Content uploaded by Minh Nguyen
Author content
All content in this area was uploaded by Minh Nguyen on Jul 28, 2021
Content may be subject to copyright.
WISH: User-centric Bitrate Adaptation for HTTP
Adaptive Streaming on Mobile Devices
Minh Nguyen, Ekrem C¸ etinkaya, Hermann Hellwagner, Christian Timmerer
Christian Doppler Laboratory ATHENA, Institute of Information Technology (ITEC)
Alpen-Adria-Universit¨
at Klagenfurt, Klagenfurt, Austria
Abstract—Recently, mobile devices have become paramount
in online video streaming. Adaptive bitrate (ABR) algorithms of
players responsible for selecting the quality of the videos face
critical challenges in providing a high Quality of Experience
(QoE) for end users. One open issue is how to ensure the optimal
experience for heterogeneous devices in the context of extreme
variation of mobile broadband networks. Additionally, end users
may have different priorities on video quality and data usage
(i.e., the amount of data downloaded to the devices through the
mobile networks). A generic mechanism for players that enables
specification of various policies to meet end users’ needs is still
missing. In this paper, we propose a weighted sum model, namely
WISH, that yields high QoE of the video and allows end users to
express their preferences among different parameters (i.e., data
usage, stall events, and video quality) of video streaming. WISH
has been implemented into ExoPlayer, a popular player used in
many mobile applications. The experimental results show that
WISH improves the QoE by up to 17.6% while saving 36.4%
of data usage compared to state-of-the-art ABR algorithms and
provides dynamic adaptation to end users’ requirements.
Index Terms—ABR Algorithms, HTTP Adaptive Streaming,
ITU-T P.1203, WISH
I. INTRODUCTION
The COVID-19 pandemic has resulted in a remarkable
increase in video consumption due to the lockdown time.
Video content watched globally has increased by 60% [19].
In addition, Cisco [6] estimates that video data will occupy
82% of the Internet traffic by 2022. Mobile devices, including
smartphones and tablets, will also become predominant in
video streaming with more than 70% of watch time [20].
HTTP-based Adaptive Streaming (HAS) is a crucial tech-
nology for delivering videos over the Internet that brings
fewer stall events and higher video quality. In HAS, a video
at the server is encoded in multiple representations. Each
representation is further split into temporal segments. The
client executes an adaptive bitrate (ABR) algorithm to get
the most suitable representation of a segment before sending
an HTTP request to the server. Despite its popularity, video
streaming over HAS on mobile devices faces some challenges
such as severe throughput fluctuations in mobile broadband
networks [14] and the heterogeneity of mobile devices. The
former challenge makes it difficult for ABR algorithms to
select suitable bitrates to avoid stall events while keeping the
video in high quality. The latter means an ABR algorithm
might not be a one-size-fits-all solution to provide the optimal
Quality of Experience (QoE) for all mobile devices.
In this paper, we address the aforementioned problems by an
ABR algorithm based on a WeIghted Sum model for HAS,
namely WISH. WISH takes into account three cost factors
of each video representation: (a) throughput cost,(b) buffer
cost, and (c) quality cost. The throughput cost indicates the
amount of downloaded data which refers to the data usage of
mobile devices. The buffer cost and quality cost represent the
stalling risk and the deterioration in quality of a representation,
respectively. A low-quality representation with a small bitrate
provides a low throughput cost due to less data delivered in
broadband networks and a low buffer cost because of less
time to download, with low risk of stalling events. However, it
suffers from high quality cost due to a low quality. An overall
cost, which is a weighted sum of these penalties, represents
each video representation. Our proposed method provides
the optimal segment bitrate by choosing the representation
with the lowest overall cost for a certain time. A solution
determining the weights of the cost factors based on end users’
preferences and mobile devices will be introduced in this work.
WISH’s source code is available on Github1.
In summary, the contributions of this paper are three-fold:
1) WISH. We propose a novel ABR algorithm with the
objective of trading off the video quality, rebuffering,
and mobile devices’ data usage. Our approach considers
three categories of costs of each video representation
and selects the lowest-cost representation for the next
segment.
