Adaptive wireless video streaming and teleconferencing
Date
2017
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
With the popularity of and the advances in wireless networking technologies, wireless
multimedia tra c has grown dramatically in recent years. Despite having many advantages,
wireless multimedia services, particularly video services, still pose a number of challenges
due to the time-varying, error-prone and bandwidth-fluctuating channels in the wireless networks.
Therefore, provisioning end-to-end Quality of Service and Quality of Experience
(QoS and QoE) of video transmission over wireless channels is of great importance. ☐ Video transmission is often described to be bursty as video is basically a sequence of
frames transmitted at a particular frame rate. A video frame cannot be decoded or played out
at the receiver side until all or most of its transmitted constituent packets are received in time.
Depending on the application scenarios, video services may have different emphases in terms
of QoE and QoS. While video streaming (e.g., Netflix and YouTube) allows for modest delay
(on the order of a few seconds) at the beginning of the playout, video teleconferencing (e.g.,
FaceTime and WebRTC) is much more delay constrained (less than a few hundred milliseconds).
This is because in real-time video systems, each frame must be delivered and decoded
by its playback time, and any packet that is retransmitted due to loss in the last transmission
or arriving late becomes useless when its stringent decoding and display deadline cannot be
met. In this dissertation, we propose several optimization algorithms to improve the QoE
and QoS for both video streaming (non real-time) and video teleconferencing (real-time)
over wireless networks. ☐ In optimizing wireless video streaming, we focus on MPEG-DASH (ISO/IEC Standard
23009-1), the current standard for video streaming. We optimize video streaming by
leveraging a technique called User Adaptive Video (UAV), which exploits the perceptual
limits of the human visual system to modulate a video stream’s bit rate based on the viewing
conditions, such as viewing distance and ambient illuminance, resulting in significant
bandwidth saving without perceived loss of quality to the user. UAV presents an opportunity
to significantly improve the e ciency of DASH by not requesting unnecessarily high
bit rate videos. We design UAV-enabled DASH (UDASH) and evaluate its performance in
Wi-Fi networks. Simulation results show that UDASH in a Wi-Fi network has the benefits
of not only significantly improving the video streaming performance such as reducing the
rebu ering probability, but also enhancing the performance of cross traffc. ☐ In addition, the MPEG-DASH standard uses TCP as the underlying transport layer
protocol, and more importantly, TCP is one type of dominant tra c in the Internet. Therefore,
we investigate how to improve TCP performance in wireless networks. We identify two
issues of TCP performance degradation due to common channel errors via both analytical
study and simulations in a typical Wi-Fi network. Motivated by these issues, a MAC layer
optimization technique is proposed, which is based on the adaptation of the Retry Limit parameter
after considering TCP traffic characteristics and throughput model. The evaluation
results confirm that the proposed technique achieves higher performance gain. ☐ In optimizing video teleconferencing, we considerWebRTC, which is Google’s open
source real-time communication framework. In wireless networks such as those based on
IEEE 802.11, packet losses due to fading and interference are often misinterpreted as indications
of congestion, causing unnecessary decrease in the data sending rate due to congestion
control by the RTCP protocol working beneath WebRTC and above RTP. For delayconstrained
applications such as video teleconferencing, packet losses may result in excessive
artifacts or freeze in the decoded video. We propose a simple and yet effective
mechanism to detect and reduce channel-caused packet losses by dynamically adjusting the
retry limit parameter of the IEEE 802.11 protocol. Since the retry limit is left configurable
in the IEEE 802.11 standard, and does not require cross-layer coordination, the proposed
scheme can be easily implemented and incrementally deployed. We also propose to use a
delay constrained retry limit adaptation algorithm to control transmission delays so that delay
constraints required by different application scenarios can be met. Experimental results
of applying the proposed scheme to aWebRTC based real-time video communication prototype
show significant performance gain compared to the case where retry limit is configured
statically. ☐ In addition to the optimization techniques proposed for the IEEE 802.11 protocol, we
also propose a cross-layer approach to optimize video teleconferencing, termed early packet
loss feedback (EPLF). In EPLF, if a packet loss is due to channel errors, the MAC layer
directly feeds back the loss information to the RTP layer with a spoofed RTCP packet that
carries a NACK message so that the RTP layer can retransmit the lost RTP packet. Since
the whole feedback process takes place in the same device (the video sender), the latency is
negligible in relation to the RTT, and hence the term ’early’ in EPLF. Theoretical analysis and
prototype-based experimental results show that EPLF almost completely eliminates channelcaused
video freezes in the decoded video while improving congestion control. ☐ Furthermore, we also apply the technique of UAV to video teleconferencing to further
reduce bandwidth consumption, and build a prototype based on WebRTC and Licode (a video
teleconferencing hub platform) to validate the bandwidth savings.