Rate Adaptation for Buffer Underflow Avoidance in Multimedia Signal Streaming

In media transmission over packet networks, one of the most challenging issues is the avoidance of receiver buffer underflows. Among the approaches proposed to solve this problem, the source rate adaptation is promising, due to the availability of multi-rate encoders. In order to succeed, rate-adaptive approaches should take into account not only the network throughput, but also the playout buffer fullness. Using this information, it is possible to determine at which rate the source should encode the stream, in order to avoid underflows. In this paper, we derive and analyze the expression of the underflow probability; we show that it can be written in closed form as a function of the source rate, the receiver buffer fullness and the channel statistics. In particular, we study the dependency on the media rate, and we show how to achieve infinitesimal underflow probabilities even in presence of a high channel variance. Furthermore, we specialize this formulation for a CBR channel. In case of channel variations during the playout, the buffered data may not suffice to ensure zero underflow probability; based on the previous formulation, we present an algorithm to recompute the source rate that allows continuous media playout. This algorithm can be safely iterated at every channel throughput change, and proved to be effective in avoiding buffer underflows.