2) Weights determination. Each cost has a specific impact
on the overall cost function illustrated by a weight. End
users express their preferences by setting corresponding
weights. We propose a mathematical formula to map
the end user’s needs to the weights’ values, rather than
determining them by conjecture and empiricism.
3) Evaluation. Extensive experiments with various test se-
quences have been conducted to compare WISH with
state-of-the-art ABR algorithms. Moreover, WISH’s abil-
ity to reflect end user preferences in video streaming is
investigated in our experiments.
The remainder of this paper is structured as follows. Sec-
tion II provides an overview of related work. The details of
WISH are presented in Section III. The experimental results
are discussed in Section IV and the conclusions are provided
in Section V.
1The URL of the source code will be provided upon acceptance.
II. RE LATE D WOR K
Many ABR algorithms have been proposed in the literature,
which can be classified into (i) throughput-based,(ii) buffer-
based, and (iii) hybrid-based [4]. A simple throughput-based
ABR algorithm that selects the highest bitrate but smaller than
the estimated throughput is used as a default in ExoPlayer2–
a media player commonly used in thousands of applications,
including YouTube and WhatsApp [13]. As a buffer-based
algorithm, BBA-0 [7] relies on an increasing function of
instant buffer status to determine segments’ qualities. If the
current buffer is less than a low threshold, the lowest quality
level is chosen. If it is more than a high threshold, BBA-
0 selects the highest quality level. Otherwise, the value of
that function demonstrates the next segment’s representation.
Wang et al. [23] propose a hybrid-based ABR algorithm,
namely SQUAD, that takes into account the available through-
put and buffer occupancy. SQUAD starts the streaming session
with multiple low-quality segments. Then, this approach uses
a so-called spectrum, which is the variation of the segment
bitrates, to minimize the number of switches and maximize the
average video bitrate. The selected representation should be
sustainable (i.e., the estimated download time does not exceed
the segment duration) and provides the lowest spectrum (i.e.,
the least bitrate variation).
Many approaches have been introduced for adaptive video
streaming to mobile devices. Ara´
ujo et al. [2] propose the
Energy-Saving Based Adaptation Algorithm (ESBA) that has
the goal to utilize the energy of the devices efficiently. To
determine the segment quality, ESBA not only considers
throughput estimation and buffer information but also takes
into account battery voltage and the energy power usage.
Yan et al. [25] design a Rate and Brightness (RnB) adaptation
framework. The relationship among bitrate, video quality,
and devices’ screen brightness is investigated. A suitable
bitrate and corresponding brightness are then selected accord-
ing to the throughput and representation’s quality to mini-
mize the power consumption of video streaming. This strat-
egy requires additional representations, namely luminance-
compensated versions, to those that are already encoded at
the server, to enable energy adaptation. It might lead to more
storage consumption at the server and decrease the scalability
of the ABR algorithm.
Weighted sum models have been utilized recently in HAS.
Nguyen et al. [16] propose a cost function as a weighted
sum of buffer cost and request cost to select the number of
segments to be downloaded by a single request in adaptive
video streaming over HTTP/2 [3], [15]. A high number of
segments may harm the buffer when the throughput drops
as the client has to download all of these segments before
switching to a lower quality level. In return, there are fewer
requests sent to the server, which reduces network overhead;
in other words, the request cost goes down. Similar to this
work, Le et al. [10] add a smoothness-related cost to penalize
a quality level if it differs from the quality of the current
2https://developer.android.com/guide/topics/media/exoplayer.
TABLE I: Notations used in this paper.
