Performance evaluation of variable-bit-rate voice in packet-switched networks

One of several congestion-control methods in integrated packet networks is a bit-dropping scheme in which voice is encoded, using the embedded coding principle, and packetized with blocks of bits organized according to their order of significance. In a state of overload, the less significant blocks are dropped in the multiplexer queue to relieve congestion. Fluctuations in voice quality occur due to dynamically varying bit rate during a call. In this paper, we present methodologies for evaluating the performance of variable-bit-rate voice under fixed and variable loads. We have developed a Markov chain model to describe the probabilistic bit-dropping pattern corresponding to any specified traffic conditions. The model is used with a software tool to emulate packetized voice under various loading conditions. The resulting bit-dropped voice packet stream is then decoded, converted to an analog signal, and presented to listeners in a subjective test. We conclude that, with prudent traffic engineering of the network, voice quality will remain robust to temporal variations in bits per sample.