A framework for the analysis of adaptive voice over IP

We present a framework for the analysis of a set of adaptive variable-bit-rate voice sources in a packet network. The instantaneous bit rate of each source is determined by an end-to-end control mechanism that, based on measurements of packet delay and loss rate, selects the rate that best matches current network conditions. Several such algorithms can be analyzed with the proposed framework, which consists of a detailed Markovian model of the source behavior and of an approximate description of the interaction between the sources and the underlying network. The model of the source takes into account time intervals during which a connection is active as well as intervals of inactivity; within a given conversation, it also models on/off (speech/silence) periods. The interaction of a source with the rest of the system is derived through an iterative procedure that evaluates the feedback that a source receives from the network. A case study presenting the results relative to an adaptive system transmitting at bit rates typical of widely used speech coding standards (64 kb/s, 13 kb/s and 8 kb/s) illustrates the proposed framework