Notation Definition
NThe number of representations
RThe set of available bitrates R={R1, R2, ..., RN}
τSegment duration
q(i)Quality of representation i
QnQuality of segment n
TnAverage throughput measured at segment n
Ts
nSmoothed throughput at segment n
TeEstimated throughput for the next segment
BBuffer size
BlThreshold of low buffer occupancy
BnBuffer occupancy at segment n
segment. Compared to these approaches, our proposed method
considers the throughput cost and the quality cost. The former
cost is the amount of data downloaded in a second compared
to the available throughput, whereas the latter includes the
difference between a quality level and the highest representa-
tion, and between that quality level and the average quality of
recent segments. Moreover, the weights of the cost elements
in the related papers are chosen empirically. In contrast, we
introduce a mathematical method to determine dynamically
those parameters based on the video profile (i.e., bitrate ladders
and segment duration), client characteristics (i.e., buffer size),
and end users’ preferences.
III. WISH: AWEIGHTED SUM MO DE L FO R HAS
A. Throughput Estimation
The simplest method to estimate the future throughput Te
is using the average throughput Tnof the current segment n.
Due to strong throughput variations in mobile networks, this
strategy may result in serious fluctuation of Tethat causes
frequent switches in the selected qualities. To tackle this issue,
we implement the “smooth” method introduced in [1]:
Ts
n=((1 −ω)×Ts
n−1+ω×Tn,if n > 1,
Tn,if n= 1,(1)
where Ts
nis the smooth throughput for downloading seg-
ment nand ωis a weight in the range [0,1]. In our paper,
ωis set to 1/8for smoothing computation, which is similar
to [9], [18].
A disadvantage of the “smooth” method is that it might
react late with the throughput decrease [9]. Therefore, we set
the estimated throughput as the last measured throughput if
the network is unfavorable. Otherwise, the smooth throughput
is used. The estimated throughput is determined as follows:
Te= min{Ts
n, Tn}.(2)
B. WISH ABR Algorithm
In this section, we describe our proposed WISH algorithm
that selects the quality of the next segment according to the
lowest-cost outcome. The notations used in our algorithm are
shown in Table I.
1) Start-up phase:If the current buffer level Bnis less than
a threshold Bl, then the buffer is in a dangerous zone with
high stall possibility. WISH selects the lowest representation
to start the playback as fast as possible at the beginning of the
streaming session and to decrease the risk of stalls.
2) Steady phase:If the current buffer exceeds Bl, the
streaming session switches to a steady phase. In this phase,
WISH calculates the cost of representation i,C(i), which is a
weighted sum of throughput cost Ct(i),buffer cost Cb(i), and
quality cost Cq(i)as presented in Eq. (3); the weights α, β, γ
are positive numbers:
C(i) = α×Ct(i) + β×Cb(i) + γ×Cq(i).(3)
The throughput cost of a representation is calculated as a
linearly increasing function of its bitrate as follows:
Ct(i) = Ri
Te.(4)
The buffer cost is defined based on the following obser-
vations. While downloading representation i, the buffer is
drained by the download time, which is Ri×τ
Te. The higher the
download time, the more the buffer is decreased. Also, with
the same download time, a buffer at a low level has a higher
risk of under-running, which would result in a stall event.
Therefore, the buffer cost of representation iis computed as:
Cb(i) = Ri×τ
(Bn−Bl)×Te.(5)
Finally, the quality cost comprises two sub-penalties: (i) a
penalty when a representation is lower than the highest-bitrate
representation, and (ii) a penalty if it is different from the
average quality of the recent segments. To make the quality
cost positive, we use an exponential function:
Cq(i) = e[q(N)−q(i)]+[Qk−q(i)]
e2×q(N)−2×q(1) ,(6)
where q(i)is the quality of representation i, and Qkis the
average quality of the last ksegments (default k= 10). In
this paper, we use the bitrate Rito calculate the quality of the
i-th representation as:
q(i) = Ri
RN
.(7)
Our proposed ABR algorithm is presented in Algorithm 1.
As the lowest bitrate is selected in the start-up phase when
the buffer occupancy is too small, it is reasonable to omit
this representation when the buffer is not in danger. Thus, the
minimum considered representation is the second-lowest one
in the steady phase (line 7 in Algorithm 1). Also, when a high
bitrate level is selected, and afterward, the throughput suddenly
drops, the buffer occupancy will be decreased dramatically,
which might result in a low bitrate for the next decisions and
QoE impairment. To avoid this issue, WISH only considers the
representations whose bitrates are less than the last throughput
with the margin µ(µ= 0.1in this paper) (lines 8 – 9 in
Algorithm 1).
