknowledged mode Packet Data Units from the RNC to both the old and the new
26
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
serving cells when the user makes a serving cell change to minimize the impact of handover interruption
times. Through bicasting, dropped packets in the source cell can be received from the target cell after
handover.
Mobile video places demands on the air interface. RTP/UDP streaming connections are usually
constrained by tight values of latency and jitter, therefore retransmissions are not usually allowed at the
radio interface to limit delay. Streaming video (employing real-time streaming protocols like RTP, RTCP or
RTSP) must address the challenges of user mobility, since the radio bearer does not usually provide any
retransmission mechanisms for error correction. Users at the cell edges may therefore experience
appreciable quality degradation if bit error rates increase due to extra interference and/or signaling
overhead when performing handovers. However, in all cases, the context is maintained together with the
IP address and quality of service attributes, as a result of which applications should not be aware of any
user mobility. HTTP Adaptive Streaming will not likely experience the issues related to jitter and delay,
since typically the client has a buffer of media to be played out and fluctuations in latency and in jitter can
be absorbed into this buffer without any degradation of the end user experience.
Progressive download video has the advantage of allowing more flexibility in traffic patterns due to the
existence of a receive buffer that compensates for fluctuations in user data due to interference or mobility.
Moreover, TCP connections guarantee error-free reception, and the radio access is allowed to perform
retransmissions in order to reduce the number of end-to-end TCP retransmissions to a minimum.
Retransmissions however come at the expense of increasing the overall delay, as a consequence of
which the player’s buffer may empty and the video playback may stalls.
HTTP Adaptive Streaming (HAS) can cope with mobility issues in a more efficient way compared to other
forms of video delivery, as the bandwidth is adaptively changed according to the instantaneous network
conditions, therefore reducing the need for retransmissions. Issues might only appear in handovers if the
instantaneous bit rate after cell change is significantly lower than the screen resolution, giving rise to
appreciable visual artifacts.
3.1.3.2.2 RETRANSMISSIONS AT THE AIR INTERFACE AND IMPACT ON MOBILE VIDEO USERS
Retransmission mechanisms for packet data exist at various layers including the MAC layer and at the IP
and application levels. End-to-end retransmissions should be avoided whenever possible due to the extra
delay incurred that will impact video quality of experience. Furthermore, TCP errors are considered as
congestion events that trigger a reduction in transmission window, thus potentially reducing throughput.
Therefore it is important to recognize the importance of MAC layer retransmissions on video delivery as
described below. Retransmissions at the radio access usually comprise two levels:

Hybrid retransmissions at the Medium Access Control (MAC) layer, usually under a mechanism
called HARQ that corrects most channel errors within a few milliseconds. This is the fastest
retransmission scheme available at present in cellular networks.

Layer-2 retransmissions above the MAC layer (when the Radio Link Control (RLC) protocol
Acknowledged Mode (AM) is used) are in the form of ARQ mechanisms and managed by radio
access nodes. ARQ is performed at the Radio Network Controller (RNC) in UMTS and HSPA, at
the eNodeBs in LTE, and avoids expensive end-to-end retransmissions. HARQ is faster than
ARQ in correcting errors, and both of them are in turn much faster than TCP retransmissions.
27
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
For these reasons, it is advisable to activate retransmissions at the radio access network when error-free
video delivery services are deployed. However, appropriate receive buffers are needed to avoid video
stalls because of the extra delay incurred. In situations with poor coverage, or increased interference
levels, the number of retransmissions can be so high that the video player stalls, and this may also
happen when the cell is so congested that the scheduling process in the base station is not able to serve
as much throughput as needed by the client to maintain an acceptable video quality.
3.1.3.2.3 PHYSICALLY FAST-MOVING USERS USING MOBILE VIDEO
Physical speeds of users can significantly impair RRM mechanisms devised between users and the base
stations. Handover processes, link adaptation processes, closed loop MIMO operation and transmission
of LTE reference signals may get compromised when the user speed is above a certain limit.
Delivery of consistently high throughput for high mobility for wireless users, including LTE and LTEadvanced users, is a challenge due to various reasons that have been described in recent publications
[14, 15]. It is therefore unlikely that wireless cellular networks are able to deliver most demanding video
services above a certain user speed. The exact value of the speed limit is dependent of multiple factors
including propagation environment, actual network configuration, etc. In these cases, it is more advisable
to switch to a lower resolution video service (if available), instead of trying to cope with increased errors
through a higher number of retransmissions.
3.1.4 SCHEDULING MECHANISMS AT THE AIR INTERFACE
Packet-based radio technologies like HSDPA or LTE bring the capability to share air interface resources
in a more efficient way among the active users, therefore posing the problem of how to distribute them
across the available resource dimensions (time and codes in the case of HSDPA, time and frequency in
LTE). This is addressed through scheduling of air interface resources.
Scheduling in both LTE and HSPA radio access networks is performed by the base stations so as to take
advantage of reduced latency and higher level of interaction with the users. The choice of scheduling
strategy affects the performance of the mobile video application that co-exists with other applications that
are competing for the same radio resources. Scheduling strategies are designed to satisfy a given
number of conflicting requirements:

Cell capacity should be maximized.

Users should not suffer from starvation due to other data-hungry users.

The scheduling process should grant a given fairness level by which all users can access
resources with the same opportunities.

Users with different radio conditions will, in principle, experience different qualities, in spite of the
intended fairness level.
Scheduling strategies pursue a given fairness level which depends on the instantaneous radio conditions
and the history of the user, with variations that lead to different scheduling strategies. Three important
dimensions can be highlighted in any scheduling strategy:

The choice of suitable scheduling strategies for different radio conditions according to the desired
fairness level, such as whether users in good radio conditions should be boosted against cell28
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
edge users, or whether cell-edge users should be favored instead. This gives rise to different
scheduling strategies among which Round-Robin, Proportional Fair and Maximum Rate are
typical examples that serve as bases for more elaborate strategies.

The management of different user subscriptions (if available), by which premium users would be
given special treatment compared to normal or low-grade users.

The management of different expectations for different services, each with particular target
requirements on maximum packet delay, maximum error rate, and guaranteed throughput (for the
case of GBR bearers), all of which configure a given Quality of Service (QoS).
While the first dimension can be dealt with by the particular scheduling strategy, the other two should be
appropriately tackled by the Policy and Charging architecture of the wireless network.

UMTS allows for some QoS control by defining so-called QoS traffic classes, namely
Conversational, Streaming, Interactive and Background classes.

LTE with its Evolved Packet Core (EPC) provides an evolved policy and charging architecture
and introduces the concept of QoS Class Identifier (QCI), which offers full control of QoS
parameters on a per-service and per-subscription basis. The various QoS attributes foreseen by
the network, are inputs to the scheduler. Section 6.2.1 describes the various QCIs and the
priorities and delay budgets associated with each that can be used to select the appropriate QCIs
for a given mobile video application.
From the above, it may seem that cellular technologies provide sufficient control over QoS attributes so
as to ensure fair and efficient resource cell utilization. However, reality proves that actual applications
may create multiple coexistence problems even in low load situations, thus making the schedulers very
sensitive to “greedy” applications. For example, consider that a YouTube mobile user is being served with
a high bit rate through progressive download in an LTE low load situation. Upon connection, the
scheduler will grant a high number of resources for that user in order to serve the high amount of
expected video data within the delay constraints imposed by the bearer characteristics. For the QoS
delay restrictions to be fulfilled, for a given QCI, high bit rate video services will require a high number of
LTE resource blocks and a high sub-frame occupancy, thus preventing other low-bandwidth users from
exploiting cell resources even in low load conditions. This may effectively overload the air interface even
with very few users, and other low-bandwidth services will suffer from lack of performance in the same
priority conditions as the greedy video user. In this case, the dynamic modification of bearer attributes
(and hence QCIs) or modification of attributes within a given QCI may be needed, or some traffic shaping
may be required at the core network gateway to limit the bit rate for a given mobile video application.
An additional scheduling dimension appears in LTE when so-called frequency selective scheduling (FSS)
is exploited. Due to the ability to grant specific frequency bands depending on the channel state
information as reported by the users, the scheduler can select a given subset of frequencies in which the
best possible channel conditions are met according to the radio environment. Mobile radio impairments
such as fading and shadowing introduce deep variations in the signal quality even in static conditions,
due to movement of the surrounding objects and obstruction of the radio link. These impairments create a
frequency-selective channel response full of dips and fades, which can be exploited by selecting only
those frequencies that present an appreciable peak response. The main implication for mobile video
delivery is that the granted bandwidth will not be constant throughout the connection, and this may
introduce impairments in constant-rate video services if the video stalls due to the player’s buffer
29
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
emptying. HTTP Adaptive Streaming can help circumvent this problem by lowering the quality of video at
the user device to an acceptable level without resulting in freezing of the video frame.
3.1.5 SIGNALING CHAL LENGES FOR VIDEO
The “signaling storm” problem caused by increased smartphone penetration in 3G and 4G wireless data
networks refers to the disproportionate amount signaling resources consumed relative to amount of data
traffic carried by the network. In cellular networks, substantial number of signaling messages is generated
prior to data transfers to activate the radio bearers that are in idle state. When most data transfer
sessions are short-lived, a substantial amount of signaling is generated in relative proportion.
Smartphone signaling traffic stems from a plethora of applications that rely on short data transfers for a
variety of reasons. Some of the common causes for short data transfers are application and TCP keep
alive messages sent to maintain long-lived application or TCP sessions, application notifications sent
from servers to the smartphone apps, status, presence or location updates, short instant messages or
Twitter type posts. However, video has distinct traffic characteristics with significant amount of data being
downloaded nearly continuously and generally does not cause disproportionately large amount of
signaling traffic.
A poor choice of the inactivity or dormancy timer that prematurely drives the radio bearer from active state
to idle state can cause unnecessary signaling in video streaming sessions. This is because highly bursty
video sources can create significant inactivity periods between chunks of data. As an example, the
YouTube video service delivers chunks of video according to a bursty pattern, characterized by significant
peaks followed by relatively large inactivity periods [16]. If the state transition timer is not larger than the
gaps in video transmission, the network can move the device to a dormant or idle state. Subsequently,
when new video chunks are to be delivered to the device, a new radio connection must be established,
with the corresponding signaling traffic and switching of states and this situation may repeat during the
whole video session thus creating many signaling events.
Furthermore, every transition between idle and connected states involves a large number of RRC
messages exchanged through the air interface. This takes an appreciable amount of time as well as
significant RNC processing. These delays increase the effective delay of the end-to-end connection, and
if accumulated over time they can force the player’s buffer to empty, thus resulting in video playback stall.
Streaming video services with limited buffer capabilities will additionally experience increased jitter and
therefore impaired user experience due to signaling storms. However, this particular issue can be easily
mitigated by setting the inactivity timer value in the range of 8 to 10 seconds since with HTTP Adaptive
Streaming, the gap between requests of chunks is a video transmission is unlikely to be larger.
In summary, signaling traffic generation is not expected to be a major challenge presented by video
applications provided the right timer value settings are configured.
3.1.5.1 IMPACT OF SIGNALING ON BATTERY CONSUMPTION
Figure 9 shows the effect that various video playout buffer algorithms have on network load and battery
consumption. This is shown through four examples.
30
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 9: Effect of video playout buffers on Relative network load (X-axis) and Relative battery consumption (Y-axis) [17]
In all four examples, the following legend applies:


