Sensitivity of bandwidth allocation algorithms for unidirectional networks to voice parameters

A comparative performance evaluation of a dynamic bandwidth allocation algorithm is made using different voice traffic models. It is shown that the voice clipping rate performance is very sensitive to characteristics of the speech arrival process and higher than previously reported. Simulation results indicate that the average clipping rate increases almost linearly with the talkspurt rate but is essentially independent of the coefficient of variation of the silence period. The hyperexponential and the exponential speech models give the same average clipping rate if their parameters are adjusted to produce the same talkspurt rates. These results suggest that, since the average clipping rate is sensitive to the average talkspurt rate of the speech model, this parameter must match accurately the desired speech input signal. On the other hand, the coefficient of variation of the silence period need not be matched to the desired speech signal with as much accuracy.<>