C. Weights determination
It is nontrivial to find suitable values for the weight set
(α, β, γ ) that meet the end users’ needs and provide a high
QoE. In this work, we introduce a strategy to dynamically set
the weights according to videos’ characteristics.
Algorithm 1: WISH ABR algorithm.
1if Bn< Blthen
2Rc=R1;
3else
4Qk=1
kPk−1
i=0 Qn−i;
5m= 2;
6M= max{i|Ri< Tn×(1 + µ), m ≤i≤N};
7for i=m, m + 1, ..., M do
8Ct(i) = Ri
Te;
9Cb(i) = Ri×τ
(Bn−Bl)×Te;
10 Cq(i) = e[q(N)−q(i)]+[Qk−q(i)]
e2×q(N)−2×q(1) ;
11 C(i) = α×Ct(i) + β×Cb(i) + γ×Cq(i);
12 jmin = arg min
j∈{m,...,M}
C(j);
13 Rc=Rjmin ;
14 return Rc;
The cost C(i)can be considered as a function f(Ri)of the
bitrate Rias follows:
f(Ri) = α×Ri
Te+β×Ri×τ
(Bn−Bl)×Te+γ×eq(N)+Qk−2×q(i)
e2×q(N)−2×q(1) .
(8)
The derivative of f(Ri)is presented in Eq. (9).
f0(Ri)=(α
Te+β×τ
(Bn−Bl)×Te)−2×γ×e1+Qk−2
RN×Ri
A×RN
,
(9)
where A=e2×q(N)−2×q(1) is a constant. It can be seen that
the value of Rithat makes f(Ri)lowest is Xif f0(X)=0.
The weights α, β, γ are determined by making the maxi-
mum bitrate RNown the lowest cost (i.e.,f0(RN)=0) under
the following conditions. Firstly, the estimated throughput Te
should be high enough, compared to the maximum bitrate RN,
to avoid buffer drainage. Secondly, the average quality of the
last segments Qkis high to indicate that the recent network
condition is favorable. Finally, the current buffer Bnneeds
to be high to prevent stall events in the case of a sudden
throughput drop. These constraints are described as follows:
f0(RN) = 0, if
Te=δ×RN
Bn=ξ×B
Qk= Q
(∗)
where δ(δ > 0) indicates the capability of the network to
deliver the maximum bitrate and ξ∈(0,1] is the proportion
of filled buffer. Qis a threshold representing the recent average
quality is in high level. In this paper, we set ξ= 0.8, which
is similar to [18], δ= 1, and Q = q(RN−1)as default.
Now, we need to find (α, β, γ ) to obtain (∗). Without loss
of generality, we can set the sum of these parameters to 1 (i.e.,
α+β+γ= 1). For relaxation, we set α, β so that throughput
and buffer costs have the same contribution in the overall cost
of the highest representation (i.e.,α×Ct(N) = β×Cb(N)).
Thus, the relationship of α, and βcan be demonstrated as
β=α×ξ×B−Bl
τ.(11)
Finally, the weights α, β, γ can be determined as follows:
α= (1 + ξ×B−Bl
τ+e3−2×R1
Rn−Q
δ)−1
β=α×ξ×B−Bl
τ
γ= 1 −α−β
(∗∗)
IV. EVALUATION AND DISCUSSION
A. Experimental Setup and Metrics
Our testbed consists of a server and a client connected via
a WIFI interface. The server hosts an Apache web server
running on a Thinkpad P1 Gen2 (i7/16GB) with Ubuntu 18.04.
WISH is implemented in ExoPlayer running on a Xiaomi Mi
11 (256 GB, Android 11). The buffer size of this player is set
to 20s (B= 20) and the threshold Blis equal to 4s.