Red lines correspond to download of video data.
Black lines correspond to signaling state transitions for the device between different WCDMA
activity states e.g. IDLE, URA/CELL-PCH, FACH and DCH.
On the main graph, the x axis is “Relative radio network load” and the y axis is “Relative battery
consumption”. The x axis in each of the four examples is “Time in seconds” and the y axis is “Throughput
in Mbps”.
Consider the various use cases shown in Figure 9. The arrows in the figure point to the respective graphs
referred to below.
1) Short Bursts (upper right corner): Here the playout client algorithm plays pulls video from the
network with multiple short bursts using multiple network accesses with small video downloads at
each network access. This “nibbling” at network resources results in multiple short transitions
between states, and small data downloads. This results in high network load and high battery
consumption due to the high signaling traffic.
2) Trickle Download (bottom right corner): Here the playout algorithm accesses the network for
longer periods of time, with a low amount of downloads during each access. Both the network
31
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
load and battery consumption is reduced compared to the upper right example due to the more
optimized network accesses.
3) Download / Progressive Download (upper left corner): In this example, the signaling
transitions are minimized, and video download is sustained. However, this algorithm is inefficient
for the majority of cases where the end user does not watch the entire video since unnecessary
video data is downloaded, thus resulting in waste of battery resources.
4) Adaptive Streaming (bottom left): This is the most efficient of all algorithms from a relative
battery consumption and network load perspective. Here the algorithm downloads an amount of
video data that is statistically adequate for the average end user, and minimizes radio bearer
state transitions while doing so.
It is therefore important that the design of video download algorithms be done prudently based on efficient
use of network and device resources.
3.3 CHALLENGES FROM THE NETWORK
Video delivery can pose considerable challenges outside the RAN, in the backhaul and core network.
3.3.1 BACKHAUL CONGE STION
The term “backhaul” refers to the transport network that connects the core network nodes with the radio
access network and represents a crucial part in any wireless architecture, as shown in Figure 10. In most
advanced wireless networks, the backhaul may become the bottleneck given the huge bandwidth
requirements that result from the aggregation of multiple cells, unless appropriate dimensioning is made
in order to limit congestion. Multiple links are usually multiplexed in the aggregation part of the transport
network prior to connecting to the core.
Figure 10: Typical backhaul architecture [18]
Backhaul capacity should be able to absorb the peak cell throughput in low load situations, where one or
few users in very good radio conditions can be granted full cell resources. Since mobile video applications
require significant data bit rates, appropriate backhaul dimensioning is critical to the quality of experience
for mobile video users. In the aggregation network, a number of backhaul links are multiplexed prior to
connection with the core network and the resulting backhaul costs would clearly be not affordable if the
peak cell throughput is to be guaranteed for all the cells. Instead, when the number of sites grows, it is
unlikely that all resources are simultaneously served at a given instant, and appropriate dimensioning
32
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
based on statistical multiplexing of the cells’ traffic is therefore applied. Backhaul congestion may
however appear whenever traffic exceeds the backhaul capacity at some point.
NGMN provides some practical guidelines for provisioning of the backhaul network. Given N number of
cells, one common approach for the minimum required backhaul is:
Backhaul provisioning for N cells = max (N x busy time mean, peak)
The busy time mean corresponds to the average throughput served by each cell in the busy hour, where
multiple users with different radio conditions are simultaneously active. The peak cell throughput is
usually the 95-percentile user throughput under light load conditions. These values should be affected by
appropriate overhead from the control plane, OAM, synchronization, transport protocol and IPSEC (if
applicable).
The above formula states that, for large values of N, the average traffic should dominate the backhaul
capacity instead of the peak traffic. The difference between them is illustrated in Figure 11. Simulations
conducted by NGMN members for typical LTE scenarios show very low throughput values in the busy
time: as an example, a typical 150 Mbps LTE cell with 2X2 MIMO in downlink would reach a mean
throughput of less than 25 Mbps per sector. This high peak-to-mean ratio suggests that significant
aggregation gains are available at the transport network.
Figure 11: Cell throughput during busy time and quiet time [18]
Backhaul congestion may have different consequences depending on the type of video application. In
streaming video services, congestion can result in instantaneous reduction of the available bit rate, thus
appearing as pixilation and other visible artifacts. In HTTP Progressive Download over TCP, the TCP
protocol would react to congestion by reducing the server bit rate, but this could be absorbed by the
client’s buffer with no appreciable quality degradation if the congestion episode is isolated. In HTTP
adaptive streaming, the instantaneous bit rate would be reduced accordingly, thus avoiding significant
degradation unless congestion forces the bit rate to decrease much below the screen resolution.
The majority of the end-to-end delay is expected at the radio access part (due to retransmissions,
scheduling, fading etc.) and much reduced delay budget values would be advisable for the backhaul
network. Based on select 4G Americas vendor implementations, the recommended delay budget values
for the backhaul are well below 20 ms for conversational and streaming video, less than 20 ms for
medium-quality videos and less than 5 ms for high-quality videos.
33
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Digital video decoders used in streaming video receivers need to receive a synchronous stream, with tight
jitter tolerances of approximately only +/- 500ns [19], in order to decode without visible impairments. Such
jitter tolerances may not be achievable natively in packet data networks. Hence, broadcast video services
use de-jitter buffers (also called playout buffers) to remove delay variations caused by the network.
Furthermore, the jitter experienced by mobile users, if not compensated by appropriate buffers, can be
alleviated at the network side by the introduction of traffic shapers at the gateway nodes located at the
boundaries of the mobile network. These traffic shapers can transform undesired bursty traffic into more
regular patterns which can be more easily managed by base station schedulers. Limitation of the
maximum bit rate according to the operator’s policies and user subscriptions can also be performed, in
order to prevent starvation of non-video users due to hungry video sources. Policy and charging rules can
be defined by the operator and stored in a core network node called PCRF (Policy and Charging Rules
Function), which interacts with other core network nodes for enforcement of the operator’s policies for the
various video users. Encryption can be performed at different layers and on different links. For instance,
the link between the RAN and the UE may be ciphered. There may also be encryption applied at the IP
layer, such as IPSEC.
3.3.2 DATA PLANE/CONTROL PLANE ENCRYPTION
Mobile video enables users to consume a variety of video content both on the go and while stationary.
For the streaming use cases, the content may be commercially produced or user generated. Because of
usage policies on commercial content as well as privacy concerns for personal content, it may be
necessary to encrypt the data and control plane traffic. Here the control plane traffic refers to both the
logical signaling channels, and also the signaling over IP used by the video service providers. In some
cases, it may be necessary to encrypt the video content to enforce usage policies and user privacy.
3.4 CONTENT BASED CH ALLENGES
3.4.1 DIGITAL RIGHTS MANAGEMENT (DRM)
All video content travels through the data plane in order to arrive to the user UE from the source. The
source may be another UE or a video server. The DRM is the use restriction based on policy. In some
cases, it uses encryption, but not in all. In majority of DRM technologies, embedded watermarking is used
to prevent copyright infringements and multiple unrestricted reuses of contents. In order to allow
legitimate use of encrypted information, the Security Key Management and Distribution must be
deployed. Without it, the confidentially protected video data and control information is reduced.
3.4.2 ENTITLEMENT/CONTENT MANAGEMENT RIGHTS
For video playback control to be implemented, a control protocol must be used. In this case, this protocol
may need to be encrypted as well for privacy concerns. This encryption can protect against rogue actions
such as: reverse engineering what programs a particular user may be watching.
Control protocols are often proprietary to each video service provider. It is generally up to the video
service provider to provide this protection to the control protocols.
34
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
4. VIDEO QUALITY OF EXPERIENCE
4.1 VIDEO USAGE BEHAVIOR AND EXPECTATION OF EXPERIENCE
An important consideration while devising the best video optimization strategy is to understand the typical
mobile user’s video usage behavior and their expectations. At home or at work, most users connect to
Wi-Fi connection and hence their video traffic is rarely seen on the mobile network. But when the person
is waiting for a train in the subway, or when the person is waiting in the queue on a local supermarket
he/she is likely to quickly take a peek at what’s happening around the world. Thus, most mobile video
sessions are of short duration (5 – 10 minutes long), and often the video is not watched in its entirety.
Similarly, from an expectation standpoint, the key metric of relevance to the user is a smooth and
uninterrupted video delivery. This alone, more than anything else, determines the subscriber’s satisfaction
rating [20].
4.2 METRICS FOR CHARACTE RIZING VIDEO QOE
Quality of Experience (QoE) is concerned with the overall experience of the video consumer. It goes
beyond the Quality of Service (QoS) used to deliver the video. To end users, quality is a subjective term –
“they know it when they see it.” In the multimedia world, the measure Mean Opinion Score (MOS) is often
used to quantify quality of experience.
MOS is a numeric value between 1 (one) and 5 (five). It was originally developed for bench marking audio
transmission and reproduction. Five meant the listener could not detect any impairment. One meant the
impairments were very annoying. Clearly, this is a very subjective measure. By taking the mean of a
large enough sample of opinions, a reasonably accurate measure of QoE can be achieved. Before
considering the application of MOS to video, it is important to identify the issues that can affect the end
users QoE.
Quality of the Encoding – The original video source can be in very high quality. However, in order to
deliver through mobile networks, it needs to be compressed. Even a standard definition uncompressed
video signal can require 270 Mbps, while high definition video can require 1.2 Gbps without compression.
This process is lossy and will introduce artefacts into the video image. Modern video compression
algorithms and associated pre and post processing steps are very effective in encoding the video.
However, typical problems include:



Blurring – This can be caused by unfocused camera lens when the material was originally
captured. It can also be introduced by the encoder. The effect is for a part of the image to have a
loss of detail that is perceptible.
Blocking – This is caused when a part, or all, of the image is made up of blocks of the same
colour. The effect is the appearance of very large pixels. It is often most noticeable on lines.
Quantisation errors – Where the original image has an expanse of similar colour but gradual
change in luminosity, the decoded and rendered image consists of jumps in colours.
Some scenes are very difficult to encode. When compression is high, scenes with a lot of information
including hard lines at an angle often show blocking errors. The sky and other areas of similar colour but
different luminosity show up quantisation errors. High motion is another challenge as it requires more bits
to describe how parts of the image have moved frame by frame. However, some motion is relatively easy
to encode because large parts of the scene move together – panning across a landscape for example.
35
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
The very hardest are scenes with a lot of information and chaotic motion (waves on water and grass/trees
moving in the wind).
Device – The screen size has a significant impact on QoE. The same encoding viewed on a smartphone
can look adequate, poor on the larger screen of a tablet and totally unsatisfactory on even a modest TV.
Viewing Distance – The distance the display device is held from the eye makes a big difference on the
ability to resolve the image. The closer the display device is to the eye, the higher the quality of the image
has to be. Note, watching a video on a tablet at conventional distance has a very similar viewing angle as
watching a large TV from the couch.
End User Expectations -- If people pay for the content (PPV, subscription, etc.), then their expectations
of video and audio quality are high. Also, expectations will be high for serious entertainment content (TV
show, film, and sports events) compared with short YouTube clips. User expectations vary depending on
the type of content and over time, they become more demanding and expect improved quality.
Playout Issues – Even if the content is prepared to meet the quality expectations for users on a
particular display device, network issues can cause problems which manifest themselves in visual and /
or audio artefacts. Buffering results in a streaming media being "preloaded" to allow for smooth playback.
Dropped IP packets can result in picture breakup unless they are masked. Network issues can cause
jerky playout (low frame rate), skipping (especially in live streams) or even picture freezing (due to
rebuffering which is the reloading of a stream to prevent skipping). The latter must be avoided if at all
