My masters thesis project was about Playout delay control algorithms for Voice over IP (VoIP). Interactive applications on the Internet are extensively used nowadays. P2P applications like Skype, VoIPBuster have implemented very successful VoIP systems. Several instant messaging applications like MSN Messenger, Yahoo Messenger, AIM Messenger among others also offer VoIP services which are free for PC-to-PC users. Pionnering research applications like NeVoT (by Henning Schulzrinne), FreePhone (by Bolot and Andrés Vega García) and Rat (by the Mice Project) invested lots of effort so as to optimize the transmission of real-time audio packets on the Internet.

Currently, the Internet still operates in a best-effort basis. Routers in the net have buffers implementing FIFO policies, which means that the first packet entering the buffer is the first packet to be serviced. So, there is no priority among packets. This has the advantage of simplicity, mainly when servicing packets, but has the disadvantage that all packets get the same priority: a voice packet will have to wait in the buffer like any other packet, even if the delay constraints for real-time applications are hard. This best-effort mechanism on the Internet causes three main phenomena, which are harmful for real-time applications like VoIP: delay, packet losses and jitter.

In this masters thesis we concentrate on Playout Delay Control Algorithms for VoIP. A playout delay control algorithm implements a buffer on the receiver side so as to store the received packets. Then, the algorithm computes a deadline for each packet. If a packet has been lost in the network or has arrived after its scheduled deadline, the packet is considered lost at the receiver. On one hand, if the deadline is long, the playout algorithm gives more chance to arrive to packets having suffered from a high level of jitter, but at the expense of loosing interaction during the VoIP call. On the other hand, if the computed deadline is short the playout algorithm is more aggressive but with the risk of loosing more packets even if they arrived at the receiver side, but having arrived after their scheduled deadline. So, a playout algorithm trades-off delay and loss so as to optimize the interactivity of the VoIP call.

In this masters thesis, we focus on the class of playout algorithms that update the playout delay at the beginning of a talkspurt. We choose to study an NLMS algorithm originally proposed by DeLeon and later modified by Shallwani and to extensively stress the working conditions of both algorithms. The results we got till today indicate from one side that the Shallwani's algorithm may be buggy, and from the other side that delay spike detection may be improved. So, we choose to improve the delay spike detection algorithm proposed by Shallwani and compare the performance of our (so called) NLMS-mod algorithm with the DeLeon's and Shallwani's algorithms. We found that, for most of the cases, with real-audio traces extensively used in other works (kindly offered by Sue Moon) our algorithm performs better than the formers.

©2007 All Rights Reserved.

Designed by Free CSS Templates.