The network conditions are controlled by the Linux traffic
control tool, tc, on the server. In our experiments, a 4G
network trace [21] collected on bus rides is used to emulate
the mobile network conditions. The average bandwidth in this
trace is 3787 Kbps with a standard deviation of 3193 Kbps,
and the maximum bandwidth is 17361 Kbps. The round trip
time is set to 20ms that is reasonable in 4G networks3.
We use four videos as the test sequences stored at the server,
each of which is 5 minutes long: (i) Tears of Steel (the first 5
mintues) – ToS1, (ii) Gameplay4[24], (iii) Rally 5,(iv) Tears
of Steel (the last 5 minutes) – ToS2. We want to choose
videos with different characteristics as our test sequences.
The Rally sequence consists of natural captured content, the
Gameplay sequence consists of computer-generated content,
and the ToS1 and ToS2 sequences contain both natural and
computer-generated content. These characteristics make them
a suitable set of test sequences. Moreover, these videos have
different spatial information (SI) and temporal information
(TI) as shown in Fig. 1. The test sequences are encoded into
seven quality levels R={107, 240, 346, 715, 1347, 2426,
4121}Kbps at {144, 240, 360, 480, 720, 1080, 1440}p res-
olutions, respectively, using the state-of-the-art H.265/HEVC
codec. HEVC is chosen as the codec since market-adaption
of the H.266/VVC is still too limited, and using H.264/AVC
would not have been suitable as it is more than a decade
old. The bitrate ladder is based on the ToS video stored in
Youtube6. Each quality level is then divided into segments of
4s (τ= 4) as recommended in [5]. WISH is compared with
three state-of-the-art approaches described in Section II, i.e.,
BBA-0, SQUAD, and the default ABR algorithm of ExoPlayer,
each of which belongs to a class of ABR algorithms [4].
The performance of those ABR algorithms is evaluated with
common QoE-related metrics as follows:
3https://iwl.com/idocs/emulating-3g-4g-networks
4https://www.youtube.com/watch?v=gkIYZCmD-40
5https://www.youtube.com/watch?v=OQbDi3PnB2g
6https://www.youtube.com/watch?v=OHOpb2fS-cM
Fig. 1: Spatial and temporal information of the test sequences.
•Average bitrate: The average bitrate of all the segments
downloaded by the client.
•Number of switches: The number of segments whose
bitrate is lower than that of their previous one.
•Video instability: The average change in the level of
representation of contiguous segments.
•Number of stalls and stall duration: The number of
rebuffering and the total video freeze time, respectively.
•QoE scores: ITU-T Rec.P.1203 mode 0 [8] is used
to estimate the perception QoE of the end users. The
implementation of this model [22] with its extension for
up to 4K HEVC videos is available on Github7.
For accuracy, each ABR scheme is executed five times
and the averages and standard deviations are presented in the
following.
B. Experimental Results
1) WISH performance compared to state-of-the-art ABR
algorithms: Fig. 2–Fig. 5 show the performance of WISH
with the default setup and existing ABR schemes for different
videos. It can be seen that WISH achieves the highest QoE
scores for all test sequences, as illustrated in Fig. 2. Its QoE
scores are in the range of 3.46 (GamePlay) to 3.71 (ToS2),
whereas this figure for other related work is no more than
3.40 in most cases except ExoPlayer in the ToS2 video. Thus,
WISH’s QoE scores are higher than others by up to 17.6%.
An interesting finding is that the built-in ABR algorithm of
ExoPlayer yields better QoE scores than BBA-0 and SQUAD.
It can be attributed to the number of stalls and stall duration
of those ABR algorithms in Fig. 3. WISH has the fewest
number of stalls with at most one stall. In particular, there is
not any stall in ToS1 and ToS2 videos. Regarding GamePlay
and Rally videos, WISH experiences on average 0.4 stall
events that last for 1.1s and 1.2s, respectively. On the other
hand, BBA-0 and SQUAD suffer severely from stalls with
on average more than two stall events, and the stall duration
varies from 15s to 30s, which may make the end users
quit the video watching [11], [12]. This is because BBA-0
only takes the current buffer occupancy into account without
considering the network condition, and SQUAD makes an
effort to provide stable-quality videos. These strategies react
slowly to the severe throughput fluctuation, which results in
7https://github.com/Telecommunication-Telemedia-Assessment/itu-
p1203-codecextension.