possible [20]. Usually less annoying, but still important to the overall QoE, is lip sync. If the audio is out of
synchronization with the video, the experience is of watching a badly dubbed film.
System Responsiveness – There is a certain maximum acceptable time beyond which if there is a
significant delay between requesting a video and it is starting to play out, then users will experience
dissatisfaction. Likewise, the system needs to be adequately responsive to channel change in order to
meet user expectations. Hence a streamed solution is generally preferred over (progressive) download.
Ease of Use – The complete experience involves discovering the content to be viewed and optionally
paying for the content. Providing a user interface making it quick and easy to find the desired content is
essential to a good QoE.
Varying Image Quality – To handle differences in available bandwidth, many video solutions use some
form of adaptive streaming. The consequence is that when the available bandwidth is low, the image
quality will become worse. Conversely, improvement in bandwidth will result in improvement of the image
quality. Both the uncertainty and change in image quality will impact the QoE. It is still an area of research
on how much temporary poor image quality impacts QoE.
Finally, IEEE reference [21] demonstrates the relationship between viewer QoE (quality of experience)
and MOS (Mean Opinion Score) and the HTTP Adaptive Streaming Video Quality (VQ) rate. Rather than
using mean MOS scores to judge the Video Quality (VQ) rate, it is shown that MOS scores are impacted
by the variance (or standard deviation-SD) of the VQ rate more so than the mean VQ level. In other
words, users care more about maintaining a consistent VQ level than they do in a higher bit rate. It is
shown that one can accurately predict MOS scores incorporating both the mean VQ level and the
standard deviation of the VQ level.
36
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
4.3 BEHAVIOR OF VIDEO CH UNKS WITH TCP/IP
HTTP adaptive steaming (HAS) is becoming the ubiquitous and reliable method of streaming video
content over service provider networks to a sizable number of classes of devices, including smartphones,
wireless tablets, laptops and desktop personal computers. The HAS client continually senses the
available bandwidth and adjusts the bit rate of the requested chunks to ensure the playout buffer does not
run dry. While HAS provides a better implementation over the traditional HTTP progressive download
mechanism by dynamically adjusting the video quality to match the available bandwidth (see section 2.6),
the nature of the protocol has several impacts on the network.
The performance of HAS depends on the client implementation, which is not at all defined by the
standards. A well-designed HAS client will try to get the highest quality possible, but it will also back off
when there is competing traffic, including other video streaming clients. A well-designed HAS client
should provide the best experience possible given current network conditions, but should not try to
provide better playback than is possible based on current network conditions. The HAS client is based on
HTTP, and thus cannot be any more greedy than any other HTTP-based application. If the HAS client
tries to be greedy and download a bit rate that is too high to be supported by the current download rate
available using HTTP over the network, then the end user will have a bad experience. This is because the
content will not be downloaded fast enough to playback in full time, and thus the playback will stall or
skip. Thus, a well-designed HAS client will adjust which bit rate it is downloading to match the current
download rate of HTTP and provide a good user playback experience. A badly designed HAS client will
provide a poor end user experience, but will not congest the network since it is using standard HTTP -same as all other applications downloading content over the network. This is one of the reasons that
HTTP Adaptive Streaming is especially well-suited for streaming to mobile devices.
Policies should be based on providing proportional fairness (however that is defined) to the mobile clients,
and probably most, if not all of them, are using HTTP to download content. A HAS client is subject to
these policies, just like all other applications, and if the HAS client tries to be greedy and download too
high of a bit rate of content then the end user suffers, not the network, as the playback of content will skip
and/or stall. However since the HAS client is using HTTP to download, it will not be able to greedily take
over the network, and the network can control the total amount of traffic that the mobile device receives.
The performance of HTTP Adaptive Streaming with some implementations that include poorly designed
HAS clients may suffer due to the following issues.
Unfairness and Instability: The HAS client continually senses the available bandwidth and adjusts the
bit rate of the requested chunks to ensure the playout buffer does not run dry. The problem is that a client
only senses the available bandwidth when it is downloading a chunk. This can result in unfairness. For
example, if one client is already downloading a chunk at a high bit rate, then the second client will sense
only limited capacity and will chose to download a small chunk. This imbalance in bandwidths can remain
for some time. Similarly, two clients can oscillate in their selection of chunk bit rate requests as their
downloads periodically overlap and are distinct. Proper network based policies can mitigate these
problems.
Under Utilised TCP throughput: Some issues still exist with the efficiency of delivering HAS over the
reliable TCP transport layer that will need to be overcome to better utilize the resources of the bearer
channels. When comparing TCP versus HAS throughput, it appears that TCP theoretical throughput is
often greater than HAS TCP throughput, in some cases, by as much as twice. The unutilized TCP
37
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
capacities exist in HAS thus limiting the performance of HAS, thereby reducing the quality of experience
for viewers. Looking at the HAS performance with respect to round trip time delay (RTT), RTT
dramatically impacts the TCP throughput and further reduces the HAS bandwidth HAS client.
Figure 12 compares the performance of HAS TCP versus theoretical TCP [22].
Figure 12: Performance of HAS TCP versus Theoretical TCP [22]
The reason for the lower HAS TCP performance is the interaction between the HAS rate determining
algorithm (RDA) and the TCP congestion avoidance algorithm. Since HAS relies on the measure of the
available bandwidth to decide which bit rates chunk to request, HAS clients either present a delay before
starting playing the video or start immediately playing the lowest available quality. As video chunks get
downloaded, the bandwidth estimation is adjusted. This slow start estimation coupled with a possibly
conservative approach by the HAS RDA greatly impacts the end-user experience, since reaching the
optimum rate for the current network conditions can take up to 20-30 seconds. After the initialization
phase, HAS is then prone to suffer from the “downward spiral” [23], shown in Figure 13, whereby the
client ends up requesting the lowest bit rate when competing flows or network congestion result in
bandwidth fluctuations. This spiral towards the worst quality comes from a constant under-estimation of
the available throughput due to the combination of several factors, like periodic request pauses to let
drain the playback buffer, repetitive TCP congestion window reset, slow start TCP ramp-up and smaller
video chunks requests over time. Determining the fair share of band-width available at the bottleneck is
precisely the role of TCP, but this information is lacking at the HTTP layer.
38
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 13: Downward spiral behavior of video throughput [23]
Clearly, the performance of HAS is very dependent on specific client implementations, and well designed
client HAS implementations can get around some of the issues and help provide good Quality of
Experience for mobile video users.
Finally, some versions of TCP cannot distinguish between packet loss caused by congestion and packet
loss that results from noise (transport errors). As a result, TCP often cuts window size in half when it
should not, thus decreasing its efficiency over noisy channels such as wireless links. Service providers
could also take advantage of Early Congestion Notification (ECN) and Selective Acknowledgements
(SACK) mechanisms to make TCP more robust. ECN helps improve efficiency in a wireless environment
by allowing TCP to partially distinguish between packet loss due to errors and packet loss due to
congestion. Instead of waiting for packet loss to signal congestion, TCP is notified when a buffer is filling
up. With Selective Acknowledgements (SACK), to avoid excessive retransmission of packets, the TCP
receiver more carefully indicates which packets were lost, rather than retransmitting everything, starting
with the first lost packet that was reported. This approach is especially valuable when the packet’s roundtrip time (RTT) is large. Additional details are available in RFC 3481 [30].
Wireless links can also be protected by link layer mechanisms that are designed for unreliable links.
Forward Error Control (FEC) and Hybrid Automatic Repeat ReQuest (HARQ) help reduce packet loss
between the endpoints of the link. However, they also introduce additional delay and overhead bit rate.
Service providers need to evaluate how well this trade-off meets their own unique requirements.
39
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
HAS/TCP interactions is an area of active research by various authors including [11], and an interested
reader is referred to various contributions from these authors and others. Careful network conditioning,
new mechanisms/policy and improved client rate determining algorithms can help overcome many of
these issues to reduce their effect and loss of performance on E2E QoE. Through subjective testing, it
has been shown that HAS delivers an acceptable video quality over a range of typical network conditions.
Thus, HAS offers a cost-effective way to deliver more video content to more customers, while reducing
costs through efficient use of network resources.
4.4 MONITORING OF VIDEO QOE
In order to determine a mean opinion score (MOS) value, you need to get a number of people to view and
rate a set of clips on a scale 1 (poor) through to 5 (excellent). The Video Quality Experts Group (VQEG)
has published guidelines on how to measure MOS. The problem is that this is a subjective measure, and
the radio network is a dynamic delivery network. The available radio resource assigned to a terminal
varies continually depending on the radio condition of the terminal and on all the other terminals in the
same cell. Hence, an objective measure of QoE is needed.
There are three classes of monitoring Video QoE as follows.
1) Full Reference – In this approach, the final displayed video after encoding, transmission and
decoding is compared with the original source material. This allows for very accurate
measurement of impairments. However, it requires access to the original material and involves
significant amount of processing. It is used by broadcasters to ensure the integrity of their
encoding and transmission equipment where high quality is essential.
2) Reduced Reference – In this approach, some information about the original material is used. For
example, checking the integrity of a watermark inserted in the original before any processing. The
degree to which it has survived is an indication of video quality.
3) Non-Reference – In this approach, no information about the original material is used. Quality is
assessed by analyzing the video itself and/or the underlying network transport. This can be done
in real-time with small processing effort. It can identify blocky and jerky videos.
There are two common Full Reference objective measurement techniques, PSNR and SSIM. PeakSignal-to-Noise Ratio (PSNR) is a simple measure of image quality based on the error between the
original and final decoded image. PSNR measures the fidelity or how similar the sequence is to the
original. It is an objective measure which is easy to compute and is well understood. However, the
correlation to subjective quality measure is poor. PSNR only considers the luminance component and
neglects the chrominance component, which is also important to human perception. Furthermore, it does
not do a good job of comparing different sequences.
Structural SIMilarity (SSIM) is a more advanced objective measurement. While PSNR measures errors
between the original and optimized image, SSIM measures the structural distortion, luminance and
contrast differences between the two images. The idea behind SSIM is that the human vision system is
highly specialized in extracting structural information from the viewing field but is not specialized in
extracting errors. Thus, a measurement on structural distortion should give a better correlation to the
subjective impression. However, the mathematical complexity of this calculation is comparatively high and
the SSIM calculations might vary from that of the subjective measurements.
Both PSNR and SSIM require access to the source material. Therefore, they can be used for operators
who are providing their own video service. They are not suitable for OTT video streamed through the
40
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
mobile network. This is becoming increasing popular with the extension of TV Everywhere to mobile
devices and LTE, thus making it feasible to receive good quality content.
Non-Reference approaches can give good results. For professional content, a common approach for
streamed content is with some form of HTTP Adaptive Streamed (HAS) content (DASH, Apple HLS,
Microsoft Smooth Streaming).
As described in Section 2.5, in HAS, the content is divided into short non-overlapping intervals of between
1 and 10 seconds long. Each interval is then encoded in multiple bit rates which are then sent over the
radio interface in chunks. The HAS client uses HTTP to request chunks. By monitoring the sequence of
HTTP requests, it is possible to reconstruct the chunks asked for in an individual session. From this, it is
possible to reconstruct the entire stream. Work by the NGMN P-SERQU project [24] shows how to map
the quality level of the requested chunks to a predicted MOS value for the session. This approach works
even for encrypted content.
3GPP in their recommendation 26.247 has defined the following parameters which should be monitored
for HAS content. The sections below refer to the sections in 3GPP 26.247.







