Optimizing Playback Quality of HTTP-Based Dynamic Adaptive Streaming on
Smartphones
Yayun Bao*, Lanshan Zhang
+
, Wendong Wang*, Xiangyang Gong*, Xirong Que*
* State Key Laboratory of Networking and Switching Technology,
Beijing University of Posts and Telecommunications, Beijing, China
E-mail: jimmybao0730@gmail.com, wdwang@bupt.edu.cn,
xygong@bupt.edu.cn, rongqx@bupt.edu.cn
+
Beijing Key Laboratory of Network System and Network Culture,
Beijing University of Posts and Telecommunications, Beijing, China
E-mail: zls326@sina.com
Abstract—Due to the seamlessly adaptation to variable wireless
network conditions, Dynamic Adaptive Streaming over HTTP
(DASH) has been widely used in today's video streaming
applications. Both the playback fluency and the average bitrate
are significant performance metrics for mobile streaming users.
However, the existing bitrate selection mechanisms based on
the inaccurate throughput estimation will lead to a feedback
loop, resulting in undesirably variable and low-quality video.
In this paper, we first present a basic understanding of DASH
system and make a simple analysis of the commercial players.
We then model the bitrate selection problem under constrains
of the buffer and the data requirement of users. To achieve
high quality and smoothness playback, we propose an online
dynamic video bitrate selection algorithm (DBS) based on the
instant throughput and the buffer state. Parameters (α, β) in
DBS can be set to adjust the bitrate switch frequency. The
extensive simulation demonstrates that DBS can improve
user's experience through video playback in different
situations.
Index Terms—DASH, Media Streaming Optimization, Bitrate
Selection.
I. INTRODUCTION
Video streaming is becoming a large fraction of global
mobile data traffic due to the rapid development wireless
communication (3G/LTE, WiFi). It has been estimated that
more than half of the mobile data is contributed by streaming
now and it will grow to 69% by 2019 [1]. However the
traditional video transmission method cannot fit the
changeable wireless conditions, which may cause the video
stalling frequently. To adapt to dynamic network conditions
seamlessly, Dynamic Adaptive Streaming over HTTP
(DASH) is raised and has been widely deployed in many
streaming applications, such as the Netflix's online video
streaming service. In DASH, the video is encoded into
multiple versions with different bitrates and fragmented into
small video chunks. Video players can dynamically change
the video bitrate according to the network conditions.
Significant efforts have focused on the rate adaptation
techniques of dynamic video streaming in the last decade.
Many works try to improve the accuracy of the bandwidth
estimation to choose the proper bitrate. The paper [2] has
made a detailed analysis of the bandwidth estimation using
the PID controlling. However the accurate client-side
bandwidth estimation is challenging as shown in [3]. The
inappropriate estimation will lead to a feedback loop for
users, which will result in undesirably variable and low-
quality video. Default strategy is to choose a bit-rate a bit
lower than the estimated throughput, whose bitrate switch
will be frequent. The bitrate selection of some commercial
streaming applications is based on the recorded throughput
of the last or last few samples. FESTIVE [4] uses the
harmonic mean throughput over the last 20 samples instead
of the instantaneous throughput. Using the last sample,
QDASH [5] integrates the measurement flow into the video
data flow to improve the accuracy of throughput estimation.
The throughput-based strategies can be inaccurate and
result in video stalling. In order to improve users' experience,
the buffer-based approaches are proposed, which choose the
video rate based on the current buffer level and without
relying on the bandwidth estimation [6]-[8].
In this paper, we would like to optimize the user's
experienced quality in the client side. The main metrics for
the client side to evaluate the quality are summarized in [9]-
[11]. Average bitrate: the average bitrate is used to reflect
the video quality as a whole. Users try to choose the highest
feasible bitrate to maximize their playback quality.
Rebuffering: the rebuffering will happen when the buffer is
starving due to the bandwidth competition and the
inappropriate bitrate selection. Bitrate switch frequency:
the quality of experience for watching a DASH video is
greatly influenced by the frequency of bitrate switch.
To deal with the challenges in the bitrate selection, we
first present a basic understanding of DASH system and
make a simple analysis of the existing commercial players.
We find that the bitrate selection is much lower than the
instant bandwidth of the channel sometimes due to the
average-throughput-based bitrate selection and the rate
degradation mechanism. And we get that the buffer size
limitation in the client side also influences the video quality.
To achieve high quality and smoothness playback, we then
proposes an online dynamic video bitrate selection
algorithms (DBS) based on the instant throughput and the
buffer state. We use two parameters α and β in DBS to adjust
the bitrate switch frequency to adapt to the variable network
condition. The extensive simulation demonstrates that DBS