Fig. 2: ITU-T QoE scores.
Fig. 3: Number of stalls and stall duration.
Fig. 4: Average bitrate of each segment.
Fig. 5: Video instability.
long stall duration when the throughput drops significantly. In
contrast, ExoPlayer adapts quickly to the network oscillation;
thus, it reduces the number of stalls to less than two, and the
stall duration is from 3.8s (ToS2 video) to 17.0s (GamePlay),
but still more than when using WISH. With the objective of
balancing bitrate, stalls, and segment quality, WISH does not
try to download high-quality segments, which avoids stalls
when the network is unfavorable.
The average bitrate of ABR algorithms is reported in Fig. 4.
In general, BBA-0 downloads segments with the highest
average bitrate, whereas this figure for WISH is the lowest.
Each segment in BBA-0 has an average bitrate of 2800 Kbps,
which is higher than that in WISH by 36.4%. ExoPlayer
and SQUAD provide 2198 Kbps and 2670 Kbps segments,
respectively, compared to 2053 Kbps segments of WISH. It
should be noted that the higher the average bitrate, the more
data usage. Therefore, WISH saves a large amount of data
usage from 7.1% to 36.4%. These analyses reveals that WISH
is able to trade off video quality, buffer (i.e., stalls and stall
duration), and mobile devices’ data usage.
However, to achieve such performance, WISH sacrifices the
video stability (Fig. 5). The video instability of our method
is around 0.6, compared to 0.51, 0.54, and 0.42 of BBA-0,
ExoPlayer, and SQUAD, respectively. On average, there are
less than six quality switches in BBA-0 and SQUAD as these
methods adapt gradually with the throughput drop. On the
other hand, WISH and ExoPlayer experience respectively 14.8
and 15 quality switches in each streaming session.
2) WISH performance with different end user preferences:
Table II shows the WISH’s results with different end user
preferences by changing the safe threshold ξof the buffer.
The requirement of the high video quality is increased from
preference #1 (γ= 0.68, ξ = 1.0) to preference #4 (γ=
0.84, ξ = 0.4). The QoE scores are not reported here as the
ITU-T P.1203 model provides the scores for general viewers
and it might not reflect correctly the experience of end users
with specific preferences.
In Table II, the average bitrate acquired by WISH increases
with γ, from 1841 Kbps at γ= 0.68 to 2270 Kbps at γ=
0.84. Additionally, there is a growth in the video instability
with 0.54 in #1 and 0.61 in #4, and the average number of
switches (13.4 switches to 15.0 switches). This is attributed
to the effort of selecting segments in high bitrates. When γis
set to a large value, a small increase in Cq(i)due to a low
bitrate results in a severe penalty in the overall cost C(i)and
WISH tries to download high bitrate levels though the network
conditions are unfavorable. Consequently, the buffer might be
decreased thoroughly, and lower bitrates have to be selected
TABLE II: WISH performance with different end user preferences for ToS2.
Preference #1 (ξ= 1.0) #2 (ξ= 0.8) #3 (ξ= 0.6) #4 (ξ= 0.4)
Average bitrate (Kbps) 1841 2002 2107 2270
Average video instability 0.54 0.57 0.58 0.62
Average # of switches 13.4 14.0 14.6 15.0
Average # of stall & stall duration (s) 0 & 0 0 & 0 0 & 0 0.6 & 5.3
(α, β, γ )(0.06, 0.26, 0.68) (0.07, 0.21, 0.72) (0.07, 0.15, 0.78) (0.08, 0.08, 0.84)
for upcoming segments, or the playback stalls if the buffer
is empty (e.g., in #4 with an average of 0.6 stalls that lasts
for 5.3s in each session). Because the estimated throughput is
conservative in increasing network conditions, the bitrates of
later segments are increased conservatively, and the average
video instability is higher than for the end user preferences in
which the video quality owns a lower priority.