List of HTTP Request / Response Transactions (Section 10.2.2). This records both when the
request is made and which quality level is requested. It allows a reconstruction of the quality of a
session including estimating if there was a hesitation in playout.
List of Representation Switch Events (Section 10.2.3). Quality of Experience is affected by
changes in quality. With HAS, these changes can be abrupt and very noticeable.
Average Throughput (Section 10.2.4). For any one codec, the higher the bit rate the better the
image quality. However, popular codecs such as AVC (H.264) are highly non-linear.
Initial Playout Delay (Section 10.2.5). If the delay in requesting an on-demand event and payout
is much more that 2 seconds, QoE is significantly reduced. For channel change to live content, it
is sub 1 second.
Buffer Level (Section 10.2.6). If the buffer level in the client drops to zero, it results in a freeze
which is very annoying and should be avoided.
Play List (Section 10.2.7). This is the information on the content being played.
MPD Information (Section 10.2.8). The information in the manifest file is used to reconstruct a
session.
Work is ongoing in the NGMN Project “Service Quality Definition and Measurement” P-SERQU [24] to
determine the Video Quality of HAS in a mobile environment. The aim of P-SERQU project is to evaluate
the end user acceptance of streamed video over a next-generation (LTE) wireless network using HAS
technology. Progressive download is excluded from the studies based on the observations of industry
direction that favors HAS for paid video streaming services. From the outcome of the test approaches
proposed in P-SERQU project, it is expected that it will be possible to correlate network level impairments
with end user QoE.
Monitoring video at the RAN level is generally not feasible. This is because of the interaction of TCP
congestion avoidance algorithm and the HAS rate determination algorithm. Also, it is often difficult to get
access to the necessary data in the eNodeB. Main factors that influence end user QoE of streamed video
are identified [24], including device and media factors and network factors. Monitoring of Video QoE
through monitoring the quality of HTTP chunks rather than individual radio and TCP parameters is
thought to be the preferred viable option.
41
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Since HAS uses HTTP which runs on TCP, IP packet losses due to network impairments are masked by
automatic retransmission. So there should not be any visual and audio artifacts caused by IP packet loss
unless the period of extreme link impairment resulting in total outage exceeds the buffer time. Rather, the
effect of network impairment is to reduce the TCP bandwidth. The HAS client adapts to the reduced TCP
bandwidth by selecting chunks of a lower bit rate. Hence, there should be no pause in playout while the
client rebuffers as long as the link outage doesn’t exceed the buffer time. However, the end user will see
a change in video quality as the RDA optimizes the highest bit rate is can pull through the network while
maintaining a buffer of video (typically 30 seconds deep). The lab tests in P-SERQU [24] should show
how much the impairments impact HAS/TCP bandwidth, and the mass tests that are a part of that study
will reveal how much the changes in the HAS chunk bit rate results in reduced MOS.
4.5 SPECIFIC ATTRIBUTES THAT CONTRIBUTE TO I MPROVED VIDEO QOE
In principle, a higher network bandwidth results the higher the QoE. However, RAN bandwidth is a scarce
shared resource. Therefore, it is sensible to choose encoding bit rates which provide adequate image and
audio quality on the target devices. For most content, this is 200-500 kbps for smartphones and 500 kpbs
to 1 Mbps for tablets when using H.264 video encoding and stereo audio. Of course, screen resolutions
are improving so we can expect increased demands for improved quality, especially if people use their
tablets to stream to TVs.
A video stream lasts for several minutes up to hours. This puts significant pressure on the demand for
sustained bandwidth. Playout freezing must be avoided. This means that when using adaptive streaming,
at least the lowest quality level must be maintained despite poor SINR and high cell congestion.
Maintaining an even bit rate at the RAN level can help eliminate variable quality levels in the viewed
video. Furthermore, minimising the impact of dropped IP packets which results in either image impairment
(RTP/UDP based video transports) or reduced throughout leading to reduced quality (TCP based
transports) is critical. Hence, using GBR or QoS to favor forwarding of video related packets will improve
QoE but at the expense of best effort data traffic.
Cell handover places particular demands on mobile video. The algorithm used in the buffering or playout
of video from a source in the cloud can largely influence:



network bandwidth use;
battery usage associated to usage of resources;
QoE as end users either wait for buffered video (as video plays) or suffer a delayed start in video
play (in case of larger playout buffers).
Through a combination of techniques that are available now, and are being developed, it is expected that
mobile video QoE will be acceptable and sufficient to support a large number of mobile video users.
4.6 NEW/UPCOMING CODECS (HEVC)
High-Efficiency Video Coding (HEVC) is currently being prepared as the newest video coding standard of
the ITU-T Video Coding Experts Group (VCEG) and the ISO/IEC Moving Picture Experts Group (MPEG).
The first edition of the HEVC standard was finalized in January 2013, resulting in an aligned text that will
be published by both ITU-T and ISO/IEC. Additional work is planned to extend the standard to support
several additional application scenarios including professional uses with enhanced precision and color
format support, scalable video coding, and 3D/stereo/multiview video coding.
42
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
HEVC is a successor to the current state of the art video coder - H.264/AVC video coding standard. Since
AVC was first standardized in 2003, some previous trends have sharply accelerated and totally new
trends have appeared which are creating even stronger needs for coding efficiency superior to
H.264/MPEG-4 AVC’s capabilities. The traffic caused by video applications targeting mobile devices and
tablets, as well as the transmission needs for video on demand services, are imposing severe challenges
on networks. There is an increased desire for higher quality and resolutions for mobile applications.
The results of subjective tests for Wide Video Graphics Array and HD sequences indicate that HEVC
encoders can achieve equivalent subjective reproduction quality as encoders that conform to
H.264/MPEG-4 AVC, by using approximately 50 percent less bit rate. The HEVC design is shown to be
especially effective for low bit rates, high-resolution video content, and low-delay communication
applications.
HEVC adheres to the same hybrid video coding structure as H.264/AVC. It uses spatial and temporal
prediction, transform of the prediction residual, and entropy coding of the transform and prediction
information. Improved coding efficiency gains in HEVC compared to H.264/AVC are achieved with
minimal increase in decoder complexity. It is estimated that HEVC decoder complexity increased 20
percent-30 percent comparing to High Profile of H.264/AVC, depending on architecture. In some
respects, HEVC is even more implementation friendly than H.264/AVC. One of the H.264/AVC
bottlenecks for hardware implementations is CABAC decoding. Context-adaptive binary arithmetic coding
(CABAC) is a form of entropy encoding used in H.264/MPEG-4 AVC video encoding and is a lossless
compression technique. CABAC is highly sequential and has strong data dependencies which make it
difficult to exploit pipelining. Numbers of modifications were introduced in HEVC CABAC design to enable
its parallel decoding. HEVC has been also designed to better support parallel processing architectures
by adding parallel processing tools: Tiles (picture partitioning into rectangular non-overlapping regions)
and Wavefront Coding.
5. ARCHITECTURES FOR MOB ILE VIDEO DELIVERY
5.1 CELLULAR (HET-NET) AND WI-FI OFFLOAD ARCHITECTURE S
In response to the rise in traffic demand, driven largely by the expected growth in video traffic, more radio
spectrum is becoming available and more spectrally efficient technologies like LTE are being introduced.
Operators are engineering their networks to provide increased capacity and coverage while maximizing
the use of their limited existing spectrum assets. LTE is being deployed in areas where traffic demand is
high. Such areas are typically already served by existing 2G/3G networks that support a mix of voice and
packet data. LTE allows operators to use new and wider spectrum to complement existing 2G/3G
networks with higher data rates, lower latency, and a flat IP-based architecture.
Considering the significant increase in traffic demands anticipated within the next few years, it is clear that
the improvements that LTE macrocells alone bring will not suffice to meet demand. Operators will need to
address the capacity crunch through a combination of more spectrum, increases in spectral efficiency,
cell densification via small cells and heterogeneous networks (HET-NET).
Both HSPA and LTE technologies are near the Shannon bound for the cases of slowly-varying channels
for which fading can be compensated by practical channel estimation algorithms. Link-layer performance
has therefore little room for further improvements, and present investigations are more focused on
increasing the experienced Signal to Noise ratios (SNRs) in the system. One way for improving the SNR
is to increase the density of sites by introducing a collection of small cells under the eventual coverage of
43
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
the macrocell, equipped with low power base stations at specific locations. These small cells can increase
capacity and coverage by enabling better SNR values in the nearby, resulting in heterogeneous networks
with enhanced area spectral efficiency [13].
Heterogeneous networks are characterized by the coexistence of macrocells and smaller metro cells,
potentially increasing the number of cells tenfold or more in some cases. Heterogeneous networks utilize
diverse technologies and base station types to improve spectral efficiency per unit area. They provide
ubiquitous coverage along with the high-bandwidth capacity to deliver a superior QoE.
The move towards heterogeneous networks is being supported by work done within 3GPP. Within 3GPP
specifications, LTE-Advanced wireless networks are being defined to improve spectral/spatial efficiency
and improve the overall throughput by shrinking cell size via deployment of a diverse set of base-stations.
LTE facilitates the deployment of heterogeneous networks since LTE has characteristics, as below, that
ease the deployment of small cells that will result in more ubiquitous usage of small cells:






