Jitter buffer management algorithms for voice communication.

Abstract

This thesis studies some jitter management algorithms for real-time applications. These algorithms are executed at a destination node, and assume no knowledge of the source characteristic or the impact of the network path characteristic. The work mainly focuses on prediction algorithms that make use of the information of the packets received in the past, and adjust buffer parameters in order to maintain certain level of quality of service (QoS). Two algorithms are proposed, first, to apply the least mean square method to predict the future packet interarrival time so that the buffer parameters can be dynamically changed in order to adapt to the bursty network traffic; second, to apply the fuzzy logic method on the buffer management to maintain the gap probability within acceptable level while keep the latency as low as possible. These two new algorithms have been evaluated using OPNET simulation and compared with some other algorithms such as the I-policy and the E-policy. We studied and discussed the tradeoff among the gap probability, average display latency, and packet loss probability. Towards the end, we have also made some design recommendations.