Adaptive Media Streaming - Scalable Continuous Media Streaming Systems - page 142

the sender will result in far better visual quality than attempting to recover from data loss at
the receiver.
To perform video adaptation we must tackle two fundamental challenges. First, the sender
must be able to dynamically control or convert the video bit-rate to the desired value. This can
be accomplished by means of scalable video coding [7] and transcoding [8-10] as discussed
in Chapter 2. Second, an adaptation algorithm is needed to estimate the network bandwidth
available, and subsequently determine the bit-rate to be used for converting and transmitting the
video stream. This chapter focuses on the second challenge, i.e., design of the rate adaptation
algorithm.
This problem has been studied by a number of researchers, including the studies by Rejaie
et al . [4] and Assuncao and Ghanbari [5] who adopted UDP as the network transport; and
the studies by Cuetos and Ross [1], Cuetos et al . [2], and Jacobs and Eleftheriadis [3] which
adopted TCP as the network transport.
A common property of these adaptation algorithms is the existence of a configurable op-
erating parameter [1-2], which is typically used in the feedback loop of the algorithms. Not
surprisingly, as will be illustrated in Section 8.6, the choice of this operation parameter can
significantly affect the performance of the rate adaptation algorithm. Unfortunately, to opti-
mize this parameter for the best performance will require a priori knowledge of the available
network bandwidth over the entire duration of the video session. This is clearly not possible in
practice and thus poses significant difficulties to deploying these rate adaptation algorithms.
In this chapter, we address this issue by presenting a rate adaptation algorithm that does not
have any configurable parameter. In other words, prior knowledge of the available network
bandwidth is not needed to run the rate adaptation algorithm. Our results show that compared
to the existing algorithms, the presented algorithm can achieve comparable or even better
performance and does so without the need to tweak any operating parameters.
8.2 Related Work
In this section, we review some related work on adapting video to cope with the bandwidth
fluctuations in the Internet. Assuncao and Ghanbari [5] and Kanakia et al. [6] proposed adapting
the data pumping rate and hence the bit-rate of video by adjusting the quantizer scales of the
frames. The data pumping rate is controlled dynamically based on the buffer occupancy of
the bottleneck switch. The idea is to maintain the buffer occupancy of the bottleneck switch
at a safe level to avoid losses of video packets due to buffer overflow. Their approach can
effectively adapt to network congestion and reduce packet losses. The only limitation is that
we need to know the buffer occupancy at the bottleneck router. This may require modification
to the router firmware to support this function.
Rejaie et al. [4] proposed an adaptive streaming scheme running on its proprietary RAP
congestion control protocol [11] with the support of multi-layer-encoded videos. They proposed
the criteria for adding or dropping a layer of video based on the current sending rate of its
rate-based RAP congestion control protocol. This approach requires the commercial streaming
player software to support playback of layered-encoded video.
Jacobs and Eleftheriadis [3] proposed a video adaptation scheme that controls the video
bit-rate using transcoding. The proposed system runs on a custom-designed semi-reliable
congestion control protocol. The main idea of the adaptation algorithm is to maintain the
server buffer occupancy at a certain level. The algorithm assumes that video data are input