Spectrum flexibility – LTE leverages more frequency bands so more spectrum is available;
Improved spectrum utilization - flexible carrier bandwidth, support of carrier overlapping partly or
totally with another carrier, support of carrier aggregation;
Flexible resource management – carrier resource flexibility and the easing of mixing various types
of cells using the same frequency resources;
Full IP architecture - native support of flat IP architecture with low latency and standardized
communication interfaces (S1, X2);
Improved coordination - through better handover and load balancing abilities and through
interference management features such as eICIC (enhanced Inter-Cell Interference Coordination)
and its variants in Rel-8/Rel-9 and Rel-10. These features allow more users to be offloaded to the
picocells to leverage the additional air interface resources, while still being able to address the
additional interference to those users who may be in poor SINR conditions.
Separation of the control plane from the bearer plane in LTE allows for independent scaling of
corresponding network elements, aligned with the unique traffic demands.
In a broad sense, heterogeneous networks comprise not just of 4G macros and small cells but also of 2G,
3G and 4G cellular technologies along with their respective radio access options namely, macros and
small cells, on same or different carriers, as shown in Figure 14. 3G Metro cells help alleviate congestion
by providing capacity offload in dense urban environments with existing device ecosystems. They may
also be used to provide indoor coverage, for enterprise users and for home users, where the high building
penetration loss may result in the inability of the macro to provide adequate indoor coverage. Generally,
video traffic is consumed by stationary/pedestrian users and hence the deployment of small cells to
provide the air interface access for such users will significantly help alleviate macro congestion.
44
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Operator
IP services
SGi
Rx
HSS
S6a
MME
PCRF
S1-MME
Gx
S11
LTE RAN
Macro/
Pico/Femto
Home
Macros
4G Femto
4G Pico
S1-U
RNC/
SC GW Iu SGSN
3G RAN
Macro/
Metro/Femto
SGW
Home
Macros
3G Femto
S5/S8
Gn
PGW
GGSN
3G Metro
To PCRF
ANDSF
server
WiFi APs
Figure 14: HET-NET Architecture
Through pooling of radio resources and through traffic management, the UE can be assigned to the
appropriate technology, carrier and cell type depending on its mobility, service required and its location. In
doing so, the QoE of all the various users is best met and a mix of technologies and access options can
efficiently satisfy capacity and coverage needs in all environments.
With HET-NETS, the planned deployment of macrocells with a large RF coverage area (for the coverage
of urban, suburban, or rural areas) are complemented with small cells with targeted RF coverage for
coverage extension or network capacity enhancement, mainly in hotspot scenarios. Still the available
capacity is far short of the capacity needed. To address the need for higher air interface capacity
availability, mobile network operators are under increasing pressure to embrace small cells and Wi-Fi
traffic offload solutions. Wi-Fi offloading from cellular to Wi-Fi networks with unlicensed spectrum can
further help improve the performance (QoE) of the cellular users by conserving precious cellular
resources. Users can be offloaded from cellular to Wi-Fi in an area of Wi-Fi coverage, or when the traffic
on the cellular carrier exceeds a congestion threshold, or based on some specific policy mechanism using
3GPP based ANDSF mechanisms.
5.1.1 VIDEO USERS IN A HETEROGENEOUS NETWORK
With the introduction of heterogeneous networks, cell sites will no longer just be on towers – they will be
located on lamp posts, utility poles, and on the sides of buildings. A variety of access types, fiber, copper,
and microwave, will be needed to connect cell sites to the backhaul network. New backhaul topologies
will be required to support LTE and the shift to heterogeneous networks. All must be able to support
multiple types of wireless services including HSPA+, LTE and Wi-Fi.
45
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Video users will more often than not be either stationary or move around the network at low mobility. If
near a hotspot location, such users can be offloaded to a nearby picocell thus freeing up the macro
capacity and allowing the pico to provide high data rates to the video user. The Quality of Experience for
the video users served by picocells is likely to be higher than the corresponding QoE of video users
served by macrocells due to proximity of the pico to the user and hence better RF propagation
characteristics, including lower path loss. The bit rate and the RTT variations between macro, pico and
Wi-Fi is dependent on the specifics of the network deployment including choice of backhaul, spectrum
etc. Video users who are not nearby a picocell but are in a Wi-Fi hotspot location can be offloaded to the
Wi-Fi network, depending on the operator policies, assisted by Wi-Fi ANDSF mechanisms and HotSpot
2.0. Traffic management across existing 3G and 4G networks can further facilitate the optimal use of
radio resources (both bearer and signaling) across the various radio access technologies, frequencies,
and cell types for both data intensive users such as video users, and signaling intensive users.
Video users moving around the network at low mobility will experience a drop in the SINR as they move
through the interference zones between the macro and the picocells. However, the time taken for the
messages in the handover signaling to go through will likely be acceptable, in spite of the SINR
degradation, due to the low user mobility. Hence the video quality is not likely to suffer for users moving
through the network at low speeds. At high speeds, degradation in video quality is more likely to be
pronounced due to higher packet drops which then need to be corrected at upper layers. Higher speed
video users may choose to remain on the macro cellular layer without handing over to the picocells, if it is
a priori known that the user will remain in the range of the picocell for a limited period of time. Above a
certain user speed, good video QoE may be difficult to sustain due to the demands on throughput.
5.2 LTE BROADCAST ARCHITECTURE
It is well established that there are significant spikes in mobile data traffic in many markets across the
world due to major sporting events, breaking news, highly watched TV programs and celebrity events.
LTE broadcast, also known as Evolved Multimedia Broadcast Multicast Service (eMBMS), extends
existing LTE/Evolved Packet Core (EPC) systems with an efficient point-to-multipoint (PMP) distribution
feature. It allows operators to dynamically allocate network resources to offer highly flexible broadcast
services thereby alleviating the strain caused by demand for popular content over unicast. The returns
are particularly efficient when content demand is confined to a dense venue or is oriented to real time
mass consumption. It is a Single Frequency Network (SFN) broadcast / multicast mode within LTE known
as eMBMS which was defined in 3GPP Rel-8 and Rel-9. Enhancements to certain aspects continue in
Rel-10 and Rel-11.
Using LTE broadcast, MNOs can handle these spikes in mobile usage and also offer premium content
service in venues, city centers or nationwide, to create new revenue streams with well-controlled media
quality and efficient scalability. Delivering video streams to hundreds of users in a cell site by using LTE
broadcast will utilize almost the same bandwidth as a single user with the same video quality. The
guaranteed quality of experience can increase subscriber loyalty and deliver significant service
differentiation compared with the competition. There are multiple business models that MNOs can use,
identified in Table 3 below.
46
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Table 3: Potential use cases and business models for LTE Broadcast services [31]
There are two aspects to eMBMS. There are a set of modifications to the radio physical layer that enable
SFN operation within the already defined LTE operating modes and is implemented at the modem level.
eMBMS is designed to operate within the mobile operator’s LTE network infrastructure and dedicated
LTE radio spectrum. There is also an upper layer eMBMS framework that contains tools to enable
services across the broadcast physical layer, often referred to as a broadcast service layer. The service
layer includes support for file delivery and forward error correction (FEC).
The primary benefits of eMBMS relate to the increase in capacity that can be achieved via SFN operation,
and the one to many aspect of broadcast distribution. These two aspects combined can provide a
significant capacity lift for any content that is being viewed simultaneously by more than one user per
cell/sector. The specific capacity numbers supported over eMBMS depend on sub-frames allocated for
eMBMS configuration, and network configuration. Figure 15 shows a high level architecture for eMBMS.
47
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 15: eMBMS/LTE Broadcast Architecture
The components in the architecture outlined in Figure 15 below are:




MCE - MCE use Session Control Signaling (received from MME) to initiate configuration of radio
resources to be used for broadcast transmissions.
eMBMS GW - The MBMS-GW provides functionality for sending/broadcasting of MBMS packets
to each eNodeB transmitting the service. The MBMS-GW uses IP Multicast as the means of
forwarding MBMS user plane data to the eNodeBs. The MBMS GW performs MBMS Session
Control Signaling (Session start/stop) towards the E-UTRAN via MME.
MME - The MME provides Session control of MBMS bearers to the E-UTRAN access. It transmits
Session control messages towards multiple E-UTRAN nodes eNodeB/MCE and it receives
MBMS service control messages and the IP Multicast address for MBMS data reception from
MBMS GW.
BM-SC - The BM-SC handles eMBMS sessions (start, stop) and is responsible to deliver user
plane media to the MBMS-GW.
The ecosystem associated with LTE Broadcast is under development, and is expected to accelerate
when LTE Broadcast service is commercially launched. Some of the challenges for the industry include:
- Understanding the tradeoffs between dedicating spectrum for multicast technologies like LTE Broadcast
and smart use of unicast video over RF.
- Monetizing of LTE Broadcast with a selection of video services that differentiate multicast over unicast.
- Building an ecosystem of content, spectrum and devices to address this optimized way of delivering
video over RF.
48
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
The reader is referred to other papers that address the LTE broadcast issues in further detail [25, 26].
5.3 VIDEO DELIVERY ARCHITECTURE
Figure 16 shows a representative CDN architecture. A Content Delivery Network (CDN) is a system of
distributed caches containing copies of data placed at various points in a network so as to maximize
bandwidth for access to the data from clients throughout the network. A client accesses a copy of the
data nearest to the client, as opposed to all clients accessing the same central server. This avoids
bottlenecks near that server. Content types include web objects, downloadable objects (media files,
software, and documents), applications and real-time media streams.
Video delivery begins with content capture and ingestion. The video that is created at the time an event is
captured is compressed and transmitted to one of the content ingestion sites where video is archived and
meta data related to the video is created and stored. The content is then prepared for delivery by coding
the video into multiple different versions corresponding to different codec technologies, container formats,
resolutions, and content variances (e.g. trailers or short clips, embedded captions), etc. Each of these
versions may also be encrypted according to the Digital Rights Management technology adopted.
Appropriate content portals or catalogues that exhibit the content availability are also updated to reflect
the new content. Policies are defined for access restrictions and advertising policies that, respectively,
control access to the published contents and advertisement placements.
Content distribution to local video headends follows the content ingestion step. In this step appropriate
versions of video created during content ingestion step are sent over high speed networks to different
video headends that serve the video to the end consumers. Video delivery begins in the origin server in
video head end but typically involves additional CDN nodes that cache the content that are popular for
delivering to the end users. CDN nodes may also be responsible for transcoding content into format
appropriate for a particular device. Description of transcoding is provided in Section 6.4.2.
Figure 16: Content Delivery Network (CDN) Architecture
49
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Request for content from the end device is redirected to the CDN’s delivery server. The CDN server
processes the request from end device, extracts device information from a subscription data base and
picks appropriate content format to deliver the video, transcoding if necessary. The delivery server
maintains the session with the end device streaming the video as the end device consumes the content.
The video server may invoke APIs provided by the mobile core to establish QoS for the session.
5.3.1 MOBILE OPERATORS AND THEIR MIGRATION TO IMS
Mobile operators have the unique opportunity to consolidate multimedia services, and provide converged
services by leveraging their IMS Core resources. Figure 17 shows the rollout of a 4G Americas operator’s
IPTV service.
Figure 17: Rollout of an operator’s IPTV service
The top part of this diagram represents a view where IPTV is rolled out as a separate service, with
Mobile, Cable and FTTx access networks. The bottom half of Figure 17 represents the next generation of
this strategy and direction, to converge all services to a common core and control plane, to maximize
converged services delivery, enabling advanced video applications like teleconferencing, distance
learning, video security, and easily moving video entertainment between platforms.
50
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
The mobile operator has a significant task in acquisition, processing and packaging of the videos sources
they intend to deliver to their subscribers. Along with the negotiations, agreements, and contracts, they
will need to have a Contract Rights and Management System (CRMS) to help them automate and provide
business rules to their Content Management System (CMS). The CRMS provides rules for how the
content can be used, where, when and how the content is delivered, exceptions, exclusions, and other
details beyond the scope of this white paper. Some operators will do this function with a great deal of
manual steps. However, they will need to progress to automation, or an outside service providing that
automation, to keep up with both demand and new content available.
The CMS provides Metadata Management and ingests from multiple sources, like Internet Movie Data
Base (IMDB), Rotten Tomatoes for details about the content itself. Rotten Tomatoes is a website devoted
to reviews, information, and news of films. Electronic Program Guides from sources like Tribune, Rovi,
validates, transforms, enriches, and normalizes the content and asset metadata, for use in Workflow
Automation. The CMS, using workflow automation, business rules from the CRMS, operator rules,
manages encoder and transcoder farms, manages scramblers and watermarking for Content
Access/Digital Rights Management (CA/DRM), and manages Quality Assurance (QA) systems. The rules
engines trigger workflows based on rules and triggered based on metadata.
Linear and VOD content themselves may come from many sources, such as satellite signal reception, offair, direct feed fiber, Internet, and local file servers, and transport mechanisms such as Aspera and
Signiant. Real-time processing of encoding (MPEG4 AVC, MPEG4 SD and HD Profiles, future HEVC),
transcoding of MPEG2 to MPEG4, creation of Single Program Transport Streams (SPTS), including
traditional splicing for Ad Insertion is also required. The CMS can be involved in the file based
encode/transcode to multi-bit rate (MBR) profiles, containerization using Packagers to HLS, MPEG-DASH
or other operator supported containers, as well as pre-encryption/watermarking of content.
The Content Access and Digital Rights Management System can protect the linear video services using
Bulk Encryptors by Content Access client type. Offline (file based) VOD services can use scramblers by
DRM client type. The encryption can also be applied in the packagers such that various containers could
use different encryption systems depending on the client type being served.
While some mobile systems may use multicast delivery for linear feeds, most VOD content from the
content processing is delivered to the Content Delivery Network (CDN) origin server, and then the CDN
caching system, streaming systems and content routers, are used to deliver unicast streams. While the
Mobile operator uses their wireless RAN, these same streams may be delivered over other access
networks, for example on a DOCSIS Cable/HFC network of a Cable Operator through the CMTS, or via
an FTTX network through the OLT.
The mobile operator will also need an IPTV “headend.” This Converged Video Application System
(CVAS) will support the middleware and core TV Applications. The server-side middleware will include, at
a minimum, an Electronic Program Guide and VOD title catalog, channel maps, defined packages and
entitlements, and user management and authentication. This headend or CVAS will have applications to
support Linear TV, VOD and nPVR. It will also provide remote authorization; enable advertising (system
wide or targeted). We have only mentioned a few of the 100s of features of a modern IPTV headend.
The headend or CVAS will also provide (directly or interfaces to) Service Provisioning Interface, Service
Management Interface and Service Billing Interface; these are not “seen” by the users, but are the most
important to the mobile video provider, to provision, manage, and most importantly, receive revenue.
The mobile operator could possibly have a future advantage in converged services using video as a key
element in the feature. Service policy can be part of the Server Middleware/Application, as an application
51
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
server on the IMS core, and by leveraging that operator’s IMS core and connected network policy, the
mobile operator can use SIP session management, identity management, profile management, using a
common control plane to also deliver the video service, and converged applications with voice. The
operator can also then leverage the network policy bandwidth and resource management, access control
management, security and firewall management, as well as portions of the OSS systems. The common
control plane at the clients (SIP or HTTP to SIP Proxy) helps to reduce the present barriers to multimedia
application convergence.
As can be seen by this high level description, the mobile operator must be prepared to take on a complex
video acquisition, content processing and packaging system, as well as comprehensive headend for
server- side video applications and middleware, system management, provisioning, user management
and interfaces to billing systems. Therefore, it is reasonably likely that many mobile operators will try to
buy their video content as a managed service rather than take on wholly new areas of expertise and
experience that normally only exist in Cable Multiple System Operators (MSO).
5.4 CONTENT MANAGEMENT IN CLOUD ENVIRONMENT
Content management delivery in a cloud environment leverages some of the edge caching techniques
addressed in Section 5.1. Section 5.4 addresses some of the architectural principles in content
management architecture in a cloud environment.
5.4.1 CONTENT MANAGEMENT ARCHITECTURE AND DELIVERY
From the perspective of the content consuming device, low latency and optimized user experience in
video delivery is maximized when the server providing the content is close to the access network edge.
This requires a good understanding of the market and traffic patterns related to video delivery, as
described in this section.
a) Application of traffic profile and expectations to video delivery results in location of caching
servers close to video consumption hotspots.
b) End user experience for video over wireless is maximized when the edge delivery architecture is
aligned with the location of wireless hotspots to maximize economies of scale in resources to
deliver video traffic over both content delivery networks (CDN) and radio access networks (RAN).
c) Hardware and software in the cloud is usually distributed for redundancy and scalability.
Hardware could be dedicated to CDNs or leased based on traffic patterns to allow for more
flexible architectures.
d) In networks where video content is just being introduced, efficiencies in the video delivery may
not be fully exercised. There may be multiple copies of the same stream being sent to consumer
devices. These are usually observed in emerging deployments of video. As traffic and demand
for the same stream of video increases, e.g. multiple hits on a YouTube video, edge solutions are
implemented that repeat the stream over the access to manage content so that video stream
duplication between the cloud and the access network is minimized. This usually suffices for
access networks that have not been overwhelmed with video delivery.
e) With evolution of access technology to LTE, it is now possible to use eMBMS techniques over
mobile broadband by which repeated streams in a geographical area are broadcast to reduce use
of duplicated, dedicated video bearers.
A combination of the last two approaches results in well-managed video content in a cloud environment,
where the Content Management System (CMS) in the cloud accomplishes the following objectives:
52
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013



