An Evaluation of Bitrate Adaptation Methods for HTTP Live Streaming

A Probabilistic Adaptation Method for HTTP Low-Delay Live Streaming over Mobile Networks

“…Then, each segment is delivered by an HTTP request-response transaction. To cope with network variations, many adaptation methods have been proposed, which can be divided into a throughput-based group and a buffer-based group [2], [4]. It should be noted that due to bitrate adaptation, buffer size and buffer level in adaptive streaming are measured in time units [1], [2], [4].…”

“…Because the client can download only the segments that have already been generated in real time, the client has to spend L segment durations for initial buffering. In live streaming, the total duration of segments downloaded in initial buffering is equal to the buffer size [4]. After initial buffering, if the download rate and the playout rate are equal, the buffer level will be stable at the value of buffer size; this value is also referred to as the target buffer level in this letter.…”

“…After initial buffering, if the download rate and the playout rate are equal, the buffer level will be stable at the value of buffer size; this value is also referred to as the target buffer level in this letter. More detailed explanation about the buffer size and the buffer level in live streaming can be found in [4]. Because low delay is a crucial requirement for live streaming [5], the buffer size should be reduced in this context.…”

“…A learning-based adaptation method for low-delay streaming is proposed in [8]; however, the lowest supported buffer size is still 3 segments. In [4], through an extensive evaluation, it is found that throughput-based methods are more effective to support small buffer sizes than buffer-based methods. However, when the buffer size is reduced to 2 segments, no adaptation methods evaluated in [4] could guarantee a continuous session under strong variations of a mobile connection.…”

“…In [4], through an extensive evaluation, it is found that throughput-based methods are more effective to support small buffer sizes than buffer-based methods. However, when the buffer size is reduced to 2 segments, no adaptation methods evaluated in [4] could guarantee a continuous session under strong variations of a mobile connection. Note that, besides the average video bitrate, interruptions strongly affect the quality of experience [9].…”