Dynamic adaptive playout algorithm using interarrival jitter and dual use of α

Two algorithms have been used as the basis for playout of packet-based voice. The first one identifies jitter level using the metric of the difference between the observed network delay for a new packet and the estimated average for previous packets. The new value of the estimated jitter level is a weighted combination of the previous value of the estimated jitter level and the new value of the observed jitter level. The second one is based on the difference in the arrival times of successive packets. The value of α used in the weighted combination is normally fixed. ThE paper proposes the use of two values of the weighting factor α to make the estimate of jitter level react quickly when jitter increases, but react slowly when it decreases. Four playout algorithms are evaluated using several jitter patterns obtained experimentally for 30 paths. The results indicate that the dynamic playout algorithm using α_LOW=0.995 and α_HIGH=0.998002 can produce significant reductions in the lateness loss rates, especially for high jitter conditions. It is also shown that the interarrival time jitter metric used in the dynamic playout algorithm results in better lateness loss performances especially in low jitter conditions. Finally, given the tendency to lower network delay, it is shown that increasing the buffering delay permits achieving significantly lower lateness loss, while still maintaining acceptable overall delay.