It identifies areas in access network where streams are being repeated, and optimizes by
streaming a high bandwidth instance to that area.
The access network at that location uses edge caching to minimize the number of repeated
streams towards the cloud.
The access network uses broadcast over RF to minimize use of radio resources dedicated to
copies of video content to different devices simultaneously.
5.5 POLICY MANAGEMENT AR CHITECTURE
The 3GPP policy and charging control (PCC) provides a rich policy and charging environment to support
dynamic control of QoS, charging and policy by application servers. Figure 14 shows a standard PCC
architecture. The PCC functions include the following:
The PCRF (Policy and Charging Rules Function) provides policy control and flow based charging control
decisions. The PCRF determines policy rules in a multimedia network in near real-time. It operates at the
network core and accesses subscriber databases and other specialized functions, such as a charging
system, in a centralized manner. The PCRF aggregates information to and from the network, OAM
systems and other sources (such as portals) in real-time to create rules and then automatically making
policy decisions for each subscriber active on the network. Therefore, it allows a network to offer multiple
services and QoS possibilities. Because it operates in real-time, the PCRF has an increased strategic
significance and broader potential role than traditional policy engines. PCRF can be integrated with
different platforms like billing, rating, charging, and subscriber database or can also be deployed as a
standalone entity.
The PCEF (policy and charging enforcement function) is implemented in the PDN gateway. It enforces
gating and QoS for individual IP flows on the behalf of the PCRF. It also provides usage measurement to
support charging.
In addition, an online charging system (OCS), not shown in Figure 14, provides credit management and
grants credit to the PCEF based on time, traffic volume or chargeable events. An off-line charging system
(OFCS) receives events from the PCEF and generates charging data records (CDRs) for the billing
system.
The support for dynamic control of policy and charging will enable applications to make multiple requests
per service session. Triggers associated with changes in location, network access, device status and
security will generate additional requests. The key message is that the PCC capabilities brought together
in 3GPP Rel-7 and enhanced with Rel-8 are sufficiently powerful to provide dynamic control of charging
and QoS on a per flow and per subscriber basis. This architecture will allow support of new business
models.
6. SURVEY OF OPTIMIZATION TECHNIQUES/MI TIGATION STRATEGIES
6.1 OBJECTIVES OF WIRELE SS VIDEO OPTIMIZATIONS
Video users require significant amount of resources to be made available throughout the network to
ensure that the QoE of those users is being met. Even with the introduction of LTE and LTE-Advanced
and the improved spectral efficiencies realized thereof, air interface resources will be at a premium due to
53
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
the need to support a larger number of video users at higher data rates along with the other non-video
users. Within the network, the backhaul and the core networks could also become bottlenecks.
Wireless video optimization techniques play an important role in ensuring that the QoE for both
guaranteed-bit rate and non-guaranteed-bit rate video users are being met. The goal is to maintain the
average throughput delivered to the clients over the time scale of several video chunks in the face of
changing radio conditions and congestion conditions to the best extent possible, and only allow gradual
changes to the average throughput. Different devices with different screen sizes, such as a tablet and
smartphone, but in the same radio conditions, should be treated differently. A video user should always
get better throughput than a data user when both have the same radio conditions and yet the extent of
priority for video over data client should be configurable. The resource partitioning between data and
video clients should adapt to the number of clients of each type to ensure that the data clients are not
starved by the video flows. These optimization techniques must be realized with minimal impact to the UE
battery life, must be dynamic enough to accommodate fast variations in RF and network loading, and
must be devised so they operate within multivendor network environments. Furthermore, the impact of
these optimizations must not adversely affect the QoE of other non-video users such as voice users. The
following sections describe techniques for wireless video optimizations both in the radio and in the core
networks.
6.2 END-TO-END QOS SUPPORT IN L TE FOR VIDEO DELIVER Y
6.2.1 LTE QCI AND OTHER NE TWORK QOS TECHNIQUES
QoS management in LTE is characterized by an advanced Policy and Charging Control (PCC)
architecture, whose main change with respect to UMTS is the introduction of QoS Class Identifiers (QCI).
QCIs are scalars used as reference for the definition of specific packet forwarding behavior, that link to a
set of parameters pre-configured by the operator at the eNodeB such as scheduling weights, admission
thresholds, link level protocol configuration, etc. QoS is controlled per service data flow, with each service
data flow being a virtual connection that carries data plane information. Service data flows sharing the
same QoS policy are in turn carried over bearers which represent suitable data paths between the user
and the packet data network. Applications may trigger multiple service data flows for specific needs, each
with their own QoS parameters.
In LTE Rel-8, a total of nine QCIs are standardized as shown in Table 4. Each QCI addresses a specific
data flow and resembles the PDP context of GPRS and UMTS. A bearer is always associated with a QCI.
54
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Table 4: Standardized QCI characteristics [from 3GPP TS 23.203]
QCI
Resource
Type
Priority
Packet
Delay
Budget
(NOTE 1)
Packet
Error
Loss
Rate
(NOTE 2)
1
2
100 ms
10
2
4
150 ms
10
3
3
50 ms
10
4
5
300 ms
10
5
1
100 ms
10
6
300 ms
10
7
100 ms
10
300 ms
10
Example Services
-2
Conversational Voice
-3
Conversational Video (Live Streaming)
-3
Real Time Gaming
-6
Non-Conversational Video (Buffered
Streaming)
IMS Signalling
GBR
-6
6
7
-6
Non-GBR
-3
Video (Buffered Streaming)
TCP-based (e.g., www, e-mal, chat, ftp,
p2p file sharing, progressive video,
etc.)
Voice,
Video (Live Streaming)
Interactive Gaming
8
8
9
9
-6
Video (Buffered Streaming)
TCP-based (e.g., www, e-mail, chat,
ftp, p2p file
sharing, progressive video, etc.)
Each QCI belongs to one of two groups either GBR (guaranteed bit rate) or non-GBR (non-guaranteed bit
rate). GBR bearers try to guarantee a minimum bit rate as required by the applications and usually force
eNodeBs to reserve the necessary radio resources throughout the life of the connection, after admission
control criteria have been met. GBR bearers are especially suitable for conversational and streaming
multimedia applications. Non-GBR bearers do not impose any minimum bit rate and are therefore could
be used for unpaid video traffic delivery.
QCI priorities enable differentiated treatment of packet flows when congestion appears at any point of the
network, with a higher preference for lower QCIs (QCI 1 being the highest priority). Scheduling strategies
at the eNodeBs should take these priorities into account in congestion situations where multiple data
flows compete for a limited set of radio resources. QCI 9 with the lowest priority corresponds to the
default bearer, which is always set up after initial power up thus allowing “always-on” capability. This
bearer does not preclude simultaneous establishment of additional bearers with higher-priority QCIs,
possibly for combined data services.
Mobile video delivery can greatly benefit from the traffic differentiation that can be achieved through use
of different QCIs. Streaming video that may require a minimum data rate should be assigned to one of the
GBR bearers corresponding to QCI 2 or QCI 4. Voice over IP on LTE (VoLTE) is supported through
QCI 1. Non-premium best-effort data services can be well served by QCI 9, while QCI 8 could be
reserved for premium-grade TCP-based services (like high quality progressive video), with a higher
priority but not very tight delay restrictions. Speech services with QCI 2 would be given special attention
(possibly through semi-persistent scheduling) and would not be under-served due to “hungry” video users
if taken care of by appropriate scheduling strategies. Streaming video applications with bit rate constraints
55
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
should be given any of the GBR bearers for tighter delay requirements. Premium-type video subscriptions
could be favored through the use of QCIs 6 and/or 8 for TCP-based progressive video services.
In addition to the parameters shown in Table 4, each bearer is provided with additional parameters such
as the Aggregate Maximum Bit Rate (AMBR), which limits the maximum user throughput at both the core
and radio access networks. This enables additional service differentiation between users connected to
the same Access Point Name (APN) and with the same QCI.
The network can also be protected from non-GBR data-intensive bearers that could collapse the network
through use of AMBR. The PDN-GW performs rate policing based on AMBR per APN for both UL and DL
traffic, while the eNodeB performs rate policing based on the AMBR per terminal. Maximum bit rate of
GBR bearers is also limited at the PDN-GW. Optimization of these values can prevent episodes of
network congestion due to greedy applications.
Prior to the introduction of QCIs, 3GPP defined a set of so-called traffic classes for UMTS [3GPP TS
23.107]. The four UMTS traffic classes are, in descending order of delay sensitivity: Conversational,
Streaming, Interactive and Background. Conversational and Streaming classes are more applicable for
GBR applications, such as video streaming, while Interactive and Background are reserved for best-effort
TCP services with less delay requirements (including progressive download video). HSDPA provides an
additional QoS control by defining Scheduling Priority Indicator (SPI), a number in the range [0, 1… 15]
available at the NodeB for scheduling purposes [27]. A high SPI could be allocated to video streaming
and other real time services, while other best-effort applications could be assigned a low SPI value.
However, SPI values are only visible at the NodeB and cannot be given other proprietary “meanings” from
the core network, therefore limiting its applicability.
6.2.2 LTE GUARANTEED BIT RATE BEARERS
3GPP has standardized a number of different QoS classes so that an application can invoke a bearer of
an appropriate QoS class. In order to maintain the data rate at a constant level, video should be delivered
over a Guaranteed Bit Rate (GBR) bearer. For a GBR bearer, the average data rate is guaranteed to be
above a certain target GBR value by allocating the required amount of resources to the device to maintain
the minimum target rate. For example, if the spectral efficiency drops to half, then twice the amount of
bandwidth (physical resource blocks) or time is allocated to the user over the time scale of a second to
maintain the minimum average rate over this time period. GBR bearers are assigned a higher priority
compared to the default or best effort bearer. Resources are first allocated to GBR bearers up to a
maximum allowed design limit and only then are remaining resources are allocated to best effort bearers.
Thus, the video quality of a video stream carried over a GBR bearer will be superior to a best effort video
bearer for the same radio conditions.
In addition to a target GBR value, 3GPP specifies a Maximum Bit Rate (MBR) not to be exceeded by a
particular bearer. The scheduler at the base station will avoid reserving more resources than needed in
order to not exceed the MBR value even in good radio conditions. LTE-Advanced (as defined from 3GPP
Rel-10) enables setting an MBR value higher than the GBR, but this is not allowed in LTE Rel-8 and
Rel-9 where MBR is constrained to be equal to GBR.
56
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
6.2.3 MAINTAINING CONSTANT VIDEO QUALITY WITH G BR SERVICE
As discussed in Section 4, frequent changes in video rates are detrimental to perceived video quality.
Hence for video delivery, it is likely necessary to maintain the data rate even when the wireless channel
fluctuates. Wireless channel variations result in instantaneous data rate changes because of the adaptive
coding and modulation. With adaptive streaming, the rate determination algorithm in the client may try to
follow the delivered data rates resulting in quality fluctuations. In order to maintain a constant average
data rate to the client, it is not sufficient to maintain a minimum guaranteed bit rate as in the case of a
GBR bearer because higher rates above the GBR target could be achieved when excess resources are
available. Thus the GBR bearer will experience rate variations above the GBR target rate.
For RTP/UDP streaming, the maximum bit rate (MBR) can be set for the bearer with the target MBR value
set equal to the guaranteed bit rate (GBR) value, which is presumably set to the RTP/UDP streaming
rate. With GBR equal to MBR, a constant average data rate will be delivered to the RTP/UDP client by
ensuring that the scheduler allocates an appropriate amount of resource to maintain the RTP/UDP
streaming target rate and provide a good user experience. With this solution, since the scheduler always
provides the same download rate to the RTP/UDP client, network utilization may not be optimized under
varying network conditions.
For HAS streaming, the GBR can be set to a rate that provides the minimum download rate needed for an
HAS client to provide the minimum acceptable quality play back experience to the user. A MBR can be
set for the bearer, with the target MBR value set to a greater value to cap the highest quality that the HAS
client provides to the user. The data rate delivered to the HAS client will then be within the specified GBR
to MBR range, thus ensuring that the scheduler allocates an appropriate amount of resource to maintain
at least the GBR rate and at most the MBR rate. With this solution, the GBR can generally be set lower
than what would be needed for RTP/UDP streaming, the MBR can be set higher than what would be
needed for RTP/UDP streaming, and the HAS client can choose from multiple available playback rates
between GBR and MBR. This provides the scheduler flexibility to optimize network utilization, while at the
same time allowing the HAS client to provide to the end user with the highest quality streaming solution
possible for current conditions within the GBR and MBR constraints. A well-designed HAS client will do
its own averaging and choose an appropriate playback rate at each point in time to provide the highest
quality playback while avoiding stalls or skips.
6.2.4 DIFFERENTIATION AND FAIRNESS WITH GBR SE RVICE
Video optimization using GBR and MBR also allows differentiating between different classes of users if
desired. For example, if “Gold” users pay more to watch video than “Silver” users, the GBR target of Gold
users could be set higher than that for the Silver users. This will ensure that in most cases Gold users get
better quality than Silver users when adaptive streaming is used. Furthermore, setting the same GBR
value across all users of a given class ensures that the same throughput is delivered. The GBR can be
set according to the device characteristics and screen size (specified in a HSS/SPR) to achieve the same
quality for different screen sizes. For example, for the same class of users, a tablet device could have
GBR equal to MBR set at 400 kbps while for a smartphone, the GBR target could be 200 kbps. Because
of the smaller screen size, the same quality can be achieved with the lower throughput for the
smartphone compared to the tablet.
57
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
6.2.5 LIMITATIONS OF GBR BEARER
While in principle GBR and MBR offer a means to control video quality, there are number of caveats to
consider. At the time the flow is admitted, a user may have reasonable radio conditions allowing for a
good GBR target. However if the radio conditions worsen, maintaining the GBR can consume an
excessive amount of resources leaving the remaining data flows starved. Conversely, setting the GBR
target too low can make it irrelevant because even the data flows will achieve that throughput. Thus it is
unclear how to pick an appropriate value for GBR that will result in a reasonable amount of resources
being devoted to the GBR flow. Network operators could shy away from using GBR for this reason.
To avoid the problem of excessive consumption of resources, a cap on the consumption of resources by
the GBR flows is usually imposed. This makes the GBR targets not achievable under congestion.
However, the side effect of that is that these GBR flows will get throughputs according to the resource
allocation policies of the underlying scheduler. This can result in highly unequal throughputs allocated to
the various GBR flows with some flows meeting their targets while some other flows falling substantially
below their targets. In other words, simply setting the GBR targets does not necessarily maximize the
aggregate quality of the video flows when there is congestion in the network. A more desirable solution
would be to have more equitable allocation that allows more video flows to reach a slightly lower target
thereby accommodating more video flows.
6.2.6 MICROBURSTING
Microbursting is a technique where servers shoot out data as fast as possible (i.e. at a high data rate)
within a short period of time to each end user and thus fill up the client buffers. However, challenges with
microbursting remain. The problem with microburst is to allocate the time to burst and permitted size. With
many independent flows in the same RAN, sharing the bandwidth in a fair manner is non-trivial especially when the sources are from servers outside the operator’s network. Protocols such as TCP,
which try to achieve fairness between multiple flows, could also going to be compromised by bursting.
6.3 RF SPECIFIC OPTIMIZATIONS FOR VIDEO DELI VERY
Video optimization at the air interface, typically performed at the base station or eNodeB and the video
client on the device, is designed to address a number of requirements such as:





