An application of neural networks to adaptive playout delay in VoIP

IP networks are not designed for real-time voice data. The statistical nature of data traffic and the dynamic routing techniques employed in IP networks results in a varying network delay (jitter) experienced by IP packets. As a result voice packets generated at successive and periodic intervals at the source will typically be buffered at the receiver prior to playback in order to smooth out the jitter. However, the additional delay introduced by the payout buffer degrades the quality of service. Thus forecasting the jitter is an integral part of selecting an appropriate buffer size. This paper compares several neural network based models for adaptive playout buffer selection. Specifically, a combined wavelet transform/neural network approach is proposed. The effectiveness of these algorithms is evaluated using recorded traces from Galway to Tokyo by comparing the buffering delay and the packet loss ratios of each technique. Simulation results indicate that the Haar-Wavelets-Packet MLP adaptive scheduling scheme is superior. Finally, the voice signal is reconstructed according to the packet loss information from estimation of each algorithm, and the quality of the voice is then given a score according to the PESQ MOS algorithm.