V. CONCLUSIONS AND FUTURE WORK
In this paper, we introduced a weighted sum model, named
WISH, to provide high QoE for mobile devices and to meet
end users’ requirements. WISH considers three types of cost,
i.e.,throughput cost,buffer cost, and quality cost, of each video
representation and combines them in a weighted sum as the
overall cost. The representation with the lowest overall cost
is selected. This strategy allows end users to express their
preferences by varying the weight of each cost. We also pro-
pose a mathematical solution to choose those weights based on
video characteristics (i.e., bitrate ladder, segment duration), the
client (i.e., buffer), and end users’ priorities. WISH has been
deployed into ExoPlayer, a common media player utilized
in commercial mobile applications. The experimental results
show that the proposed method needs the lowest data usage
while keeping the highest QoE scores. In the future, we will
integrate the retransmission technique H2BR [17] to improve
the video stability of WISH.
ACKNOWLEDGMENT
The financial support of the Austrian Federal Ministry for Digital
and Economic Affairs, the National Foundation for Research, Tech-
nology and Development, and the Christian Doppler Research As-
sociation, is gratefully acknowledged. Christian Doppler Laboratory
ATHENA: https://athena.itec.aau.at/.
REFERENCES
[1] S. Akhshabi, S. Narayanaswamy, A. C. Begen, and C. Dovrolis. An
experimental evaluation of rate-adaptive video players over HTTP.
Signal Processing: Image Communication, 27(4):271–287, 2012.
[2] F. Ara´
ujo, D. Ros´
ario, E. Cerqueira, and L. A. Villas. A hybrid
energy-aware video bitrate adaptation algorithm for mobile networks.
In 2019 15th Annual Conf. on Wireless On-demand Network Systems
and Services (WONS), pages 146–153. IEEE, 2019.
[3] M. Belshe, R. Peon, and M. Thomson. Hypertext Transfer Protocol
Version 2 (HTTP/2). RFC 7540, May 2015.
[4] A. Bentaleb, B. Taani, A. C. Begen, C. Timmerer, and R. Zimmermann.
A survey on bitrate adaptation schemes for streaming media over HTTP.
IEEE Communications Surveys & Tutorials, 21(1):562–585, 2018.
[5] Bitmovin Inc. Optimal Adaptive Streaming Formats MPEG-DASH
HLS Segment Length. [Online] Available: https://bitmovin.com/
mpeg-dash-hls-segment-length/, April 2020. Accessed: 10 June 2021.
[6] Cisco. Global - 2022 Forecast Highlights. [Online] Avail-
able: https://www.cisco.com/c/dam/m/en us/solutions/service-provider/
vni-forecast- highlights/pdf/Global 2022 Forecast Highlights.pdf. Ac-
cessed: 09 June 2021.
[7] T. Huang, R. Johari, N. McKeown, M. Trunnell, and M. Watson. A
buffer-based approach to rate adaptation: Evidence from a large video
streaming service. In ACM SIGCOMM Computer Communication
Review, volume 44, pages 187–198. ACM, 2014.
[8] ITU-T. Rec. P.1203. Parametric bitstream-based quality assessment of
progressive download and adaptive audiovisual streaming services over
reliable transport - video quality estimation module. http://handle.itu.
int/11.1002/ps/P1203-01. Accessed: 10 February 2021.
[9] H. T. Le, N. P. Ngoc, and T. C. Thang. Bitrate adaptation for seamless
on-demand video streaming over mobile networks. Signal Processing:
Image Communication, 65:154–164, 2018.
[10] H. T. Le, T. Vu, N. P. Ngoc, A. T. Pham, and T. C. Thang. Seamless
mobile video streaming over HTTP/2 with gradual quality transitions.
IEICE Transactions on Communications, 100(5):901–909, 2017.