maintain video quality in the face of time varying radio conditions;
manage video quality as the cell becomes increasingly congested;
differentiate video quality according to the subscription level;
manage video quality during handoffs;
ensure fairness when multiple users are served.
We will first discuss optimization at the base station, and then address client optimization. Techniques
described in this section may have been described earlier and are summarized here since they constitute
RF specific optimizations for video delivery.
Optimizing video flows at the base station has the inherent limitation that the base station does not have
information about the video characteristics such as the type of video or the kind of scene that is being
carried in a particular set of packets, the manifest file of the video stream, or the screen size of the
device. The manifest file holds all the information about the video file, its duration, the streams, the codec
info, and a list of all the available chunks. The manifest file may be encrypted and, thus even with DPI,
58
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
cannot be available to the base station. Any information related to the video has to be signaled to the
base station by upstream nodes. If proprietary signaling is not to be used to maintain interoperability, then
the only information that can be provided to the base station are the radio bearer characteristics defined
in the standards associated with the bearer that is designated for carrying the video flow for a specific
device. Several schemes in the literature, that optimize the stream based on providing higher rate or
reliability for the I frames compared to the P and B frames or the ones that re-encode the video into base
layer and enhancement layer according to channel rates, are not pragmatic.
On the other hand, radio-related information is readily available at the base station. The base station
knows for example the number of users being served, the number of packets waiting to be delivered for
the different flows, and the signal-to-interference ratio on the radio link to the different devices. The main
method of video optimization at the base station is by controlling the data rate delivered to the device.
Client side optimization has information about the video, such as the manifest file in the case of adaptive
streaming. In the case of HAS, the rate determination algorithm runs on the device and it can be
optimized specifically for the wireless environment. A well-designed HAS client should be able to work
well and share fairly with other HTTP flows without knowing about the other devices that share the
network. The network fairly provides as much data downloading capacity as it can to each end device.
The HAS client on the device then decides which bit rates to download at each point in time to provide the
best quality experience to the end user (i.e., no stalls or skips and a playback quality that pretty much
matches the current download rate of the HAS client).
6.3.1 GUARANTEEING MINIMUM VIDEO QUALITY ON THE AIR INTERFACE
Wireless channel quality is inherently time-varying because signal propagation is affected by the
movement of the transmitter, receiver or surrounding objects. Adaptive coding and modulation is a
standard feature in cellular wireless networks to deal with channel time-variations. With adaptive coding
and modulation, the data rate transmitted to the device is adjusted according to the channel quality. As a
result of these data rate variations, it is possible that the data rate dips too low to maintain the video
quality. This could result in the video stalling, or the video quality becoming unacceptably low for the case
of adaptive streaming. Well-designed HTTP Adaptive Streaming clients would be able to handle this
issue. In addition, video optimization at the base station that targets maintaining a minimum data rate to
the video clients in the face of channel variations can substantially improve video quality by preventing
stalling. Since video quality depends on only average data rate over time scales of a second it is sufficient
to maintain the average data rate delivered to the device at a constant level. This video optimization is
accomplished through setting of Guaranteed Bit Rate bearers for video streams.
6.3.2 BROADCAST/MULTICAST OF VIDEO
When the video is of interest to a number of subscribers at the same time then multicasting the video over
the air using eMBMS in LTE (or MBMS in 3G) is substantially more efficient than sending separate
unicast transmissions to each client. This is because wireless is fundamentally a broadcast medium.
Whatever is transmitted wirelessly can be received by multiple receivers in the range of the transmitter.
Additionally, the concept of Single Frequency Network (SFN), where the same signal is transmitted by
multiple adjacent base stations, improves performance by eliminating out-of-cell interference. SFN can be
employed when a number of users in multiple cells are watching the same content at the same time, as
can broadcast of live sports events or news. eMBMS is described in detail in Section 5.2.
59
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Traditionally, a number of users watch the same video in Linear TV. However, Linear TV is not commonly
used on mobile devices. The real application for eMBMS delivery of video to smartphones and tablets is
in stadia where specialized content such as action replays is watched by a large number of people at the
same time. With a number of small cells covering the stadium, eMBMS with SFN could be substantially
better than unicasting to different users.
6.4 NON-RF SPECIFIC OPTIMIZATIONS FOR VIDEO DELIVERY
There are several techniques above the RF layer, besides LTE QCI, which can be used to optimize the
wireless video viewing experience. These range from adapting the video quality to the available network
bandwidth, minimizing the impact of video on the network through to encouraging change of user
behavior to use the network when it is less heavily used. This section provides a survey of the main
techniques of non-RF specific optimizations for video delivery.
6.4.1 DASH (HAS)
All HTTP Adaptive Streaming (HAS) methods, including Dynamic Adaptive Streaming over HTTP
(DASH), allow the video quality to dynamically adapt to the available bandwidth, and is a non-RF specific
method of optimization. The client senses the available bandwidth and makes the decision on what
quality level to pull for the next chunk (e.g. 2 seconds of video), as has been described in Section 2.5.
There are several attractions to using HAS. Firstly, it is based on TCP. This means that lost packets are
automatically retransmitted, and the TCP congestion avoidance ensures a fair share of the available
bandwidth. A further attraction of HAS to the content providers is that it does not require any support from
the network. This means the same content can be delivered over any network. The only requirement from
the network is that it can sustain the bandwidth for at least the lowest video quality. However, this means
the end user will experience variable QoE depending on congestion and their radio conditions. A final
attraction is that HAS is based on HTTP. This allows reuse of web technology reducing the cost of
deployments. It also enables support for HAS content through web caches.
6.4.2 BIS SCALABLE VIDEO C ODECS (SVC)
Several codecs support the ideal goal of scalability. Scalable Video Coding (SVC) [28] allows
representing video into different versions with different frame rates, spatial resolutions and/or quality
levels (fidelity to the original video) by a single video stream. The content is encoded using a base layer
and one or more enhancement layers. The base layer provides a complete encoding of the content at the
lowest acceptable quality. The client only needs to receive the base layer encoding. However, it can
improve the quality if it also receives one or more of the enhancement layers. These can improve the
spatial (higher image quality) and/or temporal (frame rate) of the decoded image.
An important property as defined in MPEG Scalable Video Encoding (SVC) is that it is easy to separate
out the base layer and the individual enhancement layers – even where the video is encrypted. This
reduces storage overhead of SVC as a single file can contain the base layer and all the enhancement
layers. The overhead for the delivery server to deliver on-demand the base layer and any of the required
enhancement layers is low. Unfortunately, there is an overhead using SVC. For MPEG-4 SVC this is over
10 percent per enhancement layer. As 5+ enhancement layers are required for a smooth transition from
low quality to high quality, this overhead can become unacceptable, especially when this is applied to the
most congested part of the network namely, the wireless RAN.
60
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
When SVC is used in DASH, an additional level of optimization is realized. The different SVC layers can
be split and distributed into different “representations.” If users have enough throughput they can request
all the representations with all the layers, but when their throughput is limited, they can omit request to
data from the representations containing the highest layers [29].
6.4.3 TRANSCODING/TRANSRATING/RATE CAPPING
While techniques like HAS and SVC handle the dynamic variation of the network bandwidth efficiently, it
doesn’t work with majority of the video content watched on the mobile network today. This is because
majority of today’s video content is delivered over HTTP-PD (Progressive Download). So there is a need
for special techniques like dynamic transrating and dynamic transcoding on the mobile network to
address this content. Otherwise operators will not be able to efficiently address the video data tsunami.
Transcoding involves dynamically changing the encoding of the video stream to adapt to the available
bandwidth. The video coding technology is modified based on the knowledge of the device. A device
profile is maintained and, based on the requesting device and its software configuration, an appropriate
technology is selected and the video is transcoded from one format to another. The source video may be
prepared for a screen size that is large with a fast refresh rate, and thus coded at a bit rate that is
unnecessarily high for a wireless device with a small screen and lower refresh rate. Transcoding may be
required to change the resolution and frame rate of the video to make it suitable for the smaller screen.
Other factors that can be included in selecting the transcoded video frame rate and resolution are
subscription tier of the subscriber requesting the video, and other policy based compression techniques.
There are several approaches that support changing the resolution and fidelity to reducing the frame rate
without needing a full decode/encode. This is known as transrating. Bandwidth estimation is performed to
determine which video flows need transrating. Video streaming can thus adapt to congestion.
Transcoding is heavy on processing requirements to perform decode and re-encode in real-time. Also, it
requires the content to be unencrypted. It is less attractive than HAS since many professional content
providers are reluctant for uncontrolled changes of the encoding and hence end user QoE of their
content. Transrating is usually less demanding of compute power, and yet is able to give a good range of
qualities. Hence it is an attractive approach where content providers permit.
High value (i.e. paid for) content is likely to be encrypted, thus making transcoding difficult. This leaves
lower value content including user generated video sites as candidates for transcoding. The operator has
to decide on balancing the cost of transcoding against the improved QoE for such content. In addition,
they must consider how the content providers will react when more of their content is changed in quality
in the network. They may respond to providing the content in multiple bit rates to help mitigate the need
for network transcoding.
Rate capping is the process by which an intermediate node responsible for video optimization measures
the available data rate and adjusts the sending rate of a progressively downloaded video (without altering
the video content) such that the downloaded portion of the video does not run too much in advance of
what is played, thus avoiding a lot of wasted bandwidth when the user stops watching. The consequence
of this technique is that a large play-out buffer is required. During times that the video rate is higher than
the shaped rate, the video player needs to be able to rely on the video stored in this buffer.
61
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
6.4.4 CONGESTION-AWARE OPTIMIZATION
The impact of increased video traffic is felt the most on the RAN. It is in the RAN where many video
usability problems start surfacing as the cells get congested. As more users crowd a cell, the effective
network bandwidth experienced by each user decreases. As a result, video playback, particularly HTTPProgressive Download (HTTP-PD), starts experiencing buffering issues. Adaptive Streaming protocols
handle this problem by requesting a lower quality video, which by structure would need lower bandwidth
for satisfactory playback. But with HTTP-PD the entire video is delivered as a single content segment. So
if the rate of video delivery is lower than the playback rate, then the player would have to pause the video
and buffer enough frames before resuming the playback. It is this scenario that the “congestion triggered
optimization” technique hopes to address.
In simple terms, congestion triggered optimization is a mechanism where a video optimization engine
dynamically starts optimizing (transrating/transcoding) video traffic during periods of congestion. Once a
video flow is selected for dynamic optimization, the optimization engine would continuously modulate the
compression rate (up/down) to ensure a smooth video playback. Congestion trigger can be achieved in
two ways.
1) External RAN probe based trigger. With the external trigger approach, mobile operators
typically deploy a RAN probe on the Gn interface (between SGSN and GGSN) and have it
monitor the RAN usage messages flowing out of the RAN. When this probe detects a particular
cell/site to be congested, it would notify a network element about the same. Typically this
notification is propagated to interested elements via the PCRF. The PCRF would then issue a
dynamic update to the video optimization engine. Then, the video optimization engine would
identify all video flows happening on the congested cell and apply dynamic optimization to them.
It is important to note that in this model the entire cell is being labeled as “congested.” Therefore
the video optimization engine applies dynamic optimization to all flows pertaining to that cell.
2) Flow based trigger. In this model, the video optimization engine is deployed inline to the data
traffic. The optimization engine would then continuously monitor the average bit rate of each
video and compare it to the effective bandwidth experienced by the device for each flow. If the
optimization engine recognizes a disparity, where effective network bandwidth is lower than
video’s average bit rate, then the flow is immediately selected for dynamic optimization. Once a
flow is selected for optimization, the optimization engine would continuously monitor the effective
network bandwidth and tune the compression level appropriately to ensure smooth video
playback. This model provides a comprehensive solution to tackle the Quality of Experience issue
with the delivery of video traffic on mobile networks. A user who is experiencing poor network
quality due to his distance from the tower would receive dynamic video optimization even though
he or she is the only user on the cell. Similarly, a user who is a connected to a congested cell, as
per RAN metric, but is not experiencing bandwidth disparity issues, would not have his video
compressed. The user receives original video quality and does so without consuming any
additional computing resources (transcode/transrate) in the network.
6.4.5 CLIENT OPTIMIZATIONS FOR HAS
The client plays a significant role in determining how the video gets delivered. In the case of adaptive
streaming, the video rates and thus the quality of experience is also controlled by the client. The client
can thus play a role in optimizing the video delivery specifically taking into account the nature of cellular
wireless communication.
62
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
HAS clients can smooth the effect of wireless channel rate variations by increasing the averaging window
over which bandwidth estimation is done for video rate determination. By introducing longer averaging
time, fluctuations because of channel fading could be eliminated. Rate selection could be made more
conservative compared to wireline in the interest of maintaining a more stable rate. The criteria for
switching to the next higher rate could be tweaked to ensure that estimated bandwidth will really be
available over longer periods of time and a fall back to lower quality will not be required.
In the case of progressive video, the client has the flexibility to either download the whole file in one shot
and buffer it or periodically download small chunks. The advantage of the latter is that when the user
does not watch the entire video, then the network has not been used unnecessarily. However, if the
downloading of the individual chunks is spaced more than 10 seconds apart (or the number of seconds
corresponding to the eNodeB timer) then the radio bearer will be released after each download and
needs to be re-established for the next chunk. This causes unnecessary overheads which utilizes
additional resources both in the network as well as the device. For the device, the battery drains faster
when the download is done in multiple small chunks.
The client can in principle be also tweaked to control the download rate to lower value that what the
network can provide so as to maintain a steady download over a longer time. This can be done for
example, by using TCP receive window control mechanism. Once the receive window is full, then sender
will not send any more data. The client can thus pace the download of packets.
In general, having a larger playout buffer can minimize video stalling during periods when the network
data rate diminishes. Thus, having a larger playout buffer improves quality but can result in additional
start up delay if the client waits to fill up the buffer. However, the startup delay can be mitigated for
adaptive streaming by starting with low video rates that quickly fill the buffer and then gradually improving
the quality.
When the device has access to the network through multiple radio technologies, then the client could
dynamically select the access technology through which video chunks are downloaded. For example, the
client could request a low video rate chunk through the cellular interface and once that is delivered and
there is still time before playout it could request a higher rate version through a Wi-Fi interface.
6.4.6 CONTENT CACHING IN THE NETWORK
The RAN is not the only place where the video traffic has an impact. The core network and EPC has to be
large enough to handle the demands for the large and growing number of video streams. One way to
reduce the impact is to cache (i.e. store) the popular content close to the end user. A request for content
which is cached is diverted to a suitable cache server and delivered from there. This saves core network
bandwidth. It also reduces traffic on peering and transit links from the wireless network operator’s network
to the Internet in cases where the content is external.
Caching can be transparent or explicit. In the former case, all content is cached irrespective of source. It
will remain in the cache as long as there are requests for it. With explicit caching, the wireless network
operator agrees to cache content from a content provider. In this case, because the content will probably
delivered with higher quality (especially if some of the RF enhancements mentioned in the previous
section have been applied), the content provider probably pays for the storage and delivery of their
content.
It is important to note that while content caching may address the backhaul congestion, it does not solve
the RAN congestion issue. Furthermore, it requires a co-operating network core. If video content is
63
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
cached at the edge and delivered from the edge, then mechanisms to modify existing billing mechanisms
must be considered. In current architectures, billing is typically done by the PGW, which sits in the mobile
core. Content caching in the PGW does not require a change to the billing mechanisms.
6.4.7 TIME SHIFTING OR SIDELOADING SUPPORTED BY CON TENT CACHING IN THE UE
Traffic in a specific cell sector typically varies over the day. Typically, cells in business districts will be
heavily used during the day whereas cells in residential districts will see their usage peak during the
evening. In most cases, the cells will be largely only very lightly used during the night. Hence this makes
in an ideal time to deliver video files. Either based on the user’s explicit selection or by predicting their
likely interests, video files can be downloaded when the cell is little used. Time shifting or sideloading
refers to downloading video content to the device during off-peak times or when the device is connected
to a Wi-Fi access point and caching the content in the device for viewing later. While time shifting is not
on-demand instant streaming, with a good recommendation engine, the end user experience can be
almost as good. Indeed because the file is downloaded and not streamed, the quality is not affected by
variable quality radio conditions. eMBMS can be utilized to efficiently deliver popular content which would
be cached by a large number of users.
Time shifting technique is particularly suitable for non-streaming content such as YouTube videos that are
based on progressive download. Pre-fetching content allows high quality video to be downloaded.
Furthermore, the cost of downloading video will be substantially less when done over Wi-Fi or during offpeak times. The sideloading solution requires a pre-fetch client on the device that will talk to a sideloading
server in the network that can direct the client to obtain specific pieces of content at the most appropriate
time. At the direction of the sideload server, the client can then directly download the content from the
appropriate video server. The sideloading server can utilize information from network probes and other
information from the client such as proximity to Wi-Fi access point or battery life remaining to determine
the opportune time and access technology to be used for pre-fetching.
With increasing amounts of memory available on the device, caching popular content on the device is
becoming increasingly feasible. UEs such as smartphones and tablets now have substantial storage
capacity. This allows content to be delivered to the UE when the network is lightly loaded and cached.
When the content is requested, the request is directed to the local device avoiding traffic on the RAN.
This has the further advantage of improved quality. To minimize the amount of content that needs to be
cached, the users can opt in for sideloading and indicate to the sideloading server what class of content is
of interest. For example, a user may indicate that new music videos of a particular artist is of interest and
whenever there is a new video the sideloading server could include that in the pre-fetching list for that
user. A personal recommender system that studies which videos the user watches and determines
recommendations for the user can also be used by the sideloading server.
6.4.8 VIDEO USAGE BEHAVIOR
Yield management is the umbrella term for a set of strategies that enable capacity-constrained service
industries to realize optimum efficiency from operations. The core concept of yield management is to
provide the right service to the right customer at the right time for the right price. That concept involves
careful understanding of service, customer, time, and price. The service can be defined according to the
dimensions of the service, how and when it is delivered, and how, when, and whether it is reserved. The
principles of Yield management have been applied to the airline industry and could possibly be applied to
wireless networks particularly for video applications since they are bandwidth intensive. This concept
could allow for effective use of operator spectrum resources, while best attempting to maximize user QoE.
64
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
7. RECOMMENDATIONS
7.1 RECOMMENDATIONS FOR VIDEO APPLICATION DEVELOPERS FOR EFFICIE NT RADIO RESOURCE
MANAGEMENT
Video application developers need to consider the challenges in operating over a network with mobile
devices. This requires that applications are adaptable to changing access capabilities. This leads to the
development of APIs from the CDN to the access network. Some guidelines for video application
developers include:
1) Utilize APIs between the Application Layer and Access.

