Packet Voice Rate Adaptation Through Perceptual Frame Discarding

We address the problem of rate adaptation at the source, given a congested packet based voice carrying network. We propose and analyze a novel method for perceptually based frame discarding. Thus, we propose a perceptually based classifier (PBC) to do the discarding and we combine the PBC with a method for state synchronization, which exploits the knowledge of frame discards to combat the error propagation normally following a frame loss for state dependent coders. We further propose to use a combination of a queue model and an empirical speech quality model to decide on the proper discard rate for a given congestion scenario. In this design, we particularly focus on the queue model's ability to capture the characteristics of bursty traffic from voice sources in discontinuous transmission mode. Frame discarding is evaluated in a network bottleneck scenario, where objective results, based on PESQ-LQO, show significant improvements due to perceptually based frame discarding. The improvement depends on the degree of congestion without rate adaptation, the number of incoming voice flows, and the bottleneck buffer.