[11] P. Lebreton, K. Kawashima, K. Yamagishi, and J. Okamoto. Study on
viewing time with regards to quality factors in adaptive bitrate video
streaming. In 2018 IEEE 20th Int’l. Workshop on Multimedia Signal
Processing (MMSP), pages 1–6. IEEE, 2018.
[12] P. Lebreton and K. Yamagishi. Study on user quitting rate for adaptive
bitrate video streaming. In 2019 IEEE 21st Int’l. Workshop on Multi-
media Signal Processing (MMSP), pages 1–6. IEEE, 2019.
[13] C. Minseo. Playing video by ExoPlayer. [Online] Available: https:
//medium.com/fungjai/playing-video- by-exoplayer-b97903be0b33. Ac-
cessed: 20 February 2021.
[14] C. M ¨
uller, S. Lederer, and C. Timmerer. An evaluation of dynamic
adaptive streaming over HTTP in vehicular environments. In 4th
Workshop on Mobile Video (MoVid), pages 37–42. ACM, 2012.
[15] C. M ¨
uller, S. Lederer, C. Timmerer, and H. Hellwagner. Dynamic
adaptive streaming over HTTP/2.0. In 2013 IEEE Int’l. Conf. on
Multimedia and Expo (ICME), pages 1–6, 2013.
[16] D. V. Nguyen, H. T. Le, P. N. Nam, A. T. Pham, and T. C. Thang.
Adaptation method for video streaming over HTTP/2. IEICE Commu-
nications Express, 5(3):69–73, 2016.
[17] M. Nguyen, C. Timmerer, and H. Hellwagner. H2BR: An HTTP/2-
based Retransmission Technique to Improve the QoE of Adaptive Video
Streaming. In Proceedings of the 25th Packet Video Workshop, pages
1–7, 2020.
[18] T. Nguyen, N. H. Dang, N. Minh, N. P. Ngoc, H. T. Le, and T. C.
Thang. An Efficient Server Push Approach for On-Demand Video
Streaming Over HTTP/2. IEICE Transactions on Communications,
E101.B(11):2371–2379, 2018.
[19] Nielsen. COVID-19: Tracking The Impact on Media Consumption.
[Online] Available: https://www.nielsen.com/us/en/insights/article/2020/
covid-19-tracking-the-impact-on-media-consumption/. Accessed: 20
February 2021.
[20] Omnicore. YouTube by the Numbers: Stats, Demographics
Fun Facts. [Online] Available: https://www.omnicoreagency.com/
youtube-statistics/. Accessed: 20 February 2021.
[21] D. Raca, J. J. Quinlan, A. H. Zahran, and C. J. Sreenan. Beyond
throughput: a 4G LTE dataset with channel and context metrics. In
Proceedings of the 9th ACM Multimedia Systems Conference, pages
460–465, 2018.
[22] W. Robitza, S. G¨
oring, A. Raake, D. Lindegren, G. Heikkil¨
a, J. Gustafs-
son, P. List, B. Feiten, U. W¨
ustenhagen, M. N. Garcia, and K. Yamagishi.
HTTP adaptive streaming QoE estimation with ITU-T rec. P. 1203: open
databases and software. In Proceedings of the 9th ACM Multimedia
Systems Conf., pages 466–471, 2018.
[23] C. Wang, A. Rizk, and M. Zink. SQUAD: A spectrum-based quality
adaptation for dynamic adaptive streaming over HTTP. In Proceedings
of the 7th Int’l. Conf. on Multimedia Systems, pages 1–12, 2016.
[24] Y. Wang, S. Inguva, and B. Adsumilli. YouTube UGC dataset for video
compression research. In 2019 IEEE 21st Int’l. Workshop on Multimedia
Signal Processing (MMSP), pages 1–5. IEEE, 2019.
[25] Z. Yan and C. W. Chen. RnB: Rate and brightness adaptation for
rate-distortion-energy tradeoff in HTTP adaptive streaming over mobile
devices. In Proceedings of the 22nd Annual Int’l. Conf. on Mobile
Computing and Networking, pages 308–319, 2016.