Mobile operators could provide the development space and environment for realistic
simulations.

Drive the creation of APIs to demand QoS from the network.

App developers can differentiate themselves with access aware apps.
2) Understand benefits of network optimized applications and products such as:

A network aware application could operate better under busy network conditions than
competing applications. Consequently, there could be an increase in the demand for network
aware applications.

Develop applications for the maximum efficiency of the radio resources.
o

Utilize less data network resources than competing applications.
Make video applications more attractive to the end users.
3) Ensure that video servers deliver chunks of video in small regular intervals, not longer than
approximately 10 seconds. Applications running over wireless networks need to be aware of the
properties of the RAN. Both the video delivery algorithm and the network configuration must be
optimized. Apple HLS typically uses 10 second HTTP chunks. With longer chunks, as each chunk
is downloaded, there is a pause until the next one. Today’s wireless networks implement inactivity
timers about 5 seconds, and in many cases they have been optimized for more bursty types of
traffic (especially when Rel-8 Fast Dormancy is not available). However, even if shorter chunks
are used (which can be preferable for better rate switching and less end-to-end latency for live
streaming), the client can still download several chunks consecutively before pausing to
download the next set of chunks. The most optimal value of chunk size depends on, amongst
other things, the eNodeB timer value settings. Video sources with significant inactivity periods
between chunks of data can force the network to move the device to a dormant state, therefore
wasting signaling resources and device battery. This will become especially harmful in highspeed wireless networks (like LTE), where the high bit rates available would allow for significant
inactivity periods between video chunks.
4) Ensure that polling requests from the devices to the network are bundled to maximize use of
seized resources. This will require device video application developers to synchronize their
network information requests with other applications on the device. This requires that device
65
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
APIs to network access are leveraged. This also addresses the reality that video applications will
likely incorporate user-specific advertising, special offers, subscription updates, and other types
of notifications to the end user.
7.2 RECOMMENDATIONS ON BIT RATES FOR VARIOUS CODECS AND SCREEN SIZES/DEVICE TYPES
The choice of bit rate for the audio and video streams is a compromise between quality of experience and
bandwidth used for streaming. The device screen size and viewing distance are a critical aspect in
determining the bandwidth needed for encoding the video. The larger the screen and the closer it is
viewed (up to the near point of the eye), the higher the bandwidth required. A video that looks poor on the
larger screen of a tablet can be more than adequate, in terms of QoE, on a smartphone held at the same
distance. Scenes with high motion or scenes with a lot of detail will require more bandwidth to minimize
visual artefacts. Chaotic motion (grass blown in the wind, flames, waves, etc.) in scenes with a lot of detail
are the hardest to encode and hence require higher bandwidth to avoid loss of definition. Finally, the
compression efficiency of the video codec impacts the bandwidth needed to encode a video quality with a
given fidelity.
At the time of writing, the most widely used video codec for smartphones and tablets is H.264 / AVC.
Table 5 shows the recommended bit rates for different devices. 4G Americas recommends a minimum of
200 kbps for smartphone and 400 kbps for tablets. Note that these are average bit rates and variable bit
rate (VBR) encoding can also be used. This allows a higher encoding rate to be used for short sections (a
few frames) of high complexity and/or motion resulting in a higher overall quality. People often watch
movies and TV programs on their smartphones and tablets using high quality headphones. These require
a good quality stereo audio source. However, recognizing that radio spectrum is limited, 4G Americas
recommends using AAC (Advanced Audio Codec) at 96 kbps.
Note that HD rates are included because of the option of delivery of the video stream over a mobile
network and forwarding to TV using in home using DLNA or AirPlay over Wi-Fi. For example, with rural
broadband it may be more efficient to use wireless technology to serve a dispersed community.
66
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Table 5: Recommended choice of bit rates for video streams
Video Quality
Level
Video (kbps)
Total (kbps)
Comments
VQ1
200
296
Smartphone
VQ2
400
496
Tablet / Smartphone
VQ3
800
896
Tablet / Smartphone
VQ4
1400
1496
SD 480P TV screen through games console and
STBs, Connected TVs, PCs and Tablets
VQ5
3150
3246
HD 72P TV screen through games console and
STBs, Connected TVs, PCs and tablets
VQ6
7100
7196
Full HD 1080P, TV screen through games console
and STB, Connected TVs, PCs with General
Processor Units.
In many cases, the content provider will encode their content in multiple qualities. This allows the end
user to manually select the desired quality taking into account their device, desired quality, and data plan.
They may also select a lower video quality level because they will achieve more consistent video quality
irrespective of the radio conditions. HAS dynamically selects the video quality based on the available
network bandwidth. H.265 HEVC codec has just been ratified. Early tests indicate that it gives a 40-50
percent reduction in bit rate for the same image quality.
8. SUMMARY AND CONCLUSIONS
Video for entertainment and communication is experiencing significant growth, given the current plethora
of mobile handheld devices enjoying enriched multimedia capabilities and as competitive forces are
driving lower data rates on mobile carriers. The introduction of new technologies like LTE, with low
latency and high throughput, will enable the evolution of video content from SD to HD and handheld form
factors to larger screens, thus further increasing the data volumes in networks. People’s expectations of
quality and availability will also grow as the use of video on mobile devices becomes main stream. The
increase in streaming content over mobile networks will provide significant challenges to the carriers to
keep up with demand and to provide a Quality of Experience for these services that provides a
competitive advantage over their competitors. Mobile video presents challenges unique to and different
from wireline video. The growth of video will pose considerable stress on HSPA+ and LTE networks to
handle both the volume of traffic and end user expectations. While some of these problems have already
been addressed in wireline networks, wireless networks impose additional technical challenges due to
varying interference and cell handover issues.
This white paper identifies technologies, techniques and architectures to mitigate the problems and
recommends approaches to maximize the Quality of Experience in HSPA+ and LTE networks. Based on
the data consumed by different streaming applications, we note that without more licensed mobile
wireless spectrum allocations from governments throughout the Americas region, video streaming over
67
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
mobile broadband wireless technologies such as LTE will continue to be challenged as the growth of
mobile video demand increases at exponential rates. Technical solutions that can be used in tandem to
address the challenges of mobile video delivery are discussed, including CDN design, Adaptive
Streaming, cellular HET-NETs, Wi-Fi offload, eMBMS, etc. Optimization of mobile networks and QoE for
streaming content is critical and will need to be the focus of the industry for years to come. In order to
fulfill the mobile users' expectations, a set of network as well as client optimizations are being devised
(both RF and non-RF based), in addition to the introduction of new and more advanced codecs.
Delivery of video chunks, through HTTP adaptive streaming, at various bit rates based on congestion is
identified as the preferred option for mobile video delivery. Interactions between HTTP over Adaptive
Streaming and TCP mechanisms are discussed in depth. Appropriate monitoring of the Quality of
Experience in live networks can be conducted in order to optimize performance in the cases where it is
feasible. Operators are currently striving for setting up QoE monitoring tools that could provide a deeper
knowledge on the user’s degree of satisfaction. Recommendations for video applications developers can
persuade them from wasting network resources or overloading the network with excessive signaling. A
set of best practices will result in more efficient ways to utilize the scarce spectrum available.
This whitepaper provides different focus areas for different audience readers as summarized below.
For Operators
 Architecture choices and improvements in both access and content providing networks are
described and analyzed.
 To work smartly with available and forecasted mobile broadband capacity in wireless networks,
various ways to best utilize the wireless capacity are identified.
For Vendors
 Areas for cooperation and integration with other dependencies in the overall content generation
and delivery network are identified.
 Areas for improvement and optimization in devices and networks are identified.
For Application developers and architects
 The effect of suboptimal design of video applications and their effect on mobile networks is
addressed.
 Recommendations are provided on areas of improvement.
For Industry analysts
 The state of video on mobile broadband is described and trends are identified.
For Device manufacturers
 Evolution in video processing technology and codecs is addressed.
It is expected that through the support of various network architectures and optimizations, QoE monitoring
metrics and techniques, mobile video will grow by orders of magnitude in the next several years.
68
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
LIST OF ACRONYMS
AAC
ARQ
CA
CABAC
CDN
CMS
CRMS
CVAS
DASH
DRM
eMBMS
FEC
FLV
FTTX
GBR
HARQ
HAS
HEVC
HLS
HSS
HTTP-PD
IMDB
IMS
MPEG
MSO
NGMN
OTT
PCRF
PPV
PSNR
QCI
QoE
QT
RDP
RNC
RRM
RTMP
RTP
RTSP
RTT
SFN
SIP
SPR
SSIM
SVC
VOD
VQEG
WebRTC
Advanced Audio Coding
Automatic Repeat Request
Content Access
Context-Adaptive Binary Arithmetic Coding
Content Delivery Network
Content Management System
Contract Rights and Management System
Converged Video Application System
Dynamic Adaptive Streaming over HTTP
Digital Right Management
Evolved Multimedia Broadcast Service
Forward Error Correction
Flash Video
Fiber to the last mile
Guaranteed Bit Rate
Hybrid ARQ
HTTP Adaptive Streaming
High-Efficiency Video Coding
HTTP Live Streaming
Home Subscriber Server
HTTP Progressive Download
Internet Movie Database
IP Multimedia Systems
Moving Picture Experts Group
Multi System Operators
Next Generation Mobile Networks Alliance;
Over The Top
Policy Control Charging Function
Pay Per View
Peak-Signal-to-Noise Ratio
QoS Class Identifier
Quality of Experience
QuickTime
Real Data Protocol
Radio Network Controller
Radio Resource Management
Real Time Messaging Protocol
Real Time Protocol
Real Time Streaming Protocol
Round Trip Time
Single Frequency Network
Session Initiation Protocol
Subscriber Profile Repository
Structural SIMilarity Index
Scalable Video Coding
Video On Demand
Video Quality Experts Group
Web Real Time Communication
69
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
LIST OF REFERENCES
[1] Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2012–2017
http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11520862.html
[2] Ericsson Consumer Lab Report,
http://www.ericsson.com/res/docs/2012/consumerlab/consumerlab-tv-video-changing-the-game.pdf
[3] Ericsson AB, TV & Video, Changing the Game, 2012
http://www.ericsson.com/res/docs/2012/traffic_and_market_report_june_2012.pdf
[4] Ericsson Mobility Report, November 2012
http://www.ericsson.com/res/docs/2012/ericsson-mobility-report-november-2012.pdf
[5] Sandvine Global Internet Phenomena Report, 2H 2012
http://www.sandvine.com/downloads/documents/Phenomena_2H_2012/Sandvine_Global_Internet_Phen
omena_Report_2H_2012.pdf
[6] NFL Super Bowl SLVII Live Stream Sets Viewership Records, February 5, 2012
http://www.nfl.com/superbowl/story/0ap1000000136438/article/super-bowl-xlvii-live-stream-setsviewership-records
[7] BBC Player, Monthly Performance Pack, November 2012
http://downloads.bbc.co.uk/mediacentre/iplayer/iplayer-performance-nov12.pdf
[8] Ooyala, Global Video Index, Q3 2012
http://go.ooyala.com/rs/OOYALA/images/Ooyala-Global-Video-Index-Q3-2012.pdf
[9] ITU-T G.114 Telecommunication Standardization Sector of ITU (05/2003)
[10] 4G Americas White Paper, Mobile Broadband Explosion, August 2012
http://www.4gamericas.org/documents/4G%20Americas%20Mobile%20Broadband%20Explosion%20Aug
ust%2020121.pdf
[11] Balakrishnan, Hari, Padmanabhan, Venkata N., Seshan Srinivasan and Katz, Randy, H. “A
Comparison of Mechanisms for Improving TCP Performance over Wireless Links” August 1996
[12] Kuhlins, Christian, Ericsson & Thomas, Remi, FT Orange LSTI, Boosting LTE for Mobile Broadband
Deployment http://www.slideshare.net/guest99ced7/lsti-mwc-presentationfinal, page 11.
[13] 4G Americas White Paper, Developing and Integrating a High Performance HET-NET, October 2012
http://www.4gamericas.org/documents/4G%20Americas%20Developing%20Integrating%20High%20Performance%20HET-NET%20October%202012.pdf.
[14] 3G Americas White Paper, MIMO Transmission Schemes for LTE and HSPA Networks, June 2009
http://www.3gamericas.org/documents/mimo_and_smart_antennas_for_3g_and_4g_wireless_systems_M
ay%202010%20Final.pdf
[15] Sesia, I. Toufik and M. Baker, LTE, the UMTS Long Term Evolution: from Theory to Practice, 2
edition, John Wiley & Sons (2011).
nd
70
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
[16] Alcock, S. and Nelson, R., “Application Flow Control in YouTube Video Streams”, ACM SIGCOMM
Computer Communication Review, 41, 2 (2011).
[17] Ericsson Traffic Labs Measurements
[18] NGMN Alliance White Paper, Guidelines for LTE Backhaul Traffic Estimation, July 2011
http://www.ngmn.org/uploads/media/NGMN_Whitepaper_Guideline_for_LTE_Backhaul_Traffic_Estimatio
n.pdf
[19] Evans, John and Filsfils, Clarence, Deploying IP and MPLS QoS for Multiservice Networks, Theory
and Practice, March 2007
[20] Conviva Viewer Experience Report, Q1 2013, http://www.conviva.com/vxr/
[21] IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB),
Analytical method for objective scoring of HTTP Adaptive Streaming (HAS), 2012
[22] Robinson, David C, Jutras, Yves, Cracium, Viorel, Bell Labs Technical Journal, Volume 16, Issue 4,
Pages 5-23. Subjective Video Quality Assessment of HTTP Adaptive Streaming Technologies, March
2012.
th
[23] ACM SIGCOMM 12 Annual Conference on Internet Measurement, IMC, Confused, timid, and
unstable: picking a video streaming rate is hard. Boston, MA, USA, November 14-16, 2012
[24] NGMN Alliance, Plan to determine Video Quality of HAS in a Mobile Environment, NGMN-SERQU
HAS_TestPlan_Draft10_ChangesAccepted_Disclaimer” March 2012.
[25] Qualcomm Commissioned White Paper by iGR, Content for All – The Potential for LTE
Broadcast/eMBMS, January 2013
[26] Qualcomm White Paper, LTE Broadcast, A revenue enabler in the mobile media era, February 2013
[27] Holma, H. & Toskala, A (eds.), HSDPA/HSUPA for UMTS: High Speed Radio Access for Mobile
Communications, John Wiley & Sons (2006).
[28] ISO/IEC 14496-10 – MPEG-4 Part 10, Advanced Video Coding, Annex G Scalable Video Encoding
rd
[29] W3C Web & TV Workshop, 3 Annual, Scalable Video Coding based DASH for efficient usage of
network resources” position paper, September 2011
http://www.w3.org/2011/09/webtv/papers/Sanchez_et_al_W3C.pdf
[30] RFC 3481 TCP over Second (2.5G) and Third (3G) Generation Wireless Networks
[31] Ericsson Whitepaper: “LTE Broadcast”
http://www.ericsson.com/res/docs/whitepapers/wp-lte-broadcast.pdf
LIST OF FIGURES
Figure 1: Number of hours per week spent watching TV or Video on various devices [4] ........................... 7
Figure 2: Projection of mobile traffic trends in the United States .................................................................. 8
Figure 3: Global Shares of YouTube and Real-Time Entertainment Traffic[5] ............................................. 9
Figure 4: Streaming Video usage by users on different days in 2009 [FLO TV Incorporated] ................... 10
Figure 5: HTTP Adaptive Streaming ........................................................................................................... 16
71
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Figure 6: Message exchanges during HTTP Adaptive Streaming .............................................................. 17
Figure 7: LTE Spectral Efficiency as a Function of the System Bandwidth [10] ......................................... 23
Figure 8: HET-NET architecture [13] .......................................................................................................... 25
Figure 9: Effect of video playout buffers on Relative network load (X-axis) and Relative battery
consumption (Y-axis) [17] ........................................................................................................................... 31
Figure 10: Typical backhaul architecture [18] ............................................................................................. 32
Figure 11: Cell throughput during busy time and quiet time [18] ................................................................ 33
Figure 12: Performance of HAS TCP versus Theoretical TCP [22] ............................................................ 38
Figure 13: Downward spiral behavior of video throughput [23] .................................................................. 39
Figure 14: HET-NET Architecture ............................................................................................................... 45
Figure 15: eMBMS/LTE Broadcast Architecture ......................................................................................... 48
Figure 16: Content Delivery Network (CDN) Architecture .......................................................................... 49
Figure 17: Rollout of an operator’s IPTV service ........................................................................................ 50
LIST OF TABLES
Table 1: Data Consumed by Different Streaming Applications[10]............................................................. 21
Table 2: Performance comparison of 3GPP mobile technologies [10] ....................................................... 22
Table 3: Potential use cases and business models for LTE Broadcast services [31] ................................ 47
Table 4: Standardized QCI characteristics [from 3GPP TS23.203]………………………………………..…55
Table 5: Recommended choice of bit rates for video streams…….………………………………………..…67
ACKNOWLEDGEMENTS
The mission of 4G Americas is to promote, facilitate and advocate for the deployment and adoption of the
3GPP family of technologies throughout the Americas. 4G Americas' Board of Governor members include
Alcatel-Lucent, América Móvil, AT&T, BlackBerry, Cable & Wireless, CommScope, Entel, Ericsson,
Gemalto, HP, Nokia Siemens Networks, Openwave Mobility, Powerwave, Qualcomm, Rogers, T-Mobile
USA, and Telefónica.
4G Americas would like to recognize the joint project leadership, great efforts and important contributions
of Kamakshi Sridhar (Alcatel-Lucent) and Gautam Talagery (Ericsson) as well as representatives from the
other member companies on 4G Americas.
72
4G Americas – Supporting Wireless Video Growth and Trends